australian population studies 2018 | volume 2 | issue 2 | pages 1–2 © wilson, charles-edwards and corcoran 2018. published under the creative commons attribution-noncommercial licence 3.0 australia (cc by-nc 3.0 au). journal website: www.australianpopulationstudies.org editorial introduction to the special section on australia’s population policy tom wilson* charles darwin university elin charles-edwards the university of queensland jonathan corcoran the university of queensland *corresponding author. email: tom.wilson@cdu.edu.au. address: northern institute, charles darwin university, ellengowan drive, darwin, nt 0909, australia published on 12 november 2018 this issue of australian population studies includes a special commentary section on australia’s population policy. population issues and population policy debates periodically find themselves in the political and media spotlight, and recent months have seen a resurgence of interest in these topics. in part this attention was the result of australia’s population officially passing the 25 million milestone in early august (abs 2018), though population issues had been gaining momentum from earlier on in the year. a number of politicians and well-known figures have advocated for lower immigration and lower population growth, with relatively few voices supporting current or higher growth rates. while much attention has, as usual, focused on annual net overseas migration totals and the overall growth rate of the population, on this occasion many commentators have also expressed concern about the growth rates of melbourne and sydney, and argued for a greater proportion of immigrants to be directed away from those centres. sensing the increased political sensitivity of the issue, the australian government created a new minister for cities, urban infrastructure and population in addition to the established portfolio of immigration, citizenship and multicultural affairs as part of the cabinet reshuffle accompanying the change in prime minister in august (parliament of australia 2018). many will be watching for population-related policy announcements from these ministers with interest, particularly in the lead-up to the next federal election. in the special commentary section of this issue three australian demographers present their responses to the question ‘what sort of population policy should australia adopt?’ (if any). the purpose is to add some academic perspectives to the wide-ranging population policy debate. all contributions underwent the same double-blind reviewing process applied to regular journal papers. peter mcdonald argues that australia already has a comprehensive de facto population policy and that the country should retain it until at least 2026, especially in light of the labour supply crunch coming in the next decade. nick parr sets out the case for australia to adopt an official population policy and focuses particularly on the skilled stream of the migration program. importantly, he also calls for more research on demographic issues to build up a more comprehensive evidence base for australia’s population policy. liz allen also argues for an official population policy which eschews http://www.australianpopulationstudies.org/ mailto:tom.wilson@cdu.edu.au 2 editorial: wilson, charles-edwards and corcoran australian population studies 2 (2) 2018 demographic targets and instead emphasises the wellbeing of individuals and communities, and long-term sustainability. references australian bureau of statistics (abs) (2018) australia's population to reach 25 million. media release. http://www.abs.gov.au/ausstats/abs@.nsf/mediareleasesbyreleasedate/c3315f52f6219de9c a2582e1001bc66a?opendocument. accessed on 8th october 2018 parliament of australia (2018) current ministry list: the 45th parliament. https://www.aph.gov.au/about_parliament/parliamentary_departments/parliamentary_library/ parliamentary_handbook/current_ministry_list. accessed on 8th october 2018. http://www.abs.gov.au/ausstats/abs@.nsf/mediareleasesbyreleasedate/c3315f52f6219de9ca2582e1001bc66a?opendocument http://www.abs.gov.au/ausstats/abs@.nsf/mediareleasesbyreleasedate/c3315f52f6219de9ca2582e1001bc66a?opendocument https://www.aph.gov.au/about_parliament/parliamentary_departments/parliamentary_library/parliamentary_handbook/current_ministry_list https://www.aph.gov.au/about_parliament/parliamentary_departments/parliamentary_library/parliamentary_handbook/current_ministry_list australian population studies 2018 | volume 2 | issue 2 | pages 59-61 © reimondos, gray and evans 2018. published under the creative commons attribution-noncommercial licence 3.0 australia (cc by-nc 3.0 au). journal website: www.australianpopulationstudies.org how moving a box changed the religious landscape of australia anna reimondos* australian national university edith gray australian national university ann evans australian national university * corresponding author. email: anna.reimondos@anu.edu.au. address: school of demography, anu college of arts and social sciences, the australian national university, 9 fellows road, acton, act 2601, australia paper received 1 june 2018; accepted 20 july 2018; published 12 november 2018 this data visualisation is the winner of the australian population studies 2018 population data visualisation competition, kindly sponsored by geografia pty ltd. in this data visualisation we use the 2011-2016 australian census longitudinal dataset (acld) to examine how answers to the question on religious affiliation changed between 2011 and 2016. a question on religious affiliation has been included in every census since the first national census of 1911. it is the only question in the census which is optional, as outlined in the census and statistics act 1905, and it is also the question that has arguably attracted the most controversy over the years. up until 1991 the question on religion was in the form of an open text question. in 1991 the format changed to pre-defined check boxes listing several main religions including catholic, anglican (church of england) and uniting church. the check boxes were followed by a space for an open-text answer where people could write any ‘other’ religion that was not included in the pre-defined list. finally at the bottom was a check box for ‘no religion’. although the content and order of the pre-defined list of religions changed between 1991 and 2011, in all years up to and including 2011 ‘no religion’ was always on the bottom after ‘other’. in a public consultation on the content of the 2016 census, nearly half of the 915 submissions related to the question on religious affiliation (abs 2013). following a review, the format of the question changed and for the first time in 2016 ‘no religion’ was placed at the very top of the list of predefined religions. although the percentage of australians identifying as not having a religion has been increasingly steadily since the first census in 1911, between 2011 and 2016 the percentage stating ‘no religion’ increased dramatically from 22.3% to 30.1%. this 35% increase is significantly higher than the 20% increase between 2006 and 2011 and the 19% increase between 2001 and 2006. to observe the change in behaviour following the re-formatted question in the 2016 census we used the acld, which is a representative 5% sample (1.2 million records) from the 2011 census linked to the corresponding records from the 2016 census. we used the abs tablebuilder to cross-tabulate people’s responses to the religion question from 2011 and 2016. this table was re-formatted in stata d em og ra ph ic http://www.australianpopulationstudies.org/ mailto:anna.reimondos@anu.edu.au 60 reimondos, gray & evans australian population studies 2 (2) 2018 to create an individual record of each combination of possible religions in 2011 and 2016. the data was then inputted to and open source data visualisation framework at rawgraphs1. the resulting svg code was then further edited to create the desired image. the visualisation is shown in figure 1. on the left of the diagram is the distribution of religious affiliation in the 2011 census, in order of magnitude, and on the right are the answers given in 2016. in 2011, catholic was the dominant religious affiliation answered by 26% of the acld sample. in 2016 ‘no religion’ had become the prominent religious belief with catholics in second place. census respondents who answered ‘no religion’ in 2016 had previously identified with a wide variety of other religions. not surprisingly, the majority of those stating ‘no religion’ in 2016 had also stated ‘no religion’ in 2011. however among those who did state a religion in 2011, the percentage who switched to ‘no religion’ varied considerably. for example only 11% of those who previously stated they were catholic chose ‘no religion’ in 2016. in contrast a fairly large proportion (21%) of anglicans changed to ‘no religion’. of those who did not state a religion in 2011, just under half (46%) chose ‘no religion’ in 2016. for some religions such as eastern orthodox, islam and hinduism only a small proportion switched to ‘no religion’. for questions involving unordered categorical responses, such as the religion question in the census, the order in which the options are presented has been shown in previous research to be important. the primacy effect suggests that people may choose the first acceptable answer presented to them, particularly if the list of response categories is long (tourangeau et al. 2000). evidently the change in question format, and moving the box titled ‘no religion’ to the top of the list of pre-defined options had a significant impact on the way in which the 2016 question on religious affiliation was answered. references abs (2013) census of population and housing: consultation on content and procedures, 2016. catalogue no. 2007.0. canberra: abs. tourangeau r, rips l j, and rasinski k (2000) the psychology of survey response. cambridge: cambridge university press. 1 http://app.rawgraphs.io australian population studies 2 (2) 2018 reimondos, gray & evans 61 figure 1: answers to the question on religion in the 2011 and 2016 censuses source: australian bureau of statistics, 2011-2016 australian census longitudinal dataset australian population studies 2018 | volume 2 | issue 1 | pages 52–55 © trauer, freak-poli, kippen and mcneil 2018. published under the creative commons attribution-noncommercial licence 3.0 australia (cc by-nc 3.0 au). journal website: www.australianpopulationstudies.org fifty years of plummeting cardiovascular death rates and implications for the individual james m trauer* monash university rosanne freak-poli monash university rebecca kippen monash university john mcneil monash university * corresponding author. email: james.trauer@monash.edu.au. address: epidemiological modelling unit, school of public health and preventive medicine, monash university, 553 st kilda road, melbourne, victoria 3004 paper received 5 march 2018; accepted 26 march 2018; published 28 may 2018 mortality rates have been reliably reported for many decades and provide important insights into major changes in disease burden or ‘epidemiological transitions’ (freak-poli, bi and hiller 2007; omran 1977). here we present illustrations of the dramatic shift from cardiovascular disease to cancer as the leading cause of death in australia and highlight the significance of these changes for individual australians. we obtained data on the total number of deaths and population size for the period 1907–2014 in australia from the general record of incidence of mortality (grim) books (australian institute of health and welfare 2017) and used this information to calculate death rates overall and by age group attributable to all causes, cardiovascular disease and cancer. the small proportion of deaths (generally <0.1%) for which age group was missing were assumed to be distributed as for the deaths for which age group was known. these calculated death rates were used to construct life tables, under which the complement of the all-cause mortality rate for each year was used to determine the number of persons surviving to the following year. as the data are grouped into five-year age groups, karup-king interpolation (siegel and swanson 2004) was used to obtain smoothed estimates for each year of age in the construction. full methods are presented as publicly available python 2.7 code at https://github.com/jtrauer/demography. we observed dramatic falls in cardiovascular mortality, with cancer overtaking cardiovascular disease as the leading cause of death category at the start of this century (figure 1, left panels). the fall is even more impressive when presented on a log-scale, because the steady absolute decrease represents a greater proportional decrease as rates decline. cardiovascular death rates are steadily falling across all age groups, although the youngest age brackets have the greatest variation due to the noise associated with these very low rates (figures 1, right panels). while the initial reduction in cardiovascular mortality was attributable to reductions in death rates in young adults and middle age, increasingly the falls in death rates are occurring in older age groups. although the fall appears less impressive in the oldest age bracket (85 years and above), this is attributable to the decline in death rates being offset by an increase in the age distribution of this age group. d e m o g ra p h ic http://www.australianpopulationstudies.org/ mailto:james.trauer@monash.edu https://github.com/jtrauer/demography australian population studies 2 (1) 2018 trauer j, freak-poli r, kippen r and mcneil j 53 figure 1: death rates by cause and cardiovascular death rates by age group from 1964–2014 in australia source: australian institute of health and welfare (2017). notes: left two panels show death rates standardised to the age structure of the australian bureau of statistics standard australian population 2001. right two panels show age-specific death rates attributable to cardiovascular disease. these changes have translated into massive changes in survival and cause of death (figure 2). for example, a person subject to the age–cause-specific death rates of 1964 would probably be dead before their 80th birthday, with cardiovascular disease the likely cause of death. by contrast, a person subject to the age–cause-specific death rates of 2014 would probably live to celebrate their 85th birthday, and likely die of non-cardiovascular causes before turning 90. again note that death rates for those aged 85 years and above are considered as a single age group, which was used to calculate survival from 85–89 years. therefore, our findings emphasise the need for disaggregation of this oldest age group, which has grown from a very small group to a considerable size, and continues to grow. past modelling in similar developed country settings suggests that the dramatic reductions in cardiovascular mortality are attributable both to changes in the population prevalence of predisposing risk factors and medical treatments – the latter including both clinical care for episodes of disease and secondary prevention (capewell et al. 2000; capewell, morrison and mcmurray 1999). pharmacological primary prevention has not contributed significantly to these major reductions in the past, although results from large community-based randomised controlled trials into aspirin (aspree investigator group 2013) and statins (zoungas et al. 2014) as primary preventive interventions will shed light on this potent. however, while intervention studies can quantify the 54 trauer j m, freak-poli r, kippen r and mcneil j australian population studies 2 (1) 2018 relative reduction in incident cardiovascular disease, one of the most important issues to address before recommending prevention strategies to individuals is the absolute risk of new cardiovascular disease (otto 2016). while disease-specific mortality rates do not translate directly to rates of incident disease, the dramatic reductions in cardiovascular death rates over recent decades are likely to reflect decreases in cardiovascular disease to some extent. therefore, it is essential to ensure guidelines are based on risk assessments that consider modern rates of incident cardiovascular disease by demographic and comorbidity status. cancer mortality is presented for comparison as it is the other major disease category responsible for a high proportion of deaths over recent decades. although cancer mortality appears relatively stable over this period, the aggregate rates mask important underlying trends in cancer types by gender (freak-poli, bi and hiller 2007 with non-cardiovascular mortality remaining relatively stable, it is clear that the declines in overall mortality directly parallel those in cardiovascular-specific mortality (figure 1, upper left panel). this highlights how critical these improvements have been in driving australia’s recent improvements in overall life expectancy, to the extent that falling cardiovascular mortality has entirely driven the all-cause mortality decreases. figure 2: cumulative outcomes for an australian living their life with death rates as observed in 1964, 1989 and 2014 source: australian institute of health and welfare (2017). notes: survival region represents a life-table, with cumulative contributions of three cause of death categories presented as shaded regions above. data from five-year brackets are smoothed by calculating yearly estimates using karup-king interpolation. australian population studies 2 (1) 2018 trauer j, freak-poli r, kippen r and mcneil j 55 references aspree investigator group (2013) study design of aspirin in reducing events in the elderly (aspree): a randomized, controlled trial. contemporary clinical trials 36(2): 555–564. https://doi.org/10.1016/j.cct.2013.09.014. accessed on 24 january 2018. australian institute of health and welfare (2017) general record of incidence of mortality (grim) books. https://www.aihw.gov.au/reports/life-expectancy-death/grim-books/contents/grim-books. accessed on 25 january 2018. capewell s, beaglehole r, seddon m and mcmurray j (2000) explanation for the decline in coronary heart disease mortality rates in auckland, new zealand, between 1982 and 1993. circulation 102(13): 1511–1516. http://www.ncbi.nlm.nih.gov/pubmed/11004141. accessed on 16 august 2018. capewell s, morrison c e and mcmurray j j (1999) contribution of modern cardiovascular treatment and risk factor changes to the decline in coronary heart disease mortality in scotland between 1975 and 1994. heart (british cardiac society) 81(4): 380–386. http://www.ncbi.nlm.nih.gov/pubmed/10092564. accessed on 16 august 2018. freak-poli r, b p and hiller j e (2007) trends in cancer mortality during the 20th century in australia. australian health review 31(4): 557–564. omran a r (1977) epidemiologic transition in the united states: the health factor in population change. population bulletin 32(2): 1–42. http://www.ncbi.nlm.nih.gov/pubmed/12335110. accessed on 2 march 2018. otto c m (2016) statins for primary prevention of cardiovascular disease. british medical journal (clinical research edition) 355: i6334. https://doi.org/10.1136/bmj.i6334. accessed on 24 january 2018. siegel j s and swanson d a (2004) the materials and methods of demography. uk: emerald group publishing limited. zoungas s, curtis a, tonkin a and mcneil j (2014) statins in the elderly. current opinion in cardiology 29(4): 372–380. https://doi.org/10.1097/hco.0000000000000082. australian population studies 2017 | volume 1 | issue 1 | pages 1–2 © wilson, charles-edwards and corcoran 2017. published under the creative commons attribution-noncommercial licence 3.0 australia (cc by-nc 3.0 au). journal website: www.australianpopulationstudies.org editorial introducing australian population studies tom wilson* charles darwin university elin charles-edwards the university of queensland jonathan corcoran the university of queensland *corresponding author. email: tom.wilson@cdu.edu.au. address: northern institute, charles darwin university, ellengowan drive, darwin nt 0909, australia published on 20 november 2017 welcome to this inaugural issue of australian population studies. the aim of the journal is to disseminate and promote high-quality peer-reviewed research which extends our knowledge and understanding of population issues in australia. in doing so, we wish to become the leading publication for australian population research. importantly, australian population studies is open access, online and does not charge fees of any kind. academic publishing is witnessing a growth in open access journals, and we are pleased to be part of this growing movement. the benefits are numerous. it brings greater visibility and impact for research, with several studies concluding that open-access papers are cited more (e.g. atchison and bull 2015; wagner 2010, 2016; wang et al. 2015). where research has been funded by the taxpayer it is available without readers having to pay further to access findings. research is also readily available to practitioners, policy makers and members of the public who typically do not have access to subscription journals, as well as to academics in countries which cannot afford such journals. all content in australian population studies is not only open access, but may be reproduced elsewhere. it is published under the creative commons attribution-noncommercial licence 3.0 australia (cc by-nc 3.0 au), which means content may be shared and reproduced for noncommercial purposes providing the authors of the work are fully acknowledged. authors are welcome – and in fact are strongly encouraged – to upload papers published in australian population studies to researchers’ social networking sites such as researchgate. our expectation is that there will be two issues per year, published in may and november, though the annual number of issues may be reviewed in the future. the latest dates for the submission of contributions are 15th january for the may issue and 15th july for the november issue, offering relatively fast submission-to-publication turnover. contributions consist of three main types. first, we publish short research papers which should not exceed 4000 words. we do not carry full-length papers. traditional population issues, contemporary ‘hot topic’ demographic debates, policy-relevant studies, local case studies, new population datasets and relevant software and code are all suitable topics for research papers. supplementary files of data, code, detailed results and similar can accompany papers. we particularly encourage papers focusing on ‘core’ demographic processes and patterns, such as fertility, mortality, migration, http://www.australianpopulationstudies.org/ mailto:tom.wilson@cdu.edu.au 2 editorial: wilson t, charles-edwards e and corcoran j australian population studies 1 (1) 2017 population ageing, household change, population distribution, relationships, families and population policy. sometimes demography and population studies can seem like a doughnut of specialised studies with relatively little filling in the middle! second, we encourage the use of data visualisation to present demographic data in interesting, new and innovative ways. these appear as demographics. we would like this to become a key element of the journal and will announce a data visualisation competition shortly. third, we publish introductory guides which consist of an introduction to demographic methods, approaches, data, theories or software which are not well covered in the textbooks. we hope these guides will be of use to students, researchers, practitioners, teachers and others wishing to learn about demography. the common requirement for all three types of contribution is that they should be written for a wide-ranging audience and not limited to academics. many people have generously given their time and expertise to make this first issue of australian population studies possible. we wish to extend our sincere thanks to all authors, reviewers, editorial advisory board members, our copy editor ros moye, our logo designer james daley, charles darwin university library, which is providing web hosting for the journal, and the northern institute at charles darwin university, which is providing financial support. we hope you enjoy reading the first issue. we welcome feedback (via the email address shown above). finally, we encourage researchers working on population issues in australia to submit contributions for future issues. submission guidelines are available at www.australianpopulationstudies.org/index.php/aps/about/submissions. references atchison a and bull j (2015) will open access get me cited? an analysis of the efficacy of open access publishing in political science. ps: political science & politics 48(1): 129–137. wagner a b (2010) open access citation advantage: an annotated bibliography. issues in science & technology librarianship (60), viewed 30 october 2017, http://www.istl.org/10winter/article2.html. wagner a b (2016) citation impact advantages of open access (oa) articles over non-oa articles – updated 12/21/2016 with 16 new 2015–16 studies, viewed 30 october 2017, https://ubir.buffalo.edu/xmlui/handle/10477/75108. wang x, liu c, mao w and fang z (2015) the open access advantage considering citation, article usage and social media attention. scientometrics 103(2): 555–564. http://www.australianpopulationstudies.org/index.php/aps/about/submissions http://www.istl.org/10-winter/article2.html http://www.istl.org/10-winter/article2.html https://ubir.buffalo.edu/xmlui/handle/10477/75108 australian population studies 2018 | volume 2 | issue 1 | pages 56–58 © wang, corcoran, liu and sigler 2018. published under the creative commons attribution-noncommercial licence 3.0 australia (cc by-nc 3.0 au). journal website: www.australianpopulationstudies.org visualising the internal migration of the mainland china-born population between australian capital cities over time siqin wang* the university of queensland jonathan corcoran the university of queensland yan liu the university of queensland thomas sigler the university of queensland * corresponding author. email: s.wang6@uq.edu.au. address: school of earth and environmental sciences, chamberlain building, the university of queensland, st lucia 4072, queensland, australia paper received 1 march 2018; accepted 16 march 2018; published 28 may 2018 over the past three decades, migration from mainland china to australia has become increasingly significant. the 2016 australian bureau of statistics (abs) census of population and housing (census) recorded a total of 526,000 china-born migrants living in australia, accounting for 2.2 per cent of the total national population, ranking it as the largest non-commonwealth foreign-born group in the country (abs 2016a). drawing on data from three australian censuses, we visualise 5-year internal migration flows of the china-born population between australian state and territory capitals. data were retrieved online from tablebuilder pro for the 2006, 2011 and 2016 censuses (abs 2006; 2011; 2016a). the mainland china-born population was defined as those residing in australia who were born in the people’s republic of china but not the special administrative regions of macau and hong kong. capital cities were spatially defined as greater capital city statistical areas (abs 2016b). our demographic is a series of three directional circular plots (abel 2016) that depict the system of internal migration flows representing the relative size and direction of internal migration movements (figure 1, page 58). the plots were created using the r statistical software and employing the package ‘circlize’ (available at: https://www.r-project.org/about.html). the circular plots represent the volumes, origins and destinations using lines of varying width and colours to capture migration flows between capital cities (figure 1). the length of each segment for a given city represents the number of migrants moving to and/or from a particular capital. for each city the direction of migration flows is indicated by an arrow pointing from origins to destinations. each capital city and the flows originating from that city are represented by a single colour (e.g. sydney is depicted in red for each of the three plots). each tick mark on the perimeter of the circle is used to represent a total of 200 migrants. we exclude the least numerically significant 30 per cent of flows from each of the circular plots. excluding these smaller flows helps to enhance the visual clarity by focussing on the main residential migration movements. d e m o g ra p h ic http://www.australianpopulationstudies.org/ mailto:s.wang6@uq.edu.au https://www.r-project.org/about.html australian population studies 2 (1) 2018 trauer j, freak-poli r, kippen r and mcneil j 57 our demographic illustrates a number of interesting internal migration patterns over the decade to 2016. • the absolute volume of internal migration between capital cities has increased 340 per cent over the decade. this increase includes migration to and from the smaller capital cities of hobart, darwin and canberra, which did not have large china-born populations in earlier decades compared to the larger capitals of sydney and melbourne. • there appears to be some stability in the internal migration trends of china-born migrants over time. more specifically, the share of migration across capital cities has principally remained the same, with sydney the largest origin and destination followed by melbourne and then adelaide. • the proportion of overall flows by capital city shows little change over time in sydney, melbourne and canberra. however, brisbane’s in-migration flows of the china-born population increase substantially over the period to 2016, reflecting a growing chinese community in brisbane and sustained migration from the other capital cities – in particular sydney and melbourne. an increase in out-migration flows is observed in both adelaide and hobart, especially over the period 2011 to 2016. out-migration from adelaide, especially in the period 2011 to 2016 may be tied to a relatively weaker job market and economic outlook (adzuna 2016; beer 2008). • sydney and melbourne both remain the largest destinations for china-born migrants. the attraction of these two cities over the other australian capitals for china-born migrants is arguably a function of denser social ties allied with a broader range of economic opportunities. there is a need to deepen our understanding of the internal migration patterns of the china-born population by capturing these movements at a range of spatial scales and unpacking the reasons under-pinning these flows. with such knowledge we will be able to better understand internal migration pathways after settlement, and how these pathways mesh with longer-term integration in australian society. acknowledgements an earlier version of this visualisation was presented at the institute of australian geographers conference in brisbane, july 2017. we thank the organisers and participants for their comments. we also thank karen borchardt at the university of queensland for help with the data collection. references abel g (2016) updated circular plots for directional bilateral migration data. https://gjabel.wordpress.com/tag/circular-plot/. accessed on may 2016. abs (australian bureau of statistics) (2006) census of population and housing – fact sheets, 2006. cat. no. 2914.0. canberra: abs. abs (2011) census of population and housing: outcomes from the 2011 census output geography discussion paper. cat. no. 2911.0.55.003. canberra: abs. abs (2016a) migration, australia, 2015–16. cat. no. 3412.0. canberra: abs. abs (2016b) australian statistical geography standard: volume 1 – main structure and greater capital city statistical areas. cat. no. 1270.0. canberra: abs. adzuna (2016) australia job market report. https://www.adzuna.com.au/blog/wpcontent/uploads/2016/03/adzuna-march-job-report-robots-by-20301.pdf. accessed on march 2016. beer a (2008) risk and return: housing tenure and labour market adjustment after employment loss in the automotive sector in southern adelaide. policy studies 29(3): 319–330. 58 wang s, corocan j, liu y and sigler t australian population studies 2 (1) 2018 source: abs 2006, 2011 and 2016 census data. note: each tick mark on the perimeter of the circle is used to represent a total of 200 migrants. figure 1: the internal migration of the mainland china-born population between australian capital cities over three census periods australian population studies 2017 | volume 1 | issue 1 | pages 69–72 © lieske 2017. published under the creative commons attribution-noncommercial licence 3.0 australia (cc by-nc 3.0 au). journal website: www.australianpopulationstudies.org urban, suburban and rural household density trends scott n. lieske* the university of queensland *corresponding author. scott n. lieske. email: scott.lieske@uq.edu.au. address: school of earth and environmental sciences, the university of queensland, brisbane qld 4072, australia paper received 17 september 2017; accepted 23 october 2017; published 20 november 2017 census data may be used to illuminate changing patterns in urban and regional australia, to inform policy development and to help us to better understand our surroundings. when visualised, census data can provide a straightforward means for planners and decision-makers both to identify spatial and temporal patterns and to consider the implications of those patterns for the future (coffee, lange and baker 2016). the visual presentation of census data can also allow citizens to better interpret their lived experience in australia’s capital cities and suburbs. this demographic examines change over time in numbers of households and household density for the entirety of australia. dwelling density is used rather than population density as the former shows not just where we are growing, but how we are growing. an investigation into household locations reveals trends in the preferred built forms of australian life. the methods used in this analysis for understanding changes in household density are based on kolko (2015a, 2015b and 2016), who presents a method for analysing the share of households that live in census geographies with different densities. kolko’s density classifications were developed via an online survey in the united states that asked people to describe where they lived within the constructs of ‘rural’, ‘suburban’ or ‘urban’. analysis of the survey results found the best predictor of how people describe where they live is household density. kolko (2016) used these survey results to collapse households in the united states (us) into eight density categories. converted from households per square mile to households per square kilometre (km), these eight categories are: 0–40; 41–193; 194–386; 387–578; 579–854; 855–1,930; 1,931–3,861; and 3,862+ households per square km. the low end of these density categories are largely congruent with the rural living and rural fringe rural residential development densities found in australia. in planning australia: an overview of urban and regional planning, sinclair and bunker (2007) characterise household densities in the range of 100 to 250 households per square km, the lower two suburban categories presented here, as rural fringe development. they characterise household densities in the range of 1 to 100 households per square km as rural living, a notion congruent with the rural and low-density suburban household densities presented in this study. extending kolko’s approach to the australian context, using data from the australian bureau of statistics (abs) census of population and housing for 2001, 2006, 2011 and 2016, household counts by australian statistical geography standard statistical area 2 (sa2) are presented in figure 1. d e m o g ra p h ic http://www.australianpopulationstudies.org/ mailto:scott.lieske@uq.edu.au 70 lieske s australian population studies 1 (1) 2017 figure 1: density of australian households by sa2 area, 2001–2016 source: calculated from abs 2001, 2006, 2011 and 2016 census data. these results show a broad pattern of growth across all rural, suburban and urban density categories in australia. the only exception is the second highest suburban density category, which shows a decrease in households from 2001–2006. these results suggest the variety of residential choices from rural to high density urban each hold appeal to various segments of australia’s growing population. insight may also be gained by considering the spatial pattern of these results. mcguirk and argent (2011) noted that urban expansion, peri-urban growth and population increases above the national average were occurring in australia’s four mega-metropolitan areas: geelong–melbourne– mornington peninsula in victoria; sunshine coast–brisbane–gold coast in south-east queensland; newcastle–sydney–wollongong in new south wales; and wanneroo–perth–mandurah in western australia. taking advantage of the consistent spatial units for the 2006, 2011 and 2016 censuses available in the abs 2016 time series profile (abs 2017), we can illustrate how the changes described by mcguirk and argent (2011) can occur over a relatively short period of time. figure 2 presents the household densities shown in figure 1 for two time periods, 2006 and 2016, combined into just three categories: rural (0–40 households per square km); suburban (41–854 households per square km); and urban (855+ households per square km). results are presented for two of australia’s mega-metropolitan areas: geelong–melbourne–mornington peninsula in victoria; and sunshine coast–brisbane–gold coast in south-east queensland. 0 200,000 400,000 600,000 800,000 1,000,000 1,200,000 1,400,000 1,600,000 1,800,000 2,000,000 0 -4 0 r u ra l 4 1 -1 9 3 s u b u rb an 1 9 4 -3 8 6 s u b u rb an 3 8 7 -5 7 8 s u b u rb an 5 7 9 -8 5 4 s u b u rb an 8 5 5 -1 9 3 0 u rb an 1 9 3 1 -3 8 6 1 u rb an 3 8 6 2 + u rb an n u m b er o f h o u se h o ld s household density category (households per km2) 2001 2006 2011 2016 australian population studies 1 (1) 2017 lieske s 71 figure 2: neighbourhood density categories of australian households 2006 and 2016 in the mega-metro regions of geelong–melbourne–mornington peninsula and sunshine coast–brisbane–gold coast the mapped results in figure 2 support coffee, lange and baker (2016), who found that populations were increasing beyond the middle-ring suburbs of cities and towards the outlying suburban fringe. the results are congruent also with spencer, gill and schmahmann (2015, p. 1), who observed that a ‘significant proportion’ of the population of australia’s three largest cities live at low suburban densities. they further observed: ‘australia’s reputation as … highly urbanised perhaps belies the predominance of low density development across australia’s metropolitan areas. australian cities … feature pockets of high density among a background of low density’ (spencer, gill and schmahmann 2015, p. 1). abs 2016 census data show that while urban areas are expanding rapidly in terms of population and households, suburban areas are expanding rapidly in terms of space. it appears australia is rapidly becoming a suburban nation. there is a continuing need to develop and refine our understanding of the processes, drivers and consequences of urban growth and expansion. suburbanisation in australia and other changes in household density can be linked with population growth and other demographic trends such as movement of particular age cohorts, patterns of socioeconomic advantage and disadvantage, the location of housing and employment and lags in middle and outer ring infrastructure provision (o’neill 2010; mcguirk and argent 2011). they are associated also with impacts on environmental quality, agricultural production, climate change vulnerability and resilience. the results presented in this paper also raise the issue of mega-city governance put forward by steele et al. (2011). further, the considerable expansion of suburban development between 2006 and 2016 combined with decreasing housing affordability over the same time period suggests that the issue of housing affordability will not be solved by supply alone. 72 lieske s australian population studies 1 (1) 2017 references australian bureau of statistics (abs) (2017) 2003.0 – census of population and housing: time series profile, australia, 2016, viewed 21 july 2017, http://www.abs.gov.au/ausstats/abs@.nsf/mf/2003.0?opendocument. coffee n t, lange j, baker e (2016) visualising 30 years of population density change in australia’s major capital cities. australian geographer 47(4): 511–525. doi: 10.1080/00049182.2016.1220901. kolko j (2015a) data and methodological details, viewed 4 may 2016, http://jedkolko.com/wpcontent/uploads/2015/05/data-and-methodological-details-052715.pdf. kolko j (2015b) how suburban are big american cities?, viewed 4 may 2016, http://fivethirtyeight.com/features/how-suburban-are-big-american-cities/. kolko j (2016) urban revival? neighborhood data show that u.s. suburbanization continues (wonkish), viewed 4 may 2016, http://jedkolko.com/2016/03/25/neighborhood-data-show-that-u-ssuburbanization-continues/. mcguirk p and argent n (2011) population growth and change: implications for australia’s cities and regions. geographical research 49(3): 317–335. o’neill p m (2010) infrastructure financing and operation in the contemporary city. geographical research 48 (1): 3–12. doi: 10.1111/j.1745-5871.2009.00606.x sinclair i and bunker r (2007) planning for rural landscapes. in: thompson s (ed.) planning australia: an overview of urban and regional planning. port melbourne vic.: cambridge university press; 159– 177. spencer a, gill j and schmahmann l (2015) urban or suburban? examining the density of australian cities in a global context. paper presented at the state of australian cities national conference, gold coast, queensland, 9–11 december 2015. steele w, crick f, serrao-neumann s, sharma v and wadsworth j (2011) governing the australian megalopolis: the challenge of the 200km city (and beyond). paper presented at the state of australian cities national conference proceedings. melbourne, victoria, 29 november–2 december 2011, viewed 4 may 2016, http://soac.fbe.unsw.edu.au/2011/papers/soac2011_0030_final(1).pdf. http://www.abs.gov.au/ausstats/abs@.nsf/mf/2003.0?opendocument http://jedkolko.com/wp-content/uploads/2015/05/data-and-methodological-details-052715.pdf http://jedkolko.com/wp-content/uploads/2015/05/data-and-methodological-details-052715.pdf http://fivethirtyeight.com/features/how-suburban-are-big-american-cities/ http://jedkolko.com/2016/03/25/neighborhood-data-show-that-u-s-suburbanization-continues/ http://jedkolko.com/2016/03/25/neighborhood-data-show-that-u-s-suburbanization-continues/ http://soac.fbe.unsw.edu.au/2011/papers/soac2011_0030_final(1).pdf australian population studies 2018 | volume 2 | issue 2 | pages 3-11 © mcdonald 2018. published under the creative commons attribution-noncommercial licence 3.0 australia (cc by-nc 3.0 au). journal website: www.australianpopulationstudies.org commentary australia should continue its current comprehensive population policy – at least for the next decade peter mcdonald* the university of melbourne * corresponding author. email: mcdonald.p@unimelb.edu.au. address: melbourne school of population and global health, university of melbourne, vic 3010, australia paper received 15 august 2018; accepted 27 october 2018; published 12 november 2018 1. introduction the most recent official statement on a population policy for australia concluded: it is more useful for governments, businesses and communities to focus on ways of improving our wellbeing, protecting our environment and making better use of the resources we have, rather than trying to determine an absolute limit to our population and focussing efforts on restricting growth in order to not exceed this ‘limit’ (department of sustainability, environment, water, population and communities 2011: 25). in other words, australian population policy does not take the form of specification of a target population level or even a target rate of population growth. australia is not alone in this regard. while countries may have policies to increase or decrease the rate of population growth, usually because their fertility rate is regarded as too high or too low, no country in the world is aiming for a specific population number or a specific rate of population growth. does the absence of precise targets for total population and for the rate of population growth mean that australia does not have a population policy? that seems to be the popular perception based on the current debate about population, and no less an authority than the productivity commission has recently called for a population policy for australia. the commission gave its view on what such a population policy should do: the primary objective of immigration and the government’s population policy is to maximise the economic, social and environmental wellbeing of the australian community (existing australian citizens and permanent residents and their future offspring) (productivity commission 2016: 108). the first problem with this definition is that maximising the economic, social and environmental wellbeing of the australian community is the raison d’etre of all policy, including tax policy, industry policy, energy policy and education policy and, indeed, of government itself. the second problem is http://www.australianpopulationstudies.org/ mailto:mcdonald.p@unimelb.edu.au 4 mcdonald australian population studies 2 (2) 2018 that accurate modelling of all these outcomes at some undefined time in the future is impossible – that is why it hasn’t been done. the third problem is that this statement provides no concern for the wellbeing of australians and their children who arrive from the day after the modelling is done. to be practical, population policy must have more modest objectives. taking a more pragmatic approach, i argue that australia already has a sophisticated, multi-faceted and effective population policy that has five components: maximising survival; supporting families to have the number of children that they want; moderating the speed and extent of population ageing; providing the skills that australia needs to promote its economic development; and influencing population distribution. 2. maximising survival australia aims to keep its population healthy and living long lives. this policy has been particularly successful from the 1970s onwards. death rates below the age of 75 years are now so low that, for women, their total elimination would add only two years to the expectation of life of australian women. just as spectacularly, death rates for older australians have plummeted since 1970. mortality in 1970-72 was such that 54 percent of men reaching the age of 55 were dead by age 75. by 2014-16, this percentage had fallen to 20 per cent (mcdonald 2017a). this has come about through the promotion of healthier lifestyles and safer environments but also by greatly increased public expenditure on health. the national transfer accounts for australia show that, between 1981-82 and 2009-10, real per capita health expenditure on persons aged 75 years and over in australia increased by a factor of 6.4 times. as the australia population aged 75 years and over increased three times in the same period, aggregate real public expenditure on the health of persons aged 75 years and over increased by an amazing 19 times (mcdonald 2017b). australia has been able to achieve this result without undue fiscal strain because government revenue increased during this period as a result of sustained economic growth. while the policy of maximising survival leads to a larger population particularly at older ages, there are no calls to reverse the trend. 3. supporting families to have the number of children that they want today, many countries in the world express concern about the level of their fertility rate. almost all countries with a fertility rate of 4.0 births per woman or more want to reduce the rate while almost all countries with a fertility rate below 1.5 births per woman want to increase their fertility rate (unfpa 2018). most countries would like to see their fertility rate fall in the relatively narrow range of 1.5 to 2.5 births per woman. this is very sensible because sustained high fertility leads to very rapid population growth which impedes economic and social development while sustained very low fertility leads to rapid population decline and excessive population ageing (unfpa 2018). planning to keep the fertility rate within this relatively narrow band is good policy (mcdonald 2006). in the case of australia, the fertility rate has fluctuated between 1.7 and 2.0 births per woman – an ideal range – for the past 42 years. this excellent result did not just happen. its achievement has been supported by the provision to women and men at low cost of the means to control the number of children that they have, that is, access to contraception and abortion. while there remains room for improvement, australian governments have also implemented a range of policies to support families with children such as subsidised childcare, parental leave, social security payments related to australian population studies 2 (2) 2018 mcdonald 5 children, and public support for education. while paul ehrlich (bateman 2016) and sustainable population australia (white 2009) have called for a china-style, one-child policy for australia, all governments, now including china, think that a one-child policy is not a great idea, leaving aside the human rights implications. fortunately, there is strong political support in australia for policies that support families to have the number of children that they want. 4. moderating the speed and extent of population ageing in the absence of migration, reduction of mortality and a fertility rate below the replacement level of around 2.06 births per woman leads inevitably to population ageing and population decline. at 30 june 2017, 15.4 per cent of the australian population was aged 65 years and over. if net overseas migration (nom) was fixed at zero, this would rise to 26.8 per cent by 2051, a rise of 11.4 percentage points. with the level of nom of 200,000 per annum continuing until 2051, 20.5 per cent of the population would be aged 65 years and over in 2050, a rise of 5.1 percentage points1. thus, while sustained migration at the present level will not stop the population from ageing, when compared with zero migration, it reduces population ageing very substantially. put another way, with nom of 200,000 per annum between now and 2051, the number of people aged 65 and over would be only 427,000 more than if nom was zero in the same period. that is, between now and 2015, the level of nom makes almost no difference to the number of older people in 2051. in contrast, with nom of 200,000 per annum, the number of people in the working ages of 20-64 years would be 6.8 million higher than if nom was zero. those additional 6.8 million people in the working ages will make a huge difference to the standard of living of older people in 2051. nom modifies population ageing because migrants are, on average, considerably younger than the australian population but, more significantly, because migrants, before they themselves become old, will have had their children and their grandchildren. nom of 200,000 per annum from today onwards would make almost no difference to the number of deaths in australia over the next 40 years but a very large difference to the number of births. this is not because migrants have a higher birth rate than non-migrants per woman but because migrants add to the population of childbearing age. if fertility remains at 1.8 births per woman, with zero nom, the australian population would begin to fall around 2045. the question then becomes: what level of nom produces the best outcomes in relation to population ageing? to address this question, in may 2010, at the request of the department of immigration and citizenship, mcdonald and temple (2010) conducted a study to identify if there was an ‘optimal’ range for nom where the criterion for an ‘optimum’ was the impact of immigration on population ageing and, hence, upon the growth rate of gdp per capita. they concluded that this range was 165,000 to 210,000. in 2011-12, the gillard labor government set the migration planning level in the middle of this range at 185,000. a later study by the same authors set this range at 160,000 to 220,000 (mcdonald and temple 2014) and, from 2012-13 until now, the planning level has been set in the middle of this range at 190,000 making for a run of nine years in which the 1 the projections results in this section are based on the june 2016 estimated resident population and assume that fertility is constant at 1.8 births per woman and that expectation of life rises by 2015 to 85.6 years for men and 88.3 years for women 6 mcdonald australian population studies 2 (2) 2018 migration planning level has been near-constant. the study showed that between 2013 and 2053, gdp per capita would be 12 per cent higher in real terms with nom at 180,000 than it would be if nom was zero across these years. for perspective, 12 per cent is larger than the overall increase in gdp per capita in australia over the past decade. an important consideration in this modelling is that nom is not the same entity as the migration planning level. the annual planning level simply consists of the total number of people to be accepted as new permanent residents in the skilled and family streams. in addition, for many years, the australian government has set the annual permanent humanitarian intake at about 13,0002. in contrast, nom is the net outcome of numerous forms of temporary and permanent movement into and out of australia as shown in table 1. table 1: components of net overseas migration by visa/movement type, australia, 2004-05 to 2015-16 (12year aggregate) visa/movement type visa components of net overseas migration, 2004-05 to 2015-16 arrivals departures net (number) net (%) permanent 1,085,468 220,812 864,656 34.0 skilled 503,424 92,815 410,609 16.2 family 398,153 62,543 335,610 13.2 humanitarian 130,453 1,665 128,788 5.1 other 53,457 63,808 -10,351 -0.4 temporary 2,733,004 1,187,117 1,545,887 60.8 skilled 416,795 170,739 246,056 9.7 student 1,239,551 447,445 792,106 31.2 working holiday 461,312 179,371 281,941 11.1 visitor 529,629 185,910 343,719 13.5 other 85,707 203,650 -117,943 -4.6 new zealand citizens 529,722 232,060 297,662 11.7 australian citizen 904,020 1,079,413 -175,393 -6.9 other 162,600 154,269 8,331 0.3 total 5,414,814 2,873,671 2,541,143 100.0 source: australian bureau of statistics, migration, australia, 2016-17. abs catalogue no. 3412.0. notes: numbers do not add precisely due to abs randomisation process. persons may change their visa type between arrival and departure, thus affecting the net figure for any visa/movement type. for example, the relatively large negative figure for ‘other temporary’ mainly reflects people who left australia when they were on a bridging visa although they had arrived on some other temporary visa. if the measured impact of migration on gdp per capita is based on the impact of nom, how can this be reconciled with the setting of the annual permanent migration intake? the answer is that, although nom fluctuates from year to year because of surges in temporary arrivals or departures, in 2 the humanitarian program is temporarily at a higher level due to a special allocation of places to syrian refugees australian population studies 2 (2) 2018 mcdonald 7 the longer term, temporary migrants can only remain in australia if they are granted permanent residence through the permanent migration program. as the net impact of the combined movements of australian and new zealand citizens is relatively small (less than 5% of nom from 2004-05 to 2015-16, see table 1), in the long run, nom and the migration planning levels are very similar (see figure 1). nevertheless, temporary migration serves the very important purpose of providing a ready pool from which the majority of permanent residents are selected. over the past decade, by far the largest category of temporary residents has been international students and table 1 shows that this category has made the largest contribution to nom. in sum, the level of nom that has resulted from the government’s migration planning levels over the past nine years, if sustained into the future, would have a highly beneficial impact on population ageing and hence on the level of gdp per capita in australia in the years to 2051. by 2051, australia would be among the youngest countries in the oecd. the offsetting factor is that, with nom of 200,000 per annum, the total population rises by 2051 to 36 million compared with 26 million (and falling) if nom was zero – but 70 per cent of the additional 10 million people will be in the productive working ages. the implications of an additional 10 million people are considered below. figure 1: annual net overseas migration (nom) compared with the annual grants of permanent residence through the permanent migration program including humanitarian, australia, 1983-84 to 2016-17 sources: department of home affairs; australian bureau of statistics. 5. providing the skills that australia needs to promote its economic development while the total numbers in the annual migration program may be set primarily on the basis of the impact of migration on population ageing and, hence upon gdp per capita, the composition of the skilled stream is determined by perceived skill shortages in the australian labour force that cannot be met by the existing population. the skilled migration program is restricted to managers, 8 mcdonald australian population studies 2 (2) 2018 professionals, para-professionals and skilled tradespeople to ensure that immigrant workers are persons with high labour productivity. within these skill categories, the government identifies those occupations where domestic supply falls short of domestic demand and includes these occupations on the list of occupations eligible for skilled migration. a wider list of occupations is applied where the applicant is nominated by an employer or by a state or territory government. the evidence is strong that the size and composition of australia’s migration program in recent times has been associated with strong labour demand across those years. between june 2011 and june 2016, total employment in australia grew by 739,000 people. this increase was composed of 613,000 migrants who entered australia after june 2011, 278,000 additional workers over the age of 55 years who were present in australia in june 2011 and a net fall of 152,000 among people aged under 55 who were present in australia in june 2011 (mcdonald 2017c). at the same time, the unemployment rate in september 2018 was at its lowest point since april 2012. since 2011, the large immigration intake has been effectively absorbed into the australian labour force. this could not have happened if there had not been an equivalent level of labour demand. looking to the next decade, australia is facing a labour supply crunch as large numbers of babyboomers retire at the same time as the young cohorts entering the labour force ages are smaller in number than their predecessors. with zero nom from 2016 to 2026, the age group 20-34 would fall by 725,000; with nom at 70,000 it would fall by 36 thousand; and with nom at 200,000, it would rise by 413 thousand. at the same time, the population in the retirement ages would increase under all scenarios by about 1.35 million. the latest abs job vacancies data (abs 2018) show a continual rise in job vacancies from 2014 onwards, and vacancies are rising at the rate of 19.3 per cent per annum. australia has created an immigration system that is responsive to labour demand. this is nowhere more evident than in the recent trends in nom at the state and territory level. when the mining boom was at its peak, nom was historically high for the boom states of western australia and queensland. nom dropped off sharply to these states when the mining boom ended. now, nom is directed towards the boom cities of sydney and melbourne. this reflects shifts in the direction of investment by firms and a corresponding shift in labour demand. all the indications are that the number of job openings in australia over the next six years will be very high, that is, labour demand will remain very strong. using econometric modelling, shah and dixon (2018: table 4) have estimated that there will be 4.14 million job openings in australia across the eight years 2017-2024. more than half of these openings (2.27 million) represent replacement demand driven almost exclusively by retirement of the baby-boom generation. some of these openings will be filled by the next generation of young australian workers, but migration at least at the current level will be required to meet this demand. shah and dixon (2018: appendix table a1) also estimate that there will be very considerable change in the distribution of occupations over the period 2016 to 2024. this once more emphasises the need for young workers because it is much easier for younger workers, especially those who have not yet entered the labour force, to adjust to changes in demand for particular occupations. with zero nom, the number of employed 15-34 years-olds would fall by 518 thousand from 2016 to 2026 (assuming australian population studies 2 (2) 2018 mcdonald 9 constant employment participation rates); with nom continuing at 200,000, their number would increase by 425 thousand, a turnaround of close to a million young workers. thus, migration greatly enhances the capacity for the occupational composition of australia employment to respond to demand. 6. population distribution, cities and regions in australia, residents are free to move and take up residence in another location. this means that the population distribution will reflect the preferences that they make. in recent times, residents, old and new, have displayed a distinct preference to live in the major cities of sydney, melbourne and brisbane and in their satellite towns and cities. in 2016-17, 77 per cent of australia’s total population growth was in sydney, melbourne, brisbane and their satellites (newcastle, wollongong, central coast, geelong, melton, bendigo, ballarat, gold coast, sunshine coast and toowoomba). while melbourne’s population grew by more than 2,000 per week, the combined growth of the two fastest growing inland country towns (dubbo and albury-wodonga) was just 2,000 a year. as the global economy moves towards mega cities, there should not be an expectation that future growth of the australian population, in the foreseeable future, can be redirected in anything other than a minor way to regional australia. such redirection is important in keeping regional australia viable but, while the australian economy remains strong, labour demand will be strongest in sydney, melbourne, brisbane and their satellites. accordingly, most of the growth of population will be in these cities. this has been recognised appropriately in the emphasis placed by the federal and state and territory governments upon investment in city infrastructure. it should also be remembered that, in general, the provision of urban infrastructure has lagged behind population growth in australian cities. if you want the evidence, view the sombre (but highly entertaining) you tube video on the 1954 plan for melbourne (https://www.youtube.com/watch?v=mrmovhij34m). some have suggested that lowering the level of the migration program would provide a ‘breathing space’ so that infrastructure development could catch up with population growth. this is flawed logic because if, as argued above, labour demand remains very strong at least in the large cities, firms unable to fill that demand from international migration will draw instead upon the rest of australia and on new zealand. wages and opportunities will be higher in sydney and melbourne and the best talent from the rest of australia will be drawn to these cities. the infrastructure problem in the big cities would not have gone away and regional australia will lose population. and major construction of infrastructure in the big cities requires labour just as mine construction phase did in western australia and queensland during the mining boom. both labor and coalition governments over the past decade have sought to ensure that economic growth remains strong. a wealthy australia will be in a strong position to fund the services required by the growing population, to invest in new urban infrastructure and renewable energy sources, and to deal with environmental issues. in the coming years of the labour supply crunch, present australian population policy is in keeping with these objectives and should be continued at least until 2026. https://www.youtube.com/watch?v=mrmovhij34m 10 mcdonald australian population studies 2 (2) 2018 7. key messages • australia has an effective population policy – and should stick with it at least to 2026. • australia is facing a labour supply crunch in the next decade as the baby-boom retires and because the future cohorts entering the labour force at young ages are much smaller than the current young working population. • labour demand is very strong and the growth in job vacancies is approaching 20% per annum. • the occupational structure of the labour force is set to change considerably. young workers are more easily able to adapt to these changes than are older workers. • at present labour demand is strongest in the big cities. if this demand is not met by immigrants, it will be met by people moving from the rest of australia and from new zealand. • the big cities will continue to grow. there is no substitute for investment in urban infrastructure. references australian bureau of statistics (2018) job vacancies. catalogue no. 6354.0. canberra: abs. bateman d (2016) scientist calls for a one-child policy to curb population explosion. the cairns post 11 november 2016. https://www.heraldsun.com.au/news/national/scientist-calls-for-onechildpolicy-to-curb-population-explosion/news-story/6ac63d6ee1d3fff4b98a26ad5b17cabe department of sustainability, environment, water, population and communities (2011) sustainable australia – sustainable communities: a sustainable population strategy for australia. canberra: commonwealth of australia. http://apo.org.au/system/files/166281/apo-nid166281-770266.pdf mcdonald p (2006) ‘low fertility and the state: the efficacy of policy’. population and development review 32(3): 485-510. mcdonald p and temple j (2010) immigration, labour supply and per capita gross domestic product: australia 2010-2050. canberra: department of immigration and citizenship. https://www.homeaffairs.gov.au/reportsandpublications/documents/research/labour-supplygdp-2010-2050.pdf#search=peter%20mcdonald mcdonald p and temple j (2014) the long term effects of ageing and immigration upon labour supply and per capita gross domestic product: australia 2012-2062. canberra: department of immigration and border control. http://www.immi.gov.au/search/pages/results.aspx?k=the%20long%20term%20effects%20of% 20ageing%20and%20immigration%20upon%20labour%20supply%20and%20per%20capita%20g ross%20domestic%20product mcdonald p (2017a) long-term changes in mortality by age and sex in australia. powerpoint presentation to a symposium organised by the arc centre of excellence in population ageing research, melbourne town hall, 14 august 2017. mcdonald p (2017b) ageing in australia: the impact of immigration. powerpoint presentation to a seminar organised by the australian and new zealand school of government, melbourne business school, 28 november 2017. mcdonald p (2017c) international migration and employment growth in australia, 2011-2016. australian population studies 1 (1): 3-12. productivity commission (2016) migrant intake into australia. productivity commission inquiry report, 13 april 2016. shah c and dixon j (2018) future job openings for new entrants by industry and occupation. adelaide: commonwealth of australia, ncver. https://www.heraldsun.com.au/news/national/scientist-calls-for-onechild-policy-to-curb-population-explosion/news-story/6ac63d6ee1d3fff4b98a26ad5b17cabe https://www.heraldsun.com.au/news/national/scientist-calls-for-onechild-policy-to-curb-population-explosion/news-story/6ac63d6ee1d3fff4b98a26ad5b17cabe http://apo.org.au/system/files/166281/apo-nid166281-770266.pdf https://www.homeaffairs.gov.au/reportsandpublications/documents/research/labour-supply-gdp-2010-2050.pdf#search=peter%20mcdonald https://www.homeaffairs.gov.au/reportsandpublications/documents/research/labour-supply-gdp-2010-2050.pdf#search=peter%20mcdonald http://www.immi.gov.au/search/pages/results.aspx?k=the%20long%20term%20effects%20of%20ageing%20and%20immigration%20upon%20labour%20supply%20and%20per%20capita%20gross%20domestic%20product http://www.immi.gov.au/search/pages/results.aspx?k=the%20long%20term%20effects%20of%20ageing%20and%20immigration%20upon%20labour%20supply%20and%20per%20capita%20gross%20domestic%20product http://www.immi.gov.au/search/pages/results.aspx?k=the%20long%20term%20effects%20of%20ageing%20and%20immigration%20upon%20labour%20supply%20and%20per%20capita%20gross%20domestic%20product australian population studies 2 (2) 2018 mcdonald 11 unfpa (united nations population fund) (2018) state of world population 2018: the power of choice, reproductive rights and the demographic transition. new york: unfpa. white h (2009) institute one child policy for a “sustainable” australia: population control group lifesitenews.com, 24 april 2009. https://www.lifesitenews.com/news/institute-one-child-policyfor-a-sustainable-australia-population-control-g https://www.lifesitenews.com/news/institute-one-child-policy-for-a-sustainable-australia-population-control-g https://www.lifesitenews.com/news/institute-one-child-policy-for-a-sustainable-australia-population-control-g 7. key messages references australian population studies 2017 | volume 1 | issue 1 | pages 13–25 © wilson and taylor 2017. published under the creative commons attribution-noncommercial licence 3.0 australia (cc by-nc 3.0 au). journal website: www.australianpopulationstudies.org alternative methods of determining the number of house of representatives seats for australia’s territories tom wilson* charles darwin university andrew taylor charles darwin university *corresponding author. email: tom.wilson@cdu.edu.au. address: northern institute, charles darwin university, darwin nt 0909, australia paper received 15 august 2017; accepted 19 october 2017; published 20 november 2017 abstract background population size determines the number of seats each australian state and territory is entitled to in the house of representatives. the northern territory (nt) and australian capital territory (act) were allocated two and three seats, respectively, in the august 2017 determination, but by very small margins. both territories risk losing a seat at the next determination. this would result in them having considerably more people per member of parliament than any of the states. aims this paper (1) provides modelling to support the consideration of alternative rules for determining membership entitlement to the house of representatives which does not disadvantage the nt and act and (2) presents population projections for future determinations under the current and alternative rules. data and methods population projections for the states and territories were produced for three demographic scenarios. the resulting numbers of seats for the nt and act were calculated for each scenario under the current and proposed alternative seat entitlement rules. results under the existing rules the nt and act would only keep their current number of seats at the next determination if they experienced higher net in-migration than in recent years. under the alternative seat entitlement rules suggested, the nt and act would be very unlikely to lose any seats and would almost certainly gain seats in ensuing decades. conclusions there is a case for re-examining the way the states and territories are allocated seats in the house of representatives. key words commonwealth electoral act; constitution; house of representatives; australia; seat entitlements; population projections; northern territory; australian capital territory. http://www.australianpopulationstudies.org/ mailto:tom.wilson@cdu.edu.au 14 wilson t and taylor a australian population studies 1 (1) 2017 1. introduction population statistics play a central role in determining the number of seats in the house of representatives assigned to each state and territory. the seat allocation process is set out in the commonwealth electoral act 1918, and is based on the fundamental principle described in the australian constitution that each state is allocated a share of seats equal to its share of the national population (commonwealth electoral act 1918; commonwealth of australia constitution act 1900). various special provisions, along with the need to round fractional seat entitlements (e.g. 10.6 or 11.4 seats) to the nearest whole number, mean that the population–seat relationship is imprecise, but much of the basic principle is retained. because of changes to state and territory shares of the national population over time, the electoral act requires the australian electoral commissioner to periodically re-calculate seat entitlements. a year-and-a-day after the first sitting of the house of representatives following a general election, updated population numbers are obtained from the australian bureau of statistics (abs). new calculations of state and territory seat entitlements are then made. when the number of seats in a state or territory changes, a redistribution of electoral boundaries within that jurisdiction is initiated and the new boundaries come into effect at the next federal general election. sometimes seat entitlement calculations have generated controversy, especially in the case of the northern territory (nt). in the 2003 determination of state and territory entitlements for the house of representatives, the nt saw its entitlement reduced from two seats to just one because of a falling share of the national population (australian electoral commission 2003). but it was very close: if the nt had just 295 more people it would have retained two seats. the ensuing controversy resulted in an inquiry into representation of the territories in the house of representatives by the joint standing committee on electoral matters. the committee report recognised the greater uncertainty of territory estimated resident population (erp) calculations relative to those of the states and recommended both that the 2003 seat determination for the nt be set aside and a margin of error be incorporated in future seat determinations ‘when a territory falls short of quota’ (joint standing committee on electoral matters 2003, pp. viii–ix). the electoral act was amended the following year to incorporate the report’s recommendations and the nt retained its two seats. in the years 2013–2016 the nt again experienced population growth below the national rate (wilson 2017a) resulting in a declining share of the national population. in the recent 2017 determination the nt retained two seats, though by a very small margin (australian electoral commission 2017). if the nt were to lose a seat in the near future, about a quarter of a million people would be represented by a single member of parliament, a population to seat ratio much higher than for any of the states. the 2017 determination also gave the act an additional seat, but by a very small margin. if the act were to lose a seat, it too would be subject to a much higher population to seat ratio than any of the states. this raises two key questions: (1) might the nt and act lose a seat in the next determination?; (2) what could be changed to ensure the territories do not lose a seat in the future? this paper proposes two alternative models for seat entitlement rules and presents population and associated seat projections for the nt and act under the current and proposed alternative seat entitlement rules. section 2 of the paper describes the process for calculating seat entitlements prescribed by current legislation, and models two alternative schemes in which the territories are not australian population studies 1 (1) 2017 wilson t and taylor a 15 subject to inferior representation in the house of representatives. population projections based on various scenarios for the three schemes (current and alternative models) are presented in section 3, while brief conclusions form the final section of the paper. 2. current and alternative seat entitlement calculations 2.1 current scheme the current method for seat entitlement calculations is specified in the commonwealth electoral act 1918 as amended and in force on 21 october 2016. the 2016 amendments mean that the calculations differ slightly from those described earlier by abs (2005), corr (2016), payer and taylor (2015) and wilson, beneforti and barnes (2005). the key elements of the process are summarised in the flow diagram in figure 1. seat entitlement calculations for each of the states are relatively straightforward. first, the quota is calculated. the quota is the population required for a jurisdiction to be assigned a seat in the house of representatives. it is defined in the electoral act as the latest official population number for the six states combined, divided by twice the total number of senators for the six states (2 × 72 = 144). the number of senators is defined by legislation. while it can change in theory, for practical purposes the number can be assumed to be fixed at 72. the latest erp of each state is divided by the quota to obtain the seat entitlement, shown in figure 1 as w.f. where w represents the whole number part of the seat entitlement and f the fractional part. the seat entitlement is rounded down if the fractional part f is less than or equal to 0.5, and rounded up if greater than 0.5. if any state (but not territory) has fewer than five seats, then the australian constitution requires the number of seats to be increased to five. in practice this now only applies to tasmania (though long ago it was relevant also to other states). the calculations are more complicated for the nt and act. the populations of the territories of the cocos (keeling) islands and christmas island are included in the nt’s population. (strictly, the electoral act states that if the cocos islands and christmas island do not have sufficient population to be allocated a seat themselves they are to be included with the nt – but in practical terms they can always be regarded as part of the nt.) a similar provision applies to norfolk island whose erp is included in the act’s population until it is entitled to separate representation1; jervis bay territory is counted as part of the act. the provisional seat entitlements for the nt and act are then calculated by dividing their populations by the quota. at this point the calculations for the territories can deviate from those of the states. if the fractional part f of the unrounded preliminary seat number for the nt or act is less than or equal to 0.5, then a check is made to determine whether the population of the territory is within a specified margin of error. this margin of error allowance was introduced to the electoral act in 2004 on the recommendation of the joint standing committee on electoral matters (2003). 1 this is the first seat determination in which norfolk island’s population (most recently 1,756) has been included as part of the act due to recent changes in norfolk island’s status. however, it did not materially affect the seat determination outcome. 16 wilson t and taylor a australian population studies 1 (1) 2017 figure 1: summary of the current house of representatives seat determination calculations source: based on commonwealth electoral act 1918. notes: w.f. = whole number.fraction; se = standard error. quota = combined erp of the 6 states 2 × total no. of senators of the 6 states no. of seats = state erp = w.f. quota is the fractional part f ≤ 0.5? no. of seats rounded down to w no. of seats rounded up to w+1 is the no. of seats < 5? yes no yes no no. of seats unchanged no. of seats set at 5 nt population = nt erp + christmas island erp + cocos islands erp act population = act erp + jervis bay erp + norfolk island erp no. of seats = territory population = w.f. quota states yes no the population is increased by 2 × se of the net undercount is the fractional part f ≤ 0.5? yes no nt act no. of seats rounded down to w no. of seats rounded up to w+1 yes no is the fractional part f ≤ 0.5 and (quota × n.5 – population) ≤ (2 × se net undercount)? is the no. of seats zero? no. of seats set at 1 no. of seats unchanged no. of seats = augmented population = w.f. quota australian population studies 1 (1) 2017 wilson t and taylor a 17 the increase in population required to obtain an entitlement with exactly a 0.5 fraction is then calculated. for example, if the whole number part of the seat entitlement is 2, then this ‘threshold’ population is 2.5 × quota. if the increase required to reach this threshold population is less than or equal to twice the standard error (se) of the net undercount estimated from the last census, then the official population of the territory is taken to be its original population plus twice the se of the net undercount. the seat entitlement calculation is re-calculated using this augmented population. the same rounding process as for the states is then applied. finally, a check is made to ensure that each territory has a minimum of one seat. table 1 presents key statistics from the 2017 seat determination. the most recent official population numbers at the time of the determination were the 31 december 2016 erps published by the abs in australian demographic statistics on 27 june 2017 (abs 2017). using these data, the quota was calculated as 23,729,561 / 144 = 164,788.6181. three jurisdictions experienced changes in their number of seat entitlements. victoria gained a seat while south australia lost one. the act also gained a seat due to the margin of error allowance for its erp. the preliminary unrounded number of seats for the act was calculated as 2.4793 (column b). because the fractional part f was less than 0.5 it was necessary to determine whether the act’s population was within the permitted margin of error to get above 2.5. twice the se of net undercount from the abs 2016 census for the act was 10,694 (column c), while the additional population required to increase the seat entitlement to 2.5 was 2.5 × 164,788.6181 – 408,562 = 3410 (column d). therefore, the population of the act was increased by twice the se of the undercount to obtain an augmented population (column e) and the entitlement calculation was repeated (column f). table 1: state and territory seat determinations, 2017 jurisdiction population (a) population/ quota (b) 2 × se net undercount (c) w.5 quota – population (d) augmented population (e) aug. popn/ quota (f) seats (g) nsw 7,797,791 47.3200 47 vic. 6,244,227 37.8923 38 qld 4,883,739 29.6364 30 sa 1,716,966 10.4192 10 wa 2,567,788 15.5823 16 tas. 519,050 3.1498 *5 nt 247,512 1.5020 2 act 408,562 2.4793 10,694 3,410 419,256 2.5442 3 total – – – – – – 151 source: based on australian electoral commission (2017). note: *5 seat minimum applied. under the current legislation the nt and act have two and three seats, respectively, by very small margins. if they were to drop to one and two seats, then the population being represented by each member of parliament in the territories would be far higher than that of any of the states. there are a number of alternative ways in which the number of seats could be calculated for the territories which would prevent them from being disadvantaged in this way. we propose two alternative seat entitlement schemes, the special quota scheme and the fixed quota scheme, discussed below. 18 wilson t and taylor a australian population studies 1 (1) 2017 2.2 special quota scheme because the australian constitution provides for a minimum of five seats in any state, tasmania receives two more seats than it would be assigned simply on the basis of its population size. the effective quota for tasmania is therefore substantially lower than the actual quota and in the 2017 determination was 519,050 erp / 5 seats = 103,810. under our proposed special quota scheme the effective tasmanian quota would be applied to the nt and act, the argument being that if a state is entitled to minimum representation due to small population size, then these two territories should be afforded the same treatment. the process for the nt and act might be amended to something like this: • the number of seats which would be allocated to each territory under the regular quota is calculated. • if the minimum 5 seat provision is applied to any state, then the effective quota for that state – which could be termed the special quota – is calculated. if the minimum 5 seat provision applies to more than one state then the lowest effective quota is deemed the special quota (though in practice this is unlikely). • if (i) a territory would be allocated fewer than 5 seats by dividing its population by the quota and (ii) the minimum 5 seat provision is applied to any state, then the seat entitlement for that territory is calculated by dividing its population by the special quota. • the number of seats calculated in this way for any territory is capped at 5. the rounding up or down of fractional parts of the seat entitlement remains the same as in the current process, as does the provision of a margin of error based on the se of the net undercount from the last census. had this alternative system been in place in the 2017 determination, then the outcome would have been as shown in table 2: the nt would have achieved two seats comfortably, while the act would have been allocated four seats. the nt’s entitlement of 2.3843 would not have put it within the margin of error to be rounded up to 3. table 2: 2017 seat determinations for the nt and act under the special quota system territory population population/special quota seats nt 247,512 2.3843 2 act 408,562 3.9357 4 source: authors’ calculations. 2.3 fixed quota scheme another alternative scheme would be to keep the current seat entitlement calculation rules but with one exception: the current method of calculating the quota based on the states’ latest populations would be replaced by a fixed number. this might be, for example, 150,000 people per seat. the argument behind this approach is that each member of the house of representatives can only effectively serve a certain number of constituents. therefore, as australia’s population increases, so will the number of seats in the house of representatives. given the constitutional requirement that the total number of seats for the six states be twice the number of senators, the number of senators australian population studies 1 (1) 2017 wilson t and taylor a 19 would also have to increase. this method also has the advantage of being very simple. the seat entitlements which would have been calculated in the 2017 determination with a quota of 150,000 are shown in table 3. the nt would comfortably have been assigned two seats, while the act would have been allocated three seats. the total number of seats in the house of representatives would be 165 under this scheme. table 3: 2017 seat determinations with a fixed quota of 150,000 jurisdiction population population/quota seats nsw 7,797,791 51.9853 52 vic. 6,244,227 41.6282 42 qld 4,883,739 32.5583 33 sa 1,716,966 11.4464 11 wa 2,567,788 17.1186 17 tas. 519,050 3.4603 *5 nt 247,512 1.6501 2 act 408,562 2.7237 3 total – 165 source: authors’ calculations. note: *5 seat minimum applied. 3. projections how might the nt’s and act’s seat entitlements change in the future under the current legislation and alternative calculation schemes described above? to provide answers, population projections were produced for the states and territories from 2016 to 2051. calculations of seat entitlements for the nt and act were then made for every projection year (2021, 2026, … , 2051) under the current, special quota and fixed quota schemes. 3.1 projection methods, data and scenarios projections were prepared using the regional population projection program (reppp) (wilson 2017b). this excel/vba program incorporates a multi-regional type cohort-component model which outputs projections by sex and five-year age group for every fifth year into the future. most input data were sourced from the abs website, primarily from abs.stat (http://stat.data.abs.gov.au/). these included preliminary rebased 30 june 2016 erps, total fertility rates (tfrs) and age-specific fertility rates, state and territory life expectancies at birth, interstate migration flows, and overseas migration flows. national mortality rate projections were created using ediev’s (2008) extrapolative method based on data extracted from the human mortality database (2017). jurisdiction-specific mortality assumptions were created by scaling national mortality projections according to recent life expectancy at birth values. for all three scenarios the average tfrs of 2011–2016 were assumed to remain constant. the populations of the small territories of christmas island, the cocos islands, jervis bay and norfolk island were assumed to remain constant at their 2016 erps. the ses of net undercount from the 2016 census for the nt and act were also assumed to remain constant. http://stat.data.abs.gov.au/ 20 wilson t and taylor a australian population studies 1 (1) 2017 three scenarios with differing migration assumptions were produced: 1. current trajectory scenario, in which the net interstate migration (nim) values of 2011–2016 for the nt and act were held constant; for the states an annual average of the last 20 years of nim was assumed (but adjusted to ensure nim summed to zero across all jurisdictions). net overseas migration (nom) values from 2011–2016 were held constant for all jurisdictions. projections under this scenario demonstrate the population outcomes if the demographic ‘settings’ of the 2011–2016 period for the nt and act were to remain unchanged. 2. low migration scenario, in which nim and nom for the nt and act were held constant at the 10th percentile of their distributions over the last 20 years. assumptions for the states were the same as in the current trajectory scenario (with adjustments made to obtain zero nim when summed across all jurisdictions). projections from this scenario show the result of long-run highly negative net migration outcomes for the two territories. 3. high migration scenario, in which nim and nom for the nt and act were held constant at the 90th percentile of their past distributions. again, assumptions for the states were the same as in the current trajectory scenario (except for adjusted nim values). this scenario’s projections reveal the outcome of consistently high net migration gains for the two territories. assumptions for the scenarios are given in the appendix (table 4). projections from the three scenarios do not cover the range of all possible demographic outcomes, of course. migration for individual fiveyear projection intervals may fall outside the high–low range, and fertility and mortality are both likely to deviate somewhat from the assumptions used. however, past analyses have shown that migration is the greatest source of uncertainty about future population for both the nt and act (wilson 2012), and the high–low ranges considered here cover a wide range of plausible outcomes for the two territories. 3.2 projection results the projected populations of the nt and act according to the three scenarios are shown in figure 2. given the wide range of historical nim and nom values for these two territories, it is not surprising to see substantially different results from the high migration and low migration assumptions. figure 2: population projections for the nt and act, 2016–2051 source: authors’ projections. note: the act’s current trajectory projection is closer to its high migration projection because nom over 2011–2016 was towards the higher end its distribution for the last 20 years. australian population studies 1 (1) 2017 wilson t and taylor a 21 northern territory australian capital territory figure 3: unrounded seat entitlements for the nt and act, 2016–2051, according to three seat calculation schemes and three demographic scenarios source: authors’ calculations. 22 wilson t and taylor a australian population studies 1 (1) 2017 figure 3 illustrates the unrounded seat entitlements calculated for the nt and act if seat determinations were made on 30 june of the projection years 2021, 2026, … , 2051. the detailed calculations behind these graphs are contained in the excel workbook accompanying this paper. under the current seat entitlement rules, each territory only keeps the number of seats assigned at the 2017 determination under the high migration scenario. this is because population growth amongst the six states results in the quota increasing over time, reaching about 260,000 by 2051. under the current trajectory scenario both territories lose a seat in the 2030s but, as the graphs show, they remain very close to losing a seat for the entire preceding period. if the low migration scenario was to eventuate, both territories would probably lose a seat as early as the next determination. under the special quota scheme, the quota is determined by the size of tasmania’s population. the seat projections are therefore less robust because of uncertainty about the future of tasmania’s population. in our projections tasmania experiences only a modest amount of population growth until the early 2030s (to about 540,000), followed by gradual decline. as a consequence, the special quota never exceeds 108,000. under the ‘worst case’ scenario of low migration the nt increases from two to three seats in the 2040s and the act retains four seats through to 2051. under the high migration scenario both territories increase to five seats: the nt at the end of the projection horizon, and the act early on (noting that the special quota scheme provides a maximum of five seats). even if the slightly more optimistic abs 2012-based series b projections for tasmania were used instead (abs 2013), the ‘worst case’ low migration scenario would keep the nt on two seats, and the act on four throughout the projection horizon. under the fixed quota scheme the seat projections are determined solely by each jurisdiction’s future population. under the low migration scenario, both the nt and act remain on two and three seats, respectively, for the whole projection horizon. under the high migration scenario the nt will have gained one seat and the act two by mid-century. 4. conclusion we believe there is a case to re-examine the legislation on representation in the commonwealth house of representatives, focusing particularly on the nt and act. not only are there precedents for this, with revisions made previously to the legislation, it is also permitted under the constitution. section 122 of the australian constitution states: the parliament may make laws for the government of any territory … and may allow the representation of such territory in either house of the parliament to the extent and on the terms which it thinks fit. under the current electoral act both the nt and act are at risk of losing a seat in the house of representatives if current population trends continue. if this were to occur, then the population represented by each of their members of parliament would be greater than in any of the states. preventing future seat losses in the two territories will require much higher net migration than experienced in recent years (or very low population growth in the six states). the alternative special quota and fixed quota seat entitlement schemes proposed in this paper would prevent the territories australian population studies 1 (1) 2017 wilson t and taylor a 23 from losing seats. even under the low migration scenario, neither the nt nor the act lose seats under these two schemes; under the high migration scenario both will gain seats. the straightforward nature of the fixed quota scheme would potentially make for simpler and clearer legislation. alternatively, the special quota scheme would place the nt and act at least on a par with tasmania, in that they would be subject to the same effective special quota. the two schemes could even be combined using a standard quota of 150,000 together with the special quota applied to the nt and act. of course, there are other seat entitlement schemes which could be considered. one would be to provide a minimum of two seats each for the nt and act, a proposal made to the joint standing committee on electoral matters inquiry in 2003 but not recommended by the committee. alternatively, the two territories could be subject to the same minimum number of seats which apply to the states (5 seats), but that might be seen as being too generous to the nt. with no change to the current rules the nt and act could well lose a seat each – possibly at the next determination. importantly, should the nt be reduced to just one seat it would then be by far the most poorly represented jurisdiction in the house of representatives, with about a quarter of a million people to one member of parliament for an electorate covering 1.4 million square kilometres. on that basis alone, there is sufficient weight to consider alternative approaches, including those proposed here. key messages • there is a case for the parliament of australia to re-examine the way seat entitlements are calculated. • in the recent 2017 determination of house of representatives seats the nt and act were allocated two and three seats, respectively. however, they avoiding being assigned just one and two seats, respectively, by the narrowest of margins. • had the nt and act lost seats at the 2017 determination, the two territories would have had a much larger number of people per elected member than any of the australian states. however, projections in this paper demonstrate that under the current seat entitlement rules, the nt and act are at risk of losing a seat each at the next determination. • this paper proposes two alternative ways in which seat entitlements could be calculated for the nt and act which would prevent them from being disadvantaged. under these proposed schemes, the nt and act would be very unlikely to lose any seats at the next determination and therefore would not be disadvantaged in their representation in respect to the states. acknowledgments the authors are most grateful for helpful advice and comments on an earlier draft by australian electoral commission staff and the anonymous reviewers. all opinions and errors remain our own. 24 wilson t and taylor a australian population studies 1 (1) 2017 references australian bureau of statistics (abs) (2005) information paper: determining seats in the house of representatives – legislative requirements for provision of abs statistics. cat. no. 3107.0.55.002. canberra: abs. australian bureau of statistics (abs) (2013) population projections, australia, 2012 (base) to 2101. cat. no. 3222.0. canberra: abs. australian bureau of statistics (abs) (2017) australian demographic statistics: december quarter 2016. cat. no. 3101.0. canberra: abs. australian electoral commission (2003) 2003 redistribution process commences. electoral newsfile 110 (april). canberra: australian electoral commission. australian electoral commission (2017) determination of membership entitlement to the house of representatives, media release, viewed 31 august 2017, http://www.aec.gov.au/media/mediareleases/2017/08-31.htm. commonwealth of australia constitution act 1900. compilation prepared on 25 july 2003 taking into account alterations up to act no. 84 of 1977, viewed 1 july 2017, https://www.legislation.gov.au/details/c2004c00469. commonwealth electoral act 1918. compilation no. 63 as amended and in force on 21 october 2016, viewed 1 july 2017, https://www.legislation.gov.au/details/c2016c01022. corr p (2016) democracy and dollars: use of demography in distribution. in: wilson t, charles-edwards e and bell m (eds) demography for planning and policy: australian case studies. applied demography series, vol. 7. cham: springer; 9–29. ediev d m (2008) extrapolative projections of mortality: towards a more consistent method. part i: the central scenario. vienna institute of demography working papers 3/2008. vienna, austria: austrian academy of sciences. human mortality database (2017) deaths and population data for australia. university of california, berkeley usa and max planck institute for demographic research, rostock, germany, viewed 7 july 2017, http://www.mortality.org/cgi-bin/hmd/country.php?cntr=aus&level=1. joint standing committee on electoral matters (2003) territory representation: report of the inquiry into increasing the minimum representation of the australian capital territory and the northern territory in the house of representatives. canberra: parliament of the commonwealth of australia. payer h and taylor a (2015) northern territory seats in the australian parliament: it’s a long way up, but not far down. research briefs 02: 1–11, viewed 17 july 2017, http://www.cdu.edu.au/sites/default/files/research-brief-2015-02.pdf. wilson t (2012) forecast accuracy and uncertainty of australian bureau of statistics state and territory population projections. international journal of population research. volume 2012. http://dx.doi.org/10.1155/2012/419824. wilson (2017a) territory population update: new population statistics for december 2016, viewed 8 july 2017, http://www.cdu.edu.au/sites/default/files/the-northerninstitute/territory_population_update_27-06-2017.pdf. wilson t (2017b) regional population projection program (reppp) model description. northern institute working papers 2017/01, viewed 8 july 2017, http://www.cdu.edu.au/sites/default/files/thenorthern-institute/docs/reppp-model-description-revised2.pdf. wilson t, beneforti m and barnes t (2005) population statistics and the number of house of representatives seats for the northern territory. people and place 13(4): 23–33. http://www.aec.gov.au/media/media-releases/2017/08-31.htm http://www.aec.gov.au/media/media-releases/2017/08-31.htm https://www.legislation.gov.au/details/c2004c00469 https://www.legislation.gov.au/details/c2016c01022 http://www.mortality.org/cgi-bin/hmd/country.php?cntr=aus&level=1 http://www.cdu.edu.au/sites/default/files/research-brief-2015-02.pdf http://dx.doi.org/10.1155/2012/419824 http://www.cdu.edu.au/sites/default/files/the-northern-institute/territory_population_update_27-06-2017.pdf http://www.cdu.edu.au/sites/default/files/the-northern-institute/territory_population_update_27-06-2017.pdf http://www.cdu.edu.au/sites/default/files/the-northern-institute/docs/reppp-model-description-revised2.pdf http://www.cdu.edu.au/sites/default/files/the-northern-institute/docs/reppp-model-description-revised2.pdf australian population studies 1 (1) 2017 wilson t and taylor a 25 appendix table 4: summary of projection assumptions used in the three projection scenarios jurisdiction tfr e0 in 2046–51 (males; females) nim per annum nom per annum current trajectory nsw 1.88 88.0; 89.9 -17,429 66,781 vic. 1.77 88.7; 90.1 2,977 58,514 qld 1.94 87.8; 89.4 19,428 29,881 sa 1.85 87.9; 89.7 -3,089 10,649 wa 1.90 88.1; 90.4 1,582 29,436 tas. 1.97 86.3; 88.0 -763 1,176 nt 2.06 82.8; 83.5 -2,375 2,233 act 1.77 88.9; 90.7 -331 2,354 low migration nsw -16,806 66,781 vic. 3,088 58,514 qld 20,148 29,881 sa as above as above -2,978 10,649 wa 1,641 29,436 tas. -736 1,176 nt -2,795 482 act -1,561 -159 high migration nsw -19,099 66,781 vic. 2,717 58,514 qld 17,730 29,881 sa as above as above -3,385 10,649 wa 1,444 29,436 tas. -836 1,176 nt 728 2,111 act 701 3,090 australian population studies 2018 | volume 2 | issue 2 | pages 12-21 © parr 2018. published under the creative commons attribution-noncommercial licence 3.0 australia (cc by-nc 3.0 au). journal website: www.australianpopulationstudies.org commentary what sort of population policy should australia adopt? suggestions for migration, fertility and population research policy nick parr* macquarie university * corresponding author. email: nick.parr@mq.edu.au. address: centre for workforce futures, faculty of business and economics, macquarie university, nsw 2109, australia. paper received 15 august 2018; accepted 27 october 2018; published 12 november 2018 1. introduction: why have a population policy? the implications of population change are wide-ranging and long-lasting. population growth, geographical distribution and compositional change affect the need for all sorts of goods and services, including housing, transport, childcare, education, health, water and energy; labour supply, income, wealth, tax revenue, and government expenditure; land use, living space, the natural environment and the effects of climate; time use; family life; national culture and identity; inequity and inequality; crime and security; and religion (e.g. argyrous et al. 2017; daley and coates 2018; hugo 2012; infrastructure australia 2018; mcdonald 2011; mcdonald and temple 2010; ong et al. 2017; pew research center 2017; piketty 2014; productivity commission 2015, 2016; treasury and home affairs 2018). the long run effect of immigration on the size and composition of australia’s population is more than just a matter of the initial ‘plus one’ per immigrant; it continues over the lifetimes of the immigrants who remain here, is added to by births to migrant parents in australia, and, over the longer run, will also reflect the survivorship, reproduction and emigration of second and higher order generation descendants (schmertmann 1992). similarly, changes to numbers of children born in any one year will affect subsequent numbers of (potential) parents, and, hence, the sizes of subsequent generations across future time. in view of breadth and duration of its various effects, the potential value of australia implementing (and progressively refining over time) an integrated national policy to direct, anticipate, monitor and respond to population change (i.e. a ‘population policy’) would appear to be considerable. 2. the scope of population policy population policy should be simultaneously broad in its range of considerations and focused in terms of its priorities. the considerations include: • the future prospects and scope for change through policy of international migration, internal movement within australia, fertility, and mortality, and the differing implications of alternative combinations of these variables for the future size and composition of national and subnational http://www.australianpopulationstudies.org/ mailto:nick.parr@mq.edu.au australian population studies 2 (2) 2018 parr 13 populations. trajectories of population change over long periods of time should be considered, and not just projected sizes for one future point in time (as occurred during the ‘big australia’ debate). this is because population change and its consequences vary over time, and because population change over any one time period will have ramifications for subsequent population change (treasury 2015; treasury and home affairs 2018; parr and guest 2014). • the implications, both more immediate and sustained over longer periods of time, of such population outcomes for key non-population outcomes, such as labour force size, participation rates and composition, income levels, housing (numbers, types and affordability), road traffic and congestion, public transport use, educational enrolments, university funding, health care need, water use and environmental degradation. • differences in the implications of population change between different geographical locations and for different population subgroups, especially disadvantaged subgroups. • compassionate and emotional grounds for particular types of immigration (i.e. humanitarian and family). • the implications of australian population change and policy for populations overseas, especially for those that are poor, endangered or persecuted. • the implications of population change for biodiversity and environmental sustainability. • population-related budgetary, planning, social, education, family, regional development, foreign aid, health, environmental and other policies. • time lags, degrees of uncertainty, potential for error, and contingency plans. • gaps in knowledge and understanding in all of the above, and how best to address them through coordinated and funded research and through development of the research community. • ministerial and departmental responsibility for the administration of population policy, and the coordination of policy formulation and implementation between australia’s major jurisdictions. for example, whether there should be a federal minister for population, and, if so, whether the portfolio will include responsibility for immigration policy, and how consultation with state and territory governments will be conducted. • the population-related implications and coherence of policy which is made primarily with considerations other than population to the fore but which also has population-related implications. • how accountability will be ensured. for example, whether political parties will provide clear indications of intended population-related direction before elections, and will adhere to such promises if elected. the rest of this brief article focuses on selected aspects of three key elements of population policy: immigration policy, fertility policy, and research policy. 14 parr australian population studies 2 (2) 2018 3. migration policy by international standards australia’s immigration is highly planned and managed (hugo 2014). current migration policy covers movement on a range of temporary visas (including student, temporary work-related, working holiday maker, and visitor), as well as eligibility for permanent visas (which is almost entirely on the basis of skill, family and humanitarian grounds), and moves in and out of australia by australian and new zealand citizens and some others (hugo 2014a, 2014b; doha 2018). not least because it is the component of population change which is the most readily changed through policy, migration policy, especially policy on the annual numbers of permanent residence visas issued, should be the cornerstone of any australian population policy. the long-run impact on australia’s population of visa grants will reflect the length of stay and longevity of the immigrants, the children they have after moving here, and the reproduction, emigration and life expectancies of their descendants in australia. thus whilst changes to temporary visa entries can have large shorterterm impacts on population, over the long run the numbers of permanent residence visas (including those issued to previous temporary visa holders) will have a much greater bearing on population growth than the number of temporary visas (abs 2018a). migration policy should align with (relatively) acceptable and feasible trajectories for future population change. it should recognise that in the more immediate future some degree of population growth and population ageing is virtually inevitable, both nationally and for australia’s largest cities, avoid zero (or negative) growth or zero ageing ‘population pipedreams’, and consider longer-run sustainability issues, as well as the more immediate implications of population change (abs 2013; martin et al. 2017; parr et al. 2016). there are both advantages and disadvantages to higher levels of immigration. higher levels of immigration, especially suitably-targeted skilled immigration, can help to increase living standards (mcdonald and temple 2010; parr and guest 2014; productivity commission 2016). whilst their perperson economic benefits tend to be smaller than those of skilled migration, there are compelling compassionate and emotional justifications for a humanitarian component and for most of the current components of australia’s family immigration stream (treasury and home affairs 2018). higher immigration and, hence, population growth, particularly in sydney and melbourne, has also been linked to ‘population problems’, including traffic congestion, housing affordability issues, localised school capacity shortage, and concerns related to the spread of higher density housing (daley and coates 2018, infrastructure australia 2018, ong et al. 2017). the economic consequences of a change to immigration numbers will depend on how that change is targeted. substantial proportions of particular occupational workforces (including accountants, and various ict professional and health professional occupations) are the product of australia’s skilled migration program. however, there are also smaller numbers of former skilled migrants working in less-skilled occupations (for example carers and aides, cleaners and laundry workers, numerical clerks, factory process workers and sales assistants and salespersons) which have never been on the lists of eligible occupations for skilled stream visas (albeit somewhat smaller percentages for primary applicants than for secondary applicants) (abs 2018b; birrell 2018; de alwis and parr 2018). it would be preferable for any reduction to the size of the skilled stream to disproportionately reduce the proportion of migrants who do not work in the stream’s eligible occupations post visa award. australian population studies 2 (2) 2018 parr 15 recent improvements in data availability, including linked visa application and census datasets, have strengthened the evidence base for assessment of the labour market outcomes of skilled stream entrants and their variation between demographic groups and geographical areas (abs 2018b). the provision of data at a finer level of detail (especially for the occupations for which the skills assessments were undertaken), facilitation of further data linkage initiatives (for example by recording tax file numbers with other visa data), and empirical analysis of the resulting data may help to inform increases in the efficiency with which skilled stream entrants fill shortages in areas of need and contribute to taxation revenue, and reduction in numbers of skilled migrants not working in skilled migration eligible occupations. the inclusion of a question measuring time-related underemployment on the census could facilitate identification of labour underutilisation at the detailed level of classification of occupations used in skilled migrant selection, and hence inform the targeting of reduction to migrant numbers. as well as its targeting, the timing of changes to total skilled stream visa numbers and the integration of migration policy with education and training policy are also important considerations. education and training is a critically important determinant of labour supply, especially for professional and other high skill occupations, such as those targeted by the skilled migration stream. in order to reduce the risk of oversupply and related graduate and immigrant skills wastage, it is important the setting of quota sizes for the occupations which are targeted by skilled migration is responsive to changes in domestic graduate supply, as well as to demand, productivity and workforce attrition trends in these occupations. however, in view of the inevitable time lags between policy initiation and change to graduate numbers, a phased approach to migrant number reduction, which is harmonised increases in graduate numbers, would be advisable. population policy should incorporate contingency plans for responses to rapid, unanticipated increases in need for critically important occupations with long education and training supply pipelines. flexible plans for increases or decreases in suitably-skilled immigrant numbers in response to sudden changes in demand for mining and other industries which operate in relatively sparsely-populated regions which lack locallyavailable labour and face volatile demand should also be formulated. the recent growth in numbers of ‘domestic’ australian higher education graduates, particularly into occupational workforces which currently rely heavily on skilled immigration, such as health, engineering and related technologies, and information technology, provides reason for reassessment of the related skilled migration occupation quotas (birrell 2018; crettenden et al. 2014; det 2018). proactive, suitably-calibrated, expansion of domestic graduate numbers, such as that recently implemented as a part of workforce planning for doctors and nurses, could potentially reduce risks of future labour shortage and thereby facilitate larger reductions to the size of the skilled migration intake (crettenden et al. 2014). however, in view of the inevitable time lags between policy initiation and change to graduate numbers, a phased approach to migrant number reduction, which is harmonised with expected future increases in graduate numbers, would be advisable. the 26% increase in australia’s annual number of births between 2001 and 2012 contributed to past increases in demand-related pressures on child care and to more recent school enrolment growth and localised school capacity shortage (abs 2014; goss 2016; guest and parr 2013; mcdonald 2011; productivity commission 2015). the prospective entry into the australian labour force of these 16 parr australian population studies 2 (2) 2018 significantly larger cohorts may contribute to supply surpluses for particular occupations in the absence of responsive changes to skilled migration numbers. the larger sizes of these cohorts could also provide a significant opportunity for strategically-targeted education and training to reduce australia’s future reliance on skilled immigration. of note is that the entry of the larger post-2006 birth cohorts into the australian labour force will roughly coincide with the baby boomer generations reaching older ages with high per capita health and aged care needs, and therefore with increased need for health care sector and aged care sector workers. slowing population growth in australia’s largest cities through a wider dispersal of new immigrant settlement should be a priority policy goal. despite australia’s large land mass and low population density, the concentration of population growth in major cities in general, and sydney, melbourne and brisbane in particular, increased between 2007 and 2017 (abs 2018c; hugo et al. 2015). in 201617 melbourne had a growth rate of 2.7% and sydney and brisbane both of 2.0%, and in combination these three cities accounted for over 70% of australia's population growth (abs 2018c). projections of the populations of sydney and melbourne reaching 8 million by 2061 appear conservative in the light of subsequent trends (abs 2013). it is the projected population sizes of the largest cities, rather than australia’s projected national population size, which i see as the more challenging prospect. a wider geographical dispersal of immigrant settlement and of population more broadly could in theory reduce the extent of transport, housing, infrastructure provision and space-related population problems in the major cities. that said, despite evidence of the improvement in immigrant labour market outcomes in regional and rural australia, with a few notable exceptions (for example byron bay, cairns, darwin, griffith, kalgoorlie, pilbara, shepparton, whitsunday), the immigrant share of population in areas which are distant from the larger cities has generally remained low (abs 2017a; hugo 2015; krivokapic-skoko and collins 2014; massey and parr 2012). the prospective geographical pattern of population ageing may assist dispersal of immigrant settlement beyond the largest cities to a degree. the working age populations of australia’s larger cities are generally younger than those in regional australia. the occupations in which larger numbers of recent skilled stream primary applicants work include accountants, nurses, certain ictrelated occupations and medical practitioners (abs 2018b). in these occupations too the percentages in the later working ages (i.e. 55 and over) are higher for regional australia than for the major cities, and higher for tasmania and generally (but usually less so) for south australia than for the other states and territories (abs 2017a). thus age-related retirement linked to the ageing of numericallylarger ‘baby boom’ cohorts, and the related need to recruit replacements, may increase the proportions of job vacancies in skilled stream eligible occupations outside sydney, melbourne, brisbane and perth. australia’s geography of population ageing will have workforce demand-side implications, as well as workforce supply-side implications, and has the potential to exacerbate health inequity. the ageing of larger ‘baby boom’ cohorts to progressively older and higher health care-using ages will necessitate a substantial expansion of the health workforce (treasury 2015; crettenden et al. 2014). older populations in regional australia, tasmania and south australia face the prospect of ‘doublewhammy’ demand-side and supply-side effects increasing the need for recruitment of immigrant (and local) health professionals beyond the larger cities. the failure of attempts to encourage australian population studies 2 (2) 2018 parr 17 australian medical practitioners to relocate from the cities makes the need for a successful migration-based solution to the prospective doctor shortage all the more pressing (carson et al. 2017). as searle (2018) argues, the most feasible approach to reducing the concentration of population growth in the largest cities involves trying to increase the share of growth of the next tier of cities by population size, such as adelaide, canberra, darwin, hobart, toowoomba, townsville, and (if reclassified to become part of ‘regional australia’ for visa purposes) newcastle. the success of ‘bullet trains’ in dispersing population growth from larger to smaller cities along rail corridors in china points to a potential population-related advantage of an australian high speed rail initiative (zheng and kahn 2014). promoting overseas student study at universities outside the largest cities and providing financial incentives for migrants (especially health professionals) to settle in regional areas of australia may help to disperse migrant settlement a little (tang et al. 2014). in terms of numbers of skilled migrant primary applicants it is the various ict-related occupations which are the most heavily skewed towards sydney and melbourne. thus reductions in quota sizes for these occupations might be expected have the greatest immediate impact in terms of reducing the concentration of numbers in the two largest cities (abs 2018b). in view of limited proven options, a coordinated and funded program of research on the experiences of immigrant settlement and retention in regional and rural australia and comparable countries overseas with a view to identifying a model for achieving a larger-scale dispersion of population is essential (collins et al. 2018; krivokapic-skoko and collins 2014; santoro and wilkinson 2015; taylor 2018). 4. fertility policy family policy and the monitoring and understanding of fertility trends should be important population policy considerations, but not for the ‘pronatalist’ reasons presented by the former howard government (parr and guest 2011). australia’s total fertility rate (tfr) has decreased since 2008 (abs 2014; 2017b). however, contrary to some the belief of some, sustained fertility at replacement level is not necessary to prevent long-run population decline. indeed, under continuation of its average tfr, immigration and life expectancy for 2011-15 over a very long time australia’s population would increase to over 130 million (parr 2018). whilst higher fertility would slow the extent of future population ageing, it would adversely affect the dependency ratio both immediately and over the longer run. higher birth rates may also adversely affect educational provision and attainment and will add to population growth (parr 2006). moreover, the feasibility of substantially increasing birth rates is questionable: the available evidence suggests the effects on birth rates of the former howard government’s introduction of the ‘baby bonus’ (and other simultaneous budgetary changes) most probably was slight (parr and guest 2011). nonetheless, whether, how many, and when to have children are life-changing decisions. the lifetime work, other financial, other time-related, and happiness-related effects of children on parents (and other family members), the lifetime effects of parents (and other family members) on child development, and the ways in which family policies, such as family benefits, child care policy, and leave-related policies, economic policies, social policies, and education policies affect them are fundamentally important issues (craig et al. 2014; guest and parr 2013; luppi and mencarini 2018; markey et al. 2015; myrskyla and margolis 2014; parr 2006; productivity commission 2015). it is also 18 parr australian population studies 2 (2) 2018 important that housing supply facilitates, rather than constrains, attainment of preferred family sizes. for all these reasons, it is supporting research to understand the drivers and family-level implications of fertility patterns, and developing evidence-based policy to enhance family wellbeing, and not birth rate change, which should be the primary fertility policy concern of an australian population strategy (mcdonald 2006). 5. conclusion: the need for population research policy in theory the policy suggestions in this article could: • help to put australia (and especially its larger cities) on a less unpalatable trajectory of population change • aid reduction (which i presume would be the more popular preference) of immigration intake and hence population growth with a view to minimising its economic cost and maximising its mitigation of population-related problems in our major cities • accentuate the value of strategically educating and training younger generations of australians, and enhance their future integration into the labour force • aid the geographical alignment of australia’s national workforce with its ageing population, and • promote the enhancement of family life in australia. however, the realisation of such outcomes will depend, at least in part, on a broad range of related research being undertaken. three key priority areas for research are: 1) forecasting fertility, mortality, trans-tasman migration, australian citizen return, international student movement, emigration, and (most importantly of all) subnational population change 2) analyses of the relationships between immigration selection criteria and immigrant labour market outcomes, workforce ageing, retirement trends, trends in education and training, and work, housing and family combination, and 3) developing experimental ideas for and analysis of immigrant settlement experience and retention outside australia’s largest cities. ensuring that the requisite research is undertaken will require establishment of a coordinated, funded program of population research and a systematic expansion of australia’s pool of suitablytrained researchers. currently, what little australian population-related research is conducted is a matter of the whim of a relatively small and ageing pool of researchers who compete against researchers from other disciplines for scarce research funding, and whose jobs are a by-product of staffing decisions which are not necessarily taken with a view to the broader national interest. the coordination and funding of the development of a next generation of demographers and other population researchers through dedicated scholarships for postgraduate and postdoctoral research is also critically important if a research-informed population policy for australia is to be sustained over time. australian population studies 2 (2) 2018 parr 19 6. key messages • in view of the wide-ranging and long-lasting implications of population change, australia should adopt a population policy. • changing the total number of permanent migration visas issued is the key policy lever though which the federal government may influence population trends. • the implementation of changes to skilled migration numbers should be phased over time and integrated with changes to education and training policy. • the growth of the population sizes of the largest cities presents a more challenging prospect than national population growth. a wider dispersal of immigrant settlement should be a priority population policy goal. • population ageing could create opportunities for achieving a wider dispersal of migrant settlement. • enhancing family wellbeing, and not birth rate change, should be the primary concern of fertility policy. references argyrous g, craig l, and rahman s (2017) the effect of a first born child on work and childcare time allocation: pre-post analysis of australian couples. social indicators research. 131(2): 831–851. australian bureau of statistics (abs) (2013) population projections, australia 2012 to 2101. catalogue number 3222.0. canberra: abs. australian bureau of statistics (abs) (2014) australian historical demographic statistics. catalogue number 3105.0.65.001. canberra: abs. australian bureau of statistics (abs) (2017a) census tablebuilder. canberra: abs. accessed 14 august 2018. australian bureau of statistics (abs) (2017b) births, australia 2017. catalogue number 3301.0. canberra: abs. australian bureau of statistics (abs) (2018a) migration, australia 2016-17. catalogue number 3412.0. canberra: abs. australian bureau of statistics (abs) (2018b) australian census and migrants integrated dataset, 2016 catalogue number 3417.0.55.001. canberra: abs. australian bureau of statistics (abs) (2018c) regional population growth, australia 2016-17. catalogue number 3218.0. canberra: abs. birrell b (2018) australia’s skilled migration program: scarce skills not required. australian population research institute research report: apri: melbourne. http://tapri.org.au/wpcontent/uploads/2016/04/final-march-8-australias-skilled-migration-program.pdf carson d, punshon k, mcgrail m, and kippen r (2017) comparing rural and regional migration patterns of australian medical general practitioners with other professions: implications for rural workforce strategies. australian population studies 1(1): 55-68. collins j, reid c, groutsis d, ozkul d, and watson k. (2018) syrian-conflict refugee settlement in australia. public seminar on responses to syrian-conflict refugee settlement in australia, germany, sweden, finland, uk, canada and new zealand, 12-14 march 2018. craig l, powell a, and smyth c (2014) towards intensive parenting? changes in the composition and determinants of mothers’ and fathers’ time with children 1992–2006. british journal of sociology 65(3): 555-579. http://tapri.org.au/wp-content/uploads/2016/04/final-march-8-australias-skilled-migration-program.pdf http://tapri.org.au/wp-content/uploads/2016/04/final-march-8-australias-skilled-migration-program.pdf 20 parr australian population studies 2 (2) 2018 crettenden if, mccarthy mv, fenech bj, heywood t, taitz mc, and tudman s (2014) how evidence-based workforce planning is informing policy development in the retention and distribution of the health workforce. human resources for health 12(7): 1-13. daley j and coates b (2018) housing affordability: reimagining the australian dream. melbourne: grattan institute. de alwis s and parr n (2018) differences in occupation of ancestry subgroups of asian birthplace groups in australia. australian journal of social issues 53(2): 173-202. department of education and training (det) (2018) higher education statistics, 2017 www.education.gov.au/higher-education-statistics accessed on 10 august 2018. department of home affairs (doha) (2018) moving to australia. www.homeaffairs.gov.au/trav/ente/movi accessed on 14 august 2018. department of immigration and border protection (dibp) (2017) 2016-17 migration programme report. australian government. https://www.homeaffairs.gov.au/reportsandpublications/documents/statistics/report-onmigration-program-2016-17.pdf. accessed on 13 august 2018. goss p (2016) should you worry about a schools shortage? it really depends on where you live. the conversation. january 22 2016. guest r and parr n (2013) family policy and couples’ labour supply: an empirical assessment. journal of population economics 26(4):1631-1660. hugo g (2012) population distribution, migration and climate change in australia: an exploration. accarnsi discussion paper. national climate change adaptation research facility. https://www.nccarf.edu.au/settlements-infrastructure/content/population-distributionmigration-and-climate-change-australia-exploration-graeme-hugo. accessed on 13 august 2018. hugo g (2014a) change and continuity in australian migration policy. international migration review 48(3): 868-890. hugo g (2014b) skilled migration in australia: policy and practice. asian and pacific migration review 23(4): 375-396. hugo g, feist h, tan g, and harris k (2015) population dynamics in regional australia. barton: regional australia institute. infrastructure australia (2018) future cities: planning for our growing population. http://infrastructureaustralia.gov.au/policy-publications/publications/future-cities.aspx. accessed on 8 august 2018. krivokapic-skoko b and collins j (2014) looking for rural idyll ‘down under’: international immigrants in rural australia. international migration 54(1): 167-179. luppi f and mencarini l (2018) parents’ subjective well-being after their first child and declining fertility expectations. demographic research 39(9): 285-314. markey r, ferris s, mcivor j, thornthwaite l, wright cf, muhidin s, o’neill s, and parr n (2016) extending employment entitlements to non-standard workers: alternative models for long service leave portability. journal of industrial relations 58(1): 46-68. martin p, ward j, and sutton p (2017) why a population of, say, 15 million makes sense for australia. the conversation. 22 july 2017. massey sjl and parr n (2012) the socioeconomic status of migrant populations in regional and rural australia and its implications for future population policy. journal of population research 29(1): 1-21. mcdonald p (2011) the determinants of australia’s future demography. chapter 2 of a ‘sustainable’ population? — key policy issues. canberra: productivity commission. http://www.education.gov.au/higher-education-statistics http://www.homeaffairs.gov.au/trav/ente/movi https://www.homeaffairs.gov.au/reportsandpublications/documents/statistics/report-on-migration-program-2016-17.pdf https://www.homeaffairs.gov.au/reportsandpublications/documents/statistics/report-on-migration-program-2016-17.pdf https://www.nccarf.edu.au/settlements-infrastructure/content/population-distribution-migration-and-climate-change-australia-exploration-graeme-hugo https://www.nccarf.edu.au/settlements-infrastructure/content/population-distribution-migration-and-climate-change-australia-exploration-graeme-hugo http://infrastructureaustralia.gov.au/policy-publications/publications/future-cities.aspx australian population studies 2 (2) 2018 parr 21 mcdonald p and temple j (2010) immigration, labour supply and per capita gross domestic product: australia 2010‒2050. canberra: dima. myrskyla m and margolis r (2014) happiness: before and after the kids. demography 51(1): 1843-1866. ong r, dalton t, gurran n, phelps c, rowley s, and wood g (2017) housing supply responsiveness in australia: distribution, drivers and institutional settings. ahuri final report no. 281. melbourne: australian housing and urban research institute. parr n (2006) do children from small families do better? journal of population research 23(1):1-25. parr n (2012) trends in differentials in the workforce participation of mothers with young children in australia 2002-08. journal of population research 29(3): 203-227. parr n (2015) population growth could go far beyond igr projections. the conversation. 10 march 2015. parr n (2018) redefining replacement: very long-run implications of the current fertility, migration and mortality levels of ten more developed countries. presentation to the european association of population studies conference in brussels, belgium, 6-9 june 2018. parr n and guest r (2011) the contribution of increases in family benefits to australia’s early 21st century fertility increase: an empirical analysis. demographic research 25(6): 214-244. parr n and guest r (2014) a method for socially evaluating the effects of long-run demographic paths on living standards. demographic research 31(11): 275-318. parr n, li j, and tickle l (2016) a cost of living longer: projections of the effects of prospective mortality improvement on economic support ratios for fourteen more advanced economies. population studies 70(2): 181-200. pew research center (2017) the changing global religious landscape. www.pewforum.org/2017/04/05/the-changing-global-religious-landscape/ accessed 13 august 2018. piketty t (2014) capital in the 21st century. cambridge: harvard university press. productivity commission (2015) childcare and early childhood learning. melbourne: australian government. productivity commission (2016) migrant intake into australia. melbourne: australian government. santoro n and wilkinson j (2016) sudanese young people building capital in rural australia: the role of mothers and community. ethnography and education 11(1):107-120. searle g (2018) making small cities bigger will help better distribute australia’s 25 million people. the conversation. august 8 2018. schmertmann cp (1992) immigrants’ ages and the structure of stationary populations with belowreplacement fertility. demography 29(4): 595‒612. tang azr, rowe f, corcoran j, and sigler t (2014) where are the overseas graduates staying on? overseas graduate migration and rural attachment in australia. applied geography 53(september): 66-76. taylor a (2018) heading north, staying north? the increasing importance of international migrants to northern and remote australia. migration and border policy working paper no 7. lowy institute. https://www.lowyinstitute.org/publications/heading-north-staying-north-increasing-importanceinternational-migrants-northern-and. accessed 13 august 2018. the treasury (2015) 2015 intergenerational report: australia in 2055. https://treasury.gov.au/publication/2015-intergenerational-report/. accessed on 15 august 2018. the treasury and home affairs (2018) shaping a nation: population growth and immigration over time. https://cdn.tspace.gov.au/uploads/sites/107/2018/04/shaping-a-nation-1.pdf. accessed on 15 august 2018. zeng s and kahn me (2013) china’s bullet trains facilitate market integration and mitigate the cost of megacity growth. pnas: proceedings of the national academy of sciences 110(14): e1248-e1253. http://www.pewforum.org/2017/04/05/the-changing-global-religious-landscape/ https://www.lowyinstitute.org/publications/heading-north-staying-north-increasing-importance-international-migrants-northern-and https://www.lowyinstitute.org/publications/heading-north-staying-north-increasing-importance-international-migrants-northern-and https://treasury.gov.au/publication/2015-intergenerational-report/ https://cdn.tspace.gov.au/uploads/sites/107/2018/04/shaping-a-nation-1.pdf 4. fertility policy 6. key messages references australian population studies 2019 | volume 3 | issue 1 | pages 1-12 © parr 2019. published under the creative commons attribution-noncommercial licence 3.0 australia (cc by-nc 3.0 au). journal website: www.australianpopulationstudies.org understanding the motivations for return migration in australia angélique parr* the university of queensland * email: angelique.parr@uqconnect.edu.au. address: po box 77, drummoyne, nsw 1470 paper received 15 february 2019; accepted 19 may 2019; published 27 may 2019 abstract background return migration is often overlooked by traditional analyses of internal migration. why people return has received even less scrutiny. relatively few migrants make a return move, so there is clearly something noteworthy about these people and their circumstances that trigger such a move. aims this paper explores why people make return moves in australia. data and methods migration histories were collected using semi-structured in-depth interviews; content analysis of interview transcripts was undertaken. results people return for a wider range of reasons than is indicated by neoclassical economic theory. some of the moves are linked to significant life events such as post-school education and employment. returns also occur for less tangible factors such as amenity and climate, connections to family, friends and the extent to which a place “feels like home” are equally important. conclusions a broader explanatory framework is required to explain why people return. the integration of migrant stories into more traditional migration analyses enriches the story of internal migration. key words internal migration; return migration; motivations; australia. http://www.australianpopulationstudies.org/ mailto:angelique.parr@uqconnect.edu.au 2 parr australian population studies 3 (1) 2019 1. introduction australia’s internal migration system is complex, comprising many types of moves over differing spatial and temporal scales. one type of move that is usually overlooked by traditional analyses is that of return migration. it is the event where people who moved to a new place of residence return to live at their former home or region at a later date (bell and hugo 2000). in some contexts it can be better conceptualised as part of circular migration rather than a discrete form of migration in its own right. return migration acts as a counterbalance to the dominant direction of migration flows between areas, and previous research has established that return migrants are a distinctly different group compared with other migrants in terms of both their demographic character and motives for moving (da vanzo and morrison 1981; long 1988; newbold and liaw 1990). our understanding of return migration is bound by the data used to observe it. in australia this is the census, and returns to a previous place of residence are observed over the five-year intercensal period. but returns may occur after an absence almost as long as a life span. while the use of geo-political boundaries can obscure its true spatial extent, migration must cover a sufficiently large distance to change the activity space within which a person functions on a daily basis (boyle et al. 1998; parr 2018). five dimensions (intensity, who moves, where, when and why) are recognised as fundamental perspectives in migration studies (bell 1992; long 1988), but one dimension, ‘why people move’, has received far less scrutiny than the other four. given that relatively few migrants make a return move (bell 1996a; parr 2018), there is clearly something noteworthy about such individuals and the specifics of their circumstances that trigger such a move. drawing on the results of a recently completed analysis of return migration in australia (parr 2018), this paper begins to fill out the missing “why” in the return migration story. the paper begins with a summary of recent research on the motivations for return migration. it then outlines the explanatory framework, the dataset and methods used in the analysis. the body of the paper recounts the stories of a group of australian residents who have made return moves. the paper concludes with a brief discussion of the implications of the findings for both migration theory and methods. 2. background substantial insight into the motivations for return has derived from studies of international migration, particularly the movement of labour migrants (lidgard and gilson 2002; reagan and olsen 2000; stalker 1994). comparatively little has been written in relation to internal migration, although the reasons for international and internal returns appear to have many similarities. king (1996) found there are many reasons, which change over the life course. a swedish study looking at people who returned to rural areas where they grew up is the most comprehensive so far (amcoff and niedomysl 2013; niedomysl and amcoff 2010). return migrants were more likely to report social reasons for moving than reasons related to employment or education. moving close to friends was important for divorcees and people returning to sparsely populated areas. studies of the impact of out-migration from rural scotland found that people return for a range of personal and employment-related reasons (stockdale 2002, 2006). other work reveals a multiplicity of motivations to return, with social connections a major factor (alexander 2005; bijker et al. 2012; stockdale 2006; wang and fan 2006). australian population studies 3 (1) 2019 parr 3 an australian study of young people who returned to tasmania noted the importance of lifestyle and being close to family (easthope 2006). having a sense of community and belonging, or ‘a place to call home’ appears to influence the decision to return (stockdale 2002). return moves may be linked to significant life events. in this paper such events are categorised into four types: economic, education, family-related and ageing. but if life events alone are responsible, the probability of returning would be considerably higher than is observed in australia (parr 2018). the reasons for return must therefore be much broader in scope than those offered by the traditional neoclassical framework, which has been criticised for not adequately explaining the reasons for internal migration (long 1991). they may range from macro-scale social, political and economic forces to micro-scale individual behaviour and preferences. the emphasis here is to gain an understanding of the motivations of the individual migrant, which can change over the life course (jeffery and murison 2011; king 1996; lundholm et al. 2004). this study draws upon four explanatory frameworks to understand the motivations for return. a life course perspective has been combined with lee’s push–pull model (lee 1966), wolpert’s (1965) place utility model and faist’s (2000) theory of meso-scale factors to argue that significant life events, in combination with other factors, will result in return migration. people tend to migrate to familiar places (boyle et al. 1998 p. 64) and the personal ties, familial ties and social networks that connect a migrant in some way to a place where they have lived in the past will have an additional strong pull (mulder and malmberg 2014). this paper proposes that such meso-scale considerations are just as important as other factors in explaining the selective nature of return migration. they create connections or moorings that act more strongly to draw a migrant back to a previous place of residence than other factors that would have them move to a new location. 3. data and methods motivational factors associated with internal migration are seldom captured in traditional data sources (stillwell and garcia coll 2000). in particular, bell (1995) pointed to the lack of information about why people move as a major deficiency of australia’s census. most previous studies of return migration have inferred the reasons from migrant characteristics (bell 1996b; currie and hallie 1988; long 1988) and it is only recently that research has begun to look at this directly. understanding more about the motivations for return migration can only be achieved by looking elsewhere to alternate data sources. this study has taken an important step by using detailed migration histories to examine this issue directly. qualitative research methodologies that study human behaviour and the social world are used to understand why people move. this paper reports on information from the migration histories of 22 people who returned to a former place of residence from elsewhere in australia. applying a purposive sampling approach, the study participants were recruited from the 50+ registry of the ageing mind initiative at the university of queensland, and networks at the new south wales department of planning and environment. eligible participants had changed residence in australia at least twice, with one of these moves being a return to a place they had lived in sometime in the past. no constraint was placed on the type of place to which they returned, such as the participant’s place of birth or where they grew up. but any change of residence had to be of at least six months’ 4 parr australian population studies 3 (1) 2019 duration and across a sufficiently large distance to alter the spatial patterns of the participant’s dayto-day life. the size of the study group ensured that the diverse range of behaviours and underlying motivations would be revealed from the migration histories. this group was not selected as a representative sample of return migrants in australia. detailed migration histories and associated stories about people’s lives were collected using a semistructured questionnaire. participants recounted the places they had lived in from birth to the time of the interview, and the life events or other circumstances associated with any moves. as an aide memoire, a life grid (bell 2005) was concurrently used during each interview to create a written record of information provided. a content analysis of the interviews revealed the migration pathways of these people, a wealth of information about the life events linked with these moves, and the underlying motivations as recounted in the migrant’s own words. information from the interviews was indexed and categorised to identify both similarities and differences. this did not follow a preset coding frame, but one that was developed while reviewing the transcripts (flowerdew and martin 2005). the study group ranged in age between 25 and 88 years and lived throughout australia, including metropolitan regions and smaller towns (see figure 1). up until the time of the interviews, the 22 migrants had made more than 230 moves in total, including 55 returns. figure 1: reference map of places where study participants lived at the time of interview source: the author australian population studies 3 (1) 2019 parr 5 4. results: reasons why people return the migration histories used in this study revealed the complex nature of human mobility and the extent to which changes of residence affected, or conversely were influenced by events, people and places. the names of all participants cited here have been changed to maintain anonymity; their age given in the text was as at the time of the interview. economic factors family and economic factors were the most frequently stated reasons for return. economic factors related mainly to employment, but employment was seldom the only reason for a return move – multiple reasons could be identified. a job often facilitated the return, rather than being the key reason for it and often explained a considerable time lag between the decision to move, and the actual return. transfers with the same employer facilitate returns. william, now 58 years old, moved frequently with his job in the armed forces to military bases scattered across australia. his employer had an expectation that staff would move about every couple of years, sometimes returning to a previous posting. natalie was one of the more mobile young people interviewed. she moved several times between melbourne and canberra, with the first return to canberra occurring after graduation. she found work using networks: ‘i had a casual job when i was studying and basically, i just walked straight back into that. so, i made sure i had that before i moved back and just moved back into the job. so, i guess it wasn’t [about going] back … unemployed and looking for a job. i went back to that casual job and then started looking for a permanent job when i got back there.’ (natalie, 33) previous research concluded that return migrants were often unemployed (morrison and da vanzo 1986). this was not evident from the interviews undertaken for this study. being unemployed may have contributed to the decision to return, but people were seldom unemployed when they moved; many had already found a job. housing was another economic factor, especially for younger participants. natalie’s recent move back to melbourne had motives other than a job transfer: ‘i had planned to get into the property market for a while. i’d wanted to buy a house and it was a bit out of my reach in canberra, whereas melbourne offered quite more options in that respect. so that was, i guess, my main motivation.’ (natalie, 33) economic necessity has forced some people to return. melanie was a young professional living in sydney who moved home in her mid-thirties because: ‘ … i was pregnant … there was no maternity leave. … there was no income going to be coming in, so … no way of supporting myself. i’m a single mum. so, it wasn’t planned [that i] move home with mum and dad at 38. that’s the last thing i wanted to do ... [but] i need their support at this point in time with babysitting and everything.’ (melanie, 41) 6 parr australian population studies 3 (1) 2019 family-related a variety of family-related reasons featured strongly in the decision to return. naomi returned to melbourne with her husband after several years in brisbane because: ‘… we were at the point in our life where we were starting to think about having kids. and we decided that it would be best to be around family … for the support network. … the only thing that would have kept us [in brisbane] would have been if my mum and dad had moved up.’ (naomi, 33) faye grew up in regional victoria and moved numerous times. her first return to ballarat was at the age of 35 with her young family, because of her ageing parents. ‘we needed to be closer to them ... they came and looked at some places [in melbourne] but they didn't want to come. they didn't want to move.’ (faye, 82) returns have also taken place because people missed family and wanted to see them more often, or as in the case of ellen, because they were homesick. ‘i came back to melbourne because i was really homesick. i didn't really adjust to sydney very well … a lot of bullying, a lot of hostility, it was quite bizarre.’ (ellen, 48) education the third dimension, education, was a contributing factor when people returned as younger adults. ellen first left melbourne aged 20 to follow a boyfriend to sydney but returned a few years later to start a writing course at deakin university. thelma left brisbane in her early twenties to attend bible college in melbourne but returned a few years later to complete her accountancy qualification. wendy returned to brisbane because of a desire to send her children to the same school that she had attended. ‘i enrolled my girls [at my old high school] when they were very young; i wanted them to get a good education, to get more attention based on their merit rather than socio-economic disadvantage.’ (wendy, 56) ageing returns related to ageing did not feature as much as expected; and returns were usually made to care for elderly parents, rather than because the return migrants themselves were aged. as well wanting to put her children through high school, wendy returned to brisbane to help care for her elderly mother. ‘i’ve got two brothers who are really busy. ... so, there’s no one really to look after mum and rather than park her in a home, which we didn’t want to do, it was about meeting her needs, and meeting my girls’ [education] needs [at the same time].’ (wendy, 58) australian population studies 3 (1) 2019 parr 7 erin was one of the few cases whose return was a lifestyle choice related to ageing, rather than a necessity. in 2003 she moved with her young family to adelaide to live closer to her father. although only in their early forties, erin and her husband were starting to think about where they wanted to be living, as they got older. they decided to settle in adelaide until ‘one day my husband came home and said, “you know what, i don’t imagine retiring down here [in adelaide]” … so he started looking for a job and six months later we moved back to sydney’. (erin, 47) the migration histories of the oldest study participants support litwak and longino’s (1987) findings that the onset of age-related disability does not result in return migration. returns had occurred some years earlier, in early retirement. the most recent moves were generally local moves into aged care facilities. ties to place a variety of factors were consistent with faist’s (2000) meso-scale explanations for return. the links to family and friends who remained behind were a particularly strong motivator. karen was a young child when her parents moved the family to dubbo, a regional city in central new south wales. but the family soon returned to sydney. ‘… [my parents] missed family and friends, and both their parents were getting on and my grandparents were getting sick, … so they wanted to go back ... they were thinking that they needed to return home to be closer with the family there.’ (karen, 46) family and friends can take pressure off the return move, providing a place to stay or some other form of support as migrants settle back in after a return move. links with other networks, particularly social ones or with previous employers also made resettling an easier process. erin and her family found their return from adelaide to sydney easier “… because we kept all the relationships going with friends and stuff. we just started up with where we left off.” (erin, 47) another strong theme to emerge was the sense that the place to which people had returned was somewhere that was very familiar and where they felt happy. such places were often where people had grown up. thinking about the places she had lived other than sydney, karen remarked: ‘sydney felt like home … the surroundings [in cairns and the central coast] were so different to what i remember in sydney as a child. i think your understanding of a place starts from when you’re a child and yeah, as you get older that anchors you to the way you think and react to certain things and i think that’s – it just was different … this isn’t where i grew up. this isn’t my familiar place’ (karen, 46) having ties to a place, or a sense of identity, was a strong theme that featured repeatedly, among migrants of all ages. ‘melbourne is my home; it’s where my friends and family are’ (leslie, 58) 8 parr australian population studies 3 (1) 2019 ‘melbourne’s my home and i love my footy’ (ellen, 48) ‘sydney is always home’ (vivian, 59) ‘once a queenslander, always a queenslander’ (henry, 74) karen is also an example of someone whose ties to a place carried across the generations. although she had many good memories of living in dubbo: ‘my parents brought the family back to sydney because dubbo didn’t feel like home. they never felt at home there. … so yeah, back to where they felt was home. … the job promotion [in dubbo] was good, young children, cheaper housing, all those things, but they just felt they wanted to come back home to the bigger city.’ (karen, 46) amenity another strong consideration was amenity. natalie’s frequent returns to canberra were not just about work or education: ‘i really liked the lifestyle in canberra. i was living quite far out of the [central] city in melbourne, so i was really feeling the distance and the lack of convenience which is not an issue at all in canberra. … i guess i’m not much of a city person so [canberra]’s got a country atmosphere but with all the advantages of living in a city … if you do feel the need for a bit of city contact, you’ve got sydney just up the road as well. so, i feel like canberra’s very wellplaced in a lot of respects … it had everything i could want … and i actually liked the weather as well, despite what everyone says about it.’ (natalie, 33) sharon (56) moved between melbourne and darwin on several occasions, for a series of short-term and contract jobs. two reasons had her looking back to these two cities time and again. the melbourne seaside suburb in which she lived had a comfortable familiarity, a sense of being home. many locals also had similar values to her. the other factor, common to both darwin and melbourne, was her affinity with the natural environment. she loved being near the ocean and living in neighbourhoods with lots of palm trees. on another occasion, sharon had lived in canberra but soon returned to melbourne because “… two to three years was enough. i was disoriented as i was too far away from the sea.” the climate of a location was mentioned by other participants as an influence on the decision to return. for example, william and his wife were happy to have a return posting to darwin as they liked the city’s climate. but climate could also push people away, encouraging a return to a previous place of residence. in 2009 leanne and her husband moved from perth to brisbane to be with their granddaughter but left two years later because “it rained all the time … so we were glad to come back to perth. the weather played a huge part on our decision to come back” (leanne, 55). another reason why karen’s parents brought their young family back to sydney from dubbo was that they did not like the temperature extremes. australian population studies 3 (1) 2019 parr 9 ‘summer, dad used to say was stinking hot and the winter freezing, and they remembered the toilet would block up, it would freeze.’ (karen, 46) other factors besides climate, a desire for a change of pace can underpin a return move. thelma spent more than 20 years living in sydney and was finally able to return home to brisbane with her family: ‘i married a sydney fellow. i produced three sydney children. and there was no way, even though i would have loved to come back to queensland, there was no way i would have initiated that. i didn't feel i had the right, to do that. but my husband was a service technician and he was driving all over sydney. it drove him barmy and he wanted to get out.’ (thelma, 72) for her part, karen was very pleased to return to sydney: ‘i was very happy to move back to sydney because i grew up there and it was familiar and it was convenient and i was young and i had the culture and the food and the shops and my work friends and that whole lifestyle. it was good.’ (karen, 46). 5. discussion and conclusions the aim of this paper was to supplement the often overlooked, but now growing, body of work about the reasons for return migration. migration histories are a rich source of information concerning return moves of people living in australia. findings revealed that people return for a wider range of reasons than is indicated by neoclassical economic theory and support the suggestions from other researchers (bell and hugo 2000; cromartie and stack 1989) that a broader more inclusive conceptual framework is needed to explain the selectivity of people who return. the framework used in this paper, which combines a life course framework with macro-scale and meso-scale explanations, is a good starting point. results reveal that the reasons why people make a return move are as varied as the people themselves. some of the moves are linked to significant life events such as post-school education and employment. returns are also linked to other, less tangible, factors such as amenity and climate. the connections to family, friends and other networks and the extent to which a place “feels like home” are just as important. these less tangible factors are much harder to measure using conventional data sources and are outside the realm of the frameworks normally used to explain return migration in developed countries such as australia. the meso-scale framework developed by faist (2000) is a very useful way to explain more of the reasons for return migration revealed in the migration histories. the motivations for return have similarities with the motives for internal migration in general. work is not always a critical reason for moving in either case. social and environmental considerations are important and the reasons for moving can be complicated. some of the motivations revealed in this paper may also explain why return migrant selectivity is not always distinct. the reasons for the 10 parr australian population studies 3 (1) 2019 return may bear no connection to the socio-demographic character of the return migrant. for example, returning to care for elderly or sick family members may not be related to age, gender, occupation or income. it may have more to do with the strength of family ties and a sense of duty to care for others in the family. the histories used in this study were not intended to be representative of the people and patterns revealed by census analyses. rather, the aim was to illustrate what could be revealed about return migration in australia using alternative methods and different data sources. the collection of migration histories is always fraught with issues. difficulties with recall of the timing of events were evident with some of the older participants and post-facto rationalisation of the choices people make is a known problem. it has been suggested that people look at their past with rose-tinted glasses to justify their decisions. there is no obvious way to get around this, but these issues need to be recognised. limitations aside, the wealth of information provided by migration histories demonstrates that this is an extremely rich source of data about return migration. it also complements the precision of quantitative analysis habitually undertaken by demographers and population geographers. the value of integrating this material with the findings derived from conventional census-based analyses is strongly encouraged and broadens our understanding of australia’s internal migration system. key messages  understanding why people return to a place helps explains the patterns of movement.  the reasons why people make a return move are as varied as the people themselves.  a job is not the only reason why people return. some returns are linked to significant life events. ties to place, family and friends, the climate and environment are also important factors.  that a place “feels like home” is also an important reason why people return.  migration histories are a rich source of information about why people return. they enhance research that traditionally relies on data from australia’s census. acknowledgements this research forms part of a larger study supported by the australian research council discovery grant dp 0451399: understanding the structure of internal migration in australia. i am grateful for the helpful comments of the anonymous reviewers on an earlier draft of this paper. references alexander j t (2005) they're never here more than a year: return migration in the southern exodus 19401970. journal of social history 38(3): 653-671. amcoff j and niedomysl t (2013) back to the city: internal return migration to metropolitan regions in sweden. environment and planning a 45(10): 2477-2494. bell a j (2005) 'oh yes i remember it well!' reflections on using the life grid in qualitative interviews with couples. qualitative sociology review 1(1): 51-67. australian population studies 3 (1) 2019 parr 11 bell m (1992) internal migration in australia: 1981-1986. canberra: australian government publishing service. bell m (1995) internal migration in australia 1986-91: overview report. canberra: australian government publishing service. bell m (1996a) repeat and return migration. in newton p w and bell m (eds.) population shift: mobility and change in australia. canberra: australian government publishing service; 147-164. bell m (1996b) understanding internal migration. canberra: bureau of immigration, multicultural and population research. bell m and hugo g (2000) internal migration in australia 1991-1996: overview and the overseas born. canberra: ausinfo. bijker r a, haartsen t and strijker d (2012) migration to less popular areas in the netherlands: exploring the motivations. journal of rural studies 28(4): 490-498. boyle p, halfacree k h and robinson v (1998) exploring contemporary migration. harlow, uk: addison wesley longman. cromartie j and stack c b (1989) reinterpretation of black return and nonreturn migration to the south 1975-1980. geographical review 79(3): 297-310. currie r f and halli s s (1988) mixed motivations for migration in the urban prairies: a comparative approach. social indicators research 1(5): 481-499. da vanzo j s and morrison p a (1981) return and other sequences of migration in the united states. demography 18(1): 85-101. easthope h (2006) returning to place: the return migration of young adults to tasmania. phd thesis, university of tasmania, hobart. faist t (2000) the volume and dynamics of international migration and transnational social spaces. oxford: oxford university press. flowerdew r and martin d (eds.) (2005) methods in human geography: a guide for students doing a research project. harlow, england: prentice hall. jeffery l and murison j (2011) the temporal, social, spatial and legal dimensions of return and onward migration. population, space and place 17(2): 131-139. king r (1996) a celebration of migration. research papers in geography, report number 25, university of sussex, brighton, united kingdom. lee e s (1966) a theory of migration. demography 3(1): 47-57. lidgard j and gilson c (2002) return migration of new zealanders: shuttle and circular migrants. new zealand population review 28(1): 99-128. litwak e and longino c f (1987) migration patterns among the elderly: a developmental perspective. the gerontologist 27(3): 266-272. long l (1988) migration and residential mobility in the united states. new york: russell sage foundation. long l (1991) residential mobility differences among developed countries. international regional science review 14(2): 133-147. lundholm e, garvill j, malmberg g and westin k (2004) forced or free movers? the motives, voluntariness and selectivity of interregional migration in the nordic countries. population, space and place 10(1): 59-72. morrison p a and da vanzo j s (1986) the prism of migration: dissimilarities between return and onward movers. social science quarterly 67(3): 504-516. mulder c h and malmberg g (2014) local ties and family migration. environment and planning a 46(9): 2195-2211. newbold k b and liaw k (1990) characterisation of primary, return and onward interprovincial migration in canada: overall and age-specific patterns. canadian journal of regional science 13(1): 17-34. 12 parr australian population studies 3 (1) 2019 niedomysl t and amcoff j (2011) why return migrants return: survey evidence on motives for internal migration in sweden. population space and place, 17(5):656-673. parr a (2018) from here to there and back again: a study of return migration in australia. phd thesis, the university of queensland, brisbane. reagan p r and olsen r j (2000) you can go home again: evidence from longitudinal data. demography 37(3): 339-50. stalker p (1994) the work of strangers: a survey of international labour migration. geneva: international labour office. stillwell j and garcia coll a (2000) inter-provincial migration of the spanish workforce in 1988 and 1994. regional studies 34(7): 693-711. stockdale a (2002) out migration from rural scotland: the importance of family and social networks. sociologica ruralis 42(1):41-64. stockdale a (2006) migration: pre-requisite for rural economic regeneration? journal of rural studies 22(3): 354-366. wang w w and fan c c (2006) success or failure: selectivity and reasons of return migration in sichuan and anhui, china. environment and planning a: 38(5): 939-958. wolpert j (1965) behavioural aspects of the decision to migrate. papers of the regional science association 15(1): 159-169. australian population studies 2018 | volume 2 | issue 1 | pages 39–51 © maertens and taylor 2018. published under the creative commons attribution-noncommercial licence 3.0 australia (cc by-nc 3.0 au). journal website: www.australianpopulationstudies.org improving population retention in northern australia: clues from german-born territorians anita maertens* charles darwin university andrew taylor charles darwin university * corresponding author. email: anita.maertens@cdu.edu.au. address: northern institute, charles darwin university, ellengowan drive, darwin, nt 0909 paper received 9 march 2018; accepted 4 may 2018; published 28 may 2018 abstract background population growth rates in many parts of northern australia have slowed considerably in recent years. governments are interested in identifying northern migration ‘markets’ as potential targets for a mix of marketing and policy-based approaches to improve population attraction and retention. in the northern territory (nt), german-born residents present an interesting case study. many are long-term residents (‘sticky’), highly educated, in professional jobs and say they are likely to stay. aims we profile and report on a study of german-born nt residents as one important international market for offsetting population losses. understanding factors which have contributed to the attraction and retention of this group may help to inform policies and initiatives to improve the population position of the nt and northern australia more broadly. data and methods data for the paper is sourced from the 2016 abs census of population and housing (census) and the 2017 german territorian survey (gts) conducted by charles darwin university. results german-born residents are a relatively immobile (‘sticky’) and educated population group in the nt with a high ratio of females. many of those surveyed, in particular those who had arrived as working holiday makers or tourists, exhibited little or no intention of leaving. lifestyle factors, climate and job opportunities ranked highly in decisions to stay. conclusions the study of german-born territorians holds promise for developing targetted niche migration initiatives to address skills and population deficits in the nt and northern australia. analysis of responses to the gts highlighted opportunities for recruiting skilled women and the importance of tourism as a source for labour supply and population growth. key words population retention; northern territory; northern australia; overseas migrants; german-born; german migration; female migration; labour market. http://www.australianpopulationstudies.org/ mailto:anita.maertens@cdu.edu.au 40 maertens a and taylor a australian population studies 2 (1) 2018 1. introduction northern australia, the region north of the tropic of capricorn, comprises 40 per cent of the continent’s land mass yet is home to only 5 per cent of the population. the north has been assumed by various proponents to hold untapped resources with capacity for significant economic growth and population growth to assist in underwriting national prosperity. it is the focus currently of national policies for growth and development under the developing northern australia agenda (australian government 2015). however, aspirations for the north historically have fallen short in successive ventures, and population growth rates in many parts have slowed considerably recently (figure 1). in the period 2002–2003 to 2009–2010, for example, growth was beyond 2 per cent, some 0.5 per cent higher than the rest of australia. however, by 2015–2016 population growth was estimated as in decline. while peaks and troughs have been evidenced in the past, and often associated with resource industry fairings, the current downturn is prolonged and substantial. figure 1: population growth rates for northern australia and the rest of australia, 2001–2002 to 2015–2016 source: abs cat. no. 3218.0 – regional population growth, australia (various years); note: northern australia is defined here as all of the nt and all areas north of the tropic of capricorn in queensland and western australia (australian government 2015 p. 132). complex mixes of factors, which vary between regions, have likely contributed to subdued population growth lately across northern australia. a recent report for the northern territory (nt) government (taylor and carson 2017) identified some key trends and issues impacting population change including: • major (up to 50%) absolute declines in the attraction of people from some interstate migration markets, including families with children and early career workers (especially women) • increased departures for early career women and pre-retirees, with the former accounting for 60 per cent of female resident flows to and from the nt • a long-term decline in the preference of internal migrants to gravitate to australia’s northern jurisdictions and a growing perception of high risks in doing so • the end of australia’s ‘mining boom’ • technological advances circumventing the need for some resource, construction and agricultural jobs to be located in situ australian population studies 2 (1) 2018 maertens a and taylor a 41 • structural changes to the demographic characteristics of northern populations, including workforce and general population ageing • changed labour practices for resource and other sectors with the increasing engagement of nonresident workers who are not necessarily counted in official population estimates. emergent literature on the economic demography of northern regions has highlighted similar issues affecting populations in remote and sparsely populated areas of other developed nations, such as sweden, greenland, canada and norway (see, for example, taylor 2016; carson et al. 2011). the root cause of population downturns for such places has been a loss in their capacity to attract new residents in the same numbers as during high growth periods, combined with increased departures of residents of certain age or life stages (e.g. youth in northern sweden). the loss of appeal of the north to some source markets, as evidenced particularly in the nt, has contributed to net negative migration flows for northern australia. while national migration policies help to determine the volume of international migrants settling initially in northern regions, targeted approaches need to align local ‘unique selling points’ with the mobility motivations and aspirations of identified intake markets (taylor, payer and brokensha 2015; taylor 2018). although the availability of jobs is important for population retention, the issues identified by taylor and carson (2017) highlight the increasing interplay of other factors which must be considered by governments in developing marketing and policy-based approaches to support population migration and growth. in this sense, markets may be source regions or nations or sub-populations characterised by life-stage or other segmentation (e.g. early career workers, pre-retirees or skilled migrants for specific occupations). overseas-born residents offer the potential to bolster population growth and development. while northern areas generally attract an annual share of overseas migrants broadly equivalent to their share of the national population, immigrants from some source countries appear to have a greater disposition for residency in the north than other internal or international migrants. of particular interest are migrants from source countries who continue to arrive, are present in substantial numbers and have a propensity to ‘stick’ in the north ( i.e. a longer average term of residence compared both to other migrant groups and the overall population). with a long history in the nt, german-born residents are one source of international migrants who appear to meet the criteria of relatively large numbers, continued arrivals and relative ‘stickiness’ as residents. data from the australian bureau of statistics (abs) 2016 census of population and housing (census) showed that their length of residence in the territory far exceeds most overseas-born migrant groups, with 38 per cent in 2016 having arrived between 1941 and 1980 compared with 18 per cent for other overseas-born groups. understanding the complex factors contributing to this migrant sub-group coming to and staying in the nt may provide insights for strategies to attract and retain other source markets from australia and overseas. the purpose of this paper is to report on a 2017 study of german-born territorians. in the face of nine successive years of low population growth and apparent decline in attractiveness of the nt to particular market segments, overseas-born migrants offer the potential to bolster population growth and development. our aim is to explore push–pull factors that influence population attraction and retention for this particular sub-group, with a view to informing strategies and responses for attracting other niche markets and addressing population attrition in the north. prior to reporting on the study, we first profile the german-born population of the nt through analysis of australian census data. 42 maertens a and taylor a australian population studies 2 (1) 2018 2. a profile of the german-born territorians people of german descent have a long association with the nt. their roots include early european exploration and missionary activities from the late 1800s. german lutheran missionaries were prominent in establishing and running aboriginal mission settlements in central australia. two prominent proselytisers were carl strehlow, who studied and recorded arrernte (sic. ‘aranda’) and loritja (western desert) languages in the alice springs region (kimber 2008 pp. 557–559), and friedrich wilhelm albrecht, a missionary of german–polish descent who spoke against the removal of aboriginal children from their mothers (henson 2008 pp. 8–9). the migration of german-born people to the nt continued in ensuing decades and increased post1940 as a result of dislocations associated with world war ii. the german club was established in darwin in 1967 and darwin’s famous ‘beer can regatta’ was co-founded by lutz frankenfeld, a german-born long-term resident, in 1974 (german club darwin 1987; 2002). there have also been significant flows of german-born tourists to the nt. in 2017, germany was the third ranked source market for international visitors to the nt, with 31,000 german visitors, and sixth ranked source market for visitors to australia as a whole. by 2016, there were around 1,000 german-born residents in the nt, or 0.5 per cent of the total nt population. meanwhile, 1,685 residents had at least one german-born parent and 8,734 declared german ancestry (abs 2017). in 2016, almost 70 per cent of german-born residents lived in the greater darwin area, 12 per cent in alice springs and the remainder elsewhere in the nt (tourism nt 2017). data from the 2016 census indicate that more german-born women than german-born men were living in the nt in 2016 with a ratio of 124:100. this compares to an equivalent ratio of 92:100 for the nt’s non-indigenous population as a whole and 89:100 for all european-born residents. this female dominance appears to be a recent phenomenon resulting from a plateauing in male, but not female, german migration. professionals (24%), technicians and trades workers (16%) and managers (15%) were prominent occupational classifications. german-born women had higher educational attainment than their male counterparts and other resident groups: 12 per cent of german-born women had postgraduate degrees compared to 6 per cent of their male counterparts and 7 per cent of all overseasborn nt residents. in 2016 the median age of german-born residents was 51 years. this was much higher than that for all overseas-born (38 years) and australian-born (29 years) nt residents. in terms of occupations, professionals (24%), technicians and trades workers (16%) and managers (15%) were prominent. almost 30% of all german-born women were professionals. most germanborn territorians were in relationships without children (62%), compared with all overseas-born (41%) and australian-born (32%) residents. german-born women were younger on average than their male counterparts. 3. data and methods in 2017 current and former german-born nt residents were recruited for an online survey, the german territorian survey (gts), administered by researchers at charles darwin university. the survey explored a range of demographic characteristics and pull–push factors influencing nt residency. participants were recruited via german social groups in alice springs and darwin, social media (e.g. facebook), email distribution lists and the media. complete responses were received from 135 participants. australian population studies 2 (1) 2018 maertens a and taylor a 43 the survey elicited experiences of living in the nt with a mix of multiple-choice and open-ended questions which were subsequently coded. the study was limited by small participant numbers due to the recruitment methods and online survey format described above. survey respondents tended to be younger, female and more highly educated than the nt german-born population as a whole. this was attributed in part to the survey method and higher likelihood of females participating in such studies. nevertheless, the study provides important baseline information to extend and explore key themes in association with census data. 4. results gts respondents tended to be younger, female (80%) and more highly educated (53% had a bachelor degree or above) relative to the german-born nt population as a whole. most were couples with children (47%) or couples without children (30%). this is a little at odds with the findings from the 2016 census which found that most (62%) german-born nt residents were in relationships without children, and likely reflects the recruitment method (as noted). of the 135 respondents who had migrated to the nt between 1961 and 2017, 103 (76%) were living in the territory in 2017. the geographic distribution of gts participants who still resided in the nt was consistent with the census, which recorded that two-thirds were resident in greater darwin and the remainder elsewhere in the nt in 2016. 4.1 why respondents moved to the nt the primary reason for moving to the nt for 40 per cent of respondents was employment (a job offer or work opportunities). almost 25 per cent arrived as tourists on a holiday or working holiday visa who decided to stay (figure 2). other motivations included: personal connections (13% family and friends; 10% a partner); lifestyle factors (9%); seeking an escape or adventure (5%); or study or research purposes (4%). figure 2: ‘why did you move to the nt?’ source: gts 2017. 44 maertens a and taylor a australian population studies 2 (1) 2018 reasons for coming to the nt varied considerably by gender. only 37 per cent of german-born women moved for a work-related reason, compared with 61 per cent of men. interestingly, 25 per cent of women reported that they were on a working holiday and decided to stay, compared with just 11 per cent of men. while jobs featured highly in reasons reported for migration decisions, other factors were also important, such as lifestyle (over 20% for men). in addition, one-fifth of male respondents and a quarter of female respondents cited family, friends or partners as motivations for moving to the nt. 4.2 what respondents loved most about the nt lifestyle factors received the highest score for what people loved about the nt (65%), followed by work opportunities (46%), climate (42%) and social aspects (38%) (figure 3). other responses included the attractive natural environment, low population density and cultural aspects (aboriginal culture, multicultural communities, open-minded communities). figure 3: ‘what do you love most about the nt?’ source: gts 2017. both male and female respondents noted that careers in the nt were a big drawcard for women, especially for those highly qualified for whom there were diverse career opportunities (e.g. in the arts, health and social services), good leave conditions and decent salaries. other respondents thought women might be looking for adventure in the nt with its ‘frontier’ appeal, whereas males seemed to be more work driven. 4.3 what’s not to like? climate was a major concern for many respondents, both for those still resident in the nt and for those who had left. distance from family and ageing parents were prominent reasons for exiting, with others citing a better lifestyle, work opportunities and new experiences or adventures elsewhere (figure 4, next page). the gts included a question about potential areas for improvement. cost of interstate travel was noted as the biggest problem, closely followed by the cost of housing/living (figure 5, next page). interestingly, although high costs were noted as important, this would not necessarily cause many australian population studies 2 (1) 2018 maertens a and taylor a 45 respondents to move away (figure 4). in available comment fields, over 5 per cent of respondents highlighted a need to invest in renewable energy, focus on sustainability issues and protect the environment and wildlife. figure 4: ‘why did you leave?’, ‘why would you move away?’; responses combined source: gts 2017. figure 5: ‘where do you feel the nt most needs to improve?’ source: gts 2017. note: score is a weighted average where 1 = no need to improve and 5 = extreme need to improve. 4.4 likelihood of leaving the nt when asked how likely they would be to leave the nt, 42 per cent of respondents still in the nt said they had little or no intention of leaving; 29 per cent said it was ‘somewhat likely’; and 29 per cent ‘very’ or ‘extremely’ likely. of the respondents who had arrived in the nt as working holiday makers or tourists, 84 per cent said they had ‘no intention of leaving’, and none said they were ‘extremely likely’ to leave. figure 6 (next page) presents survey results for what respondents loved most about the nt crossclassified with likelihood of leaving to identify motivations for staying or leaving. over 70 per cent of 46 maertens a and taylor a australian population studies 2 (1) 2018 those with ‘no intention of leaving’ loved the nt climate, while climate scored lowest for those ‘extremely likely’ to leave. lifestyle opportunities also received a high rating (60% or above) by all respondents, except those ‘extremely likely’ to leave. work opportunities were the biggest factor for those ‘extremely’ and ‘very likely’ to leave. in a follow-up question, respondents reported that they were more likely to migrate to other parts of australia than overseas. figure 6: ‘what do you love most about the nt?’ responses by likelihood of leaving nt, current nt resident responses source: gts 2017. 4.5 regional differences a number of regional differences were evident in the responses to some questions. climate more strongly influenced likelihood of leaving for top end residents compared to residents in central australia; proximity to asia was more of a point of attraction to those in the north and interstate travel costs an issue for those in the centre. darwin residents, in particular, saw a need to improve public transport, while climate and work opportunities scored more highly as positive factors for residents in central australia. respondents across the nt rated lifestyle and nature highly. 4.6 education and employment the gts identified that 34 per cent of german-born residents were not working in a field related to their highest educational attainment. a significant proportion of these (30%) had a postgraduate degree. some commented that they were overqualified for their current roles, positions were not in their specialised field or that it was impossible to get professions recognised or accredited in australia. respondents not in the workforce stated they were retired, full-time parents or studying. 4.7 personal relationships figure 7 shows that many female respondents (55%) were in a relationship with an australian-born partner at the time of survey. conversely, most of the male respondents in a relationship had a german-born partner (50% of all males). similar proportions of women (13%) and men (15%) stated australian population studies 2 (1) 2018 maertens a and taylor a 47 they were not in a relationship. some respondents speculated that australian partners would not be able to cope as well with ‘culture shock’, language barriers and the climate if they moved to germany together. figure 7: ‘what is the country of birth of your partner or spouse’ (if applicable)? source: gts 2017. 5. discussion 5.1 opportunities for early career women results from this study indicate that german-born nt residents are a promising sub-group for improving population growth prospects in northern communities. the relatively young age profile, education and qualifications of german-born women speak to the noted declines in arrivals and increased departures of early career female workers. although women have traditionally been viewed in (now dated) migration literature as dependants, ‘moving as wives, mothers or daughters of male migrants’ (docquier et al. 2012 p. 251), the gts demonstrates the reach and embedment of skilled female migrants into even the most remote parts of australia. highly educated female professionals in our study were able to seize appealing career opportunities, especially in the health, social care and arts and culture sectors, and embed themselves in the nt lifestyle. despite the blurring of boundaries between the male and female workforce over time, women often still seek and are recruited for jobs where they enjoy more job security and a sense of contributing to a greater good, rather than focusing on extrinsic, principally financial rewards (carson et al. 2010 p. 126). gts participants suggested that the largely resource-based territory economy did not necessarily offer jobs in fields attractive to highly qualified german men, who found more lucrative opportunities interstate or overseas. the nt’s occupational and industry mix, itself reflective of the north of australia overall, appears suited to the professional aspirations of german-born female migrants. it is likely conducive to other similar-minded international female migrant groups, and therefore an ideal market for informing population initiatives and policies for the north. 48 maertens a and taylor a australian population studies 2 (1) 2018 5.1 how does the gts compare with other studies? results from the gts were compared with a german survey on population mobility, international mobil (svr 2015), the 2006 territory mobility survey (tms) (charles darwin university 2008) of current and former nt residents and a recent study on ‘lifestyle migrants’ in northern sweden (carson, carson and eimermann 2017). compared with current and former german migrants in international mobil and all non-indigenous territorians in the tms, gts respondents’ specific migration motives stood out. while international mobil’s international german migrants moved for a diverse range of reasons including cultural motivations and paid employment, a high proportion of gts participants were previously working holiday makers and tourists. the tms recorded that over 60 per cent of respondents moved for work, comparable with international mobil findings and almost 20 per cent more than the gts. tms results, like international mobil, incorporated multiple responses, which makes the lesser stating of non-work-related factors (10% or less) for mobility decisions even more striking. while nearly 25 per cent of german-born current and former residents surveyed in the gts remained in the nt, only 8 per cent of tms respondents chose to remain in the nt after a visit. a mix of work satisfaction and lifestyle, climate and social environment factors appear to make german-born territory migrants ‘stick’ once resident in the nt. in the gts, nearly half (46%) rated work opportunities as something they loved the most about the nt, while ease of obtaining work mattered to only 8 per cent of tms respondents. while the results are not directly comparable, german-born territorians were more worried about the climate (40% versus 26%) and less concerned about remoteness in terms of distance from things (other places, shops, services) than distance from people (family, ageing parents), when compared to respondents to the tms. a recent study of predominantly german-born tourism entrepreneurs and ‘lifestyle migrants’ in remote northern sweden (carson, carson and eimermann 2017) identified comparable key migration drivers to the gts. participants fulfilled their wish for counterurban lifestyles, a better work–life balance, outdoor hobbies and escaping home countries described as ‘too crowded’: ‘a desire to look for a place with more space, fewer people and a quiet environment to enjoy experiences of solitude, peace and personal freedom’ (carson, carson and eimermann 2017 p. 11). gts respondents similarly were often ‘lifestyle migrants’ tired of the big cities who enjoyed the simplicity and low population density of the nt with nature at their doorsteps, space and opportunity for adventure. they described the nt as ‘different’, fulfilling their need for adventure and a counter-urban lifestyle. unique work opportunities and a good work–life balance were also cited as reasons for remaining. german-born territorians appeared stickier (42% had little or no intention of leaving) than respondents to the international mobil study, where around 41 per cent of germans abroad planned to return home and only a third planned to remain overseas (svr 2015 p. 52). while some gts respondents considered a move away from the nt in the future, most of these thought they would stay in australia (68%) rather than returning overseas (12%). hence, although some will inevitably leave, german migrants to the nt are contributing to ‘brain circulation’ through labour mobility (svr 2015 p. 17) and embeddedness in communities. they will continue to benefit the nt, whose strength lies in its multicultural, skilled, resilient population that is committed to local communities, invested in environmental sustainability and empowers the territory through knowledge flows and international connectivity. australian population studies 2 (1) 2018 maertens a and taylor a 49 6. conclusion population challenges for the north of australia are not limited to growth per se, but include longterm structural changes that impact its attractiveness to certain migration markets. moreover, changing technologies and workforce practices for major industries and projects are reducing requirements to locate jobs in situ (taylor and carson 2017) and demographic ageing is seeing relatively large numbers of pre-retirees leave and be replaced with more transient residents. a male dominance continues in the overall population, although this varies according to location. in this context, there are clearly no ‘silver bullet’ levers for governments or industry to turn around population growth rates. the key appears to be in making incremental improvements to population growth through targeting markets within australia and overseas that are more likely to be attracted and retained by the unique lifestyle and opportunities of the north. our study has demonstrated that german-born migrants hold promise as one such international source market and may attract more long-term residents through chain migration. many have stayed, formed relationships and started families in the nt. while the availability of jobs has no doubt played a pivotal role in enabling this, a range of other factors are identified in the study. in a 2016 radio interview, darwin’s german club president ralf scharmann said many germans he knew couldn’t imagine living anywhere else (plitzco 2016). data from the gts shows that german-born nt residents have particular attributes that define their migration patterns, including diverse reasons for arrival and specific career expectations. recruitment efforts may benefit by being focused on these attributes across population groups and job types (carson et al. 2010), as well as collaborating and exchanging knowledge with other peripheral regions across the world such as northern sweden, which may attract similar migrants. there is still ample opportunity to build on existing studies that examine the attraction and retention of migrants. the internet and social media provide powerful new tools for data gathering. gts responses highlight opportunities for recruiting skilled women, in particular, by marketing attractive careers, competitive salaries, employment conditions and career progression opportunities, and adding unique nt selling points to the mix. policymakers should also consider solutions to areas for improvement raised in the gts, such as recognition of education and qualifications and a better support system for newly arrived skilled migrants who lack existing networks. the high number of gts participants who arrived as working holiday makers and tourists and expressed little or no intention of leaving also highlights the importance of tourism as a positive market force and suggests a positive association with permanent migration. finally, as remarked by respondents across the nt, the territory has the advantage of an attractive natural environment, a sunny climate and multicultural, friendly communities. while many immigrants will eventually move interstate or return overseas, these factors will ensure that a substantial number will ‘stick’ around for the long term. key messages • german-born residents are a relatively immobile (‘sticky’), highly educated and female dominated population group in the nt. • factors contributing to the attraction and retention of the german-born market segment may be applicable to other markets: not least ‘lifestyle’, multiculturalism, employment and environment. 50 maertens a and taylor a australian population studies 2 (1) 2018 • the nt’s particular occupational and industry mix seemed well suited to the professional and lifestyle aspirations of female german migrants participating in the gts. • the high number of gts participants who arrived initially in the nt as tourists and expressed little or no intention of leaving, highlights the contribution and importance of the tourism industry to population stabilisation and growth. • this study is an ideal baseline to inform further northern initiatives and policies to progress attraction and retention prospects. acknowledgements the authors would like to thank the many current and former german-born territorians who provided valuable insights in their survey responses and helped recruit other participants with great enthusiasm. danke schön! references abs (australian bureau of statistics) (2017) 2016 census of population and housing. http://www.abs.gov.au/census. accessed on 28 august 2017. australian government (2015) our north, our future: white paper on developing northern australia. canberra: australian government. http://www.northernaustralia.gov.au/files/files/nawpfullreport.pdf. accessed on 10 december 2017. carson d a, carson d b and eimermann m (2017) international winter tourism entrepreneurs in northern sweden: understanding migration, lifestyle, and business motivations. scandinavian journal of hospitality and tourism 18(2): 183–198. https://doi.org/10.1080/15022250.2017.1339503. accessed on 9 december 2017. carson d b, coe k, zander k and garnett s (2010) does the type of job matter? employee relations 32(2): 121–137. carson d, ensign p, rasmussen r and taylor a (2011) perspectives on ‘demography at the edge’. in: carson d, rasmussen r, ensign p, huskey l and taylor a (eds) (2011) demography at the edge: remote human populations in developed nations. farnham, england: ashgate publishing; 3–20. charles darwin university (2008) territory mobility survey: preliminary results for the non-indigenous population. research brief series, charles darwin university, school for social and policy research. docquier f, marfouk a, salomone s and sekkat k (2012) are skilled women more migratory than skilled men? world development 40(2): 251–265. german club darwin (1987) deutscher klub darwin. darwin: the club. german club darwin (2002) 35 years: deutscher klub darwin: 1967–2002. darwin: the club. henson b (2008) albrecht, friedrich wilhelm and minna maria margaretha nee gevers. in: carment d, edward e, james b, maynard r, powell a and wilson h (eds) northern territory dictionary of biography. darwin: charles darwin university press; 8–9. kimber r (2008) strehlow, carl freidrich theodor. in: carment d, edward e, james b, maynard r, powell a and wilson h (eds) northern territory dictionary of biography. darwin: charles darwin university press; 557–559. plitzco a (2016) treffpunkt: deutscher klub darwin. sbs german radio. 1 april. https://www.sbs.com.au/yourlanguage/german/en/audiotrack/treffpunkt-deutscher-klubdarwin. accessed on 7 october 2017. svr gmbh (forschungsbereich beim sachverständigenrat deutscher stiftungen für integration und migration) (2015) international mobil: motive, rahmenbedingungen und folgen der ausund http://www.northernaustralia.gov.au/files/files/nawp-fullreport.pdf http://www.northernaustralia.gov.au/files/files/nawp-fullreport.pdf https://doi.org/10.1080/15022250.2017.1339503 https://www.sbs.com.au/yourlanguage/german/en/audiotrack/treffpunkt-deutscher-klub-darwin https://www.sbs.com.au/yourlanguage/german/en/audiotrack/treffpunkt-deutscher-klub-darwin australian population studies 2 (1) 2018 maertens a and taylor a 51 rückwanderung deutscher staatsbürger. studie des svr-forschungsbereichs, des bundesinstituts für bevölkerungsforschung (bib) und der universität duisburg-essen, gefördert von der stiftung mercator. https://www.svr-migration.de/publikationen/international-mobil/. accessed on 9 august 2017. taylor a (2016) introduction: settlements at the edge. in: taylor a, carson d, ensign p, huskey l, rasmussen r and saxinger g (eds) (2016) settlements at the edge: remote human settlements in developed nations. gloucester, uk: edward elgar; 3–24. taylor a (2018) heading north, staying north? the increasing importance of international migrants to northern and remote australia. migration and border policy working paper, no. 7. the lowy institute for international policy, sydney. https://www.lowyinstitute.org/sites/default/files/documents/taylor%20%20increasing%20importance%20of%20international%20migrants%20to%20northern%20and%2 0remote%20australia_web.pdf. taylor a and carson d (2017) synthesising northern territory population research: a report to the northern territory department of the chief minister. darwin: charles darwin university, northern institute. taylor a, payer h and brokensha h (2015) a demographic profile of international migrants in northern australia. research brief. issue rb06, 2015. darwin: northern institute, charles darwin university. http://www.cdu.edu.au/northern-institute/ni-research-briefs. accessed on 22 february 2018. tourism nt (2017). international visitation to the northern territory: year ending september 2017. http://www.tourismnt.com.au/corporate/research/latest-visitor-data. accessed on 22 february 2018. https://www.svr-migration.de/publikationen/international-mobil/ http://www.cdu.edu.au/northern-institute/ni-research-briefs http://www.tourismnt.com.au/corporate/research/latest-visitor-data australian population studies 2018 | volume 2 | issue 1 | pages 1–13 © markham and biddle 2018. published under the creative commons attribution-noncommercial licence 3.0 australia (cc by-nc 3.0 au). journal website: www.australianpopulationstudies.org recent changes to the indigenous population geography of australia: evidence from the 2016 census francis markham* the australian national university nicholas biddle the australian national university * corresponding author. email: francis.markham@anu.edu.au. address: centre for aboriginal economic policy research, the australian national university, 24 kingsley street, acton, act 2601, australia paper received 15 february 2018; accepted 3 may 2018; published 28 may 2018 abstract background the indigenous population of australia has grown very rapidly since the first tabulation of census statistics about indigenous people in the 1971 abs census of population and housing (census). understanding the size and location of the indigenous australians is important to the state for service delivery and policy, and for indigenous peoples themselves. aims this paper summarises changes to population geography of indigenous australians between 2011 and 2016. it describes the growth in the estimated population, and its changing geographic distribution. the paper derives a measure of ‘unexpected population change’: the spatial mismatch between demographic projections from the 2011 and 2016 census counts. data and methods census data and population projections are tabulated and mapped. results indigenous people now comprise 3.3 per cent of the total australian population, or 798,381 persons. this population grew by 3.5 per cent each year between 2011 and 2016, a rate of growth 34 per cent faster than that explained by natural increase alone. both aspects of growth were concentrated in more urban parts of the country, especially coastal new south wales and southeast queensland. for the first time, fewer than 20 per cent of indigenous people were recorded as living in remote areas. conclusions indigenous population growth continues to be remarkably rapid. future research is required to understand the correlates and causes of population growth beyond that explained by natural increase. key words indigenous population growth; identification change; abs census; indigenous population geography; australia. http://www.australianpopulationstudies.org/ mailto:francis.markham@anu.edu.au 2 markham f and biddle n australian population studies 2 (1) 2018 1. introduction a detailed understanding of the indigenous population in australia is a matter of great interest to policy makers and indigenous peoples. for example, knowledge of the size, composition and characteristics of the indigenous population are used to inform government policy (e.g. the ‘closing the gap’ strategy), service delivery (e.g. the national partnership agreement on remote service delivery), resource allocation (e.g. the state-based redistribution of revenue from the federally levied goods and services tax (gst)) and advocacy for social justice (e.g. calma 2006). tabulations and estimates derived from the australian bureau of statistics (abs) census of population and housing (census) provide the best available information on the size and composition of the indigenous population of australia, albeit imperfectly. consequently, a detailed understanding of indigenous population change between 2011 and 2016 is of value to policy makers, service providers and indigenous peoples. the definition of the ‘indigenous population’ is a contentious one, and the questions in the census have arguably never met the needs of that population itself. indeed, the notion of an ‘indigenous population’ is a colonial construction which has been used to exclude indigenous people and discriminate against them (dodson 2003). for example, one primary imperative for counting the indigenous people in federated australia was the implementation of the nation’s discriminatory constitution which necessitated that indigenous persons be excluded from official population estimates produced for constitutional purposes (arcioni 2012; chesterman and galligan 1997). in practice, this meant that all people who were reached by census collectors were asked to complete a census form that included a question on ‘race’ requiring respondents to state their ‘proportion’ of aboriginal heritage. those who stated that they were ‘more than half’ aboriginal were excluded from most published population statistics (abs 2011). while this constitutional requirement was removed by a federal referendum in 1967, the census has continued to ask australian residents about their indigenous origin and is likely to do so for the foreseeable future. although questions designed to identify indigenous people have been included in every australian census, it has only been since 1971 that indigenous people have been included in the census count. the purpose, wording and administration of census questions designed to identify the indigenous population have changed considerably over time. the 2011 and 2016 census questionnaires asked about each person in the household: ‘is the person of aboriginal or torres strait islander origin?’ respondents were offered the response options of ‘no’, ‘yes, aboriginal’ and ‘yes, torres strait islander’, and were advised ‘for persons of both aboriginal and torres strait islander origin, mark both “yes” boxes’. it is important to note that this ‘standard indigenous question’ does not perfectly capture aboriginal and torres strait islander notions of indigeneity or many administrative definitions of the indigenous population. however, as walter and andersen (2013) insist, recognition of the colonial nature of these categories need not entail the abandonment of statistics about indigenous peoples. rather, they argue for the reframing of statistical collections and their interpretations in a manner that reflects indigenous standpoints. this paper adopts the abs census definition of aboriginal and/or torres strait islander peoples to explore indigenous population change in australia. the purpose of this paper is to provide an overview of the changing size and spatial distribution of the indigenous population of australia, comparing the results of the 2011 and 2016 censuses. the paper summarises three key aspects of the intercensal change: australian population studies 2 (1) 2018 markham f and biddle n 3 • the growth in the estimated population of indigenous australians • the changing geographic distribution of aboriginal and torres strait islander persons • ‘unexpected’ population change, or the spatial mismatch between demographic projections from the 2011 and 2016 census counts. 2. indigenous population increase the indigenous population increased rapidly between 2011 and 2016. in the 2016 census, 590,056 people were counted as aboriginal, 32,345 were counted as torres strait islander and 26,767 were counted as both aboriginal and torres strait islander. combined, 649,171 people were counted as indigenous, an increase of more than 100,000 people, or 18.4 per cent, since 2011. this count of the indigenous population compares with 21,341,231 persons who identified as not being of aboriginal or torres strait islander origin, resulting in the indigenous population making up 3.0 per cent of those who answered the indigenous origin census question, an increase from 2.7 per cent in 2011. the census count underestimates the number of aboriginal and torres strait islander people to a much greater degree than it underestimates the number of other australians. the abs postenumeration survey (pes), which re-interviews around 0.5 per cent of australian households nationally, is used as the basis for producing population estimates (as distinct from population counts tabulated from the census). in 2016, the pes sampled around 800 households from 33 discrete indigenous communities. the indigenous population estimate from the pes attempts to correct for at least three types of census undercount (abs 2017a): • first, a significant proportion of census records do not have a response to the indigenous status question (6.0% in 2016) and some of these individuals are likely to be of aboriginal and torres strait islander origin. records with indigenous status not stated come about either because that particular question was skipped over on the census form or because no census form was received for a dwelling that census collectors deemed to be occupied on census night. in the latter case, census collectors impute the existence of a certain number of residents, but do not answer the indigenous status question for these individuals. • second, some individuals are missed by the census – they simply do not appear on any household’s census form. this could be because they were omitted from a completed census form because census collectors missed their dwelling, or because census collectors mistakenly thought their dwelling was unoccupied. • third, a number of individuals who are listed as not being of aboriginal or torres strait islander origin on the census form later state that they are of aboriginal or torres strait islander origin when asked in the pes several weeks later, a survey that is administered face-to-face to most participants. after the pes is released, the abs further adjusts the population estimate by backdating the censusnight estimate to 30 june and estimating the number of residents temporarily overseas and thus outof-scope for the pes (abs 2017b). based on these adjustments, the abs (2017b) estimate that there were 798,381 aboriginal or torres strait islander australians in august 2016. this population estimate is an increase of 128,000 persons from 2011, or 3.5 per cent per annum. such rapid increase is not unusual for the australian indigenous population. 4 markham f and biddle n australian population studies 2 (1) 2018 as table 1 shows, both the census count and the population estimate of aboriginal and torres strait islander persons have increased considerably in almost every census since the 1967 referendum. while the annualised growth of 3.5 per cent in the aboriginal and torres strait islander population between 2011 and 2016 is substantial, it is less than the mean annual growth from 1971 to 2011 of 4.1 per cent. table 1: census count and population estimates of aboriginal and torres strait islander australians, 1901–2016 year census count population estimate undercount (%)d indigenous persons percentage of total counta,b annual growth rate (%) indigenous personsc percentage of total populationb annual growth rate (%) 2016 649,171 2.8 3.4 786,689 3.3 3.5 17.5 2011 548,370 2.5 3.8 662,335 3.0 5.2 17.2 2006 455,030 2.3 2.1 513,977 2.5 2.2 11.5 2001 410,003 2.2 3.0 460,140 2.4 4.3 10.9 1996 352,970 2.0 5.9 372,052 2.0 5.6 5.1 1991 265,489 1.5 3.1 283,631 1.6 3.4 6.4 1986 227,645 1.4 7.3 240,152 1.4 – 5.2 1981 159,897 1.1 -0.1 – – – – 1976 160,915 1.1 6.8 – – – 1971 115,953 0.9 7.5 150,076 1.1 2.6 22.7 1966e 80,750. 0.7 0.1 132,219 1.1 2.4 38.9 1961 80,526 0.8 3.8 117,495 1.1 2.3 31.5 1954 62,084 0.7 2.8 100,048 1.1 2.0 37.9 1947 51,048 0.7 0.9 87,000 1.1 1.2 41.3 1933 45,066 0.7 1.1 73,828 1.1 -0.2 39.0 1921 39,399 0.7 2.7 75,604 1.4 -1.0 47.9 1911 30,052 0.7 -4.6 83,588 1.8 -1.2 64.0 1901 48,248 1.3 – 94,564 2.4 – 49.0 sources: population estimates for 1901–1971 from abs (1986) and abs (2014a), for 1986 and 1991 from abs (1994), for 1996 from abs (1996) and for 2001–2016 from abs (2017a). census counts for 1911 from cbcs (1917), for 1921 from cbcs (1927), for 1933 from cbcs (1940), for 1947 from cbcs (1952), for 1954 from cbcs (1962), for 1961 from cbcs (1967), for 1966 from cbcs (1971), for 1901–1966 from abs (2014a), for 1971–1991 from abs (2004) and for 1996–2016 from abs community profiles. notes: all pes-derived population estimates displayed above are based on the census of that year. to the best of the authors’ knowledge, the abs did not produce estimates of the indigenous population for censuses between 1966 and 1981. the 1996 pes-derived estimate is based on the estimated population at 30 june, while all other listed figures related to the population on census night in early august. a the denominator of this percentage includes those who did not state their indigenous status in the census. b census counts and population estimates from 1901 to 1966 are for what the abs termed ‘full blood aborigines’, who were counted in the census and excluded from population statistics. because of this exclusion, denominators for these percentages have been counted by adding the indigenous census counts and population estimates to the official non-indigenous figures. persons identified in the pre-1967 censuses as ‘half-caste’ were counted as non-indigenous in these censuses and population estimates. c this column uses pes population estimates where possible (i.e. excluding those residents temporarily absent from australia on census night) in order to consistently calculate the undercount rate. d these undercount rates are calculated consistently using the method from the 2016 pes, and so do not match the published undercount rates from previous years. e excludes torres strait islanders. the indigenous census count – that is, the number of people identified as being of aboriginal or torres strait origin in the census itself, without adjustment for undercount – has also increased substantially, rising from 548,370 in 2011 to 649,171 in 2016. this is consistent with historical trends since the 1967 referendum, with the indigenous census count increasing in every census during that australian population studies 2 (1) 2018 markham f and biddle n 5 period, with the exception of the 1981 census. concerningly, the indigenous undercount rate, or difference between the population estimate and census count, is at the highest level recorded in censuses since the 1980s. the undercount rate has steadily increased from 5.2 per cent in 1986 to 17.5 per cent in 2016. while this partly reflects improvements to the population estimate, it also may indicate an increase in the number of households who have chosen not to participate in the census. the 3.5 per cent annual growth of the indigenous population between 2011 and 2016 outstripped demographic projections of population growth based on the 2011 census by a considerable degree. as figure 1 shows, around 42,000 more indigenous people were identified to be resident in australia in 2016 than the abs had predicted in projections based on the 2011 census. in this figure, the black line represents the estimate of the indigenous population based on the census of that year (i.e. the abs’s indigenous population estimates listed in table 1); the shaded region indicates the range of population estimates for that year produced by the abs based on the immediately previous census. the ‘unexplained growth’ in the indigenous population is illustrated by the gap between the black line and the shaded region. figure 1: aboriginal and torres strait islander population estimates compared with demographic population projections based on the previous census, 1996–2016 source: abs (2014b) for 2006, 2011 and 2016; abs (1998) for 2001 and 1996. notes: the population projections were produced by the abs, based on the immediately previous census (e.g. the 1996 projection is based on the 1991 census, the 2001 projection is based on the 1996 census counts, and so on). the upper bounds of the projections for 2006 and 2001 have been modified to remove predicted unexplained population growth, which was included as a component of those projections only. unexplained growth – estimated here as the difference between the 2011-based 2016 population projection and the 2016-based 2016 population estimate – accounted for around 34 per cent of intercensal population growth between 2011 and 2016. this unexplained growth is likely to result from the changing propensity of people to identify as indigenous in the census and pes, from changing rates of mixed partnering, and from changes to the methods used in administering the census and pes. figure 1 also demonstrates that the gap between the indigenous population estimates and projections based on the previous census has narrowed, compared with the growth in the indigenous population between 2006 and 2011, during which around 59 per cent of growth was unexplained. 6 markham f and biddle n australian population studies 2 (1) 2018 3. a changing spatial distribution while the number of people counted in the censuses as being of aboriginal and/or torres strait islander origin increased by 100,801 between 2011 and 2016, this growth was not evenly distributed across australia. understanding the spatial distribution of the indigenous population is important not only for local service delivery, but for resource distribution. for example, the distribution of the gst to the states and territories is influenced by the proportion of the population who are indigenous. given that the growth of the indigenous population has been most rapid in urban areas (taylor 2013), a continuation of existing trends is likely to result in a shift of the population away from parts of the country where the indigenous population is more likely to live in discrete communities, and more likely to be experiencing extreme economic disadvantage. when examined at the regional level, it is clear that growth in the indigenous census count between 2011 and 2016 is concentrated in the most heavily populated parts of new south wales and queensland. figure 2 maps indigenous population change using the 37 non-administrative indigenous regions defined by the abs, a geography that maps approximately onto the regional geography used by governments to deliver services to indigenous australians. figure 2: change in the number of indigenous people counted in the census by indigenous region, 2011–2016 source: abs 2011 and 2016 censuses. as figure 2 above shows, counts increased the most in the brisbane (17,463 persons), new south wales central and north coast (17,452 persons) and sydney–wollongong (13,852 persons) indigenous australian population studies 2 (1) 2018 markham f and biddle n 7 regions. these regions accounted for almost half of the recorded gross indigenous population growth (47.9%). at the other end of the spectrum, several regions experienced a decline in the number of people counted as being of aboriginal and torres strait islander origin in the 2016 census. declining census counts were of a much smaller magnitude than population growth, with kununurra (745 persons) and alice springs (359 persons) experiencing the largest falls in the number of indigenous people counted in the census. a similar but not identical geographic pattern of growth is apparent when indigenous population change is mapped in terms of percentage growth rather than absolute growth. as figure 3 shows, percentage population growth was high across much of southeastern australia, with the indigenous populations of victoria, tasmania, the australian capital territory, southern and eastern new south wales and southeast queensland all growing by more than 20 per cent between 2011 and 2016. the fastest growing regions were the new south wales central and north coast (33.4%), brisbane (32.8%) and melbourne (34.5%), while population decline was most rapid in kununurra (13.5%) and alice springs (7.2%). figure 3: percentage change in the number of indigenous people counted in the census by indigenous region, 2011–2016 source: abs 2011 and 2016 censuses. while these results might be interpreted as suggesting that the distribution of the indigenous population is becoming more similar to that of the non-indigenous population, the census figures do not tell a simple story of converging geographic distributions. clearly, the less urbanised indigenous population continues the long-term trend of becoming increasingly likely to live in cities. 8 markham f and biddle n australian population studies 2 (1) 2018 as table 2 shows, 36.8 per cent of the indigenous population lived in what the abs terms ‘major cities’ in 2016, an increase of 2.6 per cent from 34.2 per cent in 2011, using a consistent geographic classification. this increase consists of a combination of natural increase, migration and statistical ‘identification change’. identification change occurs when an individual is classified as being of aboriginal or torres strait islander origin in one census but not the following census, or vice versa (biddle and crawford 2015). however, during this same period, the non-indigenous population also became increasingly urbanised, with 72.3 per cent of non-indigenous australians resident in major cities in 2016, an increase of 1.3 per cent from 71.0 per cent in 2011. table 2: the indigenous and non-indigenous population count distributions by geographical remoteness, 2016 remoteness indigenous population count percentage of indigenous count change in percentage of indigenous count, 2011– 2016 nonindigenous population count percentage of nonindigenous population count change in percentage of non-indigenous population count, 2011– 2016 major cities 235,527 36.8 2.6 15,409,691 72.3 1.3 inner regional 154,087 24.1 1.7 3,858,090 18.1 -0.5 outer regional 130,976 20.5 -1.5 1,738,227 8.2 -0.6 remote 40,689 6.4 -1.0 224,485 1.1 -0.1 very remote 79,041 12.3 -1.9 90,897 0.4 -0.1 source: abs 2011 and 2016 censuses. notes: not stated population excluded. calculated on the basis of 2011 remoteness boundaries, using an area-based 2016 to 2011 statistical area level 1 (sa1) concordance. however, the starkest divergence in population dynamics lies in ‘inner regional’ areas, where 24.1 per cent of indigenous people and 18.1 per cent of non-indigenous people now live. while the proportion of the indigenous population living in inner regional areas continues to grow (up 1.7% in 2016 from 2011), the proportion of the non-indigenous population living in these locations fell by 0.5 per cent between 2011 and 2016. another difference between the two populations is that much of the growth of non-indigenous australians in urban areas is likely to be driven by international immigration. for the indigenous population the impact of international migration is negligible, with the small percentage of indigenous australians who move overseas mostly balanced by the small percentage who return. growth in urban areas is therefore likely to be made up of a combination of excess of births over deaths, changing patterns of identification and internal migration. 4. the geography of unexpected population change indigenous population growth between 2011 and 2016 was well above what could be explained by natural increase alone. as described above, the indigenous population in australia on census night, (9 august 2016) was estimated to be 786,689 on the basis of the pes, or 3.3 per cent of the total australian population (table 1). this was around 42,000 more than the upper range of the indigenous population projections based on the 2011 census. this unexplained population growth is likely to arise from identification change, and potentially also from changes to enumeration and processing methods between the 2011 and 2016 censuses. australian population studies 2 (1) 2018 markham f and biddle n 9 there is little reason to expect that the unexplained population growth is distributed evenly across australia. although too little information has been published at present to construct detailed regional population estimates, including age–sex distributions, provisional estimates can be derived using a simple, three-step procedure, as described below. • first, those records with no response to the indigenous origin question had their indigenous status imputed on the basis of the percentage of respondents who identified as indigenous in their statistical area level 1 (sa1). this increased the indigenous population from a count of 643,136 (excluding those in migratory, shipping and offshore areas) to an estimate of 692,182. • second, a provisional population estimate for each region was derived by multiplying these regional, prorated indigenous population estimates by 1.137. this multiplier was selected so that the sum of all regional population estimates equalled 786,689, the national estimate derived from the pes. • third, indigenous population projections for each region for 2016, based on the 2011 census, were subtracted from provisional population estimates for regions. the 2011-based projections were produced on the basis of natural increase and a repetition of 2006–2011 migration patterns. (for more detail, see biddle 2013.) consequently, the subtraction of the 2011-based projection from the 2016-based estimate gives a measure of unexpected population change for each region, or the spatial mismatch between projected and observed population increase. because some of this population change will be explained by changing migration patterns between 2011 and 2016, we term this population change ‘unexpected’ rather than ‘unexplained’. figure 4: unexpected population change by indigenous region, 2011–2016 source: authors’ estimates based on abs 2011 and 2016 censuses. 10 markham f and biddle n australian population studies 2 (1) 2018 estimates of unexpected population change by indigenous regions are displayed in figure 4 (above). put simply, the figure shows the difference between the estimated indigenous population in 2016, and the expected 2016 indigenous population, as projected on the basis of the 2011 census by biddle (2013). because the total indigenous population increased faster than projected, the unexpected population change in most regions was positive (i.e. the indigenous population increased faster than anticipated). however, in 10 of the 37 regions, unexpected population change was negative, meaning that the indigenous population increased less quickly than projected. much of the unexpected population increase occurred in south-eastern australia. in particular, more than 75 per cent of the unexpected population increase occurred in just five regions: new south wales central and north coast (14,844); sydney–wollongong (7,412); brisbane (5,813); riverina– orange (4,713); and northeastern new south wales (3,600). while these regions were projected to experience substantial population increase, the actual population increase substantially exceeded expectations. however, not all regions saw greater than expected population growth between 2011 and 2016. for example, the sixth most rapidly growing region, townsville–mackay, had an estimated indigenous population increase of 4,800 persons, very close to the projected increase of 4,600. other regions experienced population changes that were substantially less than projected. for example, the indigenous population of perth was projected to increase by 5,628 between 2011 and 2016, but instead grew by only 1,734 persons. similarly, the indigenous population of alice springs was projected to increase by 817 persons, but our estimates suggest that it decreased by 745 persons. 5. conclusions the quinquennial abs census remains the best source of information on indigenous population change. the aim of this paper was to provide an overview of the changing size and spatial distribution of the indigenous population of australia between 2011 and 2016, with a focus on the geography of unexpected population change. the first set of analyses focused on the growth in the estimated population of indigenous australians. we showed that the indigenous population grew rapidly between 2011 and 2016, reaching around 3.3 per cent of the total population estimate, or 786,689 persons. it is more than just a historical footnote to reflect upon the fact that this is similar to the accepted population estimate of the indigenous population at the time of european colonisation (mulvaney 2002), up from a low of around 72,000 at the time of the 1921 census (which is likely to have been a significant undercount). like other indigenous groups internationally (in particular, the united states, canada and new zealand), the indigenous population has rebounded substantially from the destruction wrought by frontier violence, disease and other forms of colonial domination. the geographic distribution of the indigenous population is also changing. between 2011 and 2016, the most rapid growth in the indigenous population count occurred in more urban parts of the country, with three regions (brisbane, the new south wales central and north coast and sydney– wollongong) accounting for almost half of the recorded gross indigenous population growth. there continues to be a policy focus on indigenous australians living in remote areas. the results of this paper suggest that a remote focus is becoming less justifiable on the basis of population geography alone, as the remote indigenous population continues to shrink as a percentage of the total australian indigenous population. australian population studies 2 (1) 2018 markham f and biddle n 11 as the economic circumstances of the remote and urban indigenous population appear to be diverging, at least when measured in terms of income (markham and biddle 2018), the spatially uneven indigenous population increase suggests that resource targeting within the indigenous population may need to become a more prominent policy consideration. indeed, while indigenous disadvantage may be most concentrated in remote areas, in absolute terms there are more disadvantaged individuals living in urban areas. consequently, it is important that regional approaches to policy do not ignore intra-regional indigenous inequality. considerable uncertainty exists regarding the growth in the population that cannot be explained by the excess of births over deaths. by comparing the 2016 census data with population projections based on 2011 data, we have shown that unexplained indigenous population growth accounts for around 34 per cent of total intercensal population growth. while considerable, this proportion is less than the equivalent for the 2011 census, for which around 59 per cent of intercensal growth was unexplained. much of this excess or unexplained growth is likely to be caused by people changing their response in successive censuses to the indigenous ‘origin’ question, or by their indigenous identification being given differently by those who respond to the census on their behalf (see biddle and markham 2018). previous research has shown that many indigenous australians are selective about the contexts in which they reveal their aboriginal or torres strait islander origin. this research suggests that the propensity for an individual to identify themselves as aboriginal often depends on their assessment of the risks and benefits of doing so, particularly with respect to racism and discrimination and whether or not identification would lead to a challenge to prove aboriginal identity (abs 2013; biddle and crawford 2015; nsw aboriginal affairs 2015). while this contextualisation is crucial to understanding identification, further research is needed to understand the causes of spatial variation in identification change, and the likelihood of continued identification change into the future. indeed, our research demonstrates that the unexplained growth in indigenous counts is neither spatially nor demographically consistent. while much of this growth occurred in three urban regions (sydney–wollongong, nsw north and central coast, and brisbane), however there were also large urban regions like perth where the population increased more slowly than expected from the population projections. further research into the components of intercensal population growth would assist in understanding the social and demographic drivers of indigenous population change. 6. key messages • 81.4 per cent of the indigenous population now live in urban or regional areas. • the indigenous population grew between 2011 and 2016 at a rate of 3.5 per cent per year, to an estimated population of 798,381. • the intercensal population increase of 128,000 persons is around 42,000 persons (or 34%) greater than can be accounted for by natural increase. this is likely to result from the changing propensity of individuals to identify as indigenous in the census or changes to the census collection and processing methods. • indigenous population growth was highest in coastal regions between melbourne and brisbane, while the indigenous population of several remote regions australia fell modestly. • identification change is very spatially concentrated, and is most rapid in tasmania, new south wales and brisbane. 12 markham f and biddle n australian population studies 2 (1) 2018 acknowledgements this manuscript is based on research funded by the australian government department of the prime minister and cabinet. all opinions expressed in the paper and any errors made therein are attributable to the authors alone. references abs (australian bureau of statistics) (1986) australian demographic trends. cat. no. 3102.0. canberra: abs. abs (1994) experimental estimates of the aboriginal and torres strait islander population, june 1986 to june 1991. cat. no. 3230.0. canberra: abs. abs (1996) australian demographic statistics. cat. no. 3101.0, march, june, september and december. canberra: abs. abs (1998) experimental projections of the aboriginal and torres strait islander population, 1996 to 2006. cat. no. 3231.0. canberra: abs. abs (2004) year book australia. cat. no. 1301.0. canberra: abs. abs (2011) reflecting a nation: stories from the 2011 census, july 2011. cat. no. 2071.0. canberra: abs. abs (2013) information paper: perspectives on aboriginal and torres strait islander identification in selected data collection contexts, 2012. cat. no. 4726.0. canberra: abs. abs (2014a) australian historical population statistics. cat. no. 3105.0.65.001. canberra: abs. abs (2014b) estimates and projections, aboriginal and torres strait islander australians, 2001 to 2026. cat. no. 3238.0. canberra: abs. abs (2017a) census of population and housing: details of overcount and undercount, australia, 2016. cat. no. 2940.0. canberra: abs. abs (2017b). australian demographic statistics, mar 2017. cat no. 3101.0. canberra: abs. arcioni e (2012) excluding indigenous australians from ‘the people’: a reconsideration of sections 25 and 127 of the constitution. federal law review 40(3): 287–316. biddle n (2013) population projections. caepr indigenous population project: 2011 census papers, paper 14, centre for aboriginal economic policy research, australian national university, canberra. biddle n and crawford h (2015) the changing aboriginal and torres strait islander population: evidence from the 2006–11 australian census longitudinal dataset. caepr indigenous population project: 2011 census papers, paper 18. centre for aboriginal economic policy research, australian national university, canberra. biddle n and markham f (2018) indigenous identification between 2011 and 2016: evidence from the australian census longitudinal dataset. caepr topical issue 1/2018. centre for aboriginal economic policy research, australian national university, canberra. calma t (2006) social justice and human rights: using indigenous socioeconomic data in policy development. in: hunter b (ed.), assessing the evidence on indigenous socioeconomic outcomes. canberra: anu press; 299–310, http://press.anu.edu.au?p=119431. cbcs (commonwealth bureau of census and statistics, australia) (1917). census of the commonwealth of australia, 1911. abs cat. no. 2112.0. melbourne: cbcs. cbcs (1927) census of the commonwealth of australia, 1921. abs cat. no. 2111.0. melbourne: cbcs. cbcs (1940) census of the commonwealth of australia, 1933. abs cat. no. 2110.0. canberra: cbcs. cbcs (1952. census of the commonwealth of australia, 1947, abs cat. no. 2109.0. canberra: cbcs. cbcs (1962) census of the commonwealth of australia, 1954, abs cat. no. 2108.0 canberra: cbcs. cbcs (1967) census of the commonwealth of australia, 1961, abs cat. no. 2107.0 canberra: cbcs. http://press.anu.edu.au/?p=119431 australian population studies 2 (1) 2018 markham f and biddle n 13 cbcs (1971) census of population and housing, 1966. abs cat. no. 2106.0. canberra: cbcs. chesterman j and galligan b (1997) citizens without rights: aborigines and australian citizenship. cambridge: cambridge university press. dodson m (2003) the end in the beginning: re(de)finding aboriginality. in: grossman m (ed.) blacklines: contemporary critical writing by indigenous australians. melbourne: melbourne university press; 25–42. markham, f and biddle n (2018) income, poverty and inequality. caepr indigenous population project: 2016 census papers, paper no. 2, centre for aboriginal economic policy research, australian national university, canberra. mulvaney j (2002) ‘difficult to found an opinion’: 1788 aboriginal population estimates. in: briscoe g and smith l (eds), the aboriginal population revisited: 70 000 years to the present. canberra: aboriginal history inc.; 1–8. nsw aboriginal affairs (2015) aboriginal identification in nsw: the way forward. an aboriginal peoples’ perspective. sydney: nsw aboriginal affairs, department of education. taylor j (2013) indigenous urbanization in australia: patterns and processes of ethnogenesis. in: peters e and andersen c (eds) indigenous in the city: contemporary identities and cultural innovation. vancouver: ubc press; 237–255. walter m and andersen c (2013) indigenous statistics: a quantitative research methodology. walnut creek, california: left coast press. australian population studies 2018 | volume 2 | issue 1 | pages 26–38 © wilson and shalley 2018. published under the creative commons attribution-noncommercial licence 3.0 australia (cc by-nc 3.0 au). journal website: www.australianpopulationstudies.org estimates of australia’s non-heterosexual population tom wilson* charles darwin university fiona shalley charles darwin university * corresponding author. email: tom.wilson@cdu.edu.au. address: northern institute, charles darwin university, ellengowan drive, darwin, nt 0909 paper received 14 february 2018; accepted 28 april 2018; published 28 may 2018 abstract background demographers have studied minority populations for many years, but relatively little attention has been paid to sexual minority groups. population estimates for sexual minorities would be useful as denominators for a range of health and socioeconomic indicators, to monitor representation in employment, assist budget planning and inform the marketing of goods and services. aim the aim of this paper is to present some approximate estimates of the non-heterosexual adult population of australia in mid-2016 by sex, broad age group and state and territory. data and methods data on sexual identity were sourced from three nationally representative surveys: the household income and labour dynamics in australia survey, the second australian study of health and relationships and the abs general social survey. use was made also of 2016 abs census of population and housing (census) data and estimated resident populations. prevalence rates of the non-heterosexual population aged 18+ were averaged over the three surveys and multiplied by erp to obtain national population estimates. census data on same-sex couples were used to distribute the national estimates by state and territory. results australia’s non-heterosexual population aged 18+ in 2016 is estimated to have been 592,000, representing about 3.2% of the adult population. new south wales is home to the largest nonheterosexual population (about 204,000) and the northern territory the smallest (4,700), while the highest prevalence is in the australian capital territory (5.1%). conclusions australia’s non-heterosexual population is a relatively small population, but its prevalence varies considerably by age and sex and between states and territories. estimates of this population should prove useful for monitoring health and wellbeing and for a variety of planning and policy purposes. key words population estimates; non-heterosexual; lesbian, gay, bisexual; states and territories; australia. http://www.australianpopulationstudies.org/ mailto:tom.wilson@cdu.edu.au australian population studies 2 (1) 2018 wilson t and shalley f 27 1. introduction minority populations have been the subject of demographic research and policy development for many decades, particularly ethnic minority populations (e.g. coleman and salt 1996; jivraj and simpson 2015) and indigenous peoples (e.g. smith 1980; taylor 2011). attention has also been paid to groups such as foreign-born populations (e.g. edmonston 2016) and minority religious populations (e.g. peach 2006; pew research center 2017). over the last two decades, and especially in recent years, demographers and statisticians have started to examine sexual and gender minority populations in western countries (e.g. aspinall 2009; baumle, compton and poston 2009; baumle 2013; black et al. 2000; ons 2010; meier and labuski 2013; ons 2017). in australia, the non-heterosexual1 population was subject to increased attention in the lead-up to the australian marriage law postal survey in 2017. following public debate about same-sex marriage over many years, the australian government decided to put the issue to the electorate in late 2017. effectively a plebiscite, the survey asked: ‘should the law be changed to allow same-sex couples to marry?’ of those who returned their forms (and excluding unclear responses), 62 per cent voted ‘yes’ (abs 2017a). shortly afterwards parliament voted overwhelmingly to amend the commonwealth marriage act 1961, and the updated law came into force in december 2017. yet while progress such as this is made towards equality for non-heterosexual people in australia, relatively little is known about the demography of this population. more demographic research could increase knowledge and understanding of this segment of society, leaving less room for stereotypes and myths (gates 2012). better demographic evidence could highlight where disadvantage and discrimination remain, and where policy and legislation need revision. sexual minority population estimates specifically could prove valuable for a variety of health, social justice and economic purposes. for example, estimates could be used to monitor representation in employment, assist budget planning for specific services for sexual minorities, form denominators for a range of health and socioeconomic indicators, and inform the marketing of goods and services (e.g. same-sex weddings). in addition, sexual minority demography can prove useful in broader analyses of demographic trends and theory, such as the influence of gender roles within couples on the labour market consequences of migration (e.g. cooke 2005). the bulk of existing research on sexual minority demography focuses on the united states and, to a lesser extent, the united kingdom (e.g. black et al. 2000; black et al. 2002; cooke and rapino 2007; flores et al. 2016; gates 2014; geary et al. 2018; ons 2017; van kampen et al. 2017). australian research in the area is limited, though a small number of contributions have been made over the last two decades (e.g. dempsey 2015; grulich et al. 2003; madeddu et al. 2006; perales and baxter 2018; prestage et al. 2008; richters, altman et al. 2014; smith et al. 2003; wilson 2004). many australianbased papers are concerned primarily with important sexual health issues, and only a few present population estimates. prestage et al. (2008) estimated the population of gay and bisexual men in australia in each of the states and territories in 2001, while madeddu et al. (2006) created estimates of gay and bisexual men across inner sydney postcodes in 2001. both multiplied gay and bisexual prevalence rates by census counts, using data from the australian bureau of statistics (abs) census 1 we use the term ‘non-heterosexual’ here, but acknowledge there is no standard terminology for sexual and gender minority groups. we discuss this issue in section 2 of the paper. 28 wilson t and shalley f australian population studies 2 (1) 2018 of population and housing (census), to obtain estimates. several other studies included estimates of the prevalence of the non-heterosexual population or parts of it (primarily gay and bisexual men). we argue that there is a need for sexual minority population estimates for australia. ideally, these would be current, available annually, contain age, sex and geographic detail, cover the whole nonheterosexual population but also include separate estimates for lesbian, gay, bisexual and other minority sexual orientations. preferably, the estimates would use estimated resident populations (erps) in their calculation rather than census counts, which suffer from undercount (particularly in the young adult ages). but creating sexual minority estimates is far from straightforward. there are so few demographic data sources which include sexual orientation (wooden 2014), and those that do so have several limitations. defining who should be included in the non-heterosexual population is also complicated because of the complex nature of sexual orientation. the aim of this paper is to present some approximate estimates of the non-heterosexual adult population of australia in mid-2016 by sex, broad age group and state and territory. the estimates are approximate in the sense that they are constructed from several imperfect data sources and necessarily involve several assumptions. at this stage we present state and territory and national estimates only, as well as those for the non-heterosexual population as a whole. finer geographical detail, annual numbers, alternative definitions (see section 2), and separate estimates for lesbian, gay, bisexual and other minority sub-populations are planned in subsequent work. following this introduction we consider how sexual orientation is conceptualised and the issue of appropriate definitions and terminology for sexual minority populations. section 3 describes the data sources and population estimation methods, while section 4 presents the results. a final section includes a short discussion and conclusion. 2. definitions and terminology sexual orientation has been defined as: an enduring pattern of emotional, romantic, and/or sexual attractions to men, women, or both sexes … [and] a person’s sense of identity based on those attractions, related behaviors, and membership in a community of others who share those attractions. (american psychological association 2008 p. 1) it is generally considered to comprise three elements (durso and gates 2013): (i) sexual attraction – the feeling of sexual desire towards others, considered by some to be the fundamental basis of sexual orientation (ii) sexual behaviour – sexual activity (iii) sexual identity – how a person describes their sexual orientation (e.g. heterosexual, gay, lesbian, bisexual). the non-heterosexual population referred to in this paper includes all those who identify as gay, homosexual, lesbian or bisexual, or construct their sexuality in other ways using non-heterosexual terminology (e.g. queer). our focus here is on sexual identity rather than behaviour or attraction. the sexual identity, behaviour and attraction of individuals are not necessarily coterminous. the venn diagram in figure 1 illustrates the relationship between the three dimensions of sexual australian population studies 2 (1) 2018 wilson t and shalley f 29 orientation for the non-heterosexual population. several surveys have found that a relatively large group of people have felt some same-sex sexual desire at some point in their adult lives; a smaller but not wholly overlapping group have engaged in same-sex sexual activity’ and a smaller group still identifies as being non-heterosexual (e.g. geary et al. 2018; richters, altman et al. 2014; smith et al. 2003). identity is shown in the diagram as ‘current’ identity because some people change the way they describe themselves over time. figure 1: the relationship between the three dimensions of sexual orientation source: loosely based on figure 1 in richters, altman et al. (2014) and figure 2 in geary et al. (2018). it is important to recognise that people counted as non-heterosexual in data sources are only those willing to identify as such in statistical collections. strictly, this is the revealed non-heterosexual population. others may consider themselves to be gay, lesbian, bisexual or other in private, but choose not to disclose their sexuality in public. this is the hidden (or ‘closeted’) population, and is not included in our population estimates (see gates 2012 for a discussion of the issue). in addition, it is important to note that sexual orientation is distinct from gender identity and intersex issues, although they are often considered together – for example, in the term lgbtqi (lesbian, gay, bisexual, transgender, queer and intersex). gender identity refers to ‘a person’s innate, deeply-felt psychological identification as a man, woman, or something else’ (fenway health 2010 p. 3). the transgender population consists of those whose gender identity or expression differs from their birth sex, while intersex refers to ‘a spectrum of conditions involving anomalies of the sex chromosomes, gonads, reproductive ducts, and/or genitalia’ (fenway health 2010 p. 9). transgender and intersex populations may be of any sexual orientation. finally, on the issue of terminology, we acknowledge that many different terms are used in the literature covering different population sub-groups, with no single term being standard (dempsey 2015). these include lgb (lesbian, gay and bisexual), lgbt (lesbian, gay, bisexual and transgender), lgbtiq (lesbian, gay, bisexual, transgender, intersex and queer), and lgbtqqip2saa (lesbian, gay, bisexual, transgender, queer, questioning, intersex, pansexual, 2-spirited, asexual and allies). we choose the shorthand term ‘non-heterosexual’ to refer to gay, lesbian, bisexual and related sexual any lifetime samesex behaviour any lifetime samesex attraction current nonheterosexual identity 30 wilson t and shalley f australian population studies 2 (1) 2018 minorities. gender minorities such as the transgender population, and sex minorities such as intersex people, are included in our population estimates only if they have identified as non-heterosexual. 3. data and methods 3.1 data sources data on people identifying as non-heterosexual were acquired from three representative national surveys: the household income and labour dynamics in australia (hilda) survey, the second australian study of health and relationships (ashr2), and the abs general social survey (gss). unpublished data on the population identifying as non-heterosexual by sex and broad age group (18–24, 25–34, 35–44, 45–54, 55–64 and 65+) were obtained and non-heterosexual proportions of the population (or prevalence) calculated for each dataset. the hilda survey is a nationally representative household-based longitudinal survey on the lives of australians (wilkins 2017). it covers a wide range of topics including family relationships, employment, education, income, health and wellbeing, life events and personal attitudes. it is conducted annually by interview and self-completion questionnaire with all those aged 15 years and over in the participating household. a question on sexual identity was included in wave 12 (2012) as part of the self-completion questionnaire (wilkins 2015; wooden 2014). in this wave interviews were conducted with 17,476 respondents and self-completion questionnaires were received from 88 per cent of the interviewed sample. the sexual identity question asked: ‘which of the following categories best describes how you think of yourself? heterosexual or straight; gay or lesbian; bisexual; other; unsure/don’t know; and prefer not to say’ (melbourne institute 2012). the ashr2 was undertaken in 2012 and 2013 to study the sexual and reproductive health, sexual practices and wellbeing of the australian population (richters, badcock et al. 2014). the survey was conducted via telephone interview using random digit dialling of landline and mobile phones. the survey covered australian residents aged 16–69 and obtained responses from 20,094 participants, achieving a response rate of 66 per cent of eligible people contacted. the sexual identity question asked: ‘do you think of yourself as: 1. heterosexual or straight; 2. homosexual (gay [asked of males]; lesbian [asked of females]); 3. bisexual’. other responses were coded as: 4. queer; 5. not sure; undecided; 6. something else/other (richters, altman et al. 2014). the gss is a household-based survey run by the abs every four years. the aim of the survey is to provide a broad range of information on an individual’s social circumstances and their relative levels of advantage and disadvantage (abs 2015a). the survey covers all people aged 15 years and over who are usual residents of private dwellings and is conducted by face-to-face interview. in the 2014 survey 12,932 out of 16,145 eligible persons responded to the survey, representing a response rate of 80 per cent. a question on sexual identity was introduced for the first time in 2014 and was asked of all participants aged 18 years and over. it asked: ‘which of the following options best describes how you think of yourself? 1. straight (heterosexual); 2. gay or lesbian; 3. bisexual; 4. other; 5. don’t know’ (abs 2015b). in addition, our study made use of some data from the 2016 census. counts of persons in same-sex couples by broad age group, sex and state and territory were extracted from the abs tablebuilder australian population studies 2 (1) 2018 wilson t and shalley f 31 pro online data service (abs 2017b). a direct question on sexual orientation is not asked in the australian census. instead, a same-sex couple variable is created by combining information from two census questions: the question on the relationship of an individual in the household to person 1 on the census form; and the question ‘is this person male or female?’. if the relationship of someone to person 1 is recorded as husband or wife or de facto partner, and both individuals record the same answer to the question on gender, then a same-sex couple is identified (abs 2018). nationally, 2016 census counts of individuals aged 18 and over living together as part of a same-sex couple totalled about 90,000 (only 0.5 per cent of adults counted in the census). this is only a subset of all non-heterosexual people, of course. aside from census undercount, non-heterosexual people who are single or in livingapart-together relationships are excluded from the census count of individuals in same-sex couples. hilda survey findings from 2012 show that, amongst those aged 25–59, only 55 per cent of nonheterosexual men and 59 per cent of women are partnered, compared to 74 per cent of heterosexual men and women in that age range (wilkins 2015). in addition, people in cohabiting same-sex couples are not always identified in the census as being in a same-sex relationship: for example, when neither is person 1 on the census form, or if they do not wish to reveal their relationship. furthermore, there is probably some error in the same-sex couple counts due to errors in responses to the question on gender. research in the united states has revealed this to be a problem with their same-sex couple census data (e.g. gates 2010; dibennardo and gates 2014), but there is no information as to whether equivalent response errors have affected australian census data. finally, mid-2016 erps by broad age group, sex and state and territory were obtained from the abs.stat online data extraction tool (abs 2017c). 3.2 methods estimates of the adult non-heterosexual population of australia in 2016 were created by multiplying non-heterosexual proportions averaged across the three surveys (hilda, ashr2, gss) by erps published by the abs for 30 june 2016. the estimates were prepared by sex and broad age group (18–24, 25–34, 35–44, 45–54, 55–64 and 65+) for australia and each of the states and territories. the decision to use proportions averaged across the three surveys was taken because of the non-trivial variations in proportions between surveys and no clear evidence that one survey was superior to another. national estimates were calculated by multiplying averaged proportions by erps: 𝑃𝐴𝑢𝑠,𝑠,𝐴 𝑁𝐻 = 𝑝𝑟𝑜𝑝𝐴𝑢𝑠,𝑠,𝐴 𝑁𝐻 𝐸𝑅𝑃𝐴𝑢𝑠,𝑠,𝐴 (1) where 𝑃 denotes population estimate, 𝐴𝑢𝑠 australia, 𝑝𝑟𝑜𝑝 the proportion of the population, 𝑁𝐻 non-heterosexual, 𝐸𝑅𝑃 estimated resident population, 𝑠 sex and 𝐴 broad age group. state and territory estimates were created by distributing national estimates according to the number of individuals in same-sex relationships identified in the 2016 census: 𝑃𝑖,𝑠,𝐴 𝑁𝐻 = 𝑃𝐴𝑢𝑠,𝑠,𝐴 𝑁𝐻 𝐶𝑖,𝑠,𝐴 𝑆𝑆𝑅 𝐶𝐴𝑢𝑠,𝑠,𝐴 𝑆𝑆𝑅 (2) where 𝑖 denotes state or territory, 𝐶 census counts, and 𝑆𝑆𝑅 the population in same-sex relationships. while the census same-sex couple data provide an imperfect geographical distribution proxy, they do not suffer from small sample size. survey estimates of state and territory non-heterosexual 32 wilson t and shalley f australian population studies 2 (1) 2018 populations were based on samples too small to be of use (e.g. there were zero non-heterosexual males in the northern territory in the hilda data). clearly, the above methods incorporate a number of assumptions which are unlikely to hold precisely. no adjustments have been made for net undercount in the census. it is assumed that the average proportions across the three surveys provide an accurate measure of the identified nonheterosexual population of australia. it is also presumed that the proportion of the population identifying with non-heterosexual identities remains unchanged between the time of the surveys in 2012 to 2014 and mid-2016. for the state and territory population estimates, the distribution of people in same-sex relationships recorded by the 2016 census is assumed to match the distribution of the wider non-heterosexual population. this last assumption is probably the most approximate. it effectively assumes that, for each age-sex group, the proportion of the non-heterosexual population in a same-sex relationship as identified by the census is the same in each state and territory. this is unlikely to be exactly the case, though we note that the census distribution of all persons in couples (irrespective of gender composition) by state and territory and broad age group is extremely close to the equivalent erp distribution. however, variations from our assumption are likely to occur due to educational composition. using survey data from california, carpenter and gates (2008) found that highly educated gay men and lesbian women were more likely to be partnered than those who were less educated. 4. non-heterosexual population estimates 4.1 new estimates the extent to which the australian population identifies as non-heterosexual, using the averaged findings from the three surveys, is shown in table 1 below. amongst the adult population of australia, 3.1 per cent of males and 3.4 per cent of females describe themselves as non-heterosexual. these percentages exclude survey participants who refused to answer the sexual identity question or who replied ‘don’t know’. the percentages are higher at younger ages and lower at older ages, with the age gradient being more pronounced for females. table 1: percentage of the australian adult population identifying as non-heterosexual averaged across three surveys, 2012–2014 age group males females 18–24 4.0 6.4 25–34 4.0 4.8 35–44 3.0 3.8 45–54 2.6 2.6 55–64 2.4 2.0 65+ 2.3 1.1 18+ 3.1 3.4 sources: gss 2014; ashr2; hilda wave 12. estimates of australia’s non-heterosexual population by age group and sex are presented in table 2. they suggest that the national non-heterosexual population aged 18+ in mid-2016 was a little under australian population studies 2 (1) 2018 wilson t and shalley f 33 600,000, representing 3.2 per cent of the total adult population. the figures indicate there were more non-heterosexual females than males in the younger adult ages, with the situation reversed in the older age groups. overall, at ages 18 and above the non-heterosexual population is younger than the australian population as a whole. table 2: estimates of the australian adult population identifying as non-heterosexual, 2016 age group males females persons 18–24 47,098 71,839 118,937 25–34 71,804 86,551 158,355 35–44 48,874 61,770 110,644 45–54 40,848 41,009 81,857 55–64 32,579 28,684 61,263 65+ 39,848 21,011 60,859 18+ 281,052 310,863 591,915 source: authors’ estimates. population estimates for the states and territories are shown in table 3. not surprisingly the most populous states, new south wales, victoria and queensland, are home to the largest non-heterosexual populations, while the australian capital territory (act), tasmania and the northern territory have the smallest populations. all jurisdictions have more females than males in their non-heterosexual populations with the one exception of new south wales. as a fraction of the total population, south australia, western australia and the northern territory have the smallest percentages identifying as non-heterosexual. the act has a relatively large non-heterosexual population at 5.1 per cent of its total adult population, a function to some extent of its comparatively young population (wilson 2016). table 3: estimates of the adult population identifying as non-heterosexual by state and territory, 2016 state/territory population aged 18+ % of erp males females persons persons nsw 106,400 98,023 204,423 3.4 vic 76,267 80,790 157,057 3.3 qld 48,996 63,596 112,592 3.0 sa 14,265 20,818 35,083 2.6 wa 21,280 29,828 51,108 2.6 tas 4,992 5,953 10,945 2.7 nt 2,147 2,596 4,743 2.6 act 6,705 9,258 15,964 5.1 australia 281,052 310,863 591,915 3.2 source: authors’ estimates. notes: nsw = new south wales; vic = victoria; qld = queensland; sa = south australia; wa = western australia; tas = tasmania; nt = northern territory; act = australian capital territory. 4.2 comparisons with other studies comparing our non-heterosexual population estimates with estimates produced by others is not straightforward because of variations in population coverage, reference dates, age ranges, data sources and social and cultural contexts. nonetheless, it is instructive to compare our figures with the few earlier estimates prepared for australia. 34 wilson t and shalley f australian population studies 2 (1) 2018 the gay and bisexual male population aged 16+ was estimated by prestage et al. (2008) to be 182,624 in 2001, representing 2.5 per cent of the male population. the prevalence estimates across states and territories varied from 0.8 per cent for tasmania and 0.9 per cent for the northern territory to 2.9 per cent for the act and 3.0 per cent for sydney. those numbers were based on the first australian survey of health and relationships (ashr). our non-heterosexual population prevalence rates for just the male population aged 18+ varied from a low of 2.2 per cent in south australia and western australia and 2.3 per cent in the northern territory up to 3.6 per cent in new south wales and 4.4 per cent in the act. it was 3.1 per cent for australia as a whole. how do our estimates compare to those produced for other countries? gates (2011) estimated that about 8 million adults identify as lesbian, gay or bisexual in the united states, representing about 3.5 per cent of the adult population. geary et al. (2018) estimated 1.2 million people aged 16–74 in britain identified as lesbian, gay, bisexual or other in 2011, or 2.7 per cent of the population. a recent estimate by the office for national statistics put the population aged 16 and over describing themselves as lesbian, gay, bisexual or other in the united kingdom in 2016 to be 1.3 million, representing 2.5 per cent of the population (ons 2017). in canada, the canadian community health survey found that 3.0 per cent of adults aged 18–59 reported themselves to be lesbian, gay or bisexual in 2014 (statistics canada 2017). in terms of age distributions, our data are consistent with most other studies, which have found a higher prevalence of non-heterosexual identities at younger ages (e.g. gates 2014; ons 2017). however, this is not a universal finding, with the prevalence estimates for britain presented by geary et al. (2018) showing little difference for ages between 16 and 64. overall, our non-heterosexual population estimates are reasonably consistent with other respectable studies, and consist of plausible and sensible numbers. 5. conclusions this paper has presented a novel set of population estimates for australia’s non-heterosexual population, with disaggregation by sex, broad age group and state and territory. the estimates should be regarded as approximate given the limitations of the data sources and the assumptions inherent in our methods. nonetheless, they provide a useful overview of the current nonheterosexual population of australia which was not previously available. it is emphasised that, conceptually, these estimates refer to the population identifying as gay, lesbian, bisexual and in other non-heterosexual ways, and not those who engage in same-sex sexual behaviour or who have ever experienced same-sex sexual attraction (figure 1). population estimates based on these alternative definitions should be useful for health surveillance and will be the subject of subsequent research. ideally, state and territory and sub-state, non-heterosexual population estimates would be based on census or large-scale survey data. the office for national statistics in the united kingdom asks a sexual identity question in its annual population survey, which is sufficiently large (covering 41,000 households each quarter) to enable sexual identity population estimates to be produced for local authorities (ons 2017). perhaps at some point the abs will follow suit and include a direct sexual identity question in a large-scale survey (or even the census). in the meantime it would be useful to investigate alternative data sources for estimating the state and territory distributions because the use of same-sex couple data from the census is imperfect. australian population studies 2 (1) 2018 wilson t and shalley f 35 it is possible that our act prevalence is marginally over-estimated due to its highly educated population. there is likely to be a higher proportion of non-heterosexual people in the act in samesex cohabiting relationships than in other jurisdictions (carpenter and gates 2008), which violates the assumption of a fixed ratio between the number of people in same-sex couples in the census and the non-heterosexual population across jurisdictions. data on the proportion of non-heterosexual adults in same-sex relationships would be ideal as it would enable census counts of persons in samesex couples to be more accurately scaled up to the total non-heterosexual population. it would also be useful to investigate the potential of new web-based data sources, such as google search data and social media profiles. for example, some facebook users reveal sexual orientation on their profiles, but even when they do not, facebook ‘likes’ can be used to accurately predict sexual orientation in 88 per cent of cases (kosinski, stillwell and graepel 2013). in a more controversial paper, wang and kosinski (2018) applied neural networks to detect sexual orientation indirectly from photographs of faces on facebook profiles. clearly, approaches such as these raise some challenging ethical questions, and also questions about data reliability and representativeness. future research should explore these new data sources while also continuing to rely on the strengths of existing survey and census data. key messages • sexual minority population estimates are useful for a variety of health, social justice and economic purposes, and for broader analyses of demographic trends and theory. • australia’s non-heterosexual population aged 18+ in 2016 is estimated to have been 592,000, representing about 3.2 per cent of the adult population. • the percentages identifying as non-heterosexual are higher at younger ages and lower at older ages, with the age gradient being more pronounced for females. • as would be expected, the most populous states, new south wales, victoria and queensland, are home to the largest non-heterosexual populations, while the act, tasmania and the northern territory have the smallest populations. • as a fraction of the total population, south australia, western australia and the northern territory have the smallest percentages identifying as non-heterosexual (2.6%), while the act has the highest (5.1%). ethics approval approval for this study was received from the charles darwin university human research ethics committee (approval h17122). contact: ethics@cdu.edu.au. acknowledgements we gratefully acknowledge helpful advice received from emeritus professor gary gates during the course of this study, and the helpful comments of the anonymous reviewers on an earlier version of this paper. our thanks and appreciation is also extended to hamish mcmanus from the kirby institute, unsw, who extracted data from the ashr2, and laura bennetts-kneebone from the commonwealth department of social services who extracted data from the hilda survey. mailto:ethics@cdu.edu.au 36 wilson t and shalley f australian population studies 2 (1) 2018 references abs (australian bureau of statistics) (2015a) general social survey, 2014. cat. no. 4159.0. canberra: abs. abs (2015b) general social survey household questionnaire. cat. no. 4159.0. canberra: abs. http://www.abs.gov.au/ausstats/abs@.nsf/detailspage/4159.02014?opendocument. accessed on 8 february 2018. abs (2017a) australian marriage law postal survey, 2017. cat. no. 1800.0. canberra: abs. abs (2017b) tablebuilder pro. canberra: abs. http://www.abs.gov.au/websitedbs/censushome.nsf/home /tablebuilder. accessed on 8 february 2018. abs (2017c) abs.stat. canberra: abs. available at: http://stat.data.abs.gov.au/. accessed on 8 february 2018. abs (2018) census of population and housing: reflecting australia – stories from the census, 2016. cat. no. 2071.0. canberra: abs. http://www.abs.gov.au/ausstats/abs@.nsf/lookup/2071.0main+featur es852016?opendocument. accessed on 8 february 2018. american psychological association (2008) answers to your questions: for a better understanding of sexual orientation and homosexuality. washington dc: american psychological association. http://www.apa.org/topics/lgbt/orientation.aspx. accessed on 8 february 2018. aspinall p (2009) estimating the size and composition of the lesbian, gay, and bisexual population in britain. equality and human rights commission research report 37. manchester: equality and human rights commission. baumle a k (ed.) (2013) international handbook on the demography of sexuality. dordrecht: springer. baumle a k, compton d and poston d l (2009) same-sex partners: the social demography of sexual orientation. albany ny: suny press. black d, gates g j, sanders s and taylor l (2000) demographics of the gay and lesbian population in the united states: evidence from available systematic data sources. demography 37(2): 139–154. black d, gates g j, sanders s and taylor l (2002) why do gay men live in san francisco? journal of urban economics 51(1): 54–76. carpenter c and gates g j (2008) gay and lesbian partnership: evidence from california. demography 45(3): 573–590. coleman d a and salt j (1996) ethnicity in the 1991 census. vol. 1. demographic characteristics of the ethnic minority population. london: hmso. cooke t (2005) migration of same-sex couples. population, space and place 11(5): 401–409. cooke t and rapino m (2007) the migration of partnered gays and lesbians between 1995 and 2000. the professional geographer 59(3): 285–297. dempsey d (2015) familiarly queer? same-sex relationships and family formation. in: heard g and arunachalam d (eds) family formation in 21st century australia. dordrecht: springer; 225–240. dibennardo r and gates g j (2014) us census same-sex couple data: adjustments to reduce measurement error and empirical implications. population research and policy review 33(4): 603–614. durso l e and gates g j (2013) best practices: collecting and analyzing data on sexual minorities. in: baumle a k (ed.) international handbook on the demography of sexuality. dordrecht: springer; 21–42. edmonston b (2016) canada’s immigration trends and patterns. canadian studies in population 43(1–2): 78–116. fenway health (2010) glossary of gender and transgender terms. boston, ma: fenway health. http://fenwayhealth.org/documents/the-fenway-institute/handouts/handout_7c_glossary_of_gender_and_transgender_terms__fi.pdf. accessed on 8 february 2018. flores a r, herman j l, gates g j and brown t n t (2016) how many adults identify as transgender in the united states? the williams institute, ucla school of law, los angeles. https://williamsinstitute.law.ucla.edu/wp-content/uploads/how-many-adults-identify-astransgender-in-the-united-states.pdf. accessed on 8 february 2018. http://www.abs.gov.au/ausstats/abs@.nsf/detailspage/4159.02014?opendocument http://www.abs.gov.au/websitedbs/censushome.nsf/home/tablebuilder http://www.abs.gov.au/websitedbs/censushome.nsf/home/tablebuilder http://stat.data.abs.gov.au/ http://www.abs.gov.au/ausstats/abs@.nsf/lookup/2071.0main+features852016?opendocument http://www.abs.gov.au/ausstats/abs@.nsf/lookup/2071.0main+features852016?opendocument http://www.apa.org/topics/lgbt/orientation.aspx http://fenwayhealth.org/documents/the-fenway-institute/handouts/handout_7-c_glossary_of_gender_and_transgender_terms__fi.pdf http://fenwayhealth.org/documents/the-fenway-institute/handouts/handout_7-c_glossary_of_gender_and_transgender_terms__fi.pdf https://williamsinstitute.law.ucla.edu/wp-content/uploads/how-many-adults-identify-as-transgender-in-the-united-states.pdf https://williamsinstitute.law.ucla.edu/wp-content/uploads/how-many-adults-identify-as-transgender-in-the-united-states.pdf australian population studies 2 (1) 2018 wilson t and shalley f 37 gates g j (2010) same-sex couples in us census bureau data: who gets counted and why. the williams institute, ucla school of law, los angeles. https://williamsinstitute.law.ucla.edu/research/census-lgbt-demographics-studies/same-sexcouples-in-us-census-bureau-data-who-gets-counted-and-why/. accessed on 8 april 2010. gates g j (2011) how many people are lesbian, gay, bisexual and transgender? the williams institute, ucla school of law, los angeles. https://williamsinstitute.law.ucla.edu/research/census-lgbtdemographics-studies/how-many-people-are-lesbian-gay-bisexual-and-transgender/. accessed on 21 august 2016. gates g j (2012) lgbt identity: a demographer’s perspective. loyola law review los angeles 42(3): 693–714. gates g j (2014) lgbt demographics: comparisons among population-based surveys. the williams institute, ucla school of law, los angeles. https://williamsinstitute.law.ucla.edu/wpcontent/uploads/lgbt-demogs-sep-2014.pdf. accessed on 8 february 2018. geary r s, tanta c, erens b, clifton s, frah p, wellings k, mitchell k r, datta j, gravningen k, fuller e, johnson a m, sonnenberg p and mercer c h (2018) sexual identity, attraction and behaviour in britain: the implications of using difference dimensions of sexual orientation to estimate the size of sexual minority populations and inform public health interventions. plos one 13(1): e0189607. grulich a e, de visser r o, smith m a, rissel c e and richters j (2003) sex in australia: homosexual experience and recent homosexual encounters. australian and new zealand journal of public health 27(2): 155–163. jivraj s and simpson l (2015) (eds) ethnic identity and inequalities in britain: the dynamics of diversity. bristol: policy press. kosinski m, stillwell d and graepel t (2013) private traits and attributes are predictable from digital records of human behaviour. pnas 110(15): 5802–5805. madeddu d, grulich a, richters j, ferris j, grierson j, smith a, allan b and prestage g (2006) estimating population distribution and hiv prevalence among homosexual and bisexual men. sexual health 3(1): 37–43. marriage act 1961 (cwth). compilation prepared on 9 december 2017 taking into account alterations up to act no. 129 of 2017. https://www.legislation.gov.au/details/c2017c00382. accessed on 28 january 2017. meier s c and labuski c m (2013) the demographics of the transgender population. in: baumle a k (ed.) international handbook on the demography of sexuality. dordrecht: springer; 289–327. melbourne institute (2012) hilda self-completion questionnaire. http://melbourneinstitute.unimelb.edu.au/assets/documents/hildaquestionnaires/selfcompletionquestionnairew12m.pdf. accessed on 8 february 2018. ons (office for national statistics) (2010) measuring sexual identity: an evaluation report. london: ons. www.ons.gov.uk/ons/guidemethod/measuring-equality/equality/sexual-identityproject/measuring-sexualidentity–an-evaluation-report.pdf. accessed on 8 february 2018. ons (office for national statistics) (2017) subnational sexual identity estimates, uk: 2013 to 2015. https://www.ons.gov.uk/peoplepopulationandcommunity/culturalidentity/sexuality/articles/sub nationalsexualidentityestimates/uk2013to2015. accessed on 8 february 2018. peach c (2006) islam, ethnicity and south asian religions in the london 2001 census. transactions of the institute of british geographers 31(3): 353–370. perales f and baxter j (2018) sexual identity and relationship quality in australia and the united kingdom. family relations 67(1): 55–69. pew research center (2017) europe’s growing muslim population. washington dc: pew research center. http://www.pewforum.org/2017/11/29/europes-growing-muslim-population/. accessed on 8 february 2018. prestage g, ferris j, grierson j, thorpe r, zablotska i, imrie j, smith a and grulich a e (2008) homosexual men in australia: population, distribution and hiv prevalence. sexual health 5(2): 97–102. https://williamsinstitute.law.ucla.edu/research/census-lgbt-demographics-studies/same-sex-couples-in-us-census-bureau-data-who-gets-counted-and-why/ https://williamsinstitute.law.ucla.edu/research/census-lgbt-demographics-studies/same-sex-couples-in-us-census-bureau-data-who-gets-counted-and-why/ https://williamsinstitute.law.ucla.edu/research/census-lgbt-demographics-studies/how-many-people-are-lesbian-gay-bisexual-and-transgender/ https://williamsinstitute.law.ucla.edu/research/census-lgbt-demographics-studies/how-many-people-are-lesbian-gay-bisexual-and-transgender/ https://williamsinstitute.law.ucla.edu/wp-content/uploads/lgbt-demogs-sep-2014.pdf https://williamsinstitute.law.ucla.edu/wp-content/uploads/lgbt-demogs-sep-2014.pdf https://www.legislation.gov.au/details/c2017c00382 http://melbourneinstitute.unimelb.edu.au/assets/documents/hilda-questionnaires/selfcompletionquestionnairew12m.pdf http://melbourneinstitute.unimelb.edu.au/assets/documents/hilda-questionnaires/selfcompletionquestionnairew12m.pdf http://www.ons.gov.uk/ons/guidemethod/measuring-equality/equality/sexual-identity-project/measuring-sexualidentity–an-evaluation-report.pdf http://www.ons.gov.uk/ons/guidemethod/measuring-equality/equality/sexual-identity-project/measuring-sexualidentity–an-evaluation-report.pdf https://www.ons.gov.uk/peoplepopulationandcommunity/culturalidentity/sexuality/articles/subnationalsexualidentityestimates/uk2013to2015 https://www.ons.gov.uk/peoplepopulationandcommunity/culturalidentity/sexuality/articles/subnationalsexualidentityestimates/uk2013to2015 http://www.pewforum.org/2017/11/29/europes-growing-muslim-population/ 38 wilson t and shalley f australian population studies 2 (1) 2018 richters j, altman d, badcock p b, smith a m a, de visser r o, grulich a e, rissel c and simpson j m (2014) sexual identity, sexual attraction and sexual experience: the second australian study of health and relationships. sexual health 11(5): 451–460. richters j, badcock p b, simpson j m, shellard d, rissel c, de visser r o, grulich a e and smith a m a (2014) design and methods of the second australian study of health and relationships. sexual health 11(5): 383–396. smith a m a, rissel c r, richters j, grulich a e and de visser r o (2003) sex in australia: sexual identity, sexual attraction and sexual experience among a representative sample of adults. australian and new zealand journal of public health 27(2): 138–145. smith l r (1980) the aboriginal population of australia. canberra: anu press. statistics canada (2017) sexual orientation. http://www.statcan.gc.ca/eng/dai/smr08/2015/smr08_203_2 015#a3. accessed on 8 february 2018. taylor j (2011) postcolonial transformation of the australian indigenous population. geographical research 49(3): 286–300. van kampen s, fornasiero m, lee w and husk k (2017) producing modelled estimates of the size of the lesbian, gay and bisexual (lgb) population of england: final report. london: public health england. wang y and kosinski m (2018) deep neural networks are more accurate than humans at detecting sexual orientation from facial images. journal of personality and social psychology 114(2): 246–257. wilkins r (2015) the household, income and labour dynamics in australia survey: selected findings from waves 1 to 14. 10th annual statistical report of the hilda survey. melbourne institute, the university of melbourne. wilkins r (2017) the household, income and labour dynamics in australia survey: selected findings from waves 1 to 15. 12th annual statistical report of the hilda survey. melbourne institute, the university of melbourne. wilson s (2004) gay, lesbian, bisexual and transgender identification and attitudes to same-sex relationships in australia and the united states. people and place 12(4): 12–21. wilson t (2016) visualising the demographic components of change shaping the age structures of australian state and territory populations. poster presented at the australian population conference, sydney, december 2016. https://tinyurl.com/y7622g4z. accessed on 8 february 2018. wooden m (2014) the measurement of sexual identity in wave 12 of the hilda survey (and associations with mental health and earnings). hilda project discussion paper series no. 1/14, faculty of business and economics, the university of melbourne. http://www.statcan.gc.ca/eng/dai/smr08/2015/smr08_203_2015#a3 http://www.statcan.gc.ca/eng/dai/smr08/2015/smr08_203_2015#a3 https://tinyurl.com/y7622g4z australian population studies 2018 | volume 2 | issue 2 | pages 22-32 © allen 2018. published under the creative commons attribution-noncommercial licence 3.0 australia (cc by-nc 3.0 au). journal website: www.australianpopulationstudies.org commentary a whole-of-government approach to population policy for australia liz allen* the australian national university * corresponding author. email: liz.allen@anu.edu.au. address: beryl rawson building (#13), centre for social research and methods, the australian national university, act 2606, australia. paper received 16 august 2018; accepted 3 november 2018; published 12 november 2018 1. introduction australia is in the midst of a popularly-constructed crisis of population. this so-called crisis, in social terms, is not dissimilar to the many which have come before in australia’s history. yet australia’s contemporary crisis differs in that the focus of the oft-referred-to population debate has come to revolve primarily around immigration and the politics of problematising population. the problematisation of population occurs in many countries across the world, and is much about immigration and the effects of migrants on settlement areas, particularly as fertility has declined. constructing population as a social problem means any consideration of population-related matters typically results in the portrayal of population, specifically immigration, as the root cause of social ills. talk of population has, unhelpfully, come to focus on race and ethnicity. for example, with ambivalence japan has widened immigration intake to allow additional visas for much-needed skilled workers (murakami 2018). despite the need for migrants, japanese people are concerned about the effects of foreign workers. concerns relate to whether migrant workers will fit in and what pressures they might place on the pension and welfare scheme. apparent concerns also relate to how immigration will change japan. in a more overt example of problematisation, the united states president donald trump has framed immigration as a threat to the american people (bennett 2017). population growth in australia, fuelled by net overseas migration, has prompted calls from politicians and social commentators across the political spectrum for a population inquiry and policy debate. questions have been raised over what might be the best, and or, sustainable population size for australia; and more importantly, whether greater intervention is needed to restrict the level of migration and the source countries of migrants. current calls for a population policy reflect wider social concerns which have come to be conflated with population growth (especially due to immigration), including housing affordability, adequacy of public infrastructure, and environmental conservation. a dominant narrative has emerged: inequality is further exacerbated by migration, conjuring nationalist and protectionist sentiments. http://www.australianpopulationstudies.org/ mailto:liz.allen@anu.edu.au australian population studies 2 (2) 2018 allen 23 the present social and political contexts are such that a population policy is likely for the first time in the country’s recent history. the government has indeed signalled a population policy (focused on migrant settlement) will be put forward with haste (benson 2018). yet exactly what constitutes an appropriate population policy for contemporary australia is still yet to be determined. the breadth of population-related issues lends itself to an approach which seeks to embed and contextualise population among wider government business – which requires a whole-of-government, joined-up approach. 2. background population policies, as a general principle, seek to address or influence components of population change, namely births, deaths and migration (demeny 1975). the concept of population policies conjures notions of restrictive and coercive government interventions designed to control various aspects of people’s lives. it is considered that australia already has a quasi or de facto population policy by way of the migration program (allen 2011; mcdonald 2003; productivity commission 2016). the migration program is currently capped at 190,000 permanent migrants per year. while this cap has gone unchanged since 2012, the actual intake has fluctuated below the cap in recent years (treasury 2018). despite the spectre of population as an ongoing feature in the australian political landscape, australia has not had an official population policy since the early-1970s (allen 2011; jones 1997). this most recent population policy was established following the second world war when a growth target of 2 per cent was set to bolster population growth, equally through births and immigration. australia’s last population policy was in response to fears australia was unable to defend itself against the threat of external foreign power (jones 1997). population inquiries are dotted throughout australia’s history (allen 2017). the most recent, the sustainable population strategy for australia, was completed in 2011 after concerns over a ‘big’ australia, and echoed the findings of previous inquiries (allen 2011). hundreds of submissions were received from a wide range of organisations and individuals. grand statements and commitments were made as part of the latest inquiry, relating to how the government of the day would address the population crisis. no population targets were set (dsewpc 2011). the year-long process and fanfare of the report release conjured a feeling of change (allen 2017). the sense of change was short-lived as it became clear the population strategy was in name only, as the existing policy environment continued. the enduring finding of australia’s population inquiries has been that preparedness and responsiveness are central for the nation to weather demographic difficulties (allen 2017). quality population data – especially from censuses, births, deaths and migration, and population projections – are vital to being prepared and responsive to the demographic needs of the future. intergenerational reports (igr) published by the federal treasury department, roughly every five years, are an important part of informing strategic direction using demographic and economic data about now and what might be into the future. the most recent igr, in 2015, was accompanied by the public awareness campaign challenge of change (treasury 2015). public backlash over the costs, lack of environmental considerations, and political tone of the campaign doomed the only real attempt australia has seen in recent time to engage the nation in a conversation about its demographic future. and what australia is presented with, in demographic terms is about managing change. 24 allen australian population studies 2 (2) 2018 popular and political discourse portrays a crisis of overpopulation in contemporary australia. focus on australia’s size and growth has undeniable political motivations, particularly in the present response from the minor and major political parties. leaders from the two major political parties have commented publicly on immigration and its impacts on australia. the leader of the opposition, bill shorten, has expressed concern that the number of overseas migrants is adversely impacting on wages and contributing to insecure and inaccessible paid work for locals. shorten has argued for a reduction in temporary migration (baxendale 2018). in contrast, the conservative prime minister, scott morrison, supports the current migration program. as treasurer, morrison promoted the economic benefits of current overseas migration levels (murphy 2018). the minister responsible for the broad home affairs portfolio, peter dutton, has called for lowering migrant intake as well as a preference for people from racially similar backgrounds as the majority australian population (koziol 2018). the minister directly responsible for immigration, david coleman, advocates the social and economic value of multiculturalism in australia (acharya 2018). conversely alan tudge, the minister responsible for population (among cities and urban infrastructure), has been colloquially appointed as the congestion-buster (elton-pym 2018). tudge has proposed australia makes more concerted efforts to encourage settlement of overseas migrants into regional areas, or in the least areas outside the three major cities of sydney, melbourne and brisbane. overpopulation in the three major cities has featured among tudge’s rhetoric. minor parties in australia have put forward a number of proposals during 2018 concerning immigration and race. an immigration plebiscite was proposed by the leader of the minor nationalist party, pauline hanson. hanson’s proposed immigration plebiscite reflects growing sentiment among those opposed to current migration intake that all australians should have a say about the level of immigration. hanson also sought recognition of the australian parliament that it is ‘ok to be white’ (karp 2018a). while the motion was inadvertently supported by the government, parliament voted narrowly against it. additionally, a failed attempt to hold a plebiscite to limit immigration from only european countries was proposed by fraser anning, a member of another minor australian nationalist party (mcculloch 2018). anning has since been expelled from his party due to his raciallymotivated public commentary (karp 2018b). data tell an interesting story, and contextualise the populist problematisation of population in australia. population growth in australia at 1.6% (for the year to 30 june 2017) is higher than the world average and considerably higher than countries among those in the organisation for economic co-operation and development (oecd), in particular canada, the united kingdom, and the united states (allen 2018a). figure 1 shows population growth in australia has been strong over the 10 years to 2017 (in the context of annual growth since 1945), despite the global financial crisis (gfc). it is not surprising that australia emerged from the gfc comparatively unscathed when compared with similar countries. strong population growth and economic measures taken by the then government were major factors. driving the majority of population growth over the past 12 years has been net overseas migration (abs 2018a). figure 2 presents the contributions of net overseas migration and natural increase to total population growth since 1975, which was when fertility dropped to replacement level before falling to below-replacement levels, where it has remained since (allen 2018a). australian population studies 2 (2) 2018 allen 25 figure 1: annual population growth, australia, 1945-2017 source: abs 2014; abs 2018a. notes: annual figures to 30 june. figure 2: components of population change, australia, 1975-2017 source: abs 2014; abs 2018a. notes: annual figures to 30 june. the definition used in calculating estimates of net overseas migration (nom) changed in 2006. caution should be taken when comparing nom over time. overseas migration has become a vital component of the population and economic landscape in australia as a response to population ageing. population ageing is increasing old-age related dependency, meaning growing pressure on the working age population to contribute income tax to fund government-provided essential services (allen 2017). recent immigration in australia has been shown to affect greater rates of labour force participation (treasury 2018; productivity commission 26 allen australian population studies 2 (2) 2018 2016), helping to moderate the adverse economic pressures of an ageing population by way of the co-called dependency burden (allen 2018b). figure 3 illustrates the increasing dependency rates over the 1975 to 2017 period. figure 3: age-related dependency ratios, australia, 1975-2017 source: abs 2014; abs 2018a. notes: annual figures to 30 june. government spending is highest for people aged over 65 years according to rice, temple and mcdonald (2014; cited in treasury 2015). increasing old-age dependency thus has clear consequences for the fiscal pressures on government spending and flow on impacts for people in the labour force contributing income tax. furthermore, a worrying risk emerges whereby young people – the future workforce – compete for government spending for more pressing and costly demographic demands. socioeconomic and demographic inequality thus becomes a risk facing australia. alongside this inequality is the unequal opportunities experienced throughout australian towns and communities. the three major cities along australia’s east coast welcome the largest share of immigration. population growth is greatest in the capital cities of melbourne (2.7% in the year to 30 june 2017), sydney (2.0%) and brisbane (2.0%). conversely, areas outside these capitals experience vastly different growth rates, and even decline (as is the case in areas outside the capital in the northern territory). furthermore, there are regions across australia with declining or struggling populations, largely due to changing labour force supply and demand. planning and infrastructure are playing catch-up in australia. rather than face head on the infrastructure and planning challenges posed by population growth, it has been suggested that the strategy in recent decades has been for the states with the majority share of immigration to ignore infrastructure needs (hartcher 2018). this has resulted in two-speed population growth story. cities australian population studies 2 (2) 2018 allen 27 are under pressure to accommodate population, whereas regional areas are under pressure to retain or attract population. unequal population distribution across australia highlights population pressure points and exacerbates the notion that population is problematic. there are have been a number of attempts in recent decades to encourage migrants to establish themselves initially in regional areas upon moving to australia, with the hope that immigrants might establish themselves and stay on in these areas. towns have initiated migrant settlements, including the highly publicised luv-a-duck business which offers a good case study into how small towns can attract population, with adequate employment opportunities. the nhill community in rural western victoria teamed up with the local duck farm and processing plant to settle karen refugees from myanmar in the area to fill much needed skills at luva-duck (ames & deloitte access economics 2015). the nhill example highlights the need for viable opportunities for migrants to maximise success. destination areas must offer essential services and infrastructure for migrants to be able to establish themselves socially, culturally and economically. similarly, such opportunities could also attract local internal migration. 3. a population policy for the future the demographic pressures of population ageing in australia demand a contemporary rethink of the approach to the development of population policy. the issue of population cannot be merely reduced to numbers or demographic statistics about births, deaths and migration. nor is population just about age composition, population distribution, and labour force participation. migration remains an ongoing feature of a population policy for australia, as does infrastructure which supports the population (physical, social, environmental or economic). australia must look to make opportunities of its demographic challenges by being prepared and responsive to population composition which might adversely impact national socioeconomic wellbeing. considerations of a population policy for australia include: health, education, gender equality, family-friendly workplaces, transportation, housing, land use, green space, water supply, food production, energy use, environmental sustainability, settlement distribution, and climate change, and many more. population policy reflects myriad complex social issues which sit across numerous ministerial portfolios, necessitating a whole-of-government approach to population policy. the reasons most recent attempts at population policy in australia have been unsuccessful might lie in the fact that a coordinated approach is necessary and has not been attempted or achieved. a reconnection of the disparately placed pieces of the population puzzle might be the best way forward for a contemporary population policy for australia. carey, mcloughlin & crammond (2015 p. 176) point to how ‘joinedup government’ is required to address complex or ‘wicked’ social problems, ‘and overcome siloed departmental approaches’. carey and colleagues demonstrate how socially complex issues which cross departmental boundaries and have interdependencies between issues can be successfully addressed through whole-ofgovernment, joined-up approaches. drawing on the australian government’s approach to the complex problem of social inclusion, carey et al. (2015) show how multidimensional, complex and contested social issues can effectively move beyond impasse via what the authors refer to as innovation narratives. 28 allen australian population studies 2 (2) 2018 innovation narratives enable a change to how a complex issue is perceived, and so shift values to facilitate the desired change (carey, mcloughlin & crammond 2015). carey and colleagues warn that innovation narratives are not about behaviour modification, but rather they can challenge thinking with the aim to change perceptions. innovation narratives are effective at gaining coherence within policyand decision-makers across government departments, but they can also be extended to community involvement. storytelling and communication are crucial to innovation narratives through the coordination of people and ideas and dissemination of information. while innovation narratives are shown to be effective among policy makers (carey 2016; carey, mcloughlin & crammond 2015), such an approach to elicit perception change could be trialled to bring the population policy domain together. it is unlikely, nor ideal, that all australians will agree on population matters. what is possible (and more helpful) is the mobilisation of the australian public with the common goal in mind that at the heart of population matters is the one aim to achieve a fair australia in which wellbeing is the paramount concern. through this process, population ceases to be the issue. what becomes the focus, instead, is the infrastructure and support for australians. more importantly, government core business is to deliver for the people, not blame people (locals or migrants). despite the failed attempt to promote the evolving needs associated with a changing demography under the guise of the challenge of change campaign, the aim of such a campaign is a positive move forward (allen 2017). a worthwhile endeavour would be to reconsider the option of innovation narratives through a change management communication strategy. this requires politicians to refrain from framing population (particularly immigration) as problematic. ideally, such an undertaking would be non-partisan. however, a non-partisan approach to population-related matters is unlikely. nonetheless, an attempt to do so would be worthwhile. because population features in many areas of government, an effective coordinated approach to policy is difficult. establishing a minister, ministerial body, or agency with oversight to ensure adequate progress within a population framework is one solution to overcome, or in the very least, minimise the difficulties of the breadth of population-related matters in policy. such a framework would not be easy, though would position australia well for the demographic challenges of the future. carey et al. (2015 p. 184) suggest the development of ‘supportive architecture’ to maximise the potential success of a whole-of-government approach. lines of accountability should be among the first things to establish. new administrative structures – and thus new ways of doing population policy – to support clear accountability can help achieve integration and a breaking down of departmental silos (carey, mcloughlin & crammond 2015). possible points of failure in a whole-ofgovernment approach to population policy include: inadequate government and/or departmental commitment, opposing ideologies among ministers, and rigid organisational structures. according to carey and colleagues (2015 p. 183) apart from a mandate for change embedded in policy, ‘establishing coherence between institutional and operational level action’ is essential to implementing a whole-of-government approach to population policy for australia. to this end, flexibility, coherence and communication are vital. this new way of approaching population policy requires evidence and a framework for assessing achievements. the development of a framework with a set of measureable indicators to report and monitor success towards population-related wellbeing could provide an opportunity for the public to remain australian population studies 2 (2) 2018 allen 29 informed of australia’s progress. such a framework, and related indictors, could be modelled after the global sustainable development goals (undp 2018) and be reported on as part of the igr process. release of five-yearly igrs could coincide with the release of australian bureau of statistics population projections. publication of data and information informing the igr and evaluation framework via a publically accessible website would help facilitate evidence-sharing for communities across australia. an overall population target, or even a target for growth is unnecessary, and has the potential to lead to coercive or restrictive measures. the current approach for setting the annual permanent migration cap is sensible, and allows for flexibility to enable a temporary migrant intake which is demand-driven. a focus on size and growth rates has been a major factor in the current population discourse and has had the effect of shifting the attention away from productive action onto matters of race and ethnicity. this numbers game is harmful. optimal or carrying capacity population size is not determinable for australia as technological advancements prohibit any sensible calculation. it may be that innovation is the major limitation for growth. additionally, there is a paucity of good quality up-to-date research identifying the best migrant intake for australia to maximise social and economic benefits. an effective migration scheme should be informed by evidence. a program of research to inform such an evidence base for australia’s migration program would guarantee the necessary investments were made in a timely and responsive manner and the intake did not lead to adverse consequences. 4. conclusion population policy is necessary in australia, not only because of the current political and populist climate, but because of the demographic challenges which lie ahead. an official population policy would address both the present problematising of demographic difficulties. the so-called population crisis is not so much a crisis of numbers or people but rather a crisis of long-term inaction. a population policy for australia should provide a blueprint for the future – a framework for what the nation aspires to be. the process to create a population policy for australia need not be by way of a formal inquiry. more importantly, consensus or majority public opinion is not necessary for the type of population policy proposed here. a public process of sorts, nonetheless, is advised. current population discourse would be well served through innovation narratives seeking to shift the perceived problem of population. such innovation narratives could be facilitated by an administrative population body within government. whole-of-government oversight to ensure communication and action is effective across ministerial portfolios is a necessary component of a population policy. evidence gathering from and by experts, alongside a process of community communication through innovation narratives, would provide a solid foundation for which government could establish a framework for the future of australia. the most effective population policy for australia is one which embeds population-related concerns and issues in the foundation of funding and policy-making processes. in doing so, population is no 30 allen australian population studies 2 (2) 2018 longer stigmatised. instead, population is placed as the basis for government plans for the future, and evidence will help guide the path forward. 5. key messages • a whole-of-government approach, reflecting population-related interdependencies across government departments, is required to achieve an effective response to the complex social issues of population. • a framework for the implementation of a whole-of-government population policy requires appropriate administrative structures enabling flexibility, coherence, and communication across siloed government departments. • research and evidence focussed on the myriad considerations of a contemporary population policy is crucial to successful policy outcomes. references abs (2014) australian historical population statistics, 2014. catalogue no. 3105.0.65.001. canberra: abs. abs (2018a) australian demographic statistics, sep 2017. catalogue no. 3101.0. canberra: abs. abs (2018b) regional population growth, australia, 2016-17. catalogue no. 3218.0. canberra: abs. allen l (2011) sustainable population strategy: public policy and implementation challenges. assa academy proceedings 2/2011. https://www.assa.edu.au/wpcontent/uploads/2017/03/ap22011.pdf. accessed on 9 august 2018. allen l (2017) australia doesn’t have a population policy – why? the conversation, 3 july. https://theconversation.com/australia-doesnt-have-a-population-policy-why-78183. accessed on 9 august 2018. allen l (2018a) factcheck: is australia’s population the ‘highest growing in the world’? the conversation, 13 july. https://theconversation.com/factcheck-is-australias-population-the-highest-growing-inthe-world-96523. accessed on 9 august 2018. allen l (2018b) migration helps balance our ageing population – we don’t need a moratorium. the conversation, 23 july. https://theconversation.com/migration-helps-balance-our-ageingpopulation-we-dont-need-a-moratorium-100030. accessed on 9 august 2018. ames & deloitte access economics (2015) small towns, big returns: economic and social impact of the karen settlement in nhill. melbourne: ames & deloitte access economics. acharya m (2018) new immigration minister hails migrants’ contribution to australia. sbs news, 7 september. https://www.sbs.com.au/yourlanguage/hindi/en/article/2018/09/07/newimmigration-minister-hails-migrants-contribution-australia. accessed on 26 october 2018. baxendale r (2018) work right visas out of control, bill shorten says amid migrant debate. the australian, 16 july. https://www.theaustralian.com.au/national-affairs/immigration/migration-programflexible-michael-sukkar-says/news-story/5498966ab70064120f41c30db50e5f3c. accessed on 15 august 2018. bennett b (2017) trump says immigration system is a national security threat, but goes easy on russia. la times, 18 december. http://www.latimes.com/politics/la-fg-trump-america-first-20171217story.html. accessed on 26 october 2018. benson s (2018) migrants to be steered to regions. the australian, 13 august. https://www.theaustralian.com.au/national-affairs/immigration/population-policy-to-steermigrants-toward-regional-centres/news-story/38d5f4b4c808089cb530ad75fe13f0ec. accessed on 13 august 2018. https://www.assa.edu.au/wp-content/uploads/2017/03/ap22011.pdf https://www.assa.edu.au/wp-content/uploads/2017/03/ap22011.pdf https://theconversation.com/australia-doesnt-have-a-population-policy-why-78183 https://theconversation.com/factcheck-is-australias-population-the-highest-growing-in-the-world-96523 https://theconversation.com/factcheck-is-australias-population-the-highest-growing-in-the-world-96523 https://theconversation.com/migration-helps-balance-our-ageing-population-we-dont-need-a-moratorium-100030 https://theconversation.com/migration-helps-balance-our-ageing-population-we-dont-need-a-moratorium-100030 https://www.sbs.com.au/yourlanguage/hindi/en/article/2018/09/07/new-immigration-minister-hails-migrants-contribution-australia https://www.sbs.com.au/yourlanguage/hindi/en/article/2018/09/07/new-immigration-minister-hails-migrants-contribution-australia https://www.theaustralian.com.au/national-affairs/immigration/migration-program-flexible-michael-sukkar-says/news-story/5498966ab70064120f41c30db50e5f3c https://www.theaustralian.com.au/national-affairs/immigration/migration-program-flexible-michael-sukkar-says/news-story/5498966ab70064120f41c30db50e5f3c http://www.latimes.com/politics/la-fg-trump-america-first-20171217-story.html http://www.latimes.com/politics/la-fg-trump-america-first-20171217-story.html https://www.theaustralian.com.au/national-affairs/immigration/population-policy-to-steer-migrants-toward-regional-centres/news-story/38d5f4b4c808089cb530ad75fe13f0ec https://www.theaustralian.com.au/national-affairs/immigration/population-policy-to-steer-migrants-toward-regional-centres/news-story/38d5f4b4c808089cb530ad75fe13f0ec australian population studies 2 (2) 2018 allen 31 carey g (2016) grassroots to government. melbourne: melbourne university press. carey g, mcloughlin p and crammond b (2016) implementing joined-up government: lessons from the australian social inclusion agenda. australian journal of public administration 74(2):176-186. demeny p (1975) population policy: the role of national governments. population and development review 1(1):147-161. dsewpc (2011) sustainable australia – sustainable communities: a sustainable population strategy for australia. canberra: department of sustainability, environment, water, population and communities. elton-pym j (2018) new visa conditions to force migrants to settle in regions under pm's population plan. sbs news, 10 october. https://www.sbs.com.au/yourlanguage/korean/en/article/2018/10/09/sewage-study-findsopioid-use-has-doubled-regional-australia. accessed on 26 october 2018. hartcher p (2018) plan or perish: management can solve population anxiety. the sydney morning herald, 11 august. www.smh.com.au/national/plan-or-perish-management-can-solve-populationanxiety-20180809-p4zwll.html. accessed on 12 august 2018. jones g (1997) an australian population policy. parliamentary library, research paper 17. canberra: australian government. karp p (2018a) 'ok to be white': australian government senators condemn 'anti-white racism'. the guardian, 15 october. https://www.theguardian.com/australia-news/2018/oct/15/ok-to-bewhite-australian-government-senators-condemn-anti-white-racism. accessed on 26 october 2018. karp p (2018b) ‘australian senator who called for 'final solution' to immigration expelled from party’, the guardian, 25 october. https://www.theguardian.com/australia-news/2018/oct/25/australiansenator-who-called-for-final-solution-to-immigration-expelled-from-party. accessed 26 october 2018. koziol m (2018) peter dutton wants a refugee intake of white south african farmers. the sydney morning herald, 15 march. https://www.smh.com.au/politics/federal/peter-dutton-wants-a-refugeeintake-of-white-south-african-farmers-20180315-p4z4gr.html. accessed on 15 august 2018. mcculloch d (2018) senate sinks anning’s ‘white australia’ plebiscite. the new daily, 18 october. https://thenewdaily.com.au/news/national/2018/10/18/senate-sinks-white-australia-plebiscite/. accessed on 26 october 2018. mcdonald p (2003) australia’s future population: population policy in low-fertility society. in: khoo s and mcdonald p (eds) the transformation of australia’s population 1970 – 2030. sydney: unsw press; 266-279. murakami s (2018) liberal democratic party’s general council approves draft immigration control bill despite members’ concerns. the japan times, 30 october. https://www.japantimes.co.jp/news/2018/10/30/national/politics-diplomacy/liberal-democraticpartys-general-council-approves-draft-immigration-control-bill-despite-membersconcerns/#.w9lhci9l1yg. accessed on 31 october 2018. murphy k (2018) peter dutton contradicts morrison and says migrant cut positive for the economy. the guardian, 15 july. https://www.theguardian.com/australia-news/2018/jul/15/peter-duttoncontradicts-morrison-and-says-migrant-cut-positive-for-the-economy. accessed on 15 august 2018. productivity commission (2016) migrant intake into australia. inquiry report no. 77. canberra: productivity commission. rice j m, temple j and mcdonald p (2014) national transfer accounts for australia. canberra: arc centre of excellence in population research. https://openresearchhttps://www.sbs.com.au/yourlanguage/korean/en/article/2018/10/09/sewage-study-finds-opioid-use-has-doubled-regional-australia https://www.sbs.com.au/yourlanguage/korean/en/article/2018/10/09/sewage-study-finds-opioid-use-has-doubled-regional-australia http://www.smh.com.au/national/plan-or-perish-management-can-solve-population-anxiety-20180809-p4zwll.html http://www.smh.com.au/national/plan-or-perish-management-can-solve-population-anxiety-20180809-p4zwll.html https://www.theguardian.com/australia-news/2018/oct/15/ok-to-be-white-australian-government-senators-condemn-anti-white-racism https://www.theguardian.com/australia-news/2018/oct/15/ok-to-be-white-australian-government-senators-condemn-anti-white-racism https://www.theguardian.com/australia-news/2018/oct/25/australian-senator-who-called-for-final-solution-to-immigration-expelled-from-party https://www.theguardian.com/australia-news/2018/oct/25/australian-senator-who-called-for-final-solution-to-immigration-expelled-from-party https://www.smh.com.au/politics/federal/peter-dutton-wants-a-refugee-intake-of-white-south-african-farmers-20180315-p4z4gr.html https://www.smh.com.au/politics/federal/peter-dutton-wants-a-refugee-intake-of-white-south-african-farmers-20180315-p4z4gr.html https://thenewdaily.com.au/news/national/2018/10/18/senate-sinks-white-australia-plebiscite/ https://www.japantimes.co.jp/news/2018/10/30/national/politics-diplomacy/liberal-democratic-partys-general-council-approves-draft-immigration-control-bill-despite-members-concerns/#.w9lhci9l1yg https://www.japantimes.co.jp/news/2018/10/30/national/politics-diplomacy/liberal-democratic-partys-general-council-approves-draft-immigration-control-bill-despite-members-concerns/#.w9lhci9l1yg https://www.japantimes.co.jp/news/2018/10/30/national/politics-diplomacy/liberal-democratic-partys-general-council-approves-draft-immigration-control-bill-despite-members-concerns/#.w9lhci9l1yg https://www.theguardian.com/australia-news/2018/jul/15/peter-dutton-contradicts-morrison-and-says-migrant-cut-positive-for-the-economy https://www.theguardian.com/australia-news/2018/jul/15/peter-dutton-contradicts-morrison-and-says-migrant-cut-positive-for-the-economy https://openresearch-repository.anu.edu.au/bitstream/1885/12260/4/nta_report_2014_r.pdf 32 allen australian population studies 2 (2) 2018 repository.anu.edu.au/bitstream/1885/12260/4/nta_report_2014_r.pdf. accessed on 9 august 2018. treasury (2015) 2015 intergenerational report: australia in 2055. canberra: the treasury. treasury (2018) shaping a nation: population growth and immigration over time. canberra: the treasury. undp (2018) sustainable development goals. http://www.undp.org/content/undp/en/home/sustainable-development-goals.html. accessed on 6 november 2018. https://openresearch-repository.anu.edu.au/bitstream/1885/12260/4/nta_report_2014_r.pdf http://www.undp.org/content/undp/en/home/sustainable-development-goals.html 5. key messages references australian population studies 2017 | volume 1 | issue 1 | pages 41–54 © bernard, forder, kendig and byles 2017. © published under the creative commons attribution-noncommercial licence 3.0 australia (cc by-nc 3.0 au). journal website: www.australianpopulationstudies.org residential mobility in australia and the united states: a retrospective study aude bernard* the university of queensland peta forder the university of newcastle hal kendig australian national university julie byles the university of newcastle *corresponding author. email: a.bernard@uq.edu.au. address: chamberlain building, queensland centre for population research, school of earth and environmental sciences, the university of queensland, st lucia qld 4072 paper received 11 august 2017; accepted 27 october 2017; published 20 november 2017 abstract background levels of internal migration vary significantly between countries. australia and the united states consistently record among the highest levels of migration anywhere in the world. very little is known, however, about the factors underlying mobility differentials. we argue that this is because existing evidence is almost exclusively based on period measures applied to cross-sectional data. aims we seek to advance understanding of cross-national variations in levels of residential mobility by drawing on a newly proposed suite of cohort migration measures, coupled with the recent release of internationally comparable retrospective residential history data. data and methods focusing on the early cohort of baby boomers born between 1947 and 1951, the paper examines residential mobility levels and patterns in early and mid-adulthood in australia and the united states and compares them with 14 european countries. differences in completed levels of residential mobility are assessed in terms of four components: the proportion of a cohort who moved at least once; mean age at first move; mean age at last move; and average interval between moves. results while cohort analysis confirms high levels of mobility in australia and the united states, it does not support the notion of a common ‘new world’ mobility regime distinct from other advanced economies. conclusion a cohort perspective offers refined insights into population mobility. the increasing availability of retrospective survey data means that researchers can now apply cohort measures to a wide range of countries. key words retrospective life histories; residential mobility; cohort migration; completed migration rate; completed migration distribution; migration progression ratio; australia; united states. http://www.australianpopulationstudies.org/ mailto:a.bernard@uq.edu.au 42 bernard a et al. australian population studies 1 (1) 2017 1. introduction ‘new world’ countries such as australia, canada, new zealand and the united states are firmly established as some of the most mobile countries in the world with 40 to 55 per cent of their populations changing residence over a five-year period (bell et al. 2017). in a seminal cross-national comparison, long (1991) attributed this high mobility to institutional frameworks, flexible housing and labour markets and peripatetic traditions inherited from immigrant forbears. more recently, in a comparison of 23 organisation for economic co-operation and development (oecd) countries, sánchez and andrews (2011) showed that australia ranked second after iceland, and the united states fourth after sweden in terms of the proportion of households changing residence over a twoyear period. similar variations in levels of internal migration were found in a global comparison of 139 countries (esipova, pugliese and ray 2013). these movements have occurred against a background of globalisation, urbanisation and population ageing (mcdonald 2017). in most countries, internal migration has become a leading agent of demographic change, shaping patterns of human settlement and affecting the age distribution of populations. this paper adopts a cohort perspective to advance understanding of levels and patterns of residential mobility in australia and the united states. a cohort approach provides distinct advantages over existing studies, which use period measures applied to cross-sectional data. first, by following individuals over their entire life course, a cohort perspective provides detailed information about the migration trajectories of different birth cohorts (bogue 1950; shyrock and larmon 1965). this can reveal how moves are distributed in the population, which cannot be achieved with the dichotomy between movers and non-movers commonly used in national censuses and surveys (xu-doeve 2006). thus, because it views migration as an incremental process where individuals progress from one move to the next, a cohort approach allows the association between the timing of migration and the number of lifetime moves to be explored, which can shed light on the demographic mechanisms underlying differences in migration levels (bernard 2017b). second, a cohort perspective has the advantage of eliminating any potential tempo effects, which inflate or deflate the period measure of a demographic event due to a rise or fall in the mean age at which the event occurs (bongaarts and feeney 2008). cohort migration is free of this interpretive difficulty. if cohort migration falls, it is a pure quantum effect: people are moving less. third, because individuals may live through periods of high and low migration, a cohort approach has the advantage of smoothing out temporal variations in migration levels. this is particularly important when comparing countries because the economic cycles, housing market conditions and government policy regimes that underpin short-term variations in migration level are unlikely to be in phase in different national contexts (bell et al. 2002). finally, comparing the migration trajectories of successive cohorts can reveal the influence of social change on the evolution of migration behaviour, while comparisons between countries can shed light on the effects of national economic, social and policy developments (kendig and nazroo 2016). to facilitate a structured approach to the cohort analysis of migration and residential mobility, bernard (2017a) has recently proposed a comprehensive suite of robust cohort measures that capture the level and distribution of completed migration and the timing and spacing of moves. application to european countries (bernard 2017b) has revealed that differences in cohort migration levels are attributable to variations in the extent of repeat movement, which is underpinned by differences in mean ages at first and last move. australian population studies 1 (1) 2017 bernard a et al. 43 we now examine the demographic mechanisms underpinning variations in residential mobility levels in two ‘new world’ countries. we draw retrospective residential history data from the life histories and health (lhh) survey in australia and the life history mail survey (lhms) in the united states. while the lhms surveyed a nationally representative sample of americans born before 1966, the lhh survey focused on early baby boomers born between 1947 and 1951 and living in new south wales at the time of the survey. the latter may therefore not be representative of all australians. the surveys collected retrospective lifetime residential mobility histories in 2011–2012 and 2015– 2016, respectively, using life-history grids. this approach involves showing respondents a schematic form that depicts the year in their life, from birth to present, alongside national and world events to help them recall past moves (belli 1998; blane 1996). while the life-history calendar only took the form of a mail survey in the united states, it was complemented in australia by computer-assisted telephone interviews to assist recall (kendig et al. 2014). despite these differences, response rates were similar in the two countries, sitting at 45 per cent in australia and 48 per cent in the united states. to position australia and the united states internationally, we complement these datasets with retrospective residential histories from the english longitudinal study of ageing (elsa) and the survey of health, ageing and retirement in europe (share). together these provide directly comparable residential histories for 14 european countries (börsch-supan 2010; marmot et al. 2016). the paper explores variations in levels and patterns of residential mobility for the cohort of baby-boomers born in each case study country between 1947 and 1951 by comparing completed migration rates, completed migration distribution and the cumulative distribution of movers by age and move order. 2. cohort measures of migration: methods and data the analysis presented here is confined to a subset of six measures recommended by bernard (2017a) to compare migration between cohorts and countries. while we apply these measures to residential mobility (i.e. changes of address), we use the term ‘cohort migration’ proposed by bernard (2017a). table 1 lists each measure in summary form, providing a definition and an algebraic representation, where m corresponds to the number of moves, p to the number of individuals and x to the age at move. subscript i refers to the order of each move (first, second, etc.) and n to an individual. thus, 𝑃𝑖 refers to the number of individuals who have moved i times, 𝑀𝑖 to the number of moves of order i for all i>0, and 𝑋𝑛 corresponds to the age at move of individual n. multiple measures are required to comprehensively quantify cohort migration: • the first of these measures is the completed migration rate (cmr), which represents the average number of moves undertaken by members of a given cohort over the course of their lives, as defined by equation (1) in table 1. it is readily comparable across countries and indicates whether the overall level of migration is high or low. • because the actual migration behaviour of individuals is more heterogeneous than this summary statistic suggests, the completed migration distribution (cmd) decomposes the population according to the number of moves individuals have made, as indicated in equation (2), and hence reveals the proportion of lifetime non-movers, infrequent movers and frequent movers. • migration progression ratios (mprs) depict the underlying, incremental process of moving by measuring the proportion of individuals who, having made a given number of moves, proceed to move at least one more time, as shown in equation (3). underpinning mprs is the idea that variation in migration behaviour depends on the number of times individuals have moved. 44 bernard a et al. australian population studies 1 (1) 2017 • mean migration age (mma) summarises migration age patterns by showing whether populations are moving early or late in life. it can be computed for all moves, as indicated by equation (4), or by move order, as shown by equation (5). of particular importance are (a) the mean age at first move, because it captures the start of the migration career of successive cohorts and (b) the mean age at last move, which indicates how early or late in life different populations stop moving. • the final measure, mean migration spacing (mms), relates to spacing between consecutive moves, which indicates the extent to which moves are close to each other or spaced further apart, as shown by equation (6). further information and worked examples can be found in bernard (2017a). table 1: cohort measures of migration source: bernard (2017a). notes: m is the number of moves; p is the number of individuals; x is the age at move; i refers to the order of the move; n refers to an individual. to maximise comparability between lhh and lhms and avoid censoring bias, the analysis is confined to individuals born between 1947 and 1951 in each case study country. to ensure that migration careers are of comparable lengths for different birth years, the paper focuses on moves undertaken during early and middle adulthood, between the ages of 17 and 50 years (inclusive), and excludes tied moves in childhood that are qualitatively different to moves made as independent adults. in both surveys, respondents were asked to report the start and end years of residence for dwellings in which they had lived for more than six months since birth (up to 29 dwellings in lhh and up to 18 dwellings in the lhms). the postcode was collected in both surveys but, to ensure confidentiality, an annual indicator of change of residence and change of state was constructed for the lhms, whereas geographic information was released at city level for the lhh. to ensure comparability, the analysis in this paper makes use of data for up to 18 moves and uses all changes of postcodes, independent of administrative units, in measuring migration. while this means that the distinction between short and long-distance moves cannot be made, the results have the advantage of not being affected by differences across countries and over time in the number and shape of spatial units, which can bias cross-national comparisons (courgeau 1973). measures definition method equation completed migration rate (cmr) average number of moves per individual by the end of their migratory life 𝐶𝑀𝑅 = 𝑀 𝑃⁄ (1) completed migration distribution (cmd) proportion of cohort who have moved exactly i times 𝐶𝑀𝐷(0,𝑖) = 𝑃𝑖 𝑃 (2) migration progression ratios (mprs) proportion of individuals who moved i times and who went on to move at least once more 𝑀𝑃𝑅(𝑖,𝑖+1) = 𝑀𝑖+1 𝑀𝑖 (3) migration mean age (mma) mean age at which individuals in cohort moved 𝑀𝑀𝐴 = ∑ ∑ 𝑋𝑛,𝑖 𝐼 𝑖=1 𝑁 𝑛=1 𝑀⁄ (4) order-specific migration mean age (mma) mean age at which individuals in cohort moved for the ith time 𝑀𝑀𝐴𝑖 = ∑ 𝑋𝑛.𝑖 𝑁 𝑛=1 𝑀𝑖⁄ (5) mean migration spacing (mms) average interval between all moves for individuals who have moved at least twice 𝑀𝑀𝑆 = ∑ (𝑀𝑀𝐴𝑖+1 − 𝑀𝑀𝐴𝑖 𝐼 𝑖=1 ) ∑ 𝑀𝑖 𝐼 𝑖=2 ⁄ (6) australian population studies 1 (1) 2017 bernard a et al. 45 3. cohort residential mobility in australia and the united states figure 1 ranks australia and the united states alongside 14 european countries from the highest to lowest completed migration rate (cmr). while both countries display a cmr well above the european mean of 2.9 moves, australia reports the highest level of residential mobility among the 16 countries, with an average of 5.1 moves per individual. the united states ranks fifth after australia, denmark, england and sweden, with one less move on average than australia. figure 1: completed migration rates: australia, united states and select european countries source: authors’ calculations from elsa, share, lhh and lhms data. note: changes of address between ages 17 to 50 years. to describe the actual range of mobility behaviours in each country, figure 2 (next page) reports the completed migration distribution (cmd), which decomposes populations according to the exact number of times people moved in their lives. the proportion of non-movers, infrequent movers and frequent movers in australia and the united states is compared to the european mean. the united states stands out for its significant proportion of non-movers. fully 15 per cent of american respondents reported having never changed residence in adulthood, which is more than twice the average proportion in europe and the second highest level after austria (bernard 2017b). in contrast, immobility in australia is very rare, with less than 2 per cent of respondents reporting no change of postcode, which is the third lowest level after denmark and sweden. both australia and the united states, however, display a high level of repeat movement. approximately five in 10 australians and four in 10 americans moved at least five times between the ages of 17 and 50, compared to an average of three in 10 people in europe. very frequent movers are especially characteristic of the australian mobility landscape, with 20 per cent of respondents reporting eight moves or more, compared with 13 per cent in the united states and a mean of 9 per cent in europe. 5.1 4.9 4.6 4.5 4.1 3.7 3.7 3.4 2.8 2.7 2.1 2.0 1.9 1.7 1.6 1.6 0 1 2 3 4 5 6 a u st ra lia d en m ar k en gl an d sw ed en u n it ed s ta te s n e th e rl an d s fr an ce sw it ze rl an d b el gi u m g e rm an y it al y sp ai n p o la n d c ze ch r ep u b lic a u st ri a g re ec e c o m p le te d m ig ra ti o n r at e 46 bernard a et al. australian population studies 1 (1) 2017 figure 2: completed migration distribution: australia, united states and european mean source: authors’ calculations from elsa, share, lhh and lhms data. notes: changes of address between ages 17 to 50 years; european mean obtained from the 14 countries in elsa and share. bernard (2017b) showed that variations in the extent of repeat movement are underpinned by differences in mean ages at first and last moves that together delineate the average length of migration careers. age at first move signals the start of an individual’s migratory life and is of particular importance because it influences the probability of subsequent moves. in countries where young adults first move early (i.e. early twenties), they subsequently move at younger ages than in countries where first-time movers are older (i.e. mid-twenties). in addition, younger adult movers are more likely to proceed to a subsequent move and consequently report higher numbers of moves throughout adulthood than late starters. we test this proportion for australia and the united states by plotting age at first move against the cmr. figure 3: age at first move by completed migration rate: australia and united states source: authors’ calculations from lhh and lhms data. note: changes of address between ages 17 to 50 years. 7 15 2 36 18 18 28 24 26 20 30 34 9 13 20 0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100% european mean united states australia 0 moves 1 to 2 moves 3 to 4 moves 5 to 7 moves 8 moves + r² = 0.7837 r² = 0.9271 0 1 2 3 4 5 6 7 17 18 19 20 21 22 23 24 25 26 27 28 29 30 c o m p le te d m ig ra ti o n r at e age at first move (years) united states australia australian population studies 1 (1) 2017 bernard a et al. 47 figure 3 reveals for both countries a clear negative association, confirming that later ages at first move in adulthood are associated with reduced lifetime mobility. the strength of this association is supported by a pearson’s correlation coefficient of -0.89 for the united states and -0.96 for australia. individuals who first moved at age 17 changed place or residence on average 5.9 times in the united states and 6.5 times in australia, compared with four times or less for individuals who first moved at age 25 or later in both countries. we further explore age differentials by reporting the cumulative distribution of movers by age and move order. we first analyse results for australia separately in figure 4, before comparing australia with the united states in figure 5. the results for australia show that the first and second moves are strongly concentrated in early adulthood, which conforms to the well-established age patterns of migration peaking in the mid-to-late twenties (bernard, bell and charles-edwards 2014a, 2014b; rogers and castro 1981). half the cohort have moved at least once by age 20 and more than 80 per cent by age 23. a year later, the same proportions have moved at least a second time. for both moves, the proportion of movers starts plateauing after age 30, which indicates that individuals who have not moved by that age never went on to move. the third and fourth moves occur a bit later in life and are spread across a broader age range. it is not until age 30 that 80 per cent of individuals who have moved twice proceed to a third move and it is not until age 36 than 80 per cent of individuals who have already moved three times undertook their fourth move. the curve for the third move plateaus around age 35, while the curve for the fourth move continues to increase at a slow rate, indicating a broader dispersion of the fourth move across the age spectrum. figure 4: cumulative distribution of movers by age and move order in australia source: authors’ calculations from lhh and lhms data. note: migration between ages 17 to 50 years. the cumulative proportion of movers at age 50 corresponds to the migration progression ratio (mpr). that is, the proportion of the cohort who having moved i times went on to move at least i +1 more times. while mprs decrease with move order, they remain high for all moves: as many as 89 per cent of individuals who moved three times moved at least one more time. this pattern underpins the high level of repeat movement identified in australia in figures 1 and 2. 0 25 50 75 100 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49 c u m u la ti ve d is tr ib u ti o n o f m o ve rs ( % ) age at move (years) first move second move 48 bernard a et al. australian population studies 1 (1) 2017 we now compare age patterns between australia and the united states. figure 5 reveals that the key difference is in the progression to the first move. while the two countries first follow a similar progression in early adulthood, in the united states the proportion of first-time movers starts to plateau at around 80 per cent from age 25 onward. thus, americans who have not moved by age 25 never proceed to a move. conversely, in australia the proportion of first-time movers continues, increasing to 98 per cent by age 40. on the other hand, second-, thirdand fourth-time movers display very similar age patterns in the two countries and comparable proportions of movers by age 40. these results confirm that it is the proportion of non-movers that is the key factor underpinning differences in completed migration rates in australia and the united states. australia united states figure 5: cumulative distribution of movers by age and move order: australia and united states source: authors’ calculations from lhh and lhms data. note: migration between ages 17 to 40 years. we now seek to quantify the relative contribution of immobility to differences in completed migration rates. following jain and mcdonald’s (1997) approach to fertility analysis, bernard (2017a) demonstrated that the cmr can be expressed mathematically as a function of four components, 𝐶𝑀𝑅 = 𝑀𝑜[1 + ( 𝐿−𝐹 𝐼 )] (7) 0 25 50 75 100 17 19 21 23 25 27 29 31 33 35 37 39 c u m u la ti ve d is tr ib u ti o n o f m o ve rs ( % ) age at move (years) first move 0 25 50 75 100 17 19 21 23 25 27 29 31 33 35 37 39 c u m u la ti ve d is tr ib u ti o n o f m o ve rs ( % ) age at move (years) second move 0 25 50 75 100 17 19 21 23 25 27 29 31 33 35 37 39 c u m u la ti ve d is tr ib u ti o n o f m o ve rs ( % ) age at move (years) third move 0 25 50 75 100 17 19 21 23 25 27 29 31 33 35 37 39 c u m u la ti ve d is tr ib u ti o n o f m o ve rs ( % ) age at move (years) fourth move australian population studies 1 (1) 2017 bernard a et al. 49 where 𝑀𝑜 is the proportion of individuals who moved at least once, 𝐹 is the mean age at first move, 𝐿 is the mean age at last move and 𝐼 is the mean length of all intervals between consecutive moves for individuals who moved at least twice. table 2: components of completed migration rate: australia and united states proportion of movers (mo) mean age at first move (f) mean age at last move (l) mean length of migration interval (i) completed migration rate (cmr) australia 99.0 20.9 37.2 4.0 5.1 united states 84.7 20.6 36.7 4.5 4.1 % difference in cmr attributed to each component in united states compared to australia -16.9 1.4 -2.7 -3.8 -22.4 source: authors’ calculations from lhh and lhms data. note: migration between ages 17 to 50 years. table 2 reports each component of equation (7) for australia and the united states and shows the estimated percentage differences in cmr in the united states compared to australia. it reveals that the lower proportion of movers in the united states is overwhelmingly the main factor underpinning migration differentials between the two countries. of the 22.4 per cent different in cmr with australia, the proportion of movers accounts for 16.9 per cent. the impact of longer mean migration intervals on the cmr is much smaller, less than 4 per cent, as is the impact of the mean age at last move. the slightly younger mean age at first move has a small counteracting effect on cmr in the united states. measured as the difference between the mean ages at first and last move, the average length of migration careers is about 16 years in both australia and the united states. this is relatively long compared to that in other countries. in southern and eastern europe, for example, individuals are mobile on average for about seven years (bernard 2017b). figure 6: proportion of adult non-movers by birth cohort in the united states source: authors’ calculations from lhms data. note: migration between ages 17 to 50 years. 0 5 10 15 20 25 30 19 27 -1 9 31 19 32 -1 9 36 19 37 -1 9 41 19 42 -1 9 46 19 47 -1 9 51 19 52 -1 9 56 19 57 -1 9 61 p e rc e n ta ge w n o n ev er m o ve d ( % ) 50 bernard a et al. australian population studies 1 (1) 2017 is the relatively high level of adult immobility a long-standing feature of the american landscape or a characteristic of the cohort born between 1947 and 1951? to answer that this question, figure 6 (previous page) examines trends in immobility between ages 17 to 50 for seven successive cohorts born between 1927–1931 and 1957–1961. it shows that immobility among the early cohort of baby boomers (born 1947–1951) forms part of a declining trend of high immobility. the proportion of adult non-movers was at its highest level for individuals born 1927–1931 (26%). it then fell for successive cohorts, reaching a low of 13 per cent for the cohort born 1952–1956, before increasing again for the 1957–1961 cohort. 4. conclusion despite wide and enduring variations in migration levels between countries around the world, explanation for the relatively high mobility observed in ‘new world’ countries remains tentative. this can be traced in part to the limitations inherent in application of cross-national data to period measures, which fail to account for heterogeneity in migration behaviour or for cohort differentials. this paper has drawn on a series of robust cohort measures of migration, coupled with internationally comparable retrospective migration histories, to compare residential mobility levels and patterns in australia and the united states and identify key differences compared with 14 european countries for early baby-boomers born between 1947 and 1951. the results have confirmed a high level of residential mobility in australia and the united states, with the former reporting the highest level of residential mobility among the 16 oecd countries in our sample and the latter ranking fifth after denmark, england and sweden. high levels of mobility in both countries are attributable to the high incidence of repeat moves. approximately five in 10 australians and four in 10 americans moved at least five times in adulthood, compared with an average of three in 10 people in the 14 european countries. very frequent moves were especially characteristic of australia, where 20 per cent of respondents reported eight moves or more. compared with australia, the united states combines a high level of repeat moves with a substantial rate of immobility. while less than 2 per cent of australians never moved in adulthood, as many as 15 per cent of americans remained immobile. this is more than twice the average proportion of nonmovers in europe and is the second highest level among the 16 oecd countries. immobility appears to be an enduring characteristic of the united states that has persisted across successive cohorts born between 1927 and 1961. although its incidence has progressively diminished, it remains close to 15 per cent for the most recent cohort. cohort analysis does not support the notion of a common ‘new world’ mobility regime distinct from other advanced economies. while australia’s very high level of mobility may be inflated as a result of the lhh sample being drawn exclusively from the state of new south wales, its mobility level and patterns correspond to the mobility regime of northern and western european countries. in these countries, high mobility is the product of the extremely low incidence of immobility and a high level of repeat movement. this, in turn, is underpinned by an early mean age at first move and a late mean age at last move, which together support long migration careers (bernard 2017b). because of its high level of immobility, the united states departs from this mobility regime and does not conform to the patterns identified in oecd countries with intermediate levels of residential mobility, such as the netherlands, france, switzerland and belgium, where immobility is very low and the level of repeat australian population studies 1 (1) 2017 bernard a et al. 51 movement moderate. this suggests that the united states has a distinct mobility regime characterised by a unique combination of a high level of repeat movement and a high rate of immobility. as with any retrospective data, residential histories face issues of recall and are based on survivors only. although survivor bias is expected to be small, the completed migration rate should, strictly speaking, be interpreted as the average number of moves undertaken by members of a cohort conditional on survival to the date of the survey. bearing these limitations in mind, a cohort perspective offers a step forward in the comparative analysis of residential mobility by revealing new insights into mobility behaviour. it provides a robust foundation for exploring the demographic mechanisms underpinning migration differentials that parallel methods long used in fertility and mortality analysis. this paper has shown that moving beyond population-level averages and considering the distribution of moves provides a realistic description of the mobility experience of each cohort and offers refined insights into mobility behaviour. of particular importance is the negative association between age at first move and completed migration, which indicates that late starters in australia and the united states move less throughout the course of their adult lives. bernard (2017a, 2017b) showed that age at first move operates to affect completed migration by influencing the likelihood of progressing to moves of higher orders and that, in turn, variations in mean age at first move underpin differences in completed migration over time and between countries. the association between age at first move and completed migration may also result from a selection effect whereby individuals who view mobility more positively choose to start moving at younger adult ages, or that the experience gained from a previous move may facilitate subsequent migrations (van arsdol, sabagh and butler 1968). in this paper, we have shown that age also matters because individuals who have not moved by the age 25 in the united states and age 30 in australia remained immobile through their adult lives. thus, moves in early adulthood seem to have a lifelong imprint on mobility behaviour. differences in population composition are likely to account for some of the variation in residential mobility observed in this paper. the cohort measures employed here can be readily applied to specific groups to explore within and between country differences, differences between socioeconomic and ethnic groups and differences between nativeand foreign-born populations. better understanding of the role of population heterogeneity would represent an important step forward in the comparative analysis and understanding of migration and residential mobility. the distinctively high proportion of non-movers in the united states also invites investigation into reasons for immobility in that country and in other national settings to identify the underpinning factors. attention to the determinants of repeat movement, which accounts for most of the differences in completed migration between the 16 oecd countries, is needed also to establish the individual-level characteristics associated with repeat movement at a range of spatial scales and in different countries. key messages • cohort migration measures facilitate a structured and systematic analysis of retrospective residential history data • application of these measures to residential histories collected in australia and the united states has shown a high level of repeat movement in both countries, and a much a higher proportion of immobility in the united states. 52 bernard a et al. australian population studies 1 (1) 2017 • individuals who had not moved by age 25 in the united states and age 30 in australia remained immobile through their adult lives. this suggests that moves in early adulthood have a lifelong imprint on mobility behaviour. acknowledgments the authors gratefully acknowledge the support of the australian research council (arc) through the arc centre of excellence in population ageing research (ce110001029), arc discovery project (dp1096778) and arc discovery early career researcher award (de160101574). the authors thank emeritus professor martin bell for providing comments on a draft version of the paper. the life history and health (lhh) survey was approved by the university of sydney human research ethics committee (#12744) and supported by an australian research council grant (dp1096778, ‘socio-economic determinants and health inequalities over the life-course: australian and english comparisons’). chief investigators were prof. hal kendig (anu), prof. julie byles (the university of newcastle) and prof. james nazroo (the university of manchester). fieldwork and the sample are described in kendig et al. (2014). this research was supported by the australian research council centre of excellence in population ageing research and by infrastructure and staff of the priority research centre for gender, health and ageing. the lhh research was completed as a sub-study of the 45 and up study. the 45 and up study is managed by the sax institute (www.saxinstitute.org.au) in collaboration with partners: cancer council nsw; national heart foundation of australia (nsw); nsw ministry of health; nsw department of family and community services; disability council nsw; and australian red cross blood service. we thank the many thousands of people participating in the 45 and up study. the 2015 life history mail survey (lhms) contains questions about residential history, education history and other important childhood and family events. the lhms is part of the university of michigan health and retirement study, supported by the national institute on aging and social security administration in the united states. the share data collection has been primarily funded by the european commission through fp5 (qlk6-ct-2001-00360), fp6 (share-i3: rii-ct-2006-062193, compare: cit5-ct-2005-028857, sharelife: cit4-ct-2006-028812) and fp7 (share-prep: n°211909, share-leap: n°227822, share m4: n°261982). additional funding from the german ministry of education and research, the max planck society for the advancement of science, the u.s. national institute on aging (u01_ag0974013s2, p01_ag005842, p01_ag08291, p30_ag12815, r21_ag025169, y1-ag-4553-01, iag_bsr06-11, ogha_04-064, hhsn271201300071c) and from various national funding sources is gratefully acknowledged (see www.share-project.org). the elsa datasets were made available through the uk data archive. elsa was developed by a team of researchers based at natcen social research, university college london and the institute for fiscal studies. the data were collected by natcen social research. the funding is provided by the national institute of aging in the united states and a consortium of uk government departments coordinated by the office for national statistics. the developers and funders of elsa and the archive do not bear any responsibility for the analyses or interpretations presented here. http://www.saxinstitute.org.au/ http://www.share-project.org/ australian population studies 1 (1) 2017 bernard a et al. 53 references bell m, blake m, boyle p, duke-williams o, rees p, stillwell j and hugo g (2002) cross-national comparison of internal migration: issues and measures. journal of the royal statistical society: series a (statistics in society) 165(3): 435–464. bell m, charles-edward e, bernard a and ueffing p (2017) global trends in internal migration. in: champion t, cooke t j and shuttleworth i (eds) internal migration in the developed world: are we becoming less mobile? london uk: routledge; 76–97. belli r f (1998) the structure of autobiographical memory and the event history calendar: potential improvements in the quality of retrospective reports in surveys. memory 6(4): 383–406. bernard a (2017a) cohort measures of internal migration: understanding long-term trends. demography. [online ahead of print] doi: 10.1007/s13524-017-0626-7. bernard a (2017b) levels and patterns of internal migration in europe: a cohort perspective. population studies 71(3): 293–311. doi: 10.1080/00324728.2017.1360932. bernard a, bell m and charles-edwards e (2014a) improved measures for the cross-national comparison of age profiles of internal migration. population studies 68(2): 179–195. bernard a, bell m and charles-edwards e (2014b) life-course transitions and the age profile of internal migration. population and development review 40(2): 231–239. blane d b (1996) collecting retrospective data: development of a reliable method and a pilot study of its use. social science & medicine 42(5): 751–757. bogue d j (1950) an exploratory study of migration and labor mobility using social security data. scripps foundation for research in population problems, miami university. bongaarts j and feeney g (2008) the quantum and tempo of life-cycle events. in: barbi e, vaupel j w and bongaarts j (eds) how long do we live? demographic models and reflections on tempo effects. berlin: springer; 29–65. börsch-supan a (2010) survey of health, ageing and retirement in europe (share) wave 3 – sharelife. release version: 5.0.0. share-eric. data set. doi: 10.6103/share.w3.500. courgeau d (1973) migrations et découpages du territoire. population [french edition] 28(3): 511–537. esipova n, pugliese a and ray j (2013) the demographics of global internal migration. migration policy practice 3(2): 3–5. jain s and mcdonald p (1997) fertility of australian birth cohorts: components and differentials. journal of the australian population association 14(1): 31–46. kendig h, byles j e, o’loughlin k, nazroo j y, mishra g, noone j, loh v and forder p m (2014) adapting data collection methods in the australian life histories and health survey: a retrospective life course study. bmj open 4(3), http://dx.doi.org/10.1136/bmjopen-2013-004476. kendig h and nazroo j (2016) life course influences on inequalities in later life: comparative perspectives. journal of population ageing 9(1–2): 1–7. long l (1991) residential mobility differences among developed countries. international regional science review 14(2): 133. marmot m, oldfield z, clemens s, blake m, phelps a, nazroo j, steptoe a, rogers n, banks j, oskala a (2016) english longitudinal study of ageing: waves 0–7, 1998–2015. [data collection]. 25th edition. uk data service. sn: 5050, http://doi.org/10.5255/ukda-sn-5050-12. mcdonald p (2017) population ageing: a demographic perspective. in: o'loughlin k, browning c and kendig h (eds) ageing in australia: challenges and opportunities. new york: springer; 41–61. rogers a and castro l j (1981) model migration schedules: a simplified formulation and an alternative parameter estimation method. iiasa working paper wp-81-063. laxenburg, austria: international institute for applied systems analysis. http://dx.doi.org/10.1136/bmjopen-2013-004476 http://doi.org/10.5255/ukda-sn-5050-12 54 bernard a et al. australian population studies 1 (1) 2017 sánchez a c and andrews d (2011) to move or not to move: what drives residential mobility rates in the oecd?. oecd economics department working papers no. 846. paris: oecd publishing, http://dx.doi.org/10.1787/5kghtc7kzx21-en. shyrock h s and larmon e a (1965) some longitudinal data on internal migration. demography 2(1): 579– 592. van arsdol m d, sabagh g and butler e w (1968) retrospective and subsequent metropolitan residential mobility. demography 5(1): 249–267. xu-doeve w l (2006) the measurement of international and internal migration in the 2010 global round of population censuses: twelve key recommendations on questions, concepts and procedures. paper presented at the european population conference of the european association for population studies (eaps), liverpool, united kingdom, 21–24 june 2006. http://dx.doi.org/10.1787/5kghtc7kzx21-en australian population studies 2019 | volume 3 | issue 2 | pages 37-40 © pribesh et al. 2019. published under the creative commons attribution-noncommercial licence 3.0 australia (cc by-nc 3.0 au). journal website: www.australianpopulationstudies.org the flow of family transitions of australian families shana pribesh* old dominion university matthew usevitch baylor college of medicine elizabeth koch sigler brigham young university kaijsa angerhofer heninger brigham young university yuanyuan yue old dominion university mikaela j dufur brigham young university jonathan a jarvis brigham young university * corresponding author. email: spribesh@odu.edu. address: department of educational foundations and leadership, old dominion university, dcoe 2307, norfolk va, usa 23529. paper received 13 june 2018; accepted 14 november 2019; published 18 november 2019 family structure disruption has been linked to negative child educational and health outcomes (perales et al. 2016). australia has relatively stable families, but income disparities between australians are widening, and single-parent families make up a large proportion of families living in poverty. cohabitation is also common in australia with approximately three-quarters of marriages preceded by cohabitation. if substantial family structure churning affects australian children this may expose a need for special policy interventions aimed at family creation and dissolution to ameliorate the negative effects of such stressful experiences. to highlight family structures and transitions australian children experience, we use sankey flow diagrams charting data from ‘growing up in australia: the longitudinal study of australian children (lsac)’ (gilding 2001). we track children from birth to 11 years old using waves 1-6 and population weights to represent 183,521 children born into australian families. we used household rosters to construct three baseline family structures in wave 1 (0-1 years old): biological married, biological cohabit, and biological single. in subsequent waves, we placed children into one of five additional categories if they experienced a family transition:  disrupted biological married – two biological parents who married after the child’s birth and live in the same household.  disrupted non-biological married – two married parents where one parent is not the biological parent of the child (stepfamilies, adoptions, etc.) and the stepparent lives in the same household.  disrupted biological cohabit – two biological parents who began cohabiting after the child’s birth and live in the same household.  disrupted non-biological cohabit – two cohabiting parents where one parent is not the biological parent of the child (social families, etc.) and the social parent lives in the same household.  disrupted single – a single-parent family where a parent exited the family after the child’s birth. d e m o g ra p h ic http://www.australianpopulationstudies.org/ mailto:spribesh@odu.edu 38 pribesh et al. australian population studies 3 (2) 2019 figure 1: paths of family transitions for children who were born to married, cohabiting and single parents australian population studies 3 (2) 2019 pribesh et al. 39 each combination of consecutive family structures defined what we call a family structure path. children with the same family structure path were grouped. we then created sankey diagrams to visualize the most common paths overall based on children’s family structure at birth (figure 1). we include only those paths that represent at least 0.6% of the population. most common family structures and transitions over three-quarters of children were born into a family with biological parents who were married, 16% into a cohabiting parent family, and 7% into a single parent family. while nearly two-thirds of children remained in a family with married biological parents, we discovered 57 distinct family structure paths after birth. but generally, family structures were stable with 75% of children experiencing no transitions and very few children experiencing more than one family transition. six primary paths represented 90% of children. still, 1% of children experienced three or more transitions. among children who experienced at least one transition, 74% were exposed to family instability before age nine. born to married parents children born into families with biological parents who were married experienced the least turbulence with approximately 84% experiencing no family changes. the next most common experience for this group was for the child’s parents to transition from married to a single-parent family representing 9.6% of children born to married parents and 7.4% overall. these two-family experiences make up almost 94 % of all children originally born into families with two biological parents who were married. born to cohabiting parents children who were born into a family with cohabiting parents took 11 distinct paths. the largest path (44%) remained in households with two biological parents who were cohabiting. a fifth of children whose biological parents were cohabiting at the child’s birth then transitioned into marriage. the third largest path in this group consists of children whose parents cohabited at birth but then split up indicating a transition into a single-parent family that persists--a path that represents 15.7% of cohabiters at birth and 2.5% of the total sample. born to a single parent forty-five percent of children born into a single-parent family remained in that path. of the 15 additional paths for children born into single-parent families, the most common was for the single parent to enter a cohabiting relationship. approximately 8% began cohabiting with the biological parent of the child and 15.7% began cohabiting with another person. marriage was less common among those who had a child while single with 7.4% of single parents marrying the biological parent of the child and 6.2% marrying someone else. 40 pribesh et al. australian population studies 3 (2) 2019 references gilding m (2001) changing families in australia. family matters 60: 6-11. perales f, o’flaherty m and baxter j (2016) early life course family structure and children’s socioemotional and behavioural functioning: a view from australia. child indicators research 9: 10031028. waldfogel j, craigie t a and brooks-gunn j (2010) fragile families and child wellbeing. future child 20(2): 87-112. australian population studies 2019 | volume 3 | issue 1 | pages 13-29 © temple, dow and baird 2019. published under the creative commons attribution-noncommercial licence 3.0 australia (cc by-nc 3.0 au). journal website: www.australianpopulationstudies.org special working arrangements to allow for care responsibilities in australia: availability, usage and barriers jeromey temple* university of melbourne; arc centre of excellence in population ageing research briony dow university of melbourne; national ageing research institute marian baird university of sydney; arc centre of excellence in population ageing research * corresponding author. email: jeromey.temple@unimelb.edu.au. address: melbourne school of population and global health, 207 bouverie st, university of melbourne, melbourne, vic, 3010. paper received 17 january 2019; accepted 10 may 2019; published 27 may 2019 abstract background population ageing is projected to reduce labour force growth and aggregate labour force participation, whilst increasing demand for informal carers. increasing the labour force participation of australians who face barriers to employment (including carers) is part of the solution to labour market pressures occurring due to demographic change and may improve the financial wellbeing of carers. aims to examine the availability, usage and barriers to accessing special working arrangements (swa) to provide care while employed in australia. data and methods the 2015 abs survey of disability, ageing and carers was used to measure the prevalence of the availability, usage and barriers to swa to care stratified by carer status and gender. results about 94% of workers reported access to at least one type of swa (n=25,094). of this group, about 22% have used swa to care in the last 6 months. the proportions using swa to care were highest among primary carers (64%) followed by other carers (43%) and non-carers (19%). of those who have used swa, about 15% wanted to use additional swa to care in the previous 6 months, but faced barriers in doing so, with higher proportions of primary carers (24.6%) and other carers (21.8%) reporting barriers. the main barriers faced by employed carers included insufficient paid leave and/or work commitments. conclusions a range of paid and unpaid arrangements are necessary for carers to combine paid work with their caregiving responsibilities. labour market legislation and workplace policies should be strengthened to reduce barriers to take up of swa. key words informal care; caring; labour force participation; barriers to employment; special working arrangements http://www.australianpopulationstudies.org/ mailto:jeromey.temple@unimelb.edu.au australian population studies 3 (1) 2019 temple et al. 14 1. introduction one aspect of population ageing that has garnered significant attention from policy makers is the implications for the australian labour market. both the productivity commission and treasury have noted that population ageing reduces aggregate (i.e. population-level) labour force participation and slows labour force growth (australian treasury 2002, 2007, 2010, 2015; productivity commission 2005, 2013). indeed, the rate of growth in labour supply is projected to decline considerably in the next 15 years, relative to the 15 years proceeding (temple and mcdonald 2017). in recent times, international migration has been key to maintaining labour force growth (mcdonald 2017). however, a further source of increasing labour supply is to improve labour force attachment among australians who face barriers to employment or who are under-employed (temple and mcdonald 2017). increasing labour force attachment of these groups is one part of the solution to declining labour supply growth. one sizable population that faces significant barriers to labour force participation are australian carers, numbering approximately 2.7 million people who contribute 60 billion dollars to the economy annually through unpaid work (abs 2016b; deloitte access economics 2015). with population ageing, the demand for informal care will increase considerably, at the same time that labour supply pressures are occurring in the formal labour market. importantly, the caregiving role has been shown to effect labour force engagement, with australian evidence showing that carers are more likely to reduce hours of work, exit the labour market and earn lower levels of income relative to non-carers (bittman et al. 2007). many female carers in particular have difficulty in accessing flexible working hours, influencing their decision to leave paid work all together (austen and ong 2013). consistent with this finding, hill and colleagues find that with the onset of caring responsibilities, flexible working arrangements (such as part time work), significantly reduced the odds of leaving the paid workforce in australia (hill et al. 2008). a systematic review of 30 years of international studies of caregiving and employment, specifically cites the importance of the further development of labour market legislation and workplace policies (in tandem with formal support services), to enable carers who wish to work, to be able to do so (lilly et al. 2007). australian evidence shows a sizeable proportion of carers wish to work – with about half of those not in paid employment with a preference to be employed (gray et al. 2008). 2. special working arrangements and caring in australia enshrined in the national employment standards of australia’s fair work act, carers, people with a disability, those aged 55 years or over and parents with the responsibility to care for a child have the legal right to request flexible working arrangements from their employer (fwo 2018). although a broad term, flexible working arrangements refer to flexibility in hours of work (e.g. start and finish times), patterns of work (e.g. job sharing or part time work) and location of work (e.g. working from home) (fwo 2018). however, evidence suggests that less than half of all workers are aware of this right (skinner and pocock 2014). moreover, there is no guarantee that the requests for flexible working arrangements are approved by employers or that they are rolled out consistently across australian workplaces (o’loughlin et al. 2017). this is important as the most commonly cited reason given by non-employed caregivers as to why they do not work is difficulty arranging flexible working hours (gray et al. 2008). australian population studies 3 (1) 2019 temple et al. 15 despite the considerable contribution of the above studies to our understanding of the integration of paid work and unpaid caring, there remains a gap in our understanding of the availability, usage and barriers to accessing flexible or special working arrangements (swa) conducive to caring responsibilities. we adopt the terminology swa, consistent with abs data collections and to avoid confusion with one component of swa, flexible working hours. previous studies have focused on either parents with care giving responsibilities, or carers whose recipient has an underlying health condition requiring assistance. in this paper, we examine these aspects of usage of swa by gender in three population groups: (1) primary carers, who provide the principal assistance for a person with a disability or long-term health condition, (2) other carers, who also care for a person with a disability or long-term health condition, but do not provide principal care, and (3) non-carers, who are defined by the australian bureau of statistics (abs) as all people who are not primary or other carers. nonetheless, they may have care responsibilities and a need to access swa, for example, parents and others (e.g. grandparents) with caring responsibilities for children. our interest is in understanding the ways in which people who care have access to, utilise or report barriers to arrangements that enable them to fulfil both their unpaid care work and paid work responsibilities. specifically, with the availability of unique nationally representative data, we seek to answer three questions: (1) what is the level of availability of swas? (2) what is the prevalence of usage of swas specifically to care?, and (3) what are the reported barriers to usage of swas to care? 3. data and methods data for this study were sourced from the 2015 abs survey of disability, ageing and carers (sdac) conducted between july and december 2015 (abs 2016c). three populations were sampled using multi-stage sampling techniques. these consisted of persons living in private dwellings, in self-care retirement villages and in care accommodation. the module on the availability of swa was administered to persons living in households. of 31,957 households originally contacted, 25,555 fully responded, yielding a response rate of 80%. 3.1. measurement of availability, usage and barriers within the employment module of sdac, a number of questions were asked regarding the availability, use and barriers to swas to support caring work. these questions were asked of people aged 15 years and over, living in households and employed, but excluding the self-employed. firstly, respondents were asked “does your employer provide you with any of the following special working arrangements, regardless of whether you have used them or not?” a prompt card was then shown to the respondent, consisting of the following list:  paid leave (e.g. annual leave, maternity leave, sick leave) excluding carer’s leave.  paid carer’s leave  ∙unpaid leave (excluding unpaid carer’s leave) australian population studies 3 (1) 2019 temple et al. 16  unpaid carer’s leave  flexible working hours  rostered day off  working from home  shift work  casual work  part time work  informal arrangement with employer  other (specify)  no special working arrangements available  don’t know. for those indicating the availability of swa, a follow up question was asked: “of those special working arrangements you have mentioned, have you used any to help look after someone in the last six months?” respondents then indicated the swa (from the list above) used to facilitate caring responsibilities and whether there was any unmet need for further access to swa. for those who used swa and indicated a barrier to further access, and those who could not access swa for caring needs at all, a prompt card was used to illicit reasons for this barrier, including:  applied or asked but was refused  do not have adequate working arrangements  didn’t apply as thought they would say no anyway  nature of work makes using flexible working arrangements difficult  work commitments  not paid for time off (e.g., casual/shift worker)  subtle or other pressure from bosses or other workers  not enough paid leave left or available  anything else (specify). 3.2. measurement of carer status using these questions, we develop measures of the availability, usage and barriers to access swa specifically for caring responsibilities. using measures of carer status available in sdac, we examine variations in availability, usage and barriers to swa reported by (1) primary carers, (2) other carers and (3) persons not defined by the abs as carers, but nonetheless have caregiving responsibilities. specifically, a primary carer is defined by the abs as “a person who provides the most informal assistance, in terms of help of supervision, to a person with one or more disabilities, with one or more of the core activities of mobility, self-care or communication” (abs 2016a). other carers are australian population studies 3 (1) 2019 temple et al. 17 defined as one who “provides informal assistance with one or more of the core activity tasks but has not been identified as the person that provides the most informal assistance” (abs 2016a). thus, both primary and other carers are defined with respect to the recipient of care having long term health conditions requiring assistance and care. however, it is not only those who care for people with disabilities or long term health conditions who utilise swa with the specific intent of caring, albeit they are not carers as defined by the abs. this group includes those who have care obligations to others not identified as primary or other carers such as parents with care responsibilities for temporarily sick children or other family members. our analysis seeks to provide population level prevalence of the use of swa to care, thus all three caring categories are included in the following stratified analysis. 4. results figure 1 displays our framework for understanding the availability, usage and barriers to swa for caring using the sdac data. of the full sample of employed people living in households (n=26,529), about 94% report access to at least one type of swa (n=25,094). of this group, about 22% have used swa to care in the last 6 months (n=5,803). of those who have used swa, about 15% wanted to use additional swa to care in the previous 6 months, but faced barriers to do so (n=891). of the 19,291 employed people who did not use swa to care, a small minority wanted to use swa to care but couldn’t (1.2% n=224). 4.1. availability of special working arrangements of the 94% reporting availability of swa, there is considerable variation in the type of swa provided (table 1). within the full working population, the most prevalent swa is paid leave (76%), followed by paid carers leave (48%), unpaid leave (43%) and flexible working hours (39%). when classified by carer status, primary or other carers are more likely to report access to informal arrangements, part time work or unpaid carers leave relative to non-carers. about 5.6% of workers were not aware of the availability of any swa (responding ‘no’ or ‘don’t know’). these data also show important gender differentials with respect to carer status and perceived availability of swa. whereas the gender split for non-carers and other carers are relatively equal, just under 70% of all primary carers are female (68.3%). moreover, although the perceived availability by swa is relatively consistent by gender, females (relative to males) are more likely to cite availability of casual work (32.2% versus 21.5%) or part time work (34.1% versus 14.7%). 4.2. usage of special working arrangements the availability of swa, although important, is distinctly different from usage. results in table 2 classify swa arrangements used in the previous 6 months specifically to care. there is a clear gradient in swa usage to care by carer status. approximately 64% of primary carers had used swa in the previous 6 months, compared with 43% of other carers and about 19% of non-carers. around 2% of primary and other carers wanted to use swa to care but couldn’t. when usage, as opposed to availability of swa is considered, about 55% of non-carers and other carers are female, as are 70% of primary carers. australian population studies 3 (1) 2019 temple et al. 18 figure 1: framework of the availability, usage and barriers to swa for caring source: designed by the authors; data from abs 2016c consistent with the data on availability (table 1), the results in table 2 show that paid leave (42%) and paid carers leave (47%) are the most commonly used swa to care. about 10% of primary carers report unpaid leave, unpaid carers leave, casual work or part-time work swa to care – which is higher than the rates of usage by non-carers. about one quarter of primary carers also cite use of flexible working hours to care (24%). again, there is relative consistency in the likelihood of using different swa to care by gender. for example, about 15.1% of males who have used swa to care have used flexible work hours as have 17% of females. however, although rates of usage are similar for males and females, the usage of specific swa to care is heavily skewed towards female workers. for example, around 80% of those using part-time work to care are female, regardless of carer status. among primary carers specifically, around 80% of those using part-time work, casual work, or unpaid carers leave to care are female, regardless of carer status. australian population studies 3 (1) 2019 temple et al. 19 table 1: availability of special working arrangements (%) by carer status and sex, 2015 males females total % female carer status total carer status total carer status total carer status total access to swa no primary other no primary other noprimary other noprimary other no 4.2 2.9 3.9 4.1 3.6 3.0 4.4 3.7 3.9 2.9 4.2 3.9 45.6 68.5 53.6 46.8 don’t know 2.0 1.0 2.8 2.0 1.5 0.3 0.8 1.4 1.7 0.5 ** 1.8 1.7 42.7 40.8 22.5 40.9 yes 93.9 96.1 93.2 93.9 94.8 96.7 94.8 94.9 94.3 96.5 ** 94.0 94.4 49.2 68.4 50.9 49.9 total 100.0 100.0 100.0 100.0 100.0 100.0 100.0 100.0 100.0 100.0 100.0 100.0 48.9 68.3 50.5 49.7 type of swa has access to: paid leave 79.9 72.1 81.1 79.8 73.1 70.0 76.6 73.3 76.6 70.7 ** 78.8 * 76.6 47.0 67.7 49.5 47.8 paid carer's leave 47.7 43.5 50.4 47.8 48.2 48.8 54.8 48.8 47.9 47.1 52.7 *** 48.3 49.5 70.8 53.0 50.4 unpaid leave 42.7 41.0 42.2 42.6 43.7 44.0 48.1 44.1 43.2 43.1 45.2 ** 43.3 49.8 69.9 54.2 50.8 unpaid carer's leave 29.2 30.0 32.4 29.5 30.3 34.5 37.0 31.1 29.8 33.1 * 34.8 ** 30.3 50.1 71.3 54.2 51.2 flexible working hours 37.2 42.6 37.8 37.3 41.1 44.9 45.1 41.6 39.1 44.2 *** 41.5 *** 39.5 51.7 69.5 55.3 52.6 rostered day off 24.6 23.7 25.7 24.6 20.1 15.0 23.0 20.1 22.4 17.7 ** 24.3 * 22.4 44.2 57.7 48.1 44.9 working from home 19.0 20.3 19.9 19.1 17.3 18.4 20.1 17.6 18.2 19.0 20.0 ** 18.3 47.0 66.2 51.1 48.0 shift work 17.3 13.7 19.4 17.4 14.9 10.9 16.1 14.8 16.1 11.7 *** 17.7 16.1 45.4 63.2 46.2 45.9 casual work 21.6 21.9 20.0 21.5 32.4 30.3 31.1 32.2 26.9 27.6 25.7 26.8 59.2 74.9 61.8 59.9 part time work 14.5 19.7 15.9 14.7 32.8 46.3 41.5 34.1 23.5 37.9 *** 29.0 *** 24.4 68.6 83.6 73.0 69.8 informal arrangement 9.2 14.9 11.8 9.5 9.7 16.1 15.3 10.4 9.4 15.8 *** 13.5 *** 10.0 50.5 70.1 57.4 52.2 other 1.0 1.7 0.8 1.0 0.8 0.7 1.3 0.8 0.9 1.0 1.1 0.9 43.0 47.3 62.6 45.1 n= 11974 263 1120 13357 11456 558 1158 13172 23430 821 2278 26529 na na na na % total (weighted) 89.9 1.9 8.2 100.0 87.4 4.1 8.5 100.0 88.7 3.0 8.4 100.0 48.9 68.3 50.5 49.7 source: abs 2016c notes: swa special working arrangements; nobase case for tests of proportions for the total category; * p<0.05 **p<0.01 ***p<0.001; % females percentage in each discrete category that are female; na not applicable. australian population studies 3 (1) 2019 temple et al. 20 table 2: whether used special working arrangements to care (%) by carer status and sex, 2015. males females total % female carer status total carer status total carer status total carer status total used swa to care? no primary other n= no primary other n= noprimary other noprimary other wanted to, but couldn't 0.7 2.7 1.1 0.8 103 0.9 2.3 1.9 1.0 121 0.8 2.4 ** 1.5 ** 0.9 53.7 65.0 63.1 55.9 no need 81.8 36.1 58.9 79.0 9838 77.1 31.9 49.9 72.9 9019 79.5 33.2 *** 54.3 *** 76.0 47.8 65.7 46.8 48.0 don't know 0.8 0.8 1.4 0.8 110 0.7 0.2 0.8 0.7 100 0.8 0.4 ** 1.1 0.8 47.0 35.6 38.1 45.7 yes, used 16.7 60.5 38.5 19.3 2503 21.3 65.7 47.4 25.4 3300 19.0 64.0 *** 43.0 *** 22.3 55.3 70.2 56.1 56.8 total 100 100 100 100 12554 100 100 100 100 12540 100 100 100 100 type of swa used to care: paid leave 42.2 42.4 42.0 42.2 1019 43.3 37.4 37.1 41.7 1326 42.8 38.9 39.3 ** 41.9 56.0 67.5 53.1 56.5 paid carer's leave 48.5 40.4 49.6 48.2 1218 46.1 42.0 50.9 46.4 1566 47.2 41.5 * 50.3 47.2 54.1 71.0 56.7 55.8 unpaid leave 4.3 9.7 6.6 5.0 123 7.8 9.6 7.7 8.0 246 6.2 9.6 * 7.3 6.7 69.4 69.9 59.9 67.8 unpaid carer's leave 1.9 7.0 5.7 2.8 65 3.4 10.4 6.7 4.7 146 2.7 9.4 *** 6.2 *** 3.9 68.5 77.8 60.0 68.3 flexible working hours 14.5 25.8 16.4 15.5 375 15.1 24.6 20.7 17.0 553 14.8 25.0 *** 18.8 ** 16.3 56.4 69.2 61.7 59.1 rostered day off 4.7 13.0 10.8 6.2 134 2.7 3.9 6.1 3.4 98 3.6 6.6 * 8.1 *** 4.6 41.5 41.3 41.9 41.6 working from home 9.7 13.3 8.7 9.7 224 7.8 7.5 7.2 7.6 241 8.6 9.2 7.9 8.5 49.9 56.8 51.3 50.7 shift work 1.4 2.0 1.3 1.4 33 1.0 1.2 1.2 1.1 35 1.2 1.4 1.2 1.2 48.3 58.3 53.5 50.2 casual work 4.1 7.1 4.7 4.4 101 6.3 12.7 6.0 6.9 212 5.3 11.0 *** 5.4 5.8 65.4 80.7 61.9 67.4 part time work 1.3 5.5 1.6 1.6 45 6.2 11.6 4.8 6.6 219 4.0 9.8 *** 3.4 4.4 85.3 83.3 78.9 84.2 informal arrangement 3.5 7.2 5.4 4.0 90 3.0 6.9 3.2 3.4 106 3.2 7.0 * 4.2 3.7 51.4 69.2 43.1 52.8 other 0.7 1.6 0.0 0.6 11 0.3 0.0 0.6 0.3 9 0.5 0.5 0.3 0.5 36.6 0.0 100.0 40.8 source: abs 2016c notes: swa special working arrangements; nobase case for tests of proportions for the total category; * p<0.05 **p<0.01 ***p<0.001; % females percentage in each discrete category that are female; n= unweighted sample count. australian population studies 3 (1) 2019 temple et al. 21 4.3. barriers to special working arrangements referring back to figure 1, of the 5,803 workers who indicated they had used swa to care, about 15.5% (n=891) reported they wanted to use more but were unable to. this group were asked what swa they wanted to use but couldn’t (table 3). once more, paid leave and paid carers leave were the most likely swa that workers wished to make more use of to care. interestingly, there is little difference in the swa type needed by carer status. however, primary carers (24.6%) and other carers (21.8%) were more likely to report a barrier to further use of swa, relative to non-carers (13.1%). types of swa needed are relatively consistent by gender. however, male other carers are more likely to cite the need for rostered day off (11.8% versus 2.5%) whereas female other carers are more likely to cite the need for paid leave (34.7% versus 23.9%). again, due simply to the overrepresentation of female primary carers, about 70% of primary carers with unmet demand for further access to swa were female. results in table 4 report the reasons for barriers to using swa to care, for two groups. the first ‘additional use of swa’ are the 15.5% (n=891) group who used swa, but wanted to use more and couldn’t. regardless of carer status, the main reasons for not using further swa to care were ‘not enough paid leave left’ (29%), ‘work commitments (23.4%) and ‘nature of work makes using flexible working arrangements difficult’ (16%). although in the minority, it is concerning to note that 8% cited ‘subtle or other pressures from bosses or other workers’ or ‘didn’t apply as thought they would say no anyway’. primary carers were slightly more likely to cite this later reason (12%) compared with non-carers. the second group in table 4, titled ‘any use of swa’ are those who had swa available to them, but could not use it to care at all. referring back to figure 1, this is the smaller group of 1.2% (n=224) of those who didn’t use swa to care. as with the previous group, about 1 in 4 report ‘work commitments’ as the main reason for not using swa to care (24.3%). approximately 15% reported ‘didn’t apply as thought they would say no’, ‘nature of work makes using flexible working arrangements difficult’ or ‘subtle or other pressure from bosses or other workers’. due to the small sample size of this population, there are few statistically significant differences in reasons given for barriers to any use of swa by carer status. moreover, there is insufficient sample size to disaggregate this table by gender. australian population studies 3 (1) 2019 temple et al. 22 table 3: unmet demand for further access to special working arrangements to care (%) by carer status and sex, 2015. males females total % female carer status total carer status total carer status total carer status total more use of swa? no primary other n= no primary other n= noprimary other noprimary other no 83.2 66.2 69.5 79.9 2010 84.6 69.6 75.9 81.6 2694 83.9 68.6 *** 73.1 ** 80.9 55.9 71.3 58.4 57.4 don’t know 3.1 8.3 5.9 3.9 96 2.8 6.3 4.4 3.4 112 2.9 6.9 ** 5.1 ** 3.6 52.6 64.2 49.1 53.8 yes 13.7 25.5 24.5 16.2 397 12.6 24.1 19.7 15.0 494 13.1 24.5 *** 21.8 *** 15.5 53.5 69.1 50.7 55.0 total 100 100 100 100 2503 100 100 100 100 3300 100 100 100 100 type of swa needed paid leave 36.6 36.1 23.9 33.4 131 35.0 32.5 34.7 34.5 163 35.7 33.6 29.4 34.0 52.4 66.8 60.0 55.8 paid carer's leave 42.6 37.4 39.8 41.4 170 42.6 44.8 44.1 43.3 221 42.6 42.5 42.0 42.4 53.5 72.8 53.3 56.1 unpaid leave 2.0 10.1 5.9 3.7 15 5.5 2.4 4.6 4.8 25 3.9 4.8 5.3 4.3 76.5 34.8 44.4 61.3 unpaid carer's leave 1.8 8.1 1.4 2.3 9 3.9 10.1 3.9 5.0 19 2.9 9.5 2.7 3.8 71.3 73.6 74.5 72.6 flexible working hours 10.4 4.0 8.5 9.3 43 7.2 13.5 8.2 8.5 48 8.7 10.6 8.3 8.9 44.4 88.2 49.9 52.7 rostered day off 1.0 6.4 11.8 4.2 13 1.0 2.1 2.5 1.5 8 1.0 3.4 7.1 ** 2.7 52.7 42.1 18.0 30.3 working from home 8.3 4.5 11.8 8.8 35 6.6 6.2 4.7 6.2 32 7.4 5.7 8.2 7.4 47.7 75.7 29.3 46.0 shift work 0.5 2.8 1.3 0.9 4 0.2 1.3 0.0 0.3 2 0.3 1.8 0.7 0.6 26.0 50.5 0.0 29.4 casual work 2.5 1.2 5.6 3.2 12 4.7 2.8 4.7 4.4 18 3.7 2.3 5.1 3.8 67.9 84.4 46.7 62.8 part time work 0.0 2.1 1.4 0.5 3 1.1 3.4 1.8 1.6 13 0.6 3.0 1.6 1.1 100.0 78.4 57.6 78.7 informal arrangement 1.5 0.7 1.1 1.3 6 1.7 1.7 0.3 1.4 8 1.6 1.4 0.7 1.4 56.2 84.0 20.2 56.0 other 0.2 0.0 0.0 0.1 1 0.1 0.0 0.0 0.0 1 0.1 0.0 0.0 0.1 27.5 0.0 0.0 27.5 source: abs 2016c notes: swa special working arrangements; nobase case for tests of proportions for the total category; * p<0.05 **p<0.01 ***p<0.001; % female percentage in each discrete category that are female; n= unweighted sample count. australian population studies 3 (1) 2019 temple et al. 23 table 4: reasons for barrier to additional or any use of special working arrangements (%) by carer status, 2015. additional use of swa [1] any use of swa [2] carer status all carer status all reason for barrier no primary other no primary other applied or asked but was refused 5.4 8.1 4.0 5.4 8.0 15.0 8.6 8.6 do not have adequate working arrangements 9.2 6.2 10.1 9.0 5.9 6.6 15.0 7.2 didn't apply as thought they would say no anyway 6.7 11.9 * 9.0 7.9 15.0 23.9 11.1 15.1 concern that using arrangements would be viewed poorly 0.9 1.0 0.0 * 0.7 0.1 0.0 3.7 0.6 nature of work makes using flexible working arrangements difficult 15.2 16.5 16.9 16.0 14.7 15.2 21.8 15.5 work commitments 22.3 30.6 # 22.0 23.4 24.9 33.9 15.0 24.3 not paid for time off (e.g. casual/shift worker) 6.6 5.2 8.2 6.8 8.8 9.9 21.0 # 10.6 subtle or other pressure from bosses or other workers 7.0 12.1 8.1 8.0 14.1 7.3 9.9 13.0 not enough paid leave left or available 31.0 26.2 25.2 29.0 6.7 18.3 10.8 8.2 own circumstances made use of arrangements unfeasible 1.3 0.0 ** 2.9 1.5 3.3 0.9 0.0 * 2.7 anything else 3.9 3.4 3.7 3.8 7.7 7.0 8.2 7.7 don't know 5.8 5.0 6.5 5.9 8.9 4.9 0.8 7.5 n 573 128 190 891 170 18 36 224 source: abs 2016c notes: swa special working arrangements; nobase case for tests of proportions; # p<0.10 * p<0.05 **p<0.01 ***p<0.001; n= unweighted sample count; [1] population of those who had already used swa, but have unmet demand for further use of swa to care; [2] population of those who have not used swa and wanted to use them to care. australian population studies 3 (1) 2019 temple et al. 24 5. discussion the clear majority of respondents reported the availability of swa in their workplace (94%) and indeed, carers were slightly more likely to cite awareness of flexible working hours, unpaid carers leave, part time work and informal arrangements with their employer than non-carers. it may be that carers are more aware of certain swa, because of their personal requirement to have them as a necessary condition of employment. that is, they self-select into positions with swa that enable a continuation of unpaid caring responsibilities. a further explanation is that they are more aware of swa regardless of occupation or industry, simply because they have a greater need for swa than non-carers. swa are taken up by a majority of primary carers (64%) and 43% of other carers (within the last six months) with paid leave and paid carers leave being the most used arrangements. however, these arrangements are not perceived as sufficient for 15.5% of the workforce surveyed, with primary carers (24.6%) and other carers (21.8%) more likely to report barriers to further use of swa than non-carers (13.1%). the majority of unmet demand is by those who already use it, but need to make more use of it. supporting previous research, our results underscore the gendered nature of care giving responsibilities in australia and show the interaction with swa usage. almost 70% of employed primary carers using swa to care were female, compared with 55% of other and non-carers. around 80% of those using part-time work to care were female, regardless of carer status. among primary carers specifically, around 80% of those using part-time work, casual work, or unpaid carers leave to care are female, regardless of carer status. while there has been considerable attention on the needs of mothers for flexible work arrangements in australia, our results also highlight the un-met needs of workers who provide care for those with a disability and long-term health problems. women also overwhelming attend to these various care needs, altering their work schedules to do so and experiencing negative consequences in terms of career and financial penalties. these findings regarding carers are supported by a growing body of research and data points to both the gendered use of swas in the general population (i.e. not just carers), including part-time work, and the gap between availability of swas and their uptake. mcdonald et al.(2005 p.41) argue that men, and women in managerial roles, are less likely to use swas due to lack of managerial support, perceptions of negative career consequences, lack of co-worker support and time pressures at work. in a study of a large telecommunications company, cooper and baird (2015) found that use of australia’s legislative right to request ‘flexible work arrangements’ was heavily dominated by working mothers when they were supported by their line managers. in a comparative review, thornthwaite (2004) shows that working mothers, more than fathers, in australia prefer part-time work and access to flexible hours to accommodate the gendered division of labour in the home, school hours and lack of suitable child care. for both male and female carers, the main barrier that carers face is insufficient paid leave and/or work commitments (requiring them to put the demands of the workplace over their need for time off to care) but to a lesser extent they face subtle or real pressure from others not to apply for leave or other flexible work arrangements. around 5% of those who wanted to access further swa for care and 9% of those who wanted to access any swa cited a rejection by the employer. australian population studies 3 (1) 2019 temple et al. 25 under the national employment standards, employers can refuse a request on ‘reasonable business grounds’. these grounds are very broad and include:  requested arrangements are too costly for the employer  there is no capacity to change the current working arrangements  impractical to change extant working arrangements  the requested arrangement would result in a ‘significant loss of efficiency or productivity’  the requested arrangement would be likely to have a significant negative impact on customer service. in the case of a refused request, the employee can seek assistance from the fair work commission. however, the fair work commission does not have the legislative power to direct an employer to agree to the employee’s request. therefore, there is limited right to appeal the employers’ refusal. moreover, there are numerous caveats to those eligible to access flexible arrangements under the act. for example, workers must have been employed for at least 12 months before making a request for flexible working arrangements. stronger legislative controls have recently come into force following a ruling from the full bench of the fair work commission. from 1 december 2018, employers must provide a detailed written response outlining the reason for the refused request. in considering the request, employers must consider (1) the needs of the employee, (2) consequences for the employee is changes in working arrangements aren’t made, as well as (3) reasonable business grounds for refusing the employees request (fwo 2019). rather than being considered as part of labour market policy alone, access to flexible working arrangements for carers should be seen as an integral part of the health system supporting australia’s ageing population and those with underlying health conditions (yeandle and cass 2013; vecchio 2015). indeed, in vecchio’s (2015) important study of the labour force behaviour of those residing with a person with a disability, they argue “policy makers need to aggressively challenge current workforce programs to encourage employers to provide more flexible work arrangements. workforce programs that allow family members opportunities to provide adequate caregiving and financial support to ill relatives are fundamental to the sustainability of health care programs” (p. 9). there is currently no overarching policy framework in australia that addresses this issue. there was a national carer strategy in australia (2011) but this was not adopted by the current government. the previous labor federal government commenced the implementation of a national carer action plan 2011-2014, which had a range of strategies for improving the economic security of carers, including strategies for enabling workforce participation. however, the current coalition government has restricted its focus to carers to carer support services accessible through the carer gateway (department of social services 2017). with the role out of the national disability insurance scheme (ndis), early evidence suggests that carers employment prospects have not improved (hamilton 2018). one possibility is to allow carers access to ndis to support employment needs. as argued by hamilton (2018) “this would be an important start in developing services that actually support carers to work – not just hoping that carers gain time for paid work by altering disability services”. australian population studies 3 (1) 2019 temple et al. 26 in interpreting the results from this study, it is important to note the limitations. first, data from the sdac are cross-sectional and the measures are collected only for the employed. we cannot, for example, determine whether carers have ceased employment due to a lack of flexible working arrangements, although evidence suggests this may be the case (austen et al. 2013). secondly, our measure of the usage of swa is retrospective, over a six month period only. third, the analyses we present herein are descriptive. further multivariable analysis is required, particularly of the industry, workplace and characteristics of carers reporting a barrier to accessing swa to care. 6. conclusions with the ageing of the australian population, there is a need for more family carers at the same time as for greater workforce participation, especially amongst women. this study suggests that to enable people to deliver their caregiving responsibilities and participate in the workforce, more paid leave is required as well as other swa strategies that encourage employers to recognize and support carers in the workplace. the business case for offering swa is growing, with a number of large employers now offering ‘flex for all’. mcdonald et al. (2005 p. 38), citing a number of studies, outline the benefits for business as ‘improving the retention or recruitment of skilled women, reducing absenteeism, increasing productivity and reducing hiring and retraining costs’. there are also numerous coted benefits to employees of being able to access swas. these include reduction in personal stressors and improvements in the mental and physical health of workers (mcdonald et al. 2005 p. 39). given that financial considerations are among the strongest unmet needs reported by australian carers, improving the fit between paid and unpaid work is one strategy by which improvements in economic wellbeing can be achieved (temple and dow 2018). more generally, greater awareness of the societal and economic benefits of caring are needed so that workplace arrangements are not just seen as an individual benefit and carers can feel legitimized in accessing swa. stronger legislative arrangements are also required to enforce employers to grant carers flexible working arrangements. recent strengthening of legislative controls have recently come into force following a ruling from the full bench of the fair work commission and will require evaluation with further data collections. another possible solution would be government subsidies to enable carers to take more leave. whatever policy and practice levers are put in place, thorough evaluation is required to improve knowledge of which swa are effective in enabling carer workforce participation. key messages  population ageing and demographic change are increasing the number of unpaid informal carers who face barriers to labour market attachment at the same time that the australian labour market is projected to experience lower growth and reduced participation.  many australian carers face barriers to employment, as well as barriers to underemployment by those already employed. although the vast majority of workers are aware of swa and about 1 in 5 have used swa to care, almost 1 in 4 primary carers and 1 in 5 other carers faced barriers to using swa to care. australian population studies 3 (1) 2019 temple et al. 27  the main barriers faced by employed carers included insufficient paid leave and/or work commitments (requiring them to put the demands of the workforce over their need for time off to care).  the results also underscore the gendered nature of care responsibilities in australia. almost 70% of employed primary carers using swa to care were female, compared with 55% of other and non-carers. among primary carers specifically, around 80% of those using part-time work, casual work, or unpaid carers leave to care were female.  stronger legislative arrangements are required to assist carers to access swa. acknowledgements data for this study were made available to the authors by the australian bureau of statistics (abs). jeromey temple is funded by the australian research council’s centre of excellence in population ageing research (ce1101029). the opinions expressed herein are those of the authors alone. references abs (2016a) disability, ageing and carers, australia: summary of findings. catalogue no. 4430.0. canberra: abs. available from: https://www.abs.gov.au/ausstats/abs@.nsf/mf/4430.0. accessed may 2018. abs (2016b) a profile of carers in australia. information sheet. canberra: abs. available from: https://www.abs.gov.au/ausstats/abs@.nsf/lookup/4430.0main+features602015. accessed may 2018. abs (2016c) microdata: disability, ageing and carers, australia, 2015. catalogue no. 4430.0.30.002. canberra: abs. austen s and ong r (2013) the effects of ill health and informal care roles on the employment retention of mid-life women: does the workplace matter? journal of industrial relations 55(5): 663-680. australian treasury (2002) intergenerational report 2002-03. canberra: australian government. australian treasury (2007) intergenerational report. canberra: australian government. australian treasury (2010) intergenerational report. australia to 2050: future challenges. canberra: australian government. australian treasury (2015) intergenerational report: australia in 2055. canberra: australian government. bittman m, hill t and thomson c (2007) the impact of caring on informal carers’ employment, income and earnings: a longitudinal approach. australian journal of social issues 42(2): 255-272. cooper r and baird m (2015) bringing the 'right to request' flexible working arrangements to life: from policies to practices. employee relations 37 (5): 568-581. cummins r, hughesj, tomyn a, gibson j, woerner j and lai, ll. (2007) the wellbeing of australians carer health and wellbeing, in australian unity wellbeing index surveys. australian centre on quality of life, deakin university. melbourne. deloitte access economics (2015) the economic value of informal care in 2015. canberra: deloitte access economics. department of social services (2017) disability and carers carer gateway. https://www.dss.gov.au/ourresponsibilities/disability-and-carers/programmes-services/for-carers/carer-gateway. accessed 25 nov 2017. fwo (2018) requests for flexible working arrangements and the national employment standards. fair work ombudsman: canberra. available from: www.fairwork.gov.au. accessed may 2018 https://www.abs.gov.au/ausstats/abs@.nsf/mf/4430.0 https://www.abs.gov.au/ausstats/abs@.nsf/lookup/4430.0main+features602015 https://www.dss.gov.au/our-responsibilities/disability-and-carers/programmes-services/for-carers/carer-gateway https://www.dss.gov.au/our-responsibilities/disability-and-carers/programmes-services/for-carers/carer-gateway http://www.fairwork.gov.au/ australian population studies 3 (1) 2019 temple et al. 28 fwo (2019) flexible working arrangements. fair work ombudsman: canberra. available from: www.fairwork.gov.au. accessed may 2019. gray m, edwards b and zmijewski n (2008) caring and women’s labour market participation. family matters 78: 28-35. hamilton m (2018) the ndis hasn’t made much difference to carers’ opportunities for paid work. the conversation. available from: https://theconversation.com/the-ndis-hasnt-made-muchdifference-to-carers-opportunities-for-paid-work-98157. accessed may 2019. hegewisch a and gornick j (2011) the impact of work-family policies on women’s employment: a review from oecd countries. community, work and family 14(2): 119-138. hill t, thomson c, bittman m and griffiths m (2008) what kinds of jobs help carers combine care and employment? family matters 80: 27-32. lilly m, laporte a and coyte p (2007) labour market work and home care’s unpaid caregivers: a systematic review of labor force participation rates, predictors of labor market withdrawal and hours of work. the milbank quarterly 85(4): 641-690. mcdonald p, brown k and bradley l (2005) explanations for the provision‐utilisation gap in work‐life policy. women in management review 20(1): 37-55. mcdonald p (2017) international migration and employment growth in australia, 2011-2016. australian population studies 1(1): 3-12. murray j (2005) work and care: new legal mechanisms for adaptation. labour and industry 15(3): 67-87. o’loughlin k, loh v, and kendig h (2017) carer characteristics and health, wellbeing and employment outcomes of older australian baby boomers. journal of cross cultural gerontology 32(3): 339356. pavalko e and henderson k (2006) combining care work and paid work: do workplace policies make a difference? research on aging 28(3): 359-374. pinquart m and sorensen s (2003) differences between caregivers and noncaregivers in psychological health and physical health: a meta-analysis. psychology and aging 18(2): 250-267. productivity commission (2005) economic implications of an ageing australia. research report. canberra: productivity commission. productivity commission (2013) an ageing australia: preparing for the future. research report. melbourne: productivity commission. schofield d, cunich m, shrestha r, passey m, kelly s, tanton r, and veermen l (2014) the impact of chronic conditions of care recipients on the labour force participation of informal carers in australia: which conditions are associated with higher rates of non-participation in the labour force? bmc public health 14(1): 1-9. skinner n and pocock b (2014) the persistent challenge: living, working and caring in australia in 2014. centre for work + life, university of south australia. available from: https://www.unisa.edu.au/siteassets/episerver-6files/documents/eass/cwl/publications/awali_2014_national_report_final.pdf. accessed may 2018. stata corporation (2017) stata/se 15.0 for windows 64 bit. college station: texas. temple j and dow b (2018) the unmet support needs of carers of older australians: prevalence and mental health. international psychogeriatrics 30(12): 1849-1860. temple j and mcdonald p (2017) population ageing and the labour force: 2000-2015 and 2015-2030. australasian journal on ageing 36(4): 264-270. thornthwaite l (2004) working time and work–family balance: a review of employees’ preferences. asiapacific journal of human resources 42(2): 165-184. http://www.fairwork.gov.au/ https://theconversation.com/the-ndis-hasnt-made-much-difference-to-carers-opportunities-for-paid-work-98157 https://theconversation.com/the-ndis-hasnt-made-much-difference-to-carers-opportunities-for-paid-work-98157 https://www.unisa.edu.au/siteassets/episerver-6-files/documents/eass/cwl/publications/awali_2014_national_report_final.pdf https://www.unisa.edu.au/siteassets/episerver-6-files/documents/eass/cwl/publications/awali_2014_national_report_final.pdf australian population studies 3 (1) 2019 temple et al. 29 vecchio n (2015) labour force participation of families coping with a disabling condition. economic analysis and policy 45(march): 1-10. yeandle s and cass b (2013) working carers of older people: steps towards securing adequate support in australia and england. in kroger t and yeandle s (eds.) combining paid work and family care: policies and experiences in international perspective. bristol: polity press; 71-87. australian population studies 2019 | volume 3 | issue 1 | pages 30-39 © norman, berrie & exeter 2019. published under the creative commons attribution-noncommercial licence 3.0 australia (cc by-nc 3.0 au). journal website: www.australianpopulationstudies.org introductory guide calculating a deprivation index using census data paul norman* university of leeds laurie berrie university of leeds daniel j exeter university of auckland * corresponding author. email: p.d.norman@leeds.ac.uk. address: school of geography and leeds institute for data analytics, university of leeds, leeds ls2 9jt, united kingdom paper received 30 october 2018; accepted 4 april 2019; published 27 may 2019 abstract background deprivation indexes have widespread use in academic research and in local and national government applications. it is useful for people to understand their construction and to be able to calculate their own measures. aims we provide an overview of the background to area based deprivation measures. we detail and explain a series of steps taken to calculate a deprivation index for small areas in australia. data and methods we use data from australia’s 2016 census of population and housing for the sa2 level of geography. after defining the set of variables used as inputs, we emulate the steps taken to calculate other census based deprivation indexes. results the resulting scheme correlates closely with an official, but more sophisticated deprivation measure, suggesting that simple schemes have utility. conclusions there are choices to be made for input variables and for some of the detail of the calculations. researchers can follow the steps we describe to develop their own measures. key words area deprivation; composite measures; small areas; census http://www.australianpopulationstudies.org/ mailto:p.d.norman@leeds.ac.uk australian population studies 3 (1) 2019 norman, berrie & exeter 31 1. introduction the use of area deprivation indexes is widespread across academic research and in local and national government applications (allik et al. 2016). the australian bureau of statistics produce socioeconomic indexes for areas (seifa) using census data, which includes the index of relative socioeconomic disadvantage (irsd). the latest version uses 2016 census data. our aim here is to demonstrate a method of calculation to people, such as academic and local government researchers, who may be new to the production of composite measures. following sections on generic aspects, this paper steps through the process of creating a deprivation index using census data for small areas (sa2s) in australia for 2016. we take a similar approach to the calculation steps used in the long-established townsend (townsend et al. 1988) and carstairs (carstairs and morris 1989) indexes. we compare the deprivation index calculated here with the seifa irsd. we provide data files so that researchers can reproduce the deprivation index and believe the method used here to be accessible to quantitative population geographers new to this kind of work who will then be able to construct their own composite measures. these measures might focus on particular dimensions of deprivation or be for other purposes for which a combination of indicator variables are applicable. 2. what is deprivation? deprivation is one of those things which we probably think we understand until we tried to define it! townsend's (1987 p. 125) definition is commonly used: “deprivation may be defined as a state of observable and demonstrable disadvantage relative to the local community or the wider society or nation to which an individual, family or group belongs. the idea has come to be applied to conditions (that is, physical, environmental and social states or circumstances) rather than resources and to specific and not only general circumstances, and therefore can be distinguished from the concept of poverty”. defining poverty itself is far from easy (paterson and gregory 2019). from a eurocentric perspective, deprived areas are places in which concentrations of people and households with the relevant characteristics exist. 3. what is a deprivation index? a deprivation index provides a single figure estimate of the level of deprivation for each location within the study area. since deprivation is both hard to define and cannot be directly measured, the index is a latent construct calculated using a combination of indicator variables believed to capture relevant dimensions. having a single figure per area simplifies applications that use the index compared to using multiple variables. the deprivation index value for any one area is usually expressed relative to a larger study region, often the national level (fu et al. 2015). 4. which variables should be used as inputs? as noble et al. (2006 p. 172) point out, if a researcher is devising their own index, there should first be a clear theoretical foundation underpinning the notion of small area deprivation, which may then 32 norman, berrie & exeter australian population studies 3 (1) 2019 be operationalised as an index. deprivation indexes have their roots in census-based work in the uk, first by holtermann (1975) and subsequently by townsend (1987) and others. we cannot turn back the clock, but speculate that there may not have been the same motivation to construct deprivation indexes had the uk, unlike many other countries, included an income question in the census (dorling 1999; boyle and dorling 2004). thus, deprivation indexes in the uk use ‘proxy’ indicators of income and other dimensions of social and/or material hardship. unemployment is ubiquitously included in schemes but there is variation in further variables used as inputs (see senior 2002). in countries such as australia, canada and new zealand, where income-related questions are asked, measures of individual and/or household income are components of their deprivation indexes. the researcher needs to obtain variables which best represent particular aspects of deprivation to ensure the index is fit for purpose. the set of chosen variables should complement each other so that the index is ‘greater than the sum of its parts’. the availability of potential variables for inclusion will be affected by decisions on the geographic unit used (below). being based on data from a census, schemes are cross-sectional (though see the paper’s concluding section). to an extent there will be overlap of the facets chosen to proxy aspects of deprivation and likely to be strong interrelationships. since the purpose is to identify multiple deprivation this collinearity may not be an issue. however, an index calculated in the manner described here would tend to have a few key variables. a more multi-dimensional approach would need a technique such as principal components analysis (noted again below). 5. which geography should be used? most deprivation indexes are devised for small areas (sub local government). as above, since the roots of schemes lie in the uk and the census has been the main source of data, then census geographies have tended to be used. initially, within england and wales, the most common geographical units used were the electoral wards (mackenzie et al. 1998), while the carstairs index, originally developed for scotland, used postal sectors. these geographies were typically chosen in order to ensure that the count of people in each ward/sector was greater than 100, thus minimising the amount of missing / suppressed data. people are often motivated to develop an index for the smallest possible areas, to aid regeneration targeting, for example. if the aim is to relate to geographical health variations, then a geography which represents people’s day-to-day lives might be more suitable. ideally, the researcher would choose the geography which best suited the relevant situation. in reality, researchers may well not have the freedom of choice of geography they might wish for, as the smallest unit for which the desired indicators are available may also dictate the level of analysis. it is important to note that all results will be affected by the specification of the zones which are used (openshaw and taylor 1981; flowerdew et al. 2008). 6. a small area deprivation index for australia we now continue with the specifics involved with developing a small area deprivation index for australia. norman et al. (2016) calculated a measure of changing deprivation using data from the 2001 and 2011 censuses in australia. we will incorporate similar variables here for a cross-sectional australian population studies 3 (1) 2019 norman, berrie & exeter 33 deprivation index for 2016. the geography will be the statistical areas level 2 (sa2s). whilst not the smallest census geography, the sa2s are local level areal units for which detailed australian bureau of statistics data are released, including tables for geographic areas by a variety of person and household attributes. sa2s generally have populations between 3,000 and 25,000 persons with an average population of around 10,000 (abs 2017). 6.1. choose a set of indicator variables table 1 lists the variables we use. these have been chosen to represent a range of deprivation dimensions: unemployment (haynes et al. 1996); low income (d’ambrosio and frick 2007); lack of good english (bertotti et al. 2012); low educational achievement (wilkinson and pickett 2007) and; lone parents (santana 2002). these are here as examples and each can be substituted or there can be additional variables should a researcher wish to use different indicators. table 1: input variables from the 2016 census in australia (sa2) 2016 census table number & title variable input to index g04 ‘age by sex’ count of persons g13 ‘language spoken at home by proficiency in spoken english language by sex’ % persons who do not speak english well g16 ‘highest year of school completed by age by sex’ % persons not completing school to year 10 g29 ‘total household income (weekly) by household composition’ % households with income less than $1,500 (median income is $1,438) g39 ‘dwelling structure by household composition and family composition’ % dwellings with lone parent households g43 ‘labour force status by age by sex’ % persons unemployed source: australian bureau of statistics note: the variables expressed as a percentage are numerator / denominator x 100. the variables listed in table 1 have been obtained from the table number stated. the numerators and denominators are sourced from the same tables so that appropriate percentages can be calculated. the inputs used for these are included in this article’s online resources (noted at the end of the paper). the percentages represent variables whereby the higher the percentage, the greater the level of the deprivation. the polarity of the variables needs to be consistent. 6.2. excluding areas with very small populations in 2016 there were 2,301 sa2s in australia but since some have very small populations, we only include those with 200 or more persons so the deprivation index developed is for 2,169 sa2s. researchers could combine an area with a small population with an adjacent area so as not to lose data. 6.3. transforming and standardising the input variables many social variables have skewed distributions and researchers may choose to transform input variables to (near) normal distributions. the townsend index, for example, log transforms the 34 norman, berrie & exeter australian population studies 3 (1) 2019 percentages of unemployment and overcrowding as these are positively skewed. the need for, and methods of transformations, are subject to debate (gilthorpe 1995; senior 2002). for simplicity, we do not transform variables here. there is more of a consensus on the need to standardise variables. underpinning the need for this step is that each of the input variables will have differences in what are inherently large and small values. for example, at the sa2 level, the percentage of persons not completing school to year 10 varies between 4.02% and 70.42% with a mean of 20.00, whereas the percentage of lone parents varies between 2.16% and 36.05% with a mean of 10.96. there is a need to have variables on a comparable scale. the most common method (as used in both the townsend and carstairs schemes) is to standardise using z-scores. the inputs to the z-score calculation for the sa2 variables are the percentages of each variable. the resulting z-scores for all variables have a mean of zero and a standard deviation of 1 so they are now comparable. higher positive z-scores are higher levels of the indicator and more negative values are lower levels. z-scores are calculated using the formula: z-score = (observation mean) / standard deviation alternatively, the ‘range standardisation’ calculation can be used (lucy and burns 2017). range standardisation rescales observations for a variable to lie between 0 and 1. 6.4. combining the standardised input variables the aim is to have a single figure deprivation index and a regular approach is to sum the z-scores of the input variables. a simple ‘additive’, unweighted approach is used in the townsend and carstairs indexes so that all variables contribute equally. the jarman index (1983) applies weights to differentially change the impact of inputs when they are summed. in reality, the weights applied by jarman more likely reflect gp workload rather than a variable’s ability to better measure deprivation (senior 2002). for deprivation in australia, we sum the five sets of z-scores unweighted to thereby have a deprivation index for the sa2 geography. once summed, in terms of interpretation, more positive index scores reflect higher levels of deprivation and more negative values are lower levels of deprivation. since the indicators aim to capture deprivation, areas found to lack deprivation should not be described as relatively wealthy or affluent. an alternative method of combining a set of indicator variables is to use principal components analysis. the new zealand index of deprivation for 2013 used nine census-derived variables of deprivation and the weights of the first principal components are used to combine the nine variables into an overall deprivation score (atkinson et al. 2014). the seifa / irsd in australia also uses principal components analysis (abs 2018a). 6.5. categorising into quantiles index scores themselves have great utility, but many applications use the scores categorised into quantiles. it is common to use quintiles or deciles whereby the distribution is divided into fifths or tenths. often this is based on the number of areas or sometimes on the count of persons so that australian population studies 3 (1) 2019 norman, berrie & exeter 35 there would be an equal number of areas or persons in each category. the impact of differently defined categorisations has had little coverage in the literature (compared with discussions of transformation and standardisation). the use of ‘population weighted’ quantiles has become common and is what we adopt here. the quantile cut-offs partition the deprivation distribution into categories with (near) equal numbers of people in each (table 2). table 2: counts of persons in each deprivation quantile deciles persons percent quintiles persons percent q1 (least deprived) 2,337,625 10.01 q1 (least deprived) 4,671,384 20.01 q2 2,333,759 9.99 q3 2,333,735 9.99 q2 4,668,398 19.99 q4 2,334,663 10.00 q5 2,336,028 10.00 q3 4,672,799 20.01 q6 2,336,771 10.01 q7 2,327,819 9.97 q4 4,660,545 19.96 q8 2,332,726 9.99 q9 2,332,560 9.99 q5 (most deprived) 4,677,180 20.03 q10 (most deprived) 2,344,620 10.04 total 23,350,306 100.00 total 23,350,306 100.00 source: based on australian bureau of statistics data 7. comparison of the deprivation index with an existing scheme the australian bureau of statistics produce socio-economic indexes for areas (seifa), which includes the index of relative socio-economic disadvantage (irsd). the most recent version uses 2016 census data. note that there is a difference in polarity from the census measure we have developed here, so a low irsd score indicates relatively greater disadvantage whereas a high score indicates a relative lack of disadvantage (abs 2018b). the deprivation index developed in our analysis correlates -0.933 with the irsd suggesting that the two schemes identify similar patterns of (non-) deprivation. a cross-tabulation of deciles from both schemes also demonstrates fair similarity in the categorisation of areas (table 3). table 3: cross-tabulation of the deprivation index and seifa irsd quantiles deprivation index deciles q1 (least) q2 q3 q4 q5 q6 q7 q8 q9 q10 (most) total se if a ir sd q1 (most) 0 0 0 0 0 0 0 1 46 169 216 q2 0 0 0 0 0 1 25 49 111 31 217 q3 0 0 0 0 7 24 52 90 40 5 218 q4 0 0 0 9 25 52 67 43 21 1 218 q5 0 0 3 20 53 75 40 25 2 1 219 q6 0 0 11 63 70 49 18 6 0 0 217 q7 2 13 70 71 41 15 4 0 0 1 217 q8 4 65 93 34 15 4 1 0 0 0 216 q9 48 112 47 7 2 0 1 1 1 0 219 q10 (least) 177 34 1 0 0 0 0 0 0 0 212 total 231 224 225 204 213 220 208 215 221 208 2169 source: based on australian bureau of statistics data 36 norman, berrie & exeter australian population studies 3 (1) 2019 8. conclusion we have developed a deprivation index for the australian sa2 geography using five indicators derived from the country’s 2016 census. we have taken similar steps to those used to calculate the townsend and carstairs indexes by calculating percentages of the variables, which are then standardised and summed (unweighted) to form a single figure deprivation index. we categorise the deprivation scores into deciles and quintiles. the similarity between an index produced in this relatively simple way with the more sophisticated seifa irsd is consistent with previous findings in australia (norman et al. 2016). researchers wishing to develop their own indexes and make choices on geographic scale and input variables should be able to follow the same steps. it could be that an input to an existing scheme is regarded by others as an outcome of deprivation so there is a need for an index without that variable. the imd for england includes a health indicator so there is risk of a self-fulfilling prophecy if relationships between deprivation and health are investigated (adams and white 2004). exeter et al. (2017) overcame this by providing a version of the imd for new zealand which removes a domain (i.e. imd no health) and redistributes the weights for the remaining six domains to reduce issues associated with data circularity. a composite index for a different application can be developed using relevant inputs. see, for example, congdon (2004) on an index of social fragmentation and lucy and burns (2017) on an index of loneliness. two aspects will be relevant to people wanting to build on this work relating to alternative data sources or extending from cross-sectional measures to those which capture changing deprivation. a variety of deprivation schemes now use administrative data sources (e.g. noble et al. 2006; exeter et al. 2017). with a move away from reliance on census data, the same approach as we use here can have administrative data inputs. a scheme for england (ajebon and norman 2016) using a small number of administrative variables as inputs relates closely to both a census derived measure as well as the more sophisticated, multiple domain imds (noble et al. 2006) now in common usage. census-based deprivation schemes are specific to the period of census data collection and administrative schemes for the year they are released (but may be based on inputs for different time points). all these are cross-sectional. however, areas change over time in their characteristics so increasingly researchers are interested in capturing this. examples of methods and applications can be found in norman (2010), exeter et al. (2011; 2019), norman et al. (2016), pearce et al. (2016), norman (2016), green et al. (2017), norman and darlington-pollock (2017) and shackleton et al. (2018). 9. resources we provide an excel file of the raw input variables plus commented files of spss syntax, stata do and r script so that people can reproduce the process we have described above. a working knowledge of these packages is needed. the files can be accessed via: http://dx.doi.org/10.17632/k957gctr9d.1. http://dx.doi.org/10.17632/k957gctr9d.1 australian population studies 3 (1) 2019 norman, berrie & exeter 37 key messages  this introductory guide explains the background concepts relating to area deprivation and to data and geography choices.  the calculation of a deprivation index is stepped through and explained.  researchers are provided with a set of resources through which to reproduce the process.  researchers are encouraged to develop their own small area indexes. acknowledgements we are grateful for the feedback from two anonymous referees. references allik m, brown d, dundas r and leyland a h (2016) developing a new small-area measure of deprivation using 2001 and 2011 census data from scotland. health & place 39: 122-130. abs (2017) australian statistical geography standard (asgs). http://www.abs.gov.au/websitedbs/d3310114.nsf/home/australian+statistical+geography+st andard+(asgs). accessed 17 may 2018. abs (2018a) technical paper socio-economic indexes for areas (seifa) 2016. catalogue no. 2033.0.55.001. canberra: abs. http://www.ausstats.abs.gov.au/ausstats/subscriber.nsf/0/756ee3dbefa869efca258259000b a746/$file/seifa%202016%20technical%20paper.pdf. accessed 22 march 2019. abs (2018b) census of population and housing: socio-economic indexes for areas (seifa), australia, 2016. catalogue no. 2033.0.55.001. canberra: abs. http://www.abs.gov.au/ausstats/abs@.nsf/mf/2033.0.55.001. accessed 22 march 2019. adams j and white m (2006) removing the health domain from the index of multiple deprivation 2004 effect on measured inequalities in census measures of health. journal of public health 28(4): 379-383. ajebon m and norman p (2016) beyond the census: a spatial analysis of health and deprivation in england. geojournal 81(3): 395-410. atkinson j, salmond c and crampton p (2014) nzdep2013 index of deprivation. university of otago (wellington) https://www.otago.ac.nz/wellington/departments/publichealth/research/hirp/otago020194.h tml. accessed 27 september 2018. bertotti m, adams-eaton f, sheridan k and renton a (2012) key barriers to community cohesion: views from residents of 20 london deprived neighbourhoods geojournal 77(2): 223-234. boyle p and dorling d (2004) the 2001 uk census: remarkable resource or bygone legacy of the ‘pencil and paper era’? area 36(2): 101-110. carstairs v and morris r (1989) deprivation: explaining differences in mortality between scotland, england and wales. british medical journal 299(6704): 886-889. congdon p (2004) commentary: contextual effects: index construction and technique. international journal of epidemiology 33(4): 741-742. d’ambrosio c and frick j r (2007). income satisfaction and relative deprivation: an empirical link. social indicators research 81(3): 497-519. dorling d (1999) who’s afraid of income inequality? environment & planning a 31(4): 571-574. http://www.abs.gov.au/websitedbs/d3310114.nsf/home/australian+statistical+geography+standard+(asgs) http://www.abs.gov.au/websitedbs/d3310114.nsf/home/australian+statistical+geography+standard+(asgs) http://www.ausstats.abs.gov.au/ausstats/subscriber.nsf/0/756ee3dbefa869efca258259000ba746/$file/seifa%202016%20technical%20paper.pdf http://www.ausstats.abs.gov.au/ausstats/subscriber.nsf/0/756ee3dbefa869efca258259000ba746/$file/seifa%202016%20technical%20paper.pdf http://www.abs.gov.au/ausstats/abs@.nsf/mf/2033.0.55.001 https://www.otago.ac.nz/wellington/departments/publichealth/research/hirp/otago020194.html https://www.otago.ac.nz/wellington/departments/publichealth/research/hirp/otago020194.html 38 norman, berrie & exeter australian population studies 3 (1) 2019 exeter d j, boyle p j and norman p (2011) deprivation (im)mobility and cause-specific premature mortality in scotland. social science & medicine 72(3): 389-397. exeter d j, zhao j, crengle s, lee a and browne m (2017) the new zealand indices of multiple deprivation (imd): a new suite of indicators for social and health research in aotearoa, new zealand. plos one 12(8): e0181260. exeter d j, feng z, zhao j, cavadino a and norman p (2019) the geographic harmonisation of scotland’s small area census data, 1981 to 2011. health & place 57: 22-26. flowerdew r, manley d j and sabel s e (2008) neighbourhood effects on health: does it matter where you draw the boundaries? social science & medicine 66(5): 1241-1255. fu m, exeter d j and anderson a (2015) “so, is that your ‘relative’ or mine?” a political-ecological critique of census-based area deprivation indices. social science & medicine 142: 27-36. gilthorpe m (1995) the importance of normalisation in the construction of deprivation indices. journal of epidemiology and community health 49(s2): s45-s50. green m a, arcaya m and subramanian s v (2017) using internal migration to estimate the causal effect of neighbourhood socioeconomic context on health: a longitudinal analysis (england, 1995-2008). annals of the association of american geographers 107(6): 1266-1278. haynes r, gale s, lovett a and bentham g (1996) unemployment rate as an updatable health needs indicator for small areas. journal of public health 18(1): 27-32. holtermann s (1975) areas of urban deprivation in great britain: an analysis of 1971 census data. social trends 6: 33-45. jarman b (1983) identification of underprivileged areas. british medical journal 286: 1705-1709. lucy l and burns l (2017) devising a composite index to analyze and model loneliness and related health risks in the united kingdom. gerontology and geriatric medicine 3: 1-10. mackenzie i f, nelder r, maconachie m and radford g (1998) my ward is more deprived than yours. journal of public health medicine 20(2): 186-190. noble m, wright g, smith g and dibben c (2006) measuring multiple deprivation at the small-area level. environment and planning a 38(1): 169-185. norman p (2010) identifying change over time in small area socio-economic deprivation. applied spatial analysis and policy 3(2-3): 107-138. norman p (2016) the changing geography of deprivation in britain, 1971 to 2011 and beyond. in champion t and falkingham j (eds.) population change in the united kingdom. london: rowman and littlefield; 193-214. norman p and darlington-pollock f (2017) the changing geography of deprivation in great britain: exploiting small area census data, 1971 to 2011. in stillwell j (ed.) routledge handbook of census resources, methods and applications. abingdon, oxon: routledge; 404-420. norman p, charles-edwards e and wilson t (2016) relationships between population change, deprivation change and health change at small area level: australia 2001-2011. in wilson t, charles-edwards e and bell m (eds.) demography for planning and policy: australian case studies. cham: springer; 197-214. openshaw s and taylor p j (1981) the modifiable areal unit problem. in wrigley n and bennett r j (eds.) quantitative geography: a british view. london: routledge and kegan paul; 335-350. paterson l l and gregory i n (2019) defining and measuring poverty. in representations of poverty and place. cham: palgrave macmillan. https://doi.org/10.1007/978-3-319-93503-4_1 pearce j, shortt n, rind e and mitchell r (2016) life course, green space and health: incorporating place into life course epidemiology. international journal of environmental research in public health 13(3): 331. santana p (2002) poverty, social exclusion and health in portugal. social science & medicine 55(1): 33-45. https://www.sciencedirect.com/science/article/pii/s0277953607006272#! https://www.sciencedirect.com/science/article/pii/s0277953607006272#! https://www.sciencedirect.com/science/article/pii/s0277953607006272#! https://www.sciencedirect.com/science/article/pii/s0277953615300526#! https://www.sciencedirect.com/science/article/pii/s0277953615300526#! https://www.sciencedirect.com/science/article/pii/s0277953615300526#! https://doi.org/10.1007/978-3-319-93503-4_1 australian population studies 3 (1) 2019 norman, berrie & exeter 39 senior m l (2002) deprivation indicators. in rees p, martin d and williamson p (eds.) the census data system. chichester: john wiley; 123-137. shackleton n, darlington-pollock f, norman p, jackson r and exeter d (2018) longitudinal deprivation trajectories and risk of cardiovascular disease in new zealand. health and place 53: 34-42. townsend p (1987) deprivation. journal of social policy 16(2): 125-146. townsend p, phillimore p and beattie a (1988) health and deprivation: inequality and the north. andover: croom helm. wilkinson r g and pickett k e (2007). the problems of relative deprivation: why some societies do better than others. social science & medicine 65(9): 1965-1978. australian population studies 2017 | volume 1 | issue 1 | pages 26–40 © raymer, biddle and guan 2017. published under the creative commons attribution-noncommercial licence 3.0 australia (cc by-nc 3.0 au). journal website: www.australianpopulationstudies.org a multiregional sources of growth model for school enrolment projections james raymer* the australian national university nicholas biddle the australian national university qing guan the australian national university *corresponding author. email: james.raymer@anu.edu.au. address: school of demography, anu college of arts and social sciences, the australian national university, 9 fellows road, acton act 2601, australia paper received 9 may 2017; accepted 17 july 2017; published 20 november 2017 abstract background education planning requires accurate and efficient projection models. current projection models either do not make use of all available information and are reliant on idiosyncratic expert judgement, or are too complex to be maintained and explained. aims to test whether a multiregional projection model performs better than current methodology in explaining and projecting school enrolments in a school system with student mobility. data and methods a multiregional cohort model was developed for projecting enrolments for multiple schools or districts simultaneously. for illustration, data were obtained for all government schools in the australian capital territory (act) for the years 2008–2016. multiregional projections were compared with a cohort transition model and the act education directorate’s own projections. results (i) there is great diversity in the sources of school enrolment growth that need to be accommodated in enrolment projections; and (ii) multiregional projections perform slightly better than traditional methods with less effort and more transparency. conclusion a sources of growth approach guides the understanding of enrolment change, which is critical for making informed projections. key words enrolment projections; cohort transition; multiregional demography; australian capital territory; australia. http://www.australianpopulationstudies.org/ mailto:james.raymer@anu.edu.au australian population studies 1 (1) 2017 raymer j, biddle n and guan q 27 1. introduction school districts require information about the expectation of future school enrolments so that they can plan the supply of classrooms, teachers, materials and facilities. school enrolments are also important for understanding the education system within a larger social and economic context (stone 1965). without accurate projections, education may be disrupted because of inadequate facilities or a lack of teachers or other resources. alternatively, education may be over-budgeted in the provision of teachers, infrastructure and resources, thus causing a misallocation of public funds. while projections are never entirely certain due to unobserved factors influencing student mobility decisions (morrison 2000), there are a variety of demographic tools that can be applied to increase projection certainty and help better judgements to be made. surprisingly, in our review we find that most education jurisdictions neglect investments in testing or improving projection methodology and instead rely on relatively crude projection models with ad-hoc assumptions built in. in this paper, we assess a method to project future enrolments for multiple schools or school districts simultaneously. our method is illustrated with data obtained for the australian capital territory (act) from the act education directorate for the period 2008–2016. our aim is to provide better understanding of the demographic drivers enrolment change, and to develop more efficient projection systems. the multiregional projections are designed to study transitional changes that have occurred over time across school levels, as well as between schools and school regions. 2. review of methods used to project school enrolments sweeney and middleton (2005, p. 366) list several distinct approaches used to project school enrolments. these include cohort survival and regression methods that use past enrolment data and birth records, as well as extended versions of these methods which incorporate other information such as migration and student drop-out rates. they also list the modifiable spatial filter method (rushton, armstrong and lolonis 1995), which combines cohort survival methods with residential address data to determine the size of future school attendance areas. these methods can incorporate birth records, past and future housing unit data and the number of students each housing type yields. webster (1971) provides a similar list of projection approaches but also includes those based directly on age-specific population counts. johnstone (1974) further distinguishes between models that are deterministic (including markov chain models), stochastic and constrainable. one key issue is whether projections should be focused at the school level or at the broader district level. this also applies to projections for subgroups in heterogeneous populations (grip 2009). a bottom-up (school level) approach can be tailored to fit each school’s composition and needs; however, this approach may not consider wider demographic changes in the region around the school. a top-down (district level) approach, on the other hand, is often considered pragmatic because projections for large areas tend to be more accurate and less prone to rapid population changes (berk & associates 2008; grip 2009; schellenberg and stephens 1987, p. 13; stronge and schultz 1981; swanson and tayman 2012, p. 281). it can also incorporate wider demographic changes which are occurring in the region of interest. the main disadvantage is that a top-down approach may produce unrealistic results for schools which have trends or student profiles that are substantially different from other schools within their district. 28 raymer j, biddle n and guan q australian population studies 1 (1) 2017 in order for projections to be understood and utilised, the methodology and assumptions need to be transparent and replicable. the projection approach must also be flexible and able to be adapted to a variety of situations and changes that occur to enrolment patterns if the time-consuming process of developing specific projection models for each school or school district is to be avoided (gould 1993; johnstone 1974; stone 1965). this includes schools in established neighbourhoods, as well as those in areas that are currently being developed or planned. at times, it may be necessary to incorporate expert judgements or scenarios to account for future change possibilities (morrison 2000). furthermore, the projection methodology should be regularly assessed in terms of accuracy (how well it predicts the truth) and efficiency (amount of time and effort required to produce results) (swanson and tayman 2012). in most developed countries, including australia, there tends to be very high rates of grade progression (close to 100 per cent) in most schools. this implies the expectation that, for example, nine years from now nearly all students enrolled in academic level 1 today will be enrolled somewhere in level 10. moreover, there tends to be high retention of students at particular schools. the exception, of course, is for students transitioning between primary and secondary school or, in jurisdictions such as the act, from a government high school (years 7 to 10) to a secondary college (years 11 and 12). current bulges or dips in student numbers can be expected to move through the academic levels over time. if primary school enrolments are lower in one year, secondary or high school enrolments will follow at predictable times in the future. projections for new schools require a relational model where auxiliary information such as births, population projections, household composition and anticipated migration is used to estimate the long-term stable size and composition of future student enrolments. there are other situations where auxiliary information is also needed. for example, newly developed areas in middle or outer growth precincts may be expected to first attract young families with preschool and primary school aged children due to the relative affordability of land or housing for newly established household units. over time, these children will progress through the school system, with the school eventually reaching a steady state in respect to the number of new and continuing enrolments (herrick 1952). one approach which has not been considered widely in projecting future education requirements is a multiregional or multistate population model, which is an extension of the cohort transition model. multiregional population models provide a general and flexible platform for modelling and analysing subnational population change over time (rogers 1975, 1995; land and rogers 1982; schoen 1988). they enable the combination of all the main components of population change by age with various transitions that each population group may experience over the life course. these transitions may be between academic levels (including primary to secondary or high school or high school to a secondary college) or between government and non-government schools. despite the many theoretical and analytical advantages, multiregional models have been relatively unexplored because of the large amount of input data needed and requirement for matrix calculations to perform the projections and analyses. one exception is sweeney and middleton (2005), who applied a multiregional cohort enrolment projection method to better understand intradistrict school mobility and evaluate existing enrolment projections in santa barbara, california. their aim was to understand intra-district school transfers in an open enrolment system for a heterogeneous population. in this paper, we build on their ideas to improve school enrolment projections in the act. australian population studies 1 (1) 2017 raymer j, biddle n and guan q 29 3. analysing the sources of enrolment change the population of the act has grown substantially over the past decade and a half from 322,000 in 2001 to 398,000 in 2016 (act government 2017). this growth has been driven both by natural increase and net internal migration. net internal migration contributed only a small amount despite large flows as levels of in-migration tended to offset out-migration. the act’s average annual population growth rate between 2001 and 2014 was 1.4 per cent with a peak of 2.2 per cent occurring in the 2006–2007 period. the act has seven population regions: belconnen, gungahlin, north canberra, south canberra, weston creek, woden valley and tuggeranong.1 the relatively new region of gungahlin, located in canberra’s north, experienced the most rapid growth during this time, increasing from 25,000 in 2001 to 62,000 in 2014 with an average annual population growth rate of 7.4 per cent. in australia, like in many other federal systems, school planning is undertaken at the state or territory level. the act is considered a geographically small system by australian standards with four main levels of public school-based education: • preschool – the level before full-time schooling, denoted as level p • primary school – the first seven years of full-time schooling, referred to as level k (kindergarten) to level 6 • high school – the next four years of full-time schooling, or levels 7 to 10 • secondary college – the last two years of full-time schooling, or levels 11 and 12. the percentage of students in the act attending a government school increased from 57.5 per cent in 20082 to 59.6 per cent in 20163. the remainder were in the non-government school sector, with transitions between both systems as documented in this paper. the act education directorate is responsible for 88 public schools in the act comprising six early childhood schools (i.e., preschool to level 2), 51 primary schools (some with preschools attached), 10 high schools, nine combined primary/high schools, eight secondary colleges and four specialist schools. 4 record-level data from 2008 to 2016 taken at each year’s census in february were provided for this study. the data represents 611,674 student observations and contains the following variables: student id (anonymised); census date; school; academic level (levels p–12); suburb; state. from these data, the transitions between academic levels were calculated, as well as movements amongst schools for all students in the act. following rees and willekens (1986), data were obtained that could be used to identify the main sources of enrolment change. this accounting framework is outlined in table 1. the variable gij represents internal migration events from one school region i to another school region j. internal migration events are excluded when i = j. instead, terms ri are entered, which are accounting balances that capture the result of subtracting all possible exit events from the starting enrolment population. 1 note, for our analyses, the canberra central region is divided into north canberra and south canberra. a map of the schools and regions is available from http://www.education.act.gov.au/__data/assets/pdf_file/0005/73319/160210-actpublic-schools-map-2016.pdf 2 http://www.abs.gov.au/ausstats/abs@.nsf/allprimarymainfeatures/ffbe2ce6d8296d21ca2576ea0011f617?opendocument 3 http://www.abs.gov.au/ausstats/abs@.nsf/detailspage/4221.02016?opendocument 4 http://www.education.act.gov.au/publications_and_policies/publications_a-z/annual_report/annual-report-20152016/section-b/b.1-organisational-overview. specialist schools are excluded from this study. http://www.education.act.gov.au/__data/assets/pdf_file/0005/73319/160210-act-public-schools-map-2016.pdf http://www.education.act.gov.au/__data/assets/pdf_file/0005/73319/160210-act-public-schools-map-2016.pdf http://www.abs.gov.au/ausstats/abs@.nsf/allprimarymainfeatures/ffbe2ce6d8296d21ca2576ea0011f617?opendocument http://www.abs.gov.au/ausstats/abs@.nsf/detailspage/4221.02016?opendocument http://www.education.act.gov.au/publications_and_policies/publications_a-z/annual_report/annual-report-2015-2016/section-b/b.1-organisational-overview http://www.education.act.gov.au/publications_and_policies/publications_a-z/annual_report/annual-report-2015-2016/section-b/b.1-organisational-overview 30 raymer j, biddle n and guan q australian population studies 1 (1) 2017 this number represents students who transition to the next academic level while remaining in the school region. total internal out-migration from each school region are denoted by gi+ and total internal in-migration to each school region are denoted by g+j. the ii variable signifies the number of migration events from outside the system of interest and oi tabulates the corresponding number of persons leaving the system of interest. mi and ni denote transfers from and to non-government schools, respectively. the graduation events, xi, and preschool, bi, complete the flows in the table. note, in this table, persons who leave the school system without graduating are included in oi. table 1: a sources of growth accounting framework for government school enrolments across regions destination region region (t+1) (t) 1 2 3 4 5 6 7 ng mig grad total 1 r1 g12 g13 g14 g15 g16 g17 m1 o1 x1 te1 2 g21 r2 g23 g24 g25 g26 g27 m2 o2 x2 te2 3 g31 g32 r3 g34 g35 g36 g37 m3 o3 x3 te3 4 g41 g42 g43 r4 g45 g46 g47 m4 o4 x4 te4 5 g51 g52 g53 g54 r5 g56 g57 m5 o5 x5 te5 6 g61 g62 g63 g64 g65 r6 g67 m6 o6 x6 te6 7 g71 g72 g73 g74 g75 g76 r7 m7 o7 x7 te7 p b1 b2 b3 b4 b5 b6 b7 0 0 0 b+ ng n1 n2 n3 n4 n5 n6 n7 0 0 0 n+ mig i1 i2 i3 i4 i5 i6 i7 0 0 0 i+ total 1 1 +te 1 2 +te 1 3 +te 1 4 +te 1 5 +te 1 6 +te 1 7 +te m+ o+ x+ notes: (1) definitions of variables and subscripts: e = school enrolment population; r = balancing terms; g = internal migrations (within the act); p (b) = preschool enrolments; ng (m, n) = transfer to/from non-government schools; mig (o, i) = migration to/from other states within australia or overseas; grad (x) = exits due to graduation from government schools; 0 = structural zeroes; t = time; i = subscript for region; and + = summation over regions. (2) assumes zero deaths. the sum of the numbers in the rows of table 1 add up to the enrolments at the beginning of the time interval, t ie . the balancing term is obtained by subtracting the total number of migrations to other regions within the act, transfers to non-government schools, migrations to places outside the act, and graduations from the population at the beginning of the time interval, i.e., 𝑟𝑟𝑖𝑖 = 𝑒𝑒𝑖𝑖𝑡𝑡 − 𝑔𝑔𝑖𝑖+ − 𝑜𝑜𝑖𝑖 − 𝑚𝑚𝑖𝑖 − 𝑥𝑥𝑖𝑖 similarly, the variables in the columns of table 1 add up to the enrolments at the end of the time interval. we can compute these by adding to the balancing term the total in-migrations, migrations from outside the act, transfers from non-government schools and preschool entrants, i.e., 𝑒𝑒𝑖𝑖𝑡𝑡+1 = 𝑟𝑟𝑖𝑖 + 𝑔𝑔+𝑖𝑖 + 𝑖𝑖𝑖𝑖 + 𝑛𝑛𝑖𝑖 + 𝑏𝑏𝑖𝑖 if we combine these two equations, the balancing term cancels out and we obtain the components of school enrolment change: 𝑒𝑒𝑖𝑖𝑡𝑡+1 = 𝑒𝑒𝑖𝑖𝑡𝑡 − 𝑔𝑔𝑖𝑖+ − 𝑜𝑜𝑖𝑖 − 𝑚𝑚𝑖𝑖 − 𝑥𝑥𝑖𝑖 + 𝑔𝑔+𝑖𝑖 + 𝑖𝑖𝑖𝑖 + 𝑛𝑛𝑖𝑖 + 𝑏𝑏𝑖𝑖 thus, the information described in table 1 provides the basis for understanding the sources of enrolment change from one year to the next at the region level. australian population studies 1 (1) 2017 raymer j, biddle n and guan q 31 consider table 2, which contains the sources of growth and cohort transitions occurring across school regions in the act for levels p–11 in 2015 to levels k–12 in 2016. the row sums equal the p–11 enrolments for schools in each of the seven act regions. the column sums equal the corresponding k–12 enrolments one year later. between 2015 and 2016, 72,694 students made a transition. this included 5,393 students who moved out of the act and 6,387 students who moved to the act. over 80 per cent of the transitions occurred within the seven regions, with non-government schools having the highest retention. the proportion of students leaving government schools between 2015 and 2016 in the act ranged from 5.6 per cent in weston creek to 10.8 per cent in woden valley. in-migration was greatest to non-government schools (n=2,142), followed by gungahlin (n=823), belconnen (n=788) and north canberra (n=696). table 2: sources of enrolment change and cohort transition proportions for school regions in the act: levels p–11 (2015) to levels k–12 (2016) region enrolment region 2016 2015 gun bel scan tugg wod west ncan ng mig total enrolment change gun 5,635 84 21 8 9 6 55 332 628 6,778 bel 124 8,582 27 18 20 12 124 331 722 9,960 scan 12 17 3,455 32 145 3 14 129 392 4,199 tugg 33 32 22 6,010 146 38 32 346 706 7,365 wod 14 11 47 98 3,177 21 132 152 441 4,093 west 8 4 10 33 121 1,612 22 77 112 1,999 ncan 56 88 89 21 19 2 4,319 144 558 5,296 ng 190 133 140 214 125 30 128 23,823 1,834 26,617 mig 823 788 576 526 609 169 754 2,142 0 6,387 total 6,895 9,739 4,387 6,960 4,371 1,893 5,580 27,476 5,393 72,694 transition proportions gun 0.831 0.012 0.003 0.001 0.001 0.001 0.008 0.049 0.093 1.000 bel 0.012 0.862 0.003 0.002 0.002 0.001 0.012 0.033 0.072 1.000 scan 0.003 0.004 0.823 0.008 0.035 0.001 0.003 0.031 0.093 1.000 tugg 0.004 0.004 0.003 0.816 0.020 0.005 0.004 0.047 0.096 1.000 wod 0.003 0.003 0.011 0.024 0.776 0.005 0.032 0.037 0.108 1.000 west 0.004 0.002 0.005 0.017 0.061 0.806 0.011 0.039 0.056 1.000 ncan 0.011 0.017 0.017 0.004 0.004 0.000 0.816 0.027 0.105 1.000 ng 0.007 0.005 0.005 0.008 0.005 0.001 0.005 0.895 0.069 1.000 mig 823 788 576 526 609 169 754 2,142 0 6,387 notes: gun = gungahlin; bel = belconnen; scan = south canberra; tugg = tuggeranong; wod = woden valley; west = weston creek; ncan = north canberra; ng = act non-government schools; mig = migration to/from other states within australia or overseas. next, consider the cohort transitions presented in table 3 (next page) from level 10 to level 11. the diagonal elements represent those who remained in schools within the region and progressed from level 10 to level 11. most regions retained the majority of their students in the transition from high school to secondary college, except weston creek (0 per cent, where no secondary college is available) and woden valley (46%). tuggeranong had the highest retention (83%). the data clearly show that substantial movements occurred across school regions in the transition from high school to secondary college. 32 raymer j, biddle n and guan q australian population studies 1 (1) 2017 table 3: cohort transition proportions for school regions in the act: level 10 (2015) to level 11 (2016) region enrolment region 2016 2015 gun bel scan tugg wod west ncan ng mig total gun 0.807 0.048 0.013 0.000 0.003 0.000 0.033 0.008 0.090 1.000 bel 0.056 0.789 0.016 0.004 0.007 0.000 0.047 0.005 0.074 1.000 scan 0.010 0.005 0.829 0.025 0.005 0.000 0.005 0.020 0.101 1.000 tugg 0.009 0.009 0.018 0.826 0.055 0.000 0.000 0.005 0.078 1.000 wod 0.006 0.006 0.091 0.360 0.457 0.000 0.000 0.000 0.080 1.000 west 0.013 0.000 0.046 0.119 0.722 0.000 0.013 0.007 0.079 1.000 ncan 0.038 0.077 0.162 0.031 0.198 0.000 0.396 0.018 0.080 1.000 ng 0.048 0.023 0.043 0.042 0.026 0.000 0.016 0.747 0.055 1.000 mig 67 99 78 95 161 0 72 87 0 659 notes: gun = gungahlin; bel = belconnen; scan = south canberra; tugg = tuggeranong; wod = woden valley; west = weston creek; ncan = north canberra; ng = act non-government schools; mig = migration to/from canberra. gungahlin tuggeranong north canberra figure 1: components of enrolment change for schools in gungahlin, tuggeranong and north canberra, 2008–2016 -1200 -800 -400 0 400 800 1200 2008-09 2009-10 2010-11 2011-12 2012-13 2013-14 2014-15 2015-16 -1500 -500 500 1500 2008-09 2009-10 2010-11 2011-12 2012-13 2013-14 2014-15 2015-16 -1000 -500 0 500 1000 2008-09 2009-10 2010-11 2011-12 2012-13 2013-14 2014-15 2015-16 preschool enrolments graduates net migration within act net migration from outside act move to non-government schools net gain australian population studies 1 (1) 2017 raymer j, biddle n and guan q 33 between 2008 and 2016 most school regions in the act, except tuggeranong, experienced growth in enrolments each year. the sources and levels of enrolment change, however, varied considerably across the regions with some regions dependent on several sources of growth (e.g., south canberra, woden valley and non-government schools) and others driven primarily by preschool enrolments (e.g., belconnen, tuggeranong and weston creek). consider, for example, the sources of enrolment change during 2008–2016 for schools in the gungahlin, tuggeranong and north canberra regions (figure 1, previous page). gungahlin grew the most, with this growth particularly evident from 2011 with 500 to 560 new students added each year. preschool enrolments were the most significant and increasing component of growth for gungahlin. however, there was a steady net loss of students to non-government schools each year. preschool enrolments similarly were the only source of enrolment growth in the tuggeranong region between 2008 and 2016, declining marginally in 2015–2016. 4. cohort transition projection models three models are introduced for projecting enrolments by academic level in this section. the first is the standard cohort transition model based on annual enrolment numbers by academic level (johnstone and philp 1973; gould 1993; webster 1970). the second projection model is a multiregional cohort transition model that integrates all of the main sources of enrolment change presented in table 1. the third projection model is that utilised by the act education directorate, which, at the individual school level, combines cohort transition information with assumptions on household compositional changes over time in relation to planned developments in the school’s priority intake area. feedback from school principals on the plausibility of the results is also incorporated. to assess the accuracy of the projections, 2008–2012 data are used to predict the observed 2013– 2016 enrolments with the assumption that preschool enrolments are known. the purpose of the analysis is to see how well the two types of data-based projections perform against the act education directorate’s projections. 4.1 cohort transition model a simple and effective way to project school enrolments is to calculate the ratios of students progressing from one academic level to the next. while this does not explain changes to enrolment numbers from one year to the next, it does provide a measure of the overall change, which can be compared over longer periods. consider, for example, the observed enrolment data for schools in the gungahlin region from 2008– 2016 (table 4). here, we see that there were 316 level 1 students in 2008 (shaded grey). in 2009, this cohort of students grew to 325 level 2 students and then to 341 level 3 students in 2010. in 2016, the same cohort represented 385 level 9 students. note, as mentioned previously, gungahlin is a relatively new school region and has been growing rapidly due to migration from outside the act (level 11 became available in 2011 and led to the first cohort of level 12 students in 2012). cohort sizes in other school regions remained very similar or declined. 34 raymer j, biddle n and guan q australian population studies 1 (1) 2017 table 4: observed enrolment data by academic level for schools in gungahlin, 2008–2016 level 2008 2009 2010 2011 2012 2013 2014 2015 2016 p 589 543 579 619 633 736 815 888 898 k 361 402 379 432 516 538 613 699 773 1 316 373 423 397 460 549 580 586 693 2 326 325 346 429 428 471 531 570 598 3 335 331 341 378 433 443 485 553 553 4 296 350 360 316 401 459 468 494 575 5 309 318 357 350 350 423 461 497 521 6 309 303 319 349 390 380 440 485 502 7 242 273 284 264 339 339 332 387 458 8 229 259 266 295 278 354 360 367 392 9 204 231 260 271 284 298 365 380 385 10 203 204 245 261 289 321 316 399 415 11 0 0 0 308 369 413 502 479 570 12 0 0 0 0 303 311 396 489 458 notes: p = preschool; k = kindergarten. transition ratios calculated from the enrolment numbers in table 4 can be used to produce projections of school enrolments. the projection model is specified, in matrix form, as 𝐄𝐄𝑡𝑡+1 = 𝐆𝐆𝐄𝐄𝑡𝑡 + 𝐁𝐁 where 𝐄𝐄𝑡𝑡 are vectors denoting enrolments at time t and in academic level x, g is the transition matrix and b is a vector including the projected preschool enrolments. in detailed form, the matrix equation looks like ⎣ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎡𝑒𝑒𝑃𝑃 𝑡𝑡+1 𝑒𝑒𝐾𝐾𝑡𝑡+1 𝑒𝑒1𝑡𝑡+1 𝑒𝑒2𝑡𝑡+1 𝑒𝑒3𝑡𝑡+1 𝑒𝑒4𝑡𝑡+1 𝑒𝑒5𝑡𝑡+1 𝑒𝑒6𝑡𝑡+1 𝑒𝑒7𝑡𝑡+1 𝑒𝑒8𝑡𝑡+1 𝑒𝑒9𝑡𝑡+1 𝑒𝑒10𝑡𝑡+1 𝑒𝑒11𝑡𝑡+1 𝑒𝑒12𝑡𝑡+1⎦ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎤ = ⎣ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎡ 0 0 0 0 0 0 0 0 0 0 0 0 0 0 𝑔𝑔𝑃𝑃 0 0 0 0 0 0 0 0 0 0 0 0 0 0 𝑔𝑔𝐾𝐾 0 0 0 0 0 0 0 0 0 0 0 0 0 0 𝑔𝑔1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 𝑔𝑔2 0 0 0 0 0 0 0 0 0 0 0 0 0 0 𝑔𝑔3 0 0 0 0 0 0 0 0 0 0 0 0 0 0 𝑔𝑔4 0 0 0 0 0 0 0 0 0 0 0 0 0 0 𝑔𝑔5 0 0 0 0 0 0 0 0 0 0 0 0 0 0 𝑔𝑔6 0 0 0 0 0 0 0 0 0 0 0 0 0 0 𝑔𝑔7 0 0 0 0 0 0 0 0 0 0 0 0 0 0 𝑔𝑔8 0 0 0 0 0 0 0 0 0 0 0 0 0 0 𝑔𝑔9 0 0 0 0 0 0 0 0 0 0 0 0 0 0 𝑔𝑔10 0 0 0 0 0 0 0 0 0 0 0 0 0 0 𝑔𝑔11 0⎦ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎤ ⎣ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎡ 𝑒𝑒𝑃𝑃 𝑡𝑡 𝑒𝑒𝐾𝐾𝑡𝑡 𝑒𝑒1𝑡𝑡 𝑒𝑒2𝑡𝑡 𝑒𝑒3𝑡𝑡 𝑒𝑒4𝑡𝑡 𝑒𝑒5𝑡𝑡 𝑒𝑒6𝑡𝑡 𝑒𝑒7𝑡𝑡 𝑒𝑒8𝑡𝑡 𝑒𝑒9𝑡𝑡 𝑒𝑒10𝑡𝑡 𝑒𝑒11𝑡𝑡 𝑒𝑒12𝑡𝑡 ⎦ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎤ + ⎣ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎡𝑏𝑏 𝑡𝑡–4𝑔𝑔0 0 0 0 0 0 0 0 0 0 0 0 0 0 ⎦ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎤ where 𝑒𝑒𝑥𝑥𝑡𝑡 denotes enrolments at time t, 𝑔𝑔𝑥𝑥 denotes transitions between school levels with the subscript marking the origin academic level, 𝑏𝑏𝑡𝑡–4 denotes births and g0 is the ratio of births at time t–4 to preschool enrolments. this model could be adapted further to include repetition rates as in johnstone and philp (1973). australian population studies 1 (1) 2017 raymer j, biddle n and guan q 35 4.2 multiregional cohort transition model the enrolment transition model presented above is designed for projecting enrolments for single regions/schools independent of other regions/schools. to allow multiple regions/schools to be projected simultaneously, the matrix projection model needs to be reworked to include transitions amongst schools as well as academic levels. this matrix projection model is specified as: et+1,x+1 = get,x+ix+1 where et,x is a vector of enrolments of regions or schools at time t and academic level x, g contains the transition probabilities between levels and schools (gij), and ix+1 is a vector of in-migration counts of enrolments at academic level x+1. in detailed form, the matrix equation for an eight school region system looks like ⎣ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎡𝑒𝑒1,𝑥𝑥+1 𝑡𝑡+1 𝑒𝑒2,𝑥𝑥+1 𝑡𝑡+1 𝑒𝑒3,𝑥𝑥+1 𝑡𝑡+1 𝑒𝑒4,𝑥𝑥+1 𝑡𝑡+1 𝑒𝑒5,𝑥𝑥+1 𝑡𝑡+1 𝑒𝑒6,𝑥𝑥+1 𝑡𝑡+1 𝑒𝑒7,𝑥𝑥+1 𝑡𝑡+1 𝑒𝑒8,𝑥𝑥+1 𝑡𝑡+1 ⎦ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎤ = ⎣ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎡ 𝑟𝑟1 𝑔𝑔21 𝑔𝑔31 𝑔𝑔41 𝑔𝑔51 𝑔𝑔61 𝑔𝑔71 𝑔𝑔81 𝑔𝑔12 𝑟𝑟2 𝑔𝑔32 𝑔𝑔42 𝑔𝑔52 𝑔𝑔62 𝑔𝑔72 𝑔𝑔82 𝑔𝑔13 𝑔𝑔23 𝑟𝑟3 𝑔𝑔43 𝑔𝑔53 𝑔𝑔63 𝑔𝑔73 𝑔𝑔83 𝑔𝑔14 𝑔𝑔24 𝑔𝑔34 𝑟𝑟4 𝑔𝑔54 𝑔𝑔64 𝑔𝑔74 𝑔𝑔84 𝑔𝑔15 𝑔𝑔25 𝑔𝑔35 𝑔𝑔45 𝑟𝑟5 𝑔𝑔65 𝑔𝑔75 𝑔𝑔85 𝑔𝑔16 𝑔𝑔26 𝑔𝑔36 𝑔𝑔46 𝑔𝑔56 𝑟𝑟6 𝑔𝑔76 𝑔𝑔86 𝑔𝑔17 𝑔𝑔27 𝑔𝑔37 𝑔𝑔47 𝑔𝑔57 𝑔𝑔67 𝑟𝑟7 𝑔𝑔87 𝑔𝑔18 𝑔𝑔28 𝑔𝑔38 𝑔𝑔48 𝑔𝑔58 𝑔𝑔68 𝑔𝑔78 𝑟𝑟8 ⎦ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎤ ⎣ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎡𝑒𝑒1,𝑥𝑥 𝑡𝑡 𝑒𝑒2,𝑥𝑥 𝑡𝑡 𝑒𝑒3,𝑥𝑥 𝑡𝑡 𝑒𝑒4,𝑥𝑥 𝑡𝑡 𝑒𝑒5,𝑥𝑥 𝑡𝑡 𝑒𝑒6,𝑥𝑥 𝑡𝑡 𝑒𝑒7,𝑥𝑥 𝑡𝑡 𝑒𝑒8,𝑥𝑥 𝑡𝑡 ⎦ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎤ + ⎣ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎡ 𝑖𝑖1,𝑥𝑥+1 𝑖𝑖2,𝑥𝑥+1 𝑖𝑖3,𝑥𝑥+1 𝑖𝑖4,𝑥𝑥+1 𝑖𝑖5,𝑥𝑥+1 𝑖𝑖6,𝑥𝑥+1 𝑖𝑖7,𝑥𝑥+1 𝑖𝑖8,𝑥𝑥+1⎦ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎤ , where the diagonal elements of the g matrix is equal to the retention proportions (ri) and the offdiagonal elements capture the movements between schools or regions (gij). this model is similar to the one used by sweeney and middleton (2005) to study enrolment transfers in the santa barbara elementary school district in california. 5. school enrolment projections 5.1 school regions to provide an assessment of different projection models, we first focus on school regions. this allows us to make broad comparisons regarding expectations of change across a range of schools, and to work with data that tend to be more stable over time and less prone to specific situations that require expert assumptions or local knowledge. for the assessment, we assume that preschool enrolments are known. normally these are estimated using observed or projected birth data (t–4). in addition to our comparison of the two projection models against the act education directorate’s projections and the observed data, we show the effects of different assumptions concerning the transitions used for the projections. the enrolment transition projection model uses the average annual transitions from 2008–2012. three different transitions assumptions are used in the multiregional projection model. the first keeps 2011–2012 transitions constant. the second uses average 2008–2012 transitions and holds them constant. the third incorporates trends based on linear regressions of the observed transitions from 2008–2009 to 2011–2012. 36 raymer j, biddle n and guan q australian population studies 1 (1) 2017 table 5: average observed and projected student enrolments by school region, 2013–2016 multiregional school region observed ed cohort transition 2011–12 2008–12 trend gun 6,941 7,195 6,229 6,663 5,971 6,911 bel 10,464 10,683 10,652 10,340 10,636 10,443 scan 4,633 4,766 4,517 4,501 4,471 4,789 tugg 7,942 7,873 7,756 8,181 7,985 8,304 wod 4,548 4,280 4,325 4,354 4,247 4,476 west 2,063 2,178 2,083 2,071 2,041 2,210 ncan 5,628 5,724 5,637 5,690 5,511 5,860 total 42,219 42,699 41,198 41,799 40,861 42,994 notes: gun = gungahlin; bel = belconnen; scan = south canberra; tugg = tuggeranong; wod = woden valley; west = weston creek; ncan = north canberra; ed = act education directorate. table 5 presents the average 2013–2016 results from the school region projections with the best performance shaded grey. overall, the multiregional projection model with 2011–2012 transitions performed the best, followed closely by the act education directorate’s projections. however, the performance for each model varied by school region: • the act education directorate’s projection model did well on average for the tuggeranong school region (-0.9% error), but poorly for gungahlin (3.7%), woden valley (-5.9%) and weston creek (5.6%). • the enrolment transition method performed the best out of all the models for south canberra (-2.4%) and north canberra (0.2%), well for weston creek (0.9%) but poorly for gungahlin (9.9%) and woden valley (-5.2%). • the multiregional model with 2011–2012 transitions performed well for weston creek (0.4%) but poorly for gungahlin (-4.5%) and woden valley (-4.5%). • the multiregional model with 2008–2012 transitions performed well for tuggeranong (0.5%) and weston creek (-1.1%), but very poorly for gungahlin (-14.6%) and poor for woden valley (-6.9%). • the multiregional model with trend transitions did very well for gungahlin (-0.5%) and belconnen (-0.2%), but not well for south canberra (3.5%), tuggeranong (4.5%), weston creek (7.2%) and north canberra (4.2%). the year-to-year projection results and observed enrolment numbers for gungahlin, tuggeranong and north canberra are presented in figure 2 (next page). in summary, the multiregional models applied to school regions appear to work best overall (especially considering the objective nature of the information required). average transition models work well when the school regions are relatively stable but poorly when regions are growing, such as gungahlin and belconnen. here, incorporating trends or auxiliary information on planning developments in the transitions makes sense. the big advantage of the multiregional projection models is that the patterns of change (and error) can be explained and linked to observed trends. there is also consistency in the projection framework for school region moves within the act. that is, a departure from one school region must be an entry into another. this is particularly important when considering the transition between primary and high school levels and between high school and secondary college levels. the other factor to consider is efficiency. australian population studies 1 (1) 2017 raymer j, biddle n and guan q 37 gungahlin tuggeranong north canberra figure 2: observed and estimated enrolments for schools in gungahlin, tuggeranong and north canberra, 2008–2016 notes: act ed = act education directorate’s projection; ct = cohort transition; mr = multiregional projection; obs = observed values. 3,500 4,000 4,500 5,000 5,500 6,000 6,500 7,000 7,500 8,000 2008 2009 2010 2011 2012 2013 2014 2015 2016 act ed ct mr 11-12 mr 08-12 mr trend obs 7,500 7,700 7,900 8,100 8,300 8,500 8,700 8,900 9,100 9,300 9,500 2008 2009 2010 2011 2012 2013 2014 2015 2016 act ed ct mr 11-12 mr 08-12 mr trend obs 4,900 5,100 5,300 5,500 5,700 5,900 6,100 6,300 2008 2009 2010 2011 2012 2013 2014 2015 2016 act ed ct mr 11-12 mr 08-12 mr trend obs % error mr 11-12 1.1 act ed 1.8 mr 8-12 -3.5 ct 0.2 mr trend 6.9 % error mr 11-12 3.3 act ed -3.2 mr 08-12 -0.3 ct -4.3 mr trend 6.8 % error mr 11-12 -7.1 act ed 0.9 mr 08-12 -23.4 ct -13.7 mr trend 0.6 38 raymer j, biddle n and guan q australian population studies 1 (1) 2017 5.2 school projections for north canberra the multiregional projection method is particularly useful for capturing students transitioning between primary and high schools and between high schools and secondary colleges. this is further illustrated with schools in north canberra, an area with a relatively stable total population in terms of age composition and growth. table 6: observed and projected enrolments for schools in north canberra, 2013–2016 2013 2014 2015 2016 school obs ed mr obs ed mr obs ed mr obs ed mr a. observed and projected enrolments a 424 436 442 410 432 434 405 443 420 396 480 412 b 315 344 331 327 335 343 350 345 345 350 336 338 c 421 429 426 432 423 432 471 456 446 480 471 439 d 567 632 591 585 662 604 598 675 614 600 656 621 e 483 425 443 554 436 447 556 456 466 611 462 466 f 77 84 79 76 84 78 70 85 74 77 84 72 g 545 587 541 553 588 544 562 600 532 573 622 515 h 718 744 711 741 797 721 723 823 719 721 842 710 i 1,068 1,064 1,059 1,054 1,064 1,082 1,050 1,064 1,073 1,065 1,064 1,084 j 848 854 809 866 841 805 919 837 824 871 833 835 total 5,466 5,599 5,433 5,598 5,662 5,490 5,704 5,784 5,510 5,744 5,850 5,491 b. difference between projected and observed a 12 18 22 24 38 15 84 16 b 29 16 8 16 -5 -5 -14 -12 c 8 5 -9 0 -15 -25 -9 -41 d 65 24 77 19 77 16 56 21 e -58 -40 -118 -107 -100 -90 -149 -145 f 7 2 8 2 15 4 7 -5 g 42 -4 35 -9 38 -30 49 -58 h 26 -7 56 -20 100 -4 121 -11 i -4 -9 10 28 14 23 -1 19 j 6 -39 -25 -61 -82 -95 -38 -36 total 133 -33 64 -108 80 -194 106 -253 notes: obs = observed values; ed = act education directorate projection; mr = multiregional projection. the school level projections for north canberra focus on the level and between-school movements across seven primary schools (schools a–g), two high schools (schools h–i) and one secondary college (school j). in this illustration, only schools in north canberra are included, but the model framework could be extended to include any set of schools. also, for illustration purposes, we apply the average 2008–2012 transitions as the basis for estimating transitions for 2013–2016. australian population studies 1 (1) 2017 raymer j, biddle n and guan q 39 the projection results for schools in north canberra are presented in table 6 (panel a). also included are the corresponding observed values and the projections made by the act education directorate (panel b) with the closest value to the observed value shown in grey. the multiregional projection model out-performed the act education directorate’s projections for seven out of the 10 schools in each of the four projection years. these differences, however, are fairly modest. the real benefits of the multiregional model are that: (i) less assumptions are required; and (ii) there is a capacity to disaggregate the projection error. 6. discussion and conclusion understanding the demographic drivers of enrolment change is essential for making accurate and informed projections. variables influencing change include the birth of children in school intake areas, movements between government schools in a city or district, migration into or out of a city or region, and movements to or from non-government schools. the multiregional cohort projection model presented in this paper provides a flexible platform for including the main transitions affecting school enrolment change. by grouping schools in a multiregional model, understanding about student movements is increased and projection bias reduced. in order to make and assess projections, some understanding of the components of enrolment change and their trends is required. there is no single projection model that performs best in all situations. school districts should consider a variety of assumptions when developing projection models and adapt them to meet local differences. they need also to incorporate proposed or new housing developments and land releases which may result in increased migration by young families into the area and rapid population growth impacting education provision for the areas affected. as projections are often heavily scrutinised, the ability to explain trends in terms of the sources of growth provides a stronger evidence base and argument for school infrastructure, personnel and resource requirements. it also allows the projections producer to understand where the sources of error occurred in relation to changing enrolments over time. this is the main motivation underlying the multiregional cohort projections. we have demonstrated the variability in the sources of student enrolment change across the act and the relative stability in the different sources of enrolment change and cohort transitions. movements within and outside the act government school system can have large effects and are more difficult to predict. the multiregional school enrolment projection model captures these movements and may be applied to a wide array of situations. however, the transition data must be available. key messages • accurate projections of school enrolments and understanding of sources of error are essential for good school planning. • the multiregional projection model used in our study performed as well or better than a cohort transition model and current projection model used by the act education directorate. • a multiregional model that interrogates enrolment change can be used for shortto mediumterm enrolment projections in the act. 40 raymer j, biddle n and guan q australian population studies 1 (1) 2017 acknowledgements this paper is drawn from a previously commissioned review of school enrolment projection methodology prepared for the act education directorate. the authors would like to thank the directorate for their comments and suggestions, and xujing bai for her research assistance. references act government (2017) estimated resident population – september quarter 2016, viewed 19 december 2016, http://apps.treasury.act.gov.au/__data/assets/pdf_file/0008/644813/erp.pdf/_recache. berk & associates (2008) office of superintendent of public instruction k–12 school enrollment projections study: final report, viewed 19 december 2016, www.k12.wa.us/schfacilities/publications/pubdoc s/enrollprojectionmethodologiesfinalreport2008.pdf. gould e (1993) an interactive flow model for projecting school enrolments. international review of education 39(4): 319–332. grip r s (2009) does projecting enrollments by race produce more accurate results in new jersey school districts? population research and policy review 28(6): 747–771. herrick j h (1952) estimating future school enrollments in rapidly growing communities. educational research bulletin 31(4): 92–94. johnstone j n (1974) mathematical models developed for use in educational planning: a review. review of educational research 44(2): 177–201. johnstone j n and philp h (1973) the application of a markov chain in educational planning. socioeconomic planning science 7(3): 283–294. land k c and rogers a (eds) (1982) multidimensional mathematical demography: proceedings of the conference on multidimensional mathematical demography held at the university of maryland, college park, maryland, march 23–25, 1981, sponsored by the national science foundation. new york: academic press. morrison p a (2000) forecasting enrollments for immigrant entry-port school districts. demography 37(4): 499–510. rees p h and willekens f j (1986) data and accounts. in: rogers a and willekens f j (eds) migration and settlement: a multiregional comparative study. dordrecht: d. reidel publishing company; 19–58. rogers a (1975) introduction to multiregional mathematical demography. new york: john wiley & sons. rogers a (1995) multiregional demography: principles, methods and extensions. chichester: wiley. rushton g, armstrong m p and lolonis p (1995) small area student enrollment projections based on a modifiable spatial filter. socio-economic planning science 29(3): 169–185. schellenberg s j and stephens c e (1987). enrollment projection: variations on a theme, paper presented at the annual meeting of the american educational research association. washington dc, 20–24 april 1987, viewed 19 december 2016, http://files.eric.ed.gov/fulltext/ed283879.pdf. schoen r (1988) modeling multigroup populations. new york: plenum press. stone r (1965) a model of the educational system. minerva 3(2): 172–186. stronge w b and schultz r r (1981) models for projecting school enrollment. educational evaluation and policy analysis 3(5): 75–81. swanson d a and tayman j (2012) subnational population estimates. dordrecht: springer. sweeney s h and middleton e j (2005) multiregional cohort enrolment projections: matching methods to enrolment policies. population, space and place 11(5): 361–380. webster w j (1970) the cohort-survival ratio method in the projection of school attendance. the journal of experimental education 39(1): 89–96. webster w j (1971) an empirical comparison of selected enrollment projection formulas. the journal of educational research 64(8): 375–381. http://apps.treasury.act.gov.au/__data/assets/pdf_file/0008/644813/erp.pdf/_recache http://www.k12.wa.us/schfacilities/publications/pubdocs/enrollprojectionmethodologiesfinalreport2008.pdf http://www.k12.wa.us/schfacilities/publications/pubdocs/enrollprojectionmethodologiesfinalreport2008.pdf http://files.eric.ed.gov/fulltext/ed283879.pdf australian population studies 2018 | volume 2 | issue 1 | pages 14–25 © charles-edwards and panczak 2018. published under the creative commons attribution-noncommercial licence 3.0 australia (cc by-nc 3.0 au). journal website: www.australianpopulationstudies.org elsewhere in australia: a snapshot of temporary mobility on the night of the 2016 census elin charles-edwards* the university of queensland radoslaw panczak the university of queensland * corresponding author. email: e.charles-edwards@uq.edu.au. address: queensland centre for population research, school of earth and environmental sciences, the university of queensland, brisbane, australia 4072 paper received 13 march 2018; accepted 2 may 2018; published 28 may 2018 abstract background temporary population mobility, moves of more than one night’s duration that do not entail a change in usual residence, are an important feature of the australian population surface. the abs census of population and housing (census) provides a snapshot of temporary movements one night every five years. aims this paper examines the intensity, age and spatial patterns of temporary movements captured at the 2016 census, and creates a classification of regions based on the age profile of movers on census nights. data and methods 2016 census data were extracted using abs tablebuilder pro. summary metrics were calculated to measure the intensity and age profile of movements. origin–destination flows were derived from a cross-classification of data on place of usual residence and place of enumeration. a classification of regions (sa4s) was constructed from the age profile of movers at origins and at destinations. results 1,142,005 individuals (about 5 per cent of the australian population) were enumerated away from home on census night 2016. mobility peaked in younger (20–30) and older (65–70) age groups. most movements were between capital city regions; however, resource regions and coastal areas were also implicated. the mobility surface was segmented by age: younger people dominated visits to cities and older movers comprise the majority of visitors to coastal areas, while remote areas had a significant proportion of visitors in the peak working ages. conclusions temporary population mobility is selective by age and sex and geographically segmented by these characteristics. improved understanding of the attribute of visitors to regions can assist to formulate and validate estimates of temporary populations from emerging data sets. key words temporary population mobility; non-residents; spatiotemporal populations; service populations; migration; census; australia. http://www.australianpopulationstudies.org/ mailto:e.charles-edwards@uq.edu.au australian population studies 2 (1) 2018 charles-edwards e and panczak r 15 1. introduction temporary population mobility drives significant short-term changes in national population distributions. population scholars, particularly in the developed world, have tended to ignore these movements in favour of permanent migration. this is changing. there is widespread recognition of the implications of short-term changes in population numbers for service provision (charles-edwards and bell 2013), emergency response and preparedness (canterford 2011; wilson et al. 2016) and disease transmission (bharti et al. 2011). data sets derived from social media (tenkanen et al. 2017), mobile telephones (silm and ahas 2010; deville et al. 2014) and other sources are providing new opportunities to monitor dynamics shifts in population numbers but are not a panacea. they are often partial, suffer from bias and lack detail on the underlying population flows and the characteristics of movers. it is against this backdrop that censuses and surveys have continued value for understanding temporary populations. this short paper explores the intensity and spatial pattern of temporary population mobility, along with the age and sex of movers, as captured by the australian bureau of statistics (abs) 2016 census of population and housing (census). it provides a short discussion of past studies of temporary population mobility in australia in section 2, then in section 3 briefly describes how temporary mobility is captured by the census and the limitation inherent in these data. section 4 reports the intensity of movements and age profile of movers as captured in 2016, and compares this to data collected at the 1996 census. in section 5, the spatial patterns of temporary flows are examined. in section 6, the sex and age characteristics of movers at origins and destinations are explored. brief conclusions are presented in section 7. 2. background temporary population mobility can be defined as moves more than one night in duration that do not entail a change in usual residence (bell and ward 2000). this separates temporary mobility from diurnal movements, such as daily commuting, and from permanent migration. there are some problems with this definition. many individuals do not have a single place of usual residence, for example, fly-in flyout (fifo)/drive-in drive-out (dido) miners and children in joint custody arrangements. the distinction between temporary and permanent mobility may also be unclear, with some ‘temporary’ movements stretching over years. notwithstanding, this definition has the advantage of being operationalised using common census definitions for place of usual residence and place of enumeration. temporary population movements are undertaken for a range of reasons. a simple distinction is between moves undertaken for work or production-related reasons (e.g. fifo/dido mining, seasonal agricultural labour and business travel) and moves undertaken for consumption-related reasons (e.g. tourism, second home travel, visits to friends and relatives). different types of mobility have distinct spatiotemporal signatures, with diverse destinations, durations, seasonal patterns and periodicities. for example, fifo mobility is spatially concentrated in resource regions and follows a common roster, such as two weeks on site followed by one week at home (mckenzie 2011). by contrast, coastal tourism can range from a few days to a few weeks in duration, is highly seasonal and concentrated in the summer months (charles-edwards and bell 2015). in addition to having distinct spatiotemporal signatures, temporary movers can be segmented according to age. in a schema proposed by mchugh, hogan and happel (1995), mobility in childhood and adolescence is tied to the family and may include both touristic mobility and mobility associated 16 charles-edwards e and panczak r australian population studies 2 (1) 2018 with family dissolution. in adulthood, mobility becomes increasingly tied to work, which can necessitate strategies such as long-distance commuting. with the transition to middle age, more financial resources enable some individuals to purchase second homes, initiating another form of temporary mobility. older age may be associated with a relaxation of work-related constraints on mobility, providing opportunities for new forms of recreational mobility and extended stays with children and grandchildren, but also additional constraints due to ill health. there have been a number of studies of temporary population mobility in australia drawing on census data (bell and brown 2006), data from the national visitor survey (charles-edwards, bell and brown 2008) and purposive surveys (mings 1997; hanson and bell 2007). bell and ward (1998) undertook the first systematic study of temporary population mobility in australia, utilising data from the 1991 census. they found that the incidence of temporary mobility remained relatively stable between the 1976 and 1991 censuses, with around 5 per cent of australians enumerated away from home on census night. the study found that temporary moves tended to cover longer distances than permanent moves, and revealed a regional pattern of net gains with inner cities, tourist regions and resource regions all attracting temporary visitors. follow up studies comparing temporary and permanent migration (bell and ward 2000) and the characteristics of temporary movers (bell and brown 2006) revealed the age profile of temporary movers, with movements selective of both younger (20–30 years) and older (65–70 years) adults. charles-edwards, bell and brown (2008) and charles-edwards and bell (2015), using data from the national visitor survey, confirmed the regionalisation of temporary movements but also the significant seasonal variation in spatial patterns tied to institutional and climatic factors. the current study provides an updated account of temporary population mobility in australia drawing on data from the 2016 census. it seeks also to extend our understanding of temporary moves by examining patterns of origin–destination flows and regional variations in the sex ratio and age profile of movers. the latter is of particular relevance to agencies interested in providing services to temporary populations. data on the age and sex of movers are also critical for the validation of emerging sources of data on temporary mobility. 3. data and methods data were sourced from the 2016 census. the census was conducted on a de facto residence basis, meaning that people were enumerated where they happened to be on census night, 9 august 2016. by combining data on place of usual residence and place of enumeration it is possible to generate a snapshot of people away from home (referred to as ‘visitors’ through the text) on one night every five years. the count of the number of people in australia enumerated away from their place of usual residence is coded in the usual address indicator (uaicp). uaicp data were extracted from the abs counting persons, place of usual residence census database within tablebuilder pro by age and sex to estimate the intensity of temporary population mobility on census night, along with the age profile of movers. while uaicp provides an estimate of the stock of people enumerated away from home on census night, it does not provide information on origin–destination flows. to explore the spatial distribution of movements, origin–destination matrices were constructed based on individuals’ place of usual residence (i.e. pur, the origin) and place of enumeration (i.e. poe, the destination) for statistical australian population studies 2 (1) 2018 charles-edwards e and panczak r 17 areas level 4 (sa4s). in tablebuilder pro, pur data and poe data are only available for crosstabulation in the counting employed persons, place of work database, which is restricted to individuals aged 15 years and older. all analyses were done in r (version 3.4.3, r core team 2018) with the circlize package used to draw the flow diagram (gu et al. 2014). maps were prepared using quantumgis software (qgis development team 2018). 4. how much movement? of 23,401,891 people enumerated on census night 2016, 1,142,005 (4.9%) were captured away from their usual residence. in relative terms, this figure has been remarkably stable over the past two decades, with 4.7 per cent of the population away from home at the 1996 census (bell and ward 2000). when we limit this to adults (aged 15 and over), 5.5 per cent of the population were away from home. while most moves were over short distances, almost a third of moves were between different states and territories. in addition to the 4.7 per cent of australians away from home on census night, the census enumerated 315,530 overseas visitors, equivalent to 1.3 per cent of the 2016 census count. like permanent migration, temporary mobility is highly selective by age and sex. the age schedule of temporary mobility is bimodal (figure 1). the highest levels of mobility are among older adults (65– 70 years) and younger adults (20–30 years). the peak in mobility at both older and younger ages reflects fewer work and family related constraints on these populations on a school night in midwinter. the lowest level of mobility is observed among school-aged children, with only 2 per cent enumerated away from home on census night. there are marked sex differentials in the mobility age schedules. mobility among young women peaks at a younger age and at a lower level than for men. it then drops to well below the male propensities during key labour force ages before it rises again to match male levels in retirement. figure 1: per cent of individuals away from home on census night by age and sex, 1996 and 2016 censuses source: abs 1996 and 2016 censuses. 18 charles-edwards e and panczak r australian population studies 2 (1) 2018 how might we explain these differences? the early peak in mobility among females is likely tied to age differentials in partnering and the absences associated with non-residential relationships. in key labour force ages, the lower mobility of women might reflect occupational structure, with women less likely to be employed in roles and industries associated with large amounts of travel (bell and brown 2006). women are also more likely than men to be constrained by childcaring roles limiting work-related mobility (abs 2016). sex differentials in movement at older ages may again reflect age differentials in partnering, with a persistent three to four year lag in mobility rates. there has been an ageing in the profile of movers since the 1996 census. mobility has decreased among younger men, with the percentage away at age 20 declining by more than 25 per cent. this decline has been offset by a substantial increase in the rate of movement across key labour force ages (25–65). for women, there has also been a drop in mobility in early adulthood. this is accompanied by a slight increase in rates between ages 25–45, but also a decline between ages 45–65. mobility in the retirement ages (65 and over) remained largely unchanged between 1996–2016, while movement intensity in infanthood dropped for both males and females. based on the schema set out by mchugh, hogan and happel (1995), these changes suggest a shift away from consumption-related moves, particularly at younger ages, towards production-related motivations for temporary movements on census night. 5. where do people move? figure 2: circular plot showing origin–destination flows between greater capital city statistical areas and sa4s (for the rest of australia), 2016 census australian population studies 2 (1) 2018 charles-edwards e and panczak r 19 figure 2 (above) shows the temporary flows recorded on census night between greater capital city statistical areas (gccsa) and sa4s (for the rest of australia). the circular plot represents all inter-sa4 flows, but only shows chords for flows greater or equal to 1,000 people. colour hues are used to define australian states and territories. the direction of the flow is defined by the colour of the origin and the presence of an arrow at the end of the chord. the width represents the number of movers. from figure 2 we learn a number of things. at the national level, we see that the most populous states of new south wales, victoria and queensland dominate the mobility system in terms of gross flows (i.e. both ins and outs). the largest flows originate and terminate in metropolitan areas, with the largest reciprocal flows between greater sydney and greater melbourne. greater brisbane is an important source of flows to greater sydney and greater melbourne, but is less attractive as a destination. another striking feature is the containment of larger flows within the states and territories, particularly in the resource states of western australia and queensland. of all states and territories, flows from victoria are the most widely dispersed across australia, with movers from victoria recorded in coastal tourism areas including the gold coast, sunshine coast and cairns (all in queensland). figure 3 shows the pattern of net mobility rate (nmr), which captures gains and losses for sa4s. nmr was calculated by subtracting the number of residents temporarily absent (i.e. out-movers) from the number of visitors (i.e. in-movers) divided by the usual resident population for each sa4. the pattern follows a north–south gradient with the greatest gains in northern australia, including outback areas of western australia, queensland and the northern territory, and the largest losses in the southern states. modest gains were also recorded in coastal destinations in southeast queensland and northern new south wales, as well as sa4s encompassing the alpine resort areas of new south wales and victoria. this is consistent with climatic factors dictating the timing and destination of some forms of temporary mobility (see charles-edwards and bell 2015). figure 3: net mobility rate (%) across sa4s, 2016 census note: number in brackets indicates number of regions in category. 20 charles-edwards e and panczak r australian population studies 2 (1) 2018 there is also a functional dimension to the mosaic of gains and losses. capital cities experienced net population losses except in their urban core, which experienced modest (e.g. greater brisbane) to large (e.g. greater sydney) gains. these gains are probably tied to business flows. gains in northwest western australia and central and southwest queensland are likely underpinned by mining and oil and gas extraction-related (fifo) mobility. 6. who moves where? the characteristics of movers are of interest to agencies involved in the planning and provision of services, as well as to researchers needing denominators for a range of social statistics. age and sex are the most critical attributes to be captured. figure 4 shows the sex ratio of visitors to sa4s across australia. the mobility surface is strongly differentiated by sex. males dominate temporary moves to resource regions including northwest western australia and central and southern queensland. the sex ratio is particularly skewed in northwest western australia with more than two male visitors for every female. sex ratios are also skewed in favour of males in the closely settled irrigated agricultural zone of southern new south wales and victoria. females dominate visitation to the suburbs and periphery of most urban areas (except greater perth), but not the central business districts. the sex ratio imbalance is in favour of females in southeast coastal districts of queensland. figure 4: sex ratio of visitors to sa4s, 2016 census note: the number in brackets indicates the number of regions in each category. we know that propensity to move varies across the life course. how does this manifest across space? to explore variations in the age structure of movers in both sending and receiving regions we grouped sa4s according to the age profiles of individuals temporarily absent from home (represented as an out-movement rate) and by the age profile of visitors to sa4s (measured as a proportion of all australian population studies 2 (1) 2018 charles-edwards e and panczak r 21 visitors to that sa4). clusters were selected using a k-means clustering algorithm implemented in the r package stats that determines the optimum partitioning of the dataset for a specified number of clusters. we derived four groups for residents away from home and four groups for visitors to sa4s. figure 5 shows the average age profile of absent residents for each of the four clusters and their spatial distribution. there are clear differences in both the level and age profile of mobility across sending regions. cluster 1 contains the most sa4s (n=28) and comprises the coastal regions of queensland and new south wales, as well as sa4s in greater adelaide, greater perth, greater darwin and other territories (not shown). outflows from this cluster are of average intensity and follow a bimodal pattern similar to the national distribution. sparsely populated outback areas comprise the majority of sa4s in cluster 2 (n=20). the out-movement rates are the highest (on average) for younger and middle-aged individuals, and the second highest for individuals aged 65 and over. figure 5: classification of sa4s based on age of absent residents, 2016 census note: the number in brackets indicates the number of regions in each category. we can surmise that the high level of mobility in part reflects trips to access goods and services located at considerable distances from home. sa4s in cluster 3 (n=16) have (on average) the highest 22 charles-edwards e and panczak r australian population studies 2 (1) 2018 proportion of individuals aged 65 and over away from home, and second largest proportion of people aged 20–29 away from home. sa4s belonging to this cluster are concentrated in the southeast corner of australia, which suggests a climatic motivation for absences on census night. cluster 4 is the second largest (n=24) and is characterised by low mobility at all ages. these sa4s are concentrated in greater sydney, greater melbourne and greater brisbane. the age profiles of movers are more varied at destinations than at origins. four clusters are shown in figure 6. note that the clusters are based on the proportion of visitors by age, and thus do not reflect the level of visitation. the biggest between-cluster differences are at younger ages (below age 35) and older ages (above age 60). sa4s in cluster 1 (n=35) and cluster 2 (n=27) have younger than average visitors. visitors to sa4s in cluster 3 and cluster 4 are older on average. figure 6: classification of sa4s based on age of visitors, 2016 census note: number in brackets indicates the number of regions in each cluster. cluster 1 is comprised of sa4s located in state and territory capital cities. visitors to these areas are more likely to be younger or in key labour force ages, suggesting employment as a key driver. cluster 2 has a similar age profile, but has a higher peak at younger ages and a lower proportion of visitors in australian population studies 2 (1) 2018 charles-edwards e and panczak r 23 key labour force ages. sa4s in this cluster are located on the peri-urban fringe of capital cities as well as southern coastal districts. the cluster includes the southern alpine ski resorts. cluster 3 (n=18) has a relatively high proportion of older visitors as well as visitors in key labour force ages. sa4s in this cluster include the mining resource regions of western australia and queensland, as well as large swathes of northern australia which are climatically most attractive during the winter months. this suggests that a combination of production and consumption motivations are driving visits to these regions. visitors to regions in the smallest cluster, cluster 4 (n=9), are older than average with relatively few individuals below age 50. sa4s belonging to this cluster are located in the coastal areas of northern new south wales and southeast queensland, and in northern queensland. 7. discussion and conclusion understanding the way in which populations shift over the course of a day, week and month due to temporary movements is important for planning and service provision, but also for the estimation of better dominators for a range of health and social statistics. this paper explored the intensity, age profile and patterns of temporary movements captured at the 2016 census. it also generated a classification of regions based on the age profile of residents absent on census night, and a separate classification of the age profile of visitors to australian regions. the results reveal that the intensity of temporary mobility in australia has remained stable over the past two decades, with a rate of around one in twenty. there have, however, been shifts in the age/sex profile, with movement increasingly dominated by males in key labour force ages. the impact of temporary population mobility on census night varies across australia. origin–destination flows connect metropolitan regions, particularly on the eastern seaboard. they distribute people from capital cities to regions in the resource rich states, and move victorians to high amenity parts of queensland. with respect to impact, the suburbs of major cities and regions in southern australia experience net losses. gains occur in remote and northern australia, selected high-amenity coastal districts and the core of major metro regions. when age is taken into account, a more differentiated mobility surface is revealed. at sending regions there is a close association between accessibility and the level of outmovement: the more remote an area, the higher the rate of out-movement. at destinations, there is a bifurcation between regions dominated by younger visitors and those more attractive to older australians. the former includes the capital cities and regions in the southeast corner of australia, while the latter incorporates the resource regions of central and northern australia and selected high-amenity coastal districts. census data provides a partial insight into temporary movements in australia, one night every five years. it provides no measure of seasonality, duration, frequency or periodicity, nor the complex circuits and motivations that underpin temporary moves. the spatial patterns observed will not be replicated at other times of the year: for example, during the january holiday period. the utility of census data for the estimation of temporary populations is therefore limited. it does, however, have the singular advantage of complete enumeration of temporary populations coupled with detailed information on the characteristics of movers. the emergence of new data sets, including those collected via social media and communication technologies, presents further opportunities for the generation of temporary population estimates. 24 charles-edwards e and panczak r australian population studies 2 (1) 2018 however, these data are often partial, biased and confounded by individual behaviour and may not reliably represent the population present. baseline data that accurately capture temporary populations and their characteristics, even if just for a single night, are essential for the production of robust estimates of non-resident populations. the integration of census data with emerging data sets forms a key aspect of an ongoing program of work (the tempo project: techniques for estimating mobile populations) aimed at developing techniques for estimating temporary populations in australia. key messages • one in twenty australians were enumerated away from their place of usual residence on census night 2016. this rate has been stable since 1996; however, the age and sex of movers has changed with males in key labour force age groups now much more likely to make temporary moves. • on census night temporary flows connect the capital cities in eastern australia; cities to resource regions in western australia and queensland; and victoria to high amenity parts of queensland and northern and outback australia. • inner cities and select coastal areas gain temporary movers, while the suburbs and southern regions lose residents. this is a winter pattern: the pattern of gains and losses is expected to look different in summer months. • the age of visitors varies across regions. understanding these patterns is important for local service provision, housing and planning. • the complete enumeration of temporary mobility by the census, even for a single night, provides an opportunity to validate estimates of temporary populations derived from new data sources, such as social media and mobile phone data. acknowledgements this research was supported through funding from australian research council linkage project, lp160100305, estimating temporary populations (tempo project). the authors thank the anonymous reviewers for their constructive comments that contributed to improving the final version of the paper. references abs (australian bureau of statistics) (2016) gender indicators, australia, feb 2016. cat. no. 4125.0. canberra: abs. bell m and brown d (2006) who are the visitors? characteristics of temporary movers in australia. population, place and space 12(2): 77–92. bell m and ward g (1998) patterns of temporary mobility in australia: evidence from the 1991 census. australian geographical studies 36(1): 58–81. bell m and ward g (2000) comparing temporary mobility with permanent migration. tourism geographies 2(1): 97–107. bharti n, tatem a j, ferrari m j, grais r f, djibo a and grenfell b t (2011) explaining seasonal fluctuations of measles in niger using nighttime lights imagery. science 334(6061): 1424–1427. canterford s (2011) locating people spatially: 2006, 2010, 2100 and 2:36pm on friday. australasian journal of regional studies 17(1): 46–59. australian population studies 2 (1) 2018 charles-edwards e and panczak r 25 charles-edwards e and bell m (2013) estimating the service population of a large metropolitan university campus. applied spatial analysis and policy 6(3): 209–228. charles-edwards e and bell m (2015) seasonal flux in australia’s population geography: linking space and time. population, space and place 21(2): 103–123. charles-edwards e, bell m and brown d (2008) where people move and when: temporary population mobility in australia. people and place 16(1): 1–30. deville p, linard c, martin s, gilbert m, stevens f r, gaughan a e, blondel, v d and tatem a j (2014) dynamic population mapping using mobile phone data. proceedings of the national academy of sciences 111(45): 15888–15893. gu z, gu l, eils r, schlesner m and brors b (2014) circlize implements and enhances circular visualization in r. bioinformatics 30(19): 2811–2812. hanson j and bell m (2007) harvest trails in australia: patterns of seasonal migration in the fruit and vegetable industry. journal of rural studies 23(1): 101–117. mchugh k e, hogan t d and happel s k (1995) multiple residence and cyclical migration: a life course perspective. professional geographer 47(3): 251–267. mckenzie f h (2011) fly-in fly-out: the challenges of transient populations in rural landscapes. in: luck g w, race d and black r (eds) demographic change in australia’s rural landscapes: implications for society and the environment. dordrecht: springer; 353–374. mings r (1997) tracking snowbirds in australia: winter sun seekers in far north queensland. australian geographical studies 35(2): 168–182. qgis development team (2018) qgis geographic information system, open source geospatial foundation http://qgis.osgeo.org. r core team (2018) r: a language and environment for statistical computing. vienna, austria, r foundation for statistical computing http://www.r-project.org. silm s and ahas r (2010) the seasonal variability of population in estonian municipalities. environment and planning a 42(10): 2527–2546. tenkanen h , minin e d, heikinheimo v, hausmann a, herbst m, kajala l and toivonen t (2017) instagram, flickr, or twitter: assessing the usability of social media data for visitor monitoring in protected areas. scientific reports 7: article no. 17615. doi: 10.1038/s41598-017-18007-4. wilson r, zu erbach-schoenberg e, albert m, power d, tudge s, gonzalez m, guthrie s, chamberlain h, brooks c, hughes c, pitonakova l, buckee c, lu x, wetter e, tatem a j and bengtsson l (2016) rapid and near real-time assessments of population displacement using mobile phone data following disasters: the 2015 nepal earthquake. plos currents disasters feb 24. edition 1. doi: 10.1371/currents.dis.d073fbece328e4c39087bc086d694b5c. http://qgis.osgeo.org/ http://www.r-project.org/ https://www.nature.com/articles/s41598-017-18007-4 https://dx.doi.org/10.1371%2fcurrents.dis.d073fbece328e4c39087bc086d694b5c australian population studies 2017 | volume 1 | issue 1 | pages 3–12 © mcdonald 2017. published under the creative commons attribution-noncommercial licence 3.0 australia (cc by-nc 3.0 au). journal website: www.australianpopulationstudies.org international migration and employment growth in australia, 2011–2016 peter mcdonald* the university of melbourne *corresponding author. email: mcdonald.p@unimelb.edu.au. address: melbourne school of population and global health, university of melbourne, vic 3010, australia paper received 26 july 2017; accepted 17 september 2017; published 20 november 2017 abstract background immigration to australia pre 1995 was largely low skilled. recessions led to competition between low-skilled domestic workers and new immigrants and subsequent cuts in migration intakes. historical changes in birth rates, increased participation in tertiary education, increasing numbers retiring and the relatively rapid restructuring of the skill level of labour demand combined to produce a skilled labour supply crisis in australia from the mid-1990s. the permanent and temporary skilled migration policies established by the australian government from 1995 played an important role in meeting that labour demand, especially in the boom years of the first decade of the 21st century. aims this paper examines the impact of immigration on employment in australia subsequent to the global financial crisis (gfc) for the five-year period from july 2011 to july 2016. data and methods data for the paper are sourced from the australian bureau of statistics. the paper uses survival methods to decompose the growth in employment in australia in the five-year period from 2011 to 2016: (1) change in age and sex distribution in the absence of migration; (2) changes in employment participation rates by age and sex; (3) net migration by age and sex. results immigration in response to strong labour demand has continued post gfc. from july 2011 to july 2016, employment in australia increased by 738,800. immigrants accounted for 613,400 of the total increase, population growth 98,900 and changes in employment participation only 26,500. migration has had a very large effect on the age structure of employment with most new immigrant workers (595,300) being under 55 years. conclusions research indicates that immigration provides major benefits to the australian economy. however, as strong labour demand is likely to sustain migration at relatively high levels in coming years, it is incumbent upon governments to plan for the effects of rapid population growth on infrastructure and resources. key words australia; labour demand; skilled migration; components of employment growth; global financial crisis (gfc). http://www.australianpopulationstudies.org/ mailto:mcdonald.p@unimelb.edu.au 4 mcdonald p australian population studies 1 (1) 2017 1. introduction this paper makes use of data from the abs labour force surveys for july 2011 (abs 2011) and july 2016 (abs 2016a) to examine changes in employment by age and sex in australia. the aim is to decompose changes in total employment in australia in this five-year period into three components: 1. change in the age and sex distribution in the absence of migration. 2. changes in employment participation rates by age and sex. 3. net migration by age and sex for the period 2011–2016. the study concludes that the third component, net migration, was clearly the largest of these three components. as this is a relatively new phenomenon, it is important to provide the historical context that has led up to this outcome. the purpose of the analysis is to provide a demographic accounting of changes in employment in the five-year period. as such, the analysis focuses on the supply of employment rather than demand. other methodologies such as computable general equilibrium (cge) modelling would be required to examine the demand side. also, though near to impossible to quantify, it should be remembered that the needs of new immigrants for goods and services creates labour demand that, in the absence of the migrants, would not have been apparent. the results provided are not net of this effect. 2. background the post-war migration program created in the late 1940s in australia had two main objectives: to provide workers for the expansion of australian industry, especially the manufacturing and construction industries; to increase the size of the australian population under the imperative of ‘populate or perish’. little attention was given to the skill level or english competency of the immigrants upon arrival because the labour demand was primarily for workers with low skills. as the various migrant groups established themselves in australia, the emphasis of migration shifted to the entry of family members of the prior immigrants. these family members, like their sponsors, were primarily people with low skills. as a consequence of these migration settings, whenever there was a downturn in the australian economy and an increase in unemployment (1961–1962, 1974–1975, 1983–1984 and 1991–1993), the migration program was wound back (figure 1). this was appropriate as the migrants competed very directly with low-skilled australians for the available jobs. this competition was heightened by the continuing expansion of the domestic labour force, which resulted both from the entry into the labour force of the baby-boom generation and the relatively low skills of the domestic population. with australia’s baby boom continuing into the early 1970s and small numbers of older workers retiring from the labour force, this situation of strong natural growth of the domestic labour force continued into the 1980s. from the 1980s onwards economic restructuring, especially the winding back of the manufacturing industry, led to the rapid decline of low-skilled jobs and a commensurate high demand for jobs requiring tertiary education and training. participation in tertiary education began to expand rapidly. in 1995 only 15 per cent of 25–29 year-old australians had a university degree. this had jumped to 37 per cent by 2015 (norton 2016, p. 68). as increasingly higher proportions of young australians australian population studies 1 (1) 2017 mcdonald p 5 pursued tertiary education and training, correspondingly, the number of young people employed in full-time jobs fell rapidly. for example, the number of 15–19 year olds employed full time fell from 463,000 in july 1989 to 240,000 in july 1993 and has continued to fall since then (abs 2016a). figure 1: net overseas migration, australia, 1947–2016(1) notes: (1) from 1 july 2006 (vertical line), the definition of net overseas migration was changed by the abs. the new definition made it more likely that temporary residents such as international students and temporary skilled workers would be counted into the population and coincided with a surge in these types of immigrants immediately after the definitional change. while net overseas migration is somewhat higher under the new definition compared with the old (especially in 2008), the average in recent years of around 200,000 per annum provides a reasonable comparison with the years prior to the definitional change. at the same time, demographic change in australia was putting an end to the natural growth of the domestic labour force. this is indicated starkly by the numbers in table 1. while in 1996 the population aged 25–29 was only 71 per cent of the size of the population aged 15–19, by 2016 the population aged 25–29 was 123 per cent of the population aged 15–19. finally, as evident also from table 1, the ratio of the near retirement age population (60–64 years) was just 41 per cent of the population aged 15–19 in 1966 but 83 per cent in 2016. to bring this situation into even sharper focus, the number of full-time workers aged 15–19 in 1978 was three times the number of full-time workers aged 60–64. in contrast, in 2016 the number of full-time workers aged 15–19 was just 39 per cent of the number aged 60–64. table 1: relative size of age groups in australia, 1966 and 2016 age group relative size of age group (%) (15–19 = 100%) 1966 2016 15–19 100 100 20–24 81 115 25–29 71 123 60–64 41 83 source: abs (2014). the long-term demographic shift resulting from historical changes in the birth rate combined with rapid restructuring of the skill level of labour demand, a hiatus of new entrants into the labour force 0 50,000 100,000 150,000 200,000 250,000 300,000 350,000 1 9 4 7 1 9 5 1 1 9 5 5 1 9 5 9 1 9 6 3 1 9 6 7 1 9 7 1 1 9 7 5 1 9 7 9 1 9 8 3 1 9 8 7 1 9 9 1 1 9 9 5 1 9 9 9 2 0 0 3 2 0 0 7 2 0 1 1 2 0 1 5 a n n u al n e t o ve rs e as m ig ra ti o n 6 mcdonald p australian population studies 1 (1) 2017 (associated with increased participation in tertiary education) and increasing numbers retiring compounded to produce a skilled labour supply challenge in australia from the mid-1990s. this was exacerbated in the early 2000s by the onset of the mining boom and its associated spin-offs into other industries. somewhat presciently, the australian government in 1995 had shifted its permanent migration program away from family migration to points-tested skilled migration and introduced an uncapped temporary skilled worker visa (the 457 visa). these well-timed changes in policy meant that as labour demand expanded in the 2000s, skilled migration could be ramped up effectively to meet the demand. the demand in the early 2000s was so strong that the australian government extended the right to apply for permanent residence to temporary residents onshore through employer sponsorship. this two-step process where migrants entered australia on a temporary visa, established themselves through employment with an employer and then applied for permanent residence as an employersponsored applicant had become the dominant pathway to permanent residence by the second decade of the 21st century. the story of these developments is told by gregory (2014) and mcdonald (2015). both authors argue that the two-step process is very effective because of the guaranteed employment of the migrant as opposed to the potentially long job search that needs to be undertaken by an independent skilled applicant. there were also substantial increases in migrants via the ‘family stream’ and ‘working holiday maker’ visa subclasses and numbers of international students. the latter two groups also have relatively high levels of employment. finally, the high australian dollar and strong labour demand led also to very large numbers of new zealand citizens taking up employment in australia under the open labour market policy of the two countries which had been established in the early 1980s (abs 2010). when the economic recession hit at the end of 2008, for the first time in australian history the australian government did not cut the immigration program. breunig, deutscher and to (2017) claim that there is no evidence to suggest that this approach had negative impacts on the labour market outcomes of domestic australian workers. with this background, the purpose of this paper is to examine post-global financial crisis (gfc) employment outcomes in australia and the ways in which immigrants have contributed to those outcomes. 3. data and methods the input data for the study were: • estimated resident population (erp) of australia by five-year age group and sex for 30 june 2011 and 30 june 2016. the erp values used were those estimated prior to the results of the 2016 census. • number of employed persons by five-year age group and sex from the july 2011 and july 2016 labour force surveys published by the australian bureau of statistics (abs 2011, 2016a). • five-year age group and sex-specific employment to population ratios from the same surveys • five-year life table survival ratios by age and sex calculated from the australian life tables for 2013–2015 published by the australian bureau of statistics (abs 2016b). step 1 in the method was to project the resident australian population in 2016 by five-year age groups and sex in july 2011 forward to july 2016 using the life table survival ratios. at the oldest australian population studies 1 (1) 2017 mcdonald p 7 ages, the population aged 85 and over was survived to 90 and over. this produced the estimated age and sex distribution of the population in 2016 in the absence of migration. in step 2 the employment to population ratios from 2011 were applied to the projected 2016 population. at the oldest ages, the published data required the use of age group 65 and over for this purpose. this step calculates the number of people who would have been employed in australia in 2016 in the absence of migration and under the assumption that there was no change in the employment to population ratios between 2011 and 2016. this yields the first component: 1. changes in numbers employed due to population growth resulting from the existing 2011 age distribution and mortality in the five-year period (referred to in the paper as ‘population growth’). step 3 applied the 2016 employment to population ratios to the 2016 projected population in the absence of migration. this calculates the number of people who would have been employed in 2016 in the absence of migration but using the actual 2016 employment to population ratios. the difference between the results of step 3 and step 2 produces the second component: 2. changes in employment numbers due to the change in the employment to population ratios. step 4, the final stage, compared the actual numbers by age and sex employed in 2016 with the result from step 3. the difference between the two, the residual, shows the effect of adding in migration along with any errors (sampling error in the two surveys; errors in the erp used to blow-up the survey results; errors in the assumed levels of mortality). this provides an estimate of the third component: 3. the effect of net migration in the period 2011–2016 on the total number of people employed in 2016. in relation to the errors, the sampling errors in the labour force survey are small and any error deriving from the survival ratios would be negligible as most of the calculations relate to younger ages. error deriving from errors in the erp are potentially larger, but the changes made to the erp following the results of the 2016 census are small by historical standards (abs 2017). it should also be noted that the 2016 census-adjusted erp was larger than the pre-census estimate, implying that the residual (the third component) would have been larger than estimated here if the 2016 censusadjusted erp had been used to adjust the 2016 labour force survey results. in the end, it is impossible to gauge the combined effect of these potential sources of error but it can be confidently concluded that any error in component 3 would be small and would be in the direction of increasing this component. while error can also apply to the estimation of components 1 and 2, the net effect of the errors would have been very small in relation to these components. there are implicit assumptions in the method that the new migrants had the same age-sex-specific employment to population ratios in 2016 as the domestic population, and that the employment to population ratios of the non-migrant population would not have changed in the absence of migration. as the age and sex distributions of both the migrant and non-migrant populations are taken into account in the calculations, and as the new migrants in 2016 constituted only 5 per cent of total employment, the effects of any errors in these implicit assumptions will be marginal. 8 mcdonald p australian population studies 1 (1) 2017 4. results figure 2 shows the results for three age groups: 15–24; 25–54 (the prime working ages); and 55 and over. total employment is also shown. looking first at the five-year change in total employment (738,800), the largest change was in the prime working ages (431,300) followed by the older ages (286,000), with only a very small positive impact evident for ages 15–24 (21,500). to provide some perspective for these numbers, the total number of employed persons in australia in july 2011 was 10,890,700. this means that employment increased by 6.8 per cent in the five years to july 2016. 4.1 youth employment the small net change at ages 15–24 was the result of a large negative component (-137,900) due to population change and a smaller negative amount (-40,100) due to falls in employment to population ratios for this age group, offset by a large gain (199,500) due to the third component, international migration. in other words, without migration, numbers employed in the age group 15–24 would have fallen substantially. employed recent migrants aged 15–19 are primarily international students and working holiday makers employed very largely in part-time shift work in the hospitality and retail industries. the other group employed in this way in the same industries, even larger in number than the migrant group, is australian students. along with the netherlands, students in australia are much more likely to be employed than students in any other oecd country (quintini 2015). in industries with very distinct peaks and troughs in demand, in terms both of the time of day and time of year, the availability of a large group willing to work under these conditions is highly beneficial to employers. furthermore, australian and international students and working holiday makers are usually much more highly skilled than the jobs require, meaning that productivity is enhanced. figure 2: decomposition of the change in the total number of people employed in australia between june 2011 and june 2016, ages 15–24, 25–54 and 55 and over -200 -100 0 100 200 300 400 500 c h an ge in e m p lo ym e n t (t h o u sa n d s) 15-24 25-54 55+ net migration 2011-2016 change in employment participation rates total population growth – no migration australian population studies 1 (1) 2017 mcdonald p 9 given that youth unemployment in australia is a serious issue, the counterfactual question arises: had there been zero migration at these ages, would australian employers have opted to employ unemployed non-migrants aged 15–24 to a much greater extent, or would total employment have been lower? in addressing this counterfactual question, it needs to be remembered that unemployed domestic job-seekers receiving the newstart allowance effectively need to find longer-term, full-time work. otherwise, they are in and out of the centrelink office every time they take a new job or their hours change and, in this situation, there is a strong chance that they will lose benefits for some periods. given the structure of the labour market at these ages, such jobs are hard to find. as the skills requirements of the australian labour market have shifted upwards, skilled workers are rarely aged under 20, or even in their early twenties, and the opportunities for low-skilled, full-time work at these ages have fallen dramatically. thus, while recent temporary migrants at this age may be in competition for jobs with full-time australian students, their absence from the australian labour market generally would not open opportunities for young australians receiving newstart payments. young unemployed australians have a low-skills problem in a skilled labour market and the social security system makes it difficult for them to take part-time, short-term jobs. while there may be some instances of displacement, the broad story is that they are not pushed out of jobs by migrants. 4.2 prime working ages employment in the prime working ages (25–54 years) in the five-year period rose a little through population growth but declined by a small amount due to changes in employment participation. the small decline related to participation was the result of an increase for women of 31,700 fully offset by a fall for men of 40,100. consequently, almost all of the substantial net growth in employment at these ages was due to migration (395,800 out of a total of 431,300, or 92 per cent). total unemployment in this age group in 2016, excluding unemployed recent immigrants, was 321,000. thus the job growth that was achieved could not have been achieved without migration, even in the extremely unlikely situation that unemployment in the domestic population at these ages had fallen to zero. for long-term unemployed australians in this age group, the issue once more was primarily inadequate skills and not competition from migrants. this is not to say that there is no competition from migrants for jobs that might be filled by domestic workers. as the 2014 independent review of the integrity of the 457 visa subclass for the australian government noted, it is important that the list of occupations open to temporary skilled workers is continuously monitored (azarias et al. 2014). 4.3 older ages migration had very little impact on the relatively substantial growth in employment at ages 55 and above. here, the main source of growth (192,900) was population growth as the baby-boom and subsequent large generations replaced the earlier smaller generations. employment participation also increased at these ages adding 75,000 additional employed persons. virtually all of these (73,600) were older women. while employment participation rates rose sharply for men at older ages in the decade 2000–2010, there was little change after 2010. for older women, the continued increase in employment participation rates through to 2016 is largely the result of cohort changes. 10 mcdonald p australian population studies 1 (1) 2017 each successive cohort of women has increased its participation at younger ages and these higher rates at younger ages have flowed on to higher rates at older ages for each successive cohort. 4.4 the age of the labour force because migrants are young, migration dramatically changes the age structure of employment in australia (table 2). without migration, employment at ages under 55 years would have fallen by 143,000 in the five-year period. instead, with migration, employment at these ages increased by 452,000, a total turnaround of 595,000. at the same time, employment would have increased at older ages without migration by 268,000 and hardly increases when migration is included. in an age where young people are the assimilators of new technology, this impact of migration on the age structure of the labour force is likely to be highly beneficial to the economy. in addition, there is a growing argument that older and younger workers are complementary rather than substitutes for each other: younger workers have the technical skills, while older workers have the experience of how these skills might be utilised. table 2: change in numbers employed by broad age group with and without migration, 2011–2016 age group without migration with migration <55 years -143,000 452,000 55 years and over 268,000 286,000 there is also a gender effect of migration. without migration, male employment would have increased by just 6,000 in the five-year period in net terms. instead, including migration, it increased by 306,000. 5. discussion and conclusion the analysis here shows that immigration had a very large impact on employment change in australia between 2011 and 2016. while this is not an analysis of the economic benefits of migration to australia, a number of relatively recent studies provide some indication of these benefits. for example, in the context of a study which showed that immigration negatively affected the wages of low-skilled workers in most oecd countries, docquier, özden and peri (2010, p. 14) found that the ‘reverse takes place in canada and australia where immigration strongly helps the wages of less educated (+3.3 and +4.5% respectively)’. a report by mcdonald and temple (2013, p. 10) on the long-term effects of ageing and immigration on labour supply and per capita gross domestic product in australia concluded: ‘to draw a more definite conclusion, if the aim is to optimize gdp per capita growth through to [year] 2053, nom [net overseas migration] between about 160,000 and 220,000 seems to do this if it is assumed that the migrants’ have the same rate of labour productivity growth as the general australian population. a broad-based study published by migration council australia utilised the independent macroeconometric model to simulate migration scenarios and concluded: the economic impact of migration flows through into every aspect of the economy. it has a profound positive impact not just on population growth, but also on labour participation and employment, on wages and incomes, on our national skills base and on net productivity. (migration council australia 2015, p. 2) australian population studies 1 (1) 2017 mcdonald p 11 the productivity commission (2016, p. 2) similarly investigated the economic impacts of immigration, finding: ‘australia’s current immigration profile is projected to deliver a demographic dividend to australia and higher economic output per person.’ a further study for the productivity commission by breunig, deutscher and to (2017, p. 1) concluded: ‘we find almost no evidence that immigration harms the labour market outcomes of those born in australia’. referring to regional employment impacts, golebiowska, elnasri and withers (2016, p. 449) found: policy design—particularly the two-step visa process for regional migrants—has allowed substantial and effective regional location encouragement for immigrants over the past decade. … the job-creation dimensions of immigration may mean that regional policies seeking greater population growth … can be benefited by the use of targeted migration visa entry conditions. on the other side of the equation, there are negative externalities created by rapid population growth such as congestion and pressure on natural resources, urban infrastructure and housing. sobels et al. (2010) concluded: ‘the macro-scale modelling found that higher levels of nom [net overseas migration] impose greater adverse impacts on the quality of our natural and built environments, other things being equal’. it is clear that strong labour demand in australia in the first decade of the 21st century led to relatively high immigration and that this had significant economic benefits. this paper has shown that immigration in response to strong labour demand has continued following the gfc. in the next decade, very strong labour demand can be expected in the caring sector with the implementation of the national disability insurance scheme and natural growth also in aged care, and in the construction sector with increased emphasis on infrastructure development. meeting these demands will have spin-offs in other sectors. the work of golebiowska, elnasri and withers (2016) indicates that migration plays an important role also in regional development. while there may be surges in temporary migration from time to time (as in 2008), in the longer term the level of immigration is determined by the setting of the federal target for the permanent migration program. the australian government has set this target at 190,000 persons per annum for 2017–2018 and the next three years. this means that relatively high population growth will continue. recognising that there exist potential negative implications of rapid population growth, particularly for sydney and melbourne, it is incumbent upon governments to plan for the effects of this inevitable population growth on both infrastructure and resources. key messages • historical changes in the birth rate, the hiatus of new entrants into the labour force as young australians increased their participation in tertiary education, increasing numbers retiring and the relatively rapid restructuring of the skill level of labour demand all combined to produce a skilled labour supply challenge in australia from the mid-1990s. • skilled migration, both permanent and temporary, played an important role in meeting this labour demand. • immigration in response to strong labour demand continued after the gfc and had a highly significant impact on the growth of employment in australia between 2011 and 2016. • it is incumbent upon governments to plan for the effects that inevitable population growth will have on infrastructure and resources in the coming years. 12 mcdonald p australian population studies 1 (1) 2017 references australian bureau of statistics (abs) (2010) new zealanders in australia. in: australian social trends, abs cat. no. 4102.2. canberra: abs. australian bureau of statistics (abs) (2011) 6291.0.55.001 – labour force, australia, detailed – electronic delivery, jul 2011, viewed 1 july 2017, http://www.abs.gov.au/ausstats/abs@.nsf/allprimarymainfeatures/e0b54015bb216045ca257 90b001620fd?opendocument. australian bureau of statistics (abs) (2014) 3105.0.65.001 – australian historical population statistics, 2014. population age-sex structure, data cube: excel spreadsheet, viewed 1 july 2017, http://www.abs.gov.au/ausstats/abs@.nsf/detailspage/3105.0.65.0012014?opendocument. australian bureau of statistics (abs) (2016a) 6291.0.55.001 – labour force, australia, detailed – electronic delivery, jul 2016, viewed 1 july 2017, http://www.abs.gov.au/ausstats/abs@.nsf/allprimarymainfeatures/5a9245950663a4f6ca258 035007fee80?opendocument. australian bureau of statistics (abs) (2016b) 3302.0.55.001 – life tables, states, territories and australia, 2013–2015, viewed 1 july 2017, http://www.abs.gov.au/ausstats/abs@.nsf/detailspage/3302.0.55.0012013-2015?opendocument australian bureau of statistics (abs) (2017) australian demographic statistics, december quarter 2016, abs cat. no. 3101.0. canberra: abs. azarias j, lambert j, mcdonald p and malyon k (2014) robust new foundations: a streamlined, transparent and responsive system for the 457 programme, viewed 1 july 2017, http://www.border.gov.au/reportsandpublications/documents/reviews-andinquiries/streamlined-responsive-457-programme.pdf. breunig r, deutscher n and to h-t (2017) the relationship between immigration to australia and the labour market outcomes of australian-born workers. economic record 93(301): 255–276. docquier f, özden c and peri g (2010) the wage effects of immigration and emigration. nber working paper no. 16646. cambridge, massachusetts: national bureau of economic research. golebiowska k, elnasri a and withers g (2016) responding to negative public attitudes towards immigration through analysis and policy: regional and unemployment dimensions. australian geographer 47(4): 435–453. gregory r (2014) cameo 3. the two-step australian immigration policy and its impacts on immigrant employment outcomes. in: chiswick b and miller p (eds) handbook of the economics of international migration, vol. 1. north holland: elsevier; 1421–1443. mcdonald p (2015) international migration and employment in australia. population review 54(2): 1–12. mcdonald p and temple j (2013) the long term effects of ageing and immigration upon labour supply and per capita gross domestic product: australia 2012–2062, final report, viewed 1 july 2017, http://www.demographicinsight.com.au/other/mcdonaldtemple%20-%20web.pdf. migration council australia (2015) the economic impact of migration. canberra: migration council australia. norton a (2016) mapping australian higher education 2016, grattan institute report no. 2016–11. melbourne: grattan institute. productivity commission (2016) migrant intake into australia. productivity commission inquiry report no. 77, 13 april 2016. canberra: commonwealth of australia. quintini g (2015) a picture of working students in oecd countries, viewed 1 july 2017, https://oecdskillsandwork.wordpress.com/2015/09/02/a-picture-of-working-students-in-oecdcountries/. sobels j, richardson s, turner g, maude a, tan y, beer a and wei z (2010) long-term physical implications of net overseas migration: australia in 2050. adelaide sa: national institute for labour studies, flinders university. http://www.abs.gov.au/ausstats/abs@.nsf/allprimarymainfeatures/e0b54015bb216045ca25790b001620fd?opendocument http://www.abs.gov.au/ausstats/abs@.nsf/allprimarymainfeatures/e0b54015bb216045ca25790b001620fd?opendocument http://www.abs.gov.au/ausstats/abs@.nsf/detailspage/3105.0.65.0012014?opendocument http://www.abs.gov.au/ausstats/abs@.nsf/allprimarymainfeatures/5a9245950663a4f6ca258035007fee80?opendocument http://www.abs.gov.au/ausstats/abs@.nsf/allprimarymainfeatures/5a9245950663a4f6ca258035007fee80?opendocument http://www.abs.gov.au/ausstats/abs@.nsf/detailspage/3302.0.55.0012013-2015?opendocument http://www.border.gov.au/reportsandpublications/documents/reviews-and-inquiries/streamlined-responsive-457-programme.pdf http://www.border.gov.au/reportsandpublications/documents/reviews-and-inquiries/streamlined-responsive-457-programme.pdf http://www.demographicinsight.com.au/other/mcdonaldtemple%20-%20web.pdf https://oecdskillsandwork.wordpress.com/2015/09/02/a-picture-of-working-students-in-oecd-countries/ https://oecdskillsandwork.wordpress.com/2015/09/02/a-picture-of-working-students-in-oecd-countries/ australian population studies 2020 | volume 4 | issue 1 | pages 1-3 © piggott 2020. published under the creative commons attribution-noncommercial licence 3.0 australia (cc by-nc 3.0 au). journal website: www.australianpopulationstudies.org introduction to the special issue on population ageing in australia john piggott* the university of new south wales * email: j.piggott@unsw.edu.au. address: arc centre of excellence in population ageing research, university of new south wales, level 3, 223 anzac parade, kensington, nsw 2033, australia. paper received 2 may 2020; accepted 4 may 2020; published 25 may 2020. it is a great privilege to have the opportunity to write an introduction to this important issue of this new and exciting journal. it is very heartening that the editors have chosen to devote an entire issue to the topic of population ageing. population ageing affects all of us and it impacts many dimensions of our lives. from changed family structures, workplace cultures, through to retirement support and health and aged care arrangements, households are facing unfamiliar, sometimes irreversible, complex and critical life choices. it is exerting unprecedented pressures on traditional social institutions, both in australia and around the world. sustained falls in fertility, sometimes to “super-low” levels, are combining with increasing life expectancy to drive global movements in labour, capital and trade in ways that are still unknown. employers will have to learn to deal with and accommodate an older workforce. and governments under fiscal stress are reducing their commitment to pensions, health and aged care, raising new private sector opportunities and challenges for regulation and governance. these adjustments will have flow-on effects to intergenerational exchange and equity and policy formulation. as the un has said, the phenomenon of population ageing is unprecedented, enduring, pervasive, and profound. research is critically required to better understand how policy and practice innovation can ease our individual and collective paths through this transition. longer life is one of humanity’s greatest achievements. it is important that this substantial contribution to human welfare is fully leveraged by moving to policy structures and private sector activity in the best possible way. this requires new knowledge and an evidence base informed by multiple disciplines. the need for cutting edge research related to demographic ageing is documented in the international literature. this is especially true when the focus is on increasing lifespan. in the 2016 application for renewed funding for the arc centre of excellence in population ageing research (cepar), we listed calls for a greater research effort in the world health organisation (2015) and the us national research council (2013), which emphasises the advantages of an inter-disciplinary approach to the issue. an especially useful benchmark is provided in national research council (2012). chapter 10 identifies the following major research needs: demographic and health measurement and projections; capacity to work longer and longer working life; changes in consumption and saving; and modelling efforts and data needs. about:blank about:blank 30 piggott australian population studies 4 (1) 2020 and there has been at least some response on the international stage. last year, the year that japan, the world’s oldest nation, hosted the g20 meetings, a task force was established under its auspices to examine population ageing (see, for example, chomik et al. 2019). the world health organisation is promoting the 2020s as the “decade of healthy aging”, leading up to the deadline for the sustainable development goals in 2030. and the us national academy of medicine has launched a commission on a “roadmap to health longevity”. but natural population increase is in some ways more important. australia remains a young country by developed country standards. partly, this is because social and family institutions are supportive of fertility. but if this were all we had, our population would still age quite rapidly and total population, and national income, would eventually decline. the total fertility rate sits well below replacement. critical to the slowing of population ageing is australia’s immigration policies, which are well targeted to keeping the population young. australia’s continuing economic prosperity depends on the maintenance of strong and well-targeted migration, and targeting depends on our international education industry. something like half of new permanent residents in australia have previously been students here. even though this is primarily a journal about australian population, it is important to recognise the impact on this nation of international demographic forces. china is currently younger than australia but is rapidly catching up – some estimates have china older than australia by the middle of this century. what does this mean for china-australia trade? for migration? for international education services? and china, while enormously important, just leads the pack of emerging economies in asia, all of them important to australia, where fertility has fallen sharply, and demographic transition is an established dynamic. it may well be that changes in international demography end up being more important influences on australia’s future than its own demographic change. in the meantime, there is no shortage of relevant and important research questions. for example: how will australia’s age structure change in the future? will covid-19 make any difference to it? how will the fertility differential shift, and what will the social implications of the shift be? how do households learn to adjust to changing demographic realities, and how does cognitive decline in older ages affect their abilities to make major life decisions? will product innovations help them in this challenge? how do organisations change their behaviour to facilitate greater participation by older workers? is retiring later good for you? does work lead to more, or less, successful ageing? and when you turn from understanding the micro-transition better, and ask broader social and economic questions, solid evidence is even harder to find. to list a few: how will an ageing demographic affect government budgets and services, and social and economic inequality, both between and within generations? how should the government best set policies to respond to demographic change? as retirement risks, and later life risks more generally, become more important, how will markets respond? will management of this kind of risk sit at all in the private sector, or will it remain for government to be the insurer? how can policy innovation help shape broader and better allocated age-related risk sharing and management between the public sector, business, and households? australian population studies 4 (1) 2020 piggott 29 there is at least some research on all these questions. but there are definitive answers to none of them. special issues such as these serve to focus research attention on these unanswered questions, and in so doing, render us all a considerable service. acknowledgements this research was supported by the australian research council centre of excellence in population ageing research (project number ce170100005). references chomik r, piggott j and yan s (2019) aging, fiscal sustainability, and adequacy of social security systems. in asian development bank institute (ed.) aging societies: policies and perspectives. tokyo: asian development bank institute; 52-64. national research council (2012) aging and the macroeconomy: long-term implications of an older population. washington, dc: the national academies press. national research council (2013) new directions in the sociology of aging. washington, dc: national academies press. world health organisation [who] (2015) world report on ageing and health. geneva: world health organisation. it is a great privilege to have the opportunity to write an introduction to this important issue of this new and exciting journal. it is very heartening that the editors have chosen to devote an entire issue to the topic of population ageing. population ageing affects all of us and it impacts many dimensions of our lives. from changed family structures, workplace cultures, through to retirement support and health and aged care arrangements, households are facing unfamiliar, sometimes irrev... research is critically required to better understand how policy and practice innovation can ease our individual and collective paths through this transition. longer life is one of humanity’s greatest achievements. it is important that this substantial... the need for cutting edge research related to demographic ageing is documented in the international literature. this is especially true when the focus is on increasing lifespan. in the 2016 application for renewed funding for the arc centre of excelle... and there has been at least some response on the international stage. last year, the year that japan, the world’s oldest nation, hosted the g20 meetings, a task force was established under its auspices to examine population ageing (see, for example, c... but natural population increase is in some ways more important. australia remains a young country by developed country standards. partly, this is because social and family institutions are supportive of fertility. but if this were all we had, our popu... critical to the slowing of population ageing is australia’s immigration policies, which are well targeted to keeping the population young. australia’s continuing economic prosperity depends on the maintenance of strong and well-targeted migration, and... even though this is primarily a journal about australian population, it is important to recognise the impact on this nation of international demographic forces. china is currently younger than australia but is rapidly catching up – some estimates have... in the meantime, there is no shortage of relevant and important research questions. for example: how will australia’s age structure change in the future? will covid-19 make any difference to it? how will the fertility differential shift, and what will... and when you turn from understanding the micro-transition better, and ask broader social and economic questions, solid evidence is even harder to find. to list a few: how will an ageing demographic affect government budgets and services, and social an... there is at least some research on all these questions. but there are definitive answers to none of them. special issues such as these serve to focus research attention on these unanswered questions, and in so doing, render us all a considerable service. acknowledgements references australian population studies 2019 | volume 3 | issue 2 | pages 29-33 © lock and pettit 2019. published under the creative commons attribution-noncommercial licence 3.0 australia (cc by-nc 3.0 au). journal website: www.australianpopulationstudies.org visualising population distribution in australia over time using rapid 3d web graphics libraries oliver lock* university of new south wales chris pettit university of new south wales * corresponding author. email: o.lock@unsw.edu.au. address: city analytics lab, faculty of built environment. red centre west wing, university of new south wales. kensington, nsw 2033. australia. paper received 1 may 2019; accepted 29 may 2019; published 18 november 2019 this paper presents an interactive 3d visualisation tool and workflow for exploring current and past population data in australia. the visualisation of small-area population data enables the exploration of population dynamics, such as scale, density, movement, age and sociodemographic information. such information is becoming increasingly accessible with the advent of open government data (jetzek et al. 2013), sensor data (sagl et al. 2012), and more sophisticated computational and visualisation tools (kashnitsky and schöley 2018; o’brien and cheshire 2016; pettit et al. 2012; pettit et al. 2017). such tools enable planners, geographers, demographers to understand the current structure of cities and how they are changing over time. there are a number of challenges in representing small-area population geographies in a userfriendly way at a large national (or, indeed, international) scale. these geographies consist of statistical boundaries which can vary significantly in area. the rendering of density and colour breaks can further be difficult, and often can skew the visibility of interconnecting regional cities and centres. due to the complexity of boundary geometries, it can be computationally slow to show national datasets in standard gis applications, and even in web applications they are particularly difficult to interpret when zoomed out as a consequence of preserving the legibility of these boundaries. several systems have been developed to address the visualisation of detailed population data at a large scale by using methods which ameliorate effects of administrative population boundaries. for example, work by smith (2016) discusses this, and creates an interactive 2d population explorer using the open release of the european commission (2019) global human settlement layer. to overcome these challenges this demographic presents a novel approach (figure 1) using rapid, open source 3d web visualisation libraries to explore how this can be achieved with national-level population data in australia. d em og ra ph ic http://www.australianpopulationstudies.org/ mailto:o.lock@unsw.edu.au 30 lock & pettit australian population studies 3 (2) 2019 figure 1: australian population explorer note: access the interactive version via https://aus3dpop.city-informatics.com webgl operates across multiple supported browsers (chrome, firefox, safari) with a royalty-free api (application programming interface), and creates fast, hardware accelerated 3d graphics (khronos group 2019). the open-source webgl-based library deckgl (uber 2019) was used due to its specialisation in processing very large geographic data sets and rendering them in three dimensions. three dimensional techniques are a sensible choice for displaying population density, which has a natural perceptual association with its impact on the height and size of the built environment to support that density. the method used in this visualisation is known as ‘hexagon binning’. a hexagon bin is a form of bivariate histogram which considers both its location and a frequency of values underneath it. the process of generating a hexagon bin is as follows: • geographic space is transformed into a regular grid of hexagons; https://aus3dpop.city-informatics.com/ australian population studies 3 (2) 2019 lock & pettit 31 • the number of points (in this instance, population) falling into each hexagon is counted; • hexagons with a count above 0 are plotted, with their colour and height varying with the density of points underneath. there are several reasons why hexagons are useful (lewin-koh 2011). hexagons are more efficient than covering a plane with squares. they are also visually less biased for displaying densities than other regular tessellations with fewer edges (and thus an implied visual directionality). as such, they provide a sound medium for communicating population and related urban densities. for this visualisation, usual residence census data from the smallest geographic unit for population data within the abs census data was used, mesh blocks. according to the abs (2016), mesh blocks contain around 30-60 dwellings, the smallest number of dwellings that can be communicated without potentially revealing sensitive information. the amount and geometry of mesh blocks has changed over time, from the 2006 census (when originally conceived) when there were approximately 314,000 to approximately 347,000 in 2011 and 358,000 in 2016. a challenge lies in their evolving geometries as their visual comparison is unclear over time. another challenge, as discussed, is in computing the complex, detailed geometries and trying to represent such granular polygons at national scale. the resulting visualisation is available through australian population explorer (lock 2019). the tool is rapid and exploratory. users can toggle between alternate years of the census and see changes in population distribution. further, changes can be made in the coverage of the hexagon to explore more densely populated areas. the lower percentile toggle can be used to see which areas fall in the highest densities of all hexagons generated across the country. it can be seen that australia’s major cities continue to operate as primate cities within their respective states/territories, however, such visualisation also highlights the country’s vast network of regional cities and the interconnectedness of urban regions such as in south-east queensland. by increasing the radius of the hexagon to 10km, and around the 95th percentile, users can begin to see australia’s network of smaller cities and how they may connect to one another, which could, for example, be useful in considering future, major infrastructure projects. with a smaller hex bin it is clearly visible that 2006 and 2011 have greater similarity in population structure and pattern than in 2016. this is likely due to a combination of factors related to the new digital version of the census in 2016, including changes in response rates, imputation challenges and new address-matching systems (the address register), documented by census independent assurance panel (harding et al. 2017). this representation difference was mitigated by increasing the hexagon sizes. by changing size dynamically, it is easier to compare cities over time and highlight overarching structures – see for example components b, c and d of figure 1. in summary, the australian population explorer presented here permits the interactive investigation of population distribution and trends over time. this contributes to our understanding of effective methods to represent detailed small-area information, which can be useful across multiple domains. such applications include planning and delivering of government services, transportation planning, 32 lock & pettit australian population studies 3 (2) 2019 estimating vulnerable communities in disaster situations, understanding communities for election preparation, estimation of population at risk for spread of diseases (wardrop et al. 2018). references australian bureau of statistics [abs] (2018) population projections, australia, 2017 (base) – 2066. catalogue no. 3222. canberra: abs. australian bureau of statistics [abs] (2016) australian statistical geography standard (asgs): volume 1 main structure and greater capital city statistical areas, july 2016. catalogue no. 1270.0.55.001. canberra: abs harding s, jackson pulver l, mcdonald p, morrison p, trewin d and voss a (2017) report on the quality of 2016 census data. http://www.abs.gov.au/websitedbs/d3310114.nsf/home/independent+assurance+panel chen c, ma j, susilo y, liu y and wang m (2016) the promises of big data and small data for travel behavior (aka human mobility) analysis. transportation research part c: emerging technologies, 68, 285–299. https://doi.org/10.1016/j.trc.2016.04.005 european commission (2019) ghsl global human settlement layer. https://ghsl.jrc.ec.europa.eu/index.php. accessed on 28 march 2019. jetzek t, avital m and bjørn-andersen n (2013) generating value from open government data. proceedings of the 34th international conference on information systems 1-20. http://aisel.aisnet.org/cgi/viewcontent.cgi?article=1181&context=icis2013. kashnitsky i and schöley j (2018) regional population structures at a glance. the lancet 392(10143): 209– 210. https://doi.org/10.1016/s0140-6736(18)31194-2 khronos group (2019) webgl opengl es for the web. https://www.khronos.org/webgl/. accessed on 25 june 2019. lewin-koh n (2011) hexagon binning: an overview. http://cran.rproject.org/web/packages/hexbin/vignettes/hexagon_binning.pdf. accessed on 28 march 2019. lock o (2019) australian population explorer. https://aus3dpop.city-informatics.com/. accessed on 25 june 2019. o’brien o and cheshire j (2016) interactive mapping for large, open demographic data sets using familiar geographical features. journal of maps 12(4): 676–683. https://doi.org/10.1080/17445647.2015.1060183 pettit c j, tanton r and hunter j (2017) an online platform for conducting spatial-statistical analyses of national census data across australia. computers, environment and urban systems 63: 68–79. https://doi.org/10.1016/j.compenvurbsys.2016.05.008 pettit c, widjaja i, russo p, sinnott r, stimson r and tomko m (2012) visualisation support for exploring urban space and place. in: isprs annals of the photogrammetry, remote sensing and spatial information sciences 1: 153–158. https://doi.org/10.5194/isprsannals-i-2-153-2012 sagl g, resch b, hawelka b and beinat e (2012) from social sensor data to collective human behaviour patterns – analysing and visualising spatio-temporal dynamics in urban environments. gi_forum 2012: geovisualization, society and learning 1: 54–63. https://doi.org/10.1080/01441647.2019.1616849 smith d a (2016) online interactive thematic mapping: applications and techniques for socio-economic research. computers, environment and urban systems 57: 106-117. https://doi.org/10.1016/j.compenvurbsys.2016.01.002 uber (2019) deck.gl. https://deck.gl/. accessed on 24 january 2019. http://www.abs.gov.au/websitedbs/d3310114.nsf/home/independent+assurance+panel https://doi.org/10.1016/j.trc.2016.04.005 https://ghsl.jrc.ec.europa.eu/index.php http://aisel.aisnet.org/cgi/viewcontent.cgi?article=1181&context=icis2013 https://doi.org/10.1016/s0140-6736(18)31194-2 https://www.khronos.org/webgl/ http://cran.rproject.org/web/packages/hexbin/vignettes/hexagon_binning.pdf https://aus3dpop.city-informatics.com/ https://doi.org/10.1080/17445647.2015.1060183 https://doi.org/10.1016/j.compenvurbsys.2016.05.008 https://doi.org/10.5194/isprsannals-i-2-153-2012 https://doi.org/10.1080/01441647.2019.1616849 https://doi.org/10.1016/j.compenvurbsys.2016.01.002 https://deck.gl/ australian population studies 3 (2) 2019 lock & pettit 33 wardrop n a, jochem w c, bird t j, chamberlain h r, clarke d, kerr d, bengtsson l, juran s, seaman v and tatem a j (2018) spatially disaggregated population estimates in the absence of national population and housing census data. proceedings of the national academy of sciences 115(14): 3529–3537. https://doi.org/10.1073/pnas.1715305115 https://doi.org/10.1073/pnas.1715305115 australian population studies 2019 | volume 3 | issue 2 | pages 1-15 © gray and evans 2019. published under the creative commons attribution-noncommercial licence 3.0 australia (cc by-nc 3.0 au). journal website: www.australianpopulationstudies.org changing education, changing fertility: a decomposition of completed fertility in australia edith gray* the australian national university ann evans the australian national university * email: edith.gray@anu.edu.au. address: school of demography, research school of social sciences, the australian national university, act 2601 paper received 23 may 2019; accepted 3 september 2019; published 18 november 2019 abstract background the expansion of education in australia, particularly for women, is one of the most significant social changes of the last five decades. the relationship between education and fertility has been widely studied, showing that increases in higher education for women are consistently associated with lower fertility. given the close link between education and fertility, this paper questions what effect the changing educational profile of australian women has had on overall fertility trends. aims this paper investigates the effect of the increase in education on completed fertility by decomposing the change in overall completed fertility into two components: (1) change in completed fertility as a result of the proportion of women in different education categories and, (2) changes in completed fertility of women in each education category. data and methods the study uses 2016 census data on the number of children ever born of five cohorts of women born between 1952 and 1976. decomposition is used to distinguish the effects of the two components. results the educational composition of women in these cohorts is dramatically different, with an increasing number of women having completed tertiary education in later cohorts. completed fertility has also changed across successive cohorts. we find that for the earliest cohorts most of the decline is due to declines in completed fertility within education categories, but for later cohorts the decline is attributable to increases in the proportion of women with higher levels of education. conclusions despite tertiary education becoming much more common, fertility within this group remains lower than other education groups. while other countries have seen a narrowing of the gap in fertility rates between education groups, this pattern is not found in australia. key words education; fertility; fertility decline; children ever born; census; decomposition. http://www.australianpopulationstudies.org/ mailto:edith.gray@anu.edu.au 2 gray & evans australian population studies 3 (2) 2019 1. introduction the expansion of education in australia is one of the most significant social changes of the last five decades. the 2016 census marked the first time in australian history that the proportion of women aged 40-44 with a university degree (36 per cent) exceeded the proportion with no post-school qualification (30 per cent). this is a remarkable increase from 30 years earlier when just 5 per cent had a university degree and 70 per cent had no post-school qualification (abs 1988). given the close link between education and fertility, we question what effect the changing educational profile of the population has had on overall fertility trends over time. as the category of women with higher educational qualifications has become less selective (adsera 2017), has the negative relationship between educational attainment and fertility weakened for younger cohorts? the relationship between education and fertility has been one of the most widely studied topics in demography. decades of research has consistently found that at both the individual and population level, higher education is almost universally associated with lower fertility (cochrane 1979; caldwell 1980; rindfuss et al. 1996; musick et al. 2009). women with higher levels of education tend to start childbearing later, have fewer children overall and are more likely to be childless. the reasons for this are numerous and include the ‘incarceration effect’ of longer time spent enrolled in education which delays the start of family formation (heck et al. 1997; ní bhrolcháin and beaujouan 2012), economic considerations including increased opportunity costs, changes in cultural or attitudinal factors such as increased knowledge and decision-making power regarding contraceptive use, and different values regarding gender and family dynamics (fort et al. 2016; neels et al. 2017). over time, the difference in fertility between those with high and low education, can develop in three ways. they can: (1) narrow, (2) widen, or (3) stay stable. recent research looking at change in educational differences across cohorts has been mixed. in south korea, yoo (2014) found that educational differences in completed fertility almost disappeared as the country went through the fertility transition. similarly, in denmark, norway and sweden education differences in cohort fertility have virtually disappeared in younger cohorts (jalovaara et al. 2018; andersson et al. 2009). however, a convergence of fertility across educational categories is not evident in all countries. a cross-national study of 25 european countries found that over time the negative gradient had remained strong in the mediterranean and post-communist countries (merz and liefbroer 2017). in great britain, women with degree-level qualifications still have smaller family sizes than those with lower levels of education and this difference in quantum appears to have widened in recent years (berrington et al. 2015). evidently the relationship between education and fertility is complex and varies not only over time, but also from country to country according to institutional contexts such as family policies, labour market characteristics and gender norms (liefbroer and corijn 1999; sobotka et al. 2017). it is suggested that educational differences in cohort fertility have all but disappeared in nordic countries because their welfare-state policies focus on supporting gender equality and allowing men and women to combine family and work (andersson et al. 2009; kravdal and rindfuss 2008). this paper investigates the effect of the increase in education on completed fertility, examining the completed fertility rate of five cohorts of women in australia born between 1952-56 and 1972-76 – cohorts which experienced large changes in education patterns. these cohorts also experienced considerable change in completed fertility. to consider these changes, we decompose the overall australian population studies 3 (2) 2019 gray & evans 3 change in completed fertility to analyse how it changed due to two factors: changes in the proportion of women in each education category, and changes in fertility of women in education categories. we note that many other changes in family formation also occurred for these cohorts including increases in non-marital childbearing, divorce and cohabitation without marriage. drawing on this analysis, we determine what proportion of the change in completed fertility across cohorts is due to the changing education patterns, and what proportion is due to changing fertility patterns within education categories. we begin by outlining recent trends in education, focusing on the period of 1960s to 1990s, the time period when most of the women in these cohorts would have been completing secondary education and gaining post-school qualifications. we then examine differences in completed fertility across education categories and how this has changed over time. 2. the expansion of education since the 1960s the 1960s marked the large-scale expansion of secondary schooling in australia. during the 1960s and 1970s there was growing political and public support for a system of comprehensive high schools which would enrol everyone regardless of merit or aptitude (campbell and proctor 2014). the influential 1973 karmel report called for reform in the education sector including establishing standards of achievement and ensuring adequate resourcing. it also addressed the existing gender inequality in male and female school completion. the report influenced a large increase in commonwealth funding during the whitlam government of the early 1970s. however, the economic crisis of the later 1970s and early 1980s, which was marred by high unemployment and inflation, led to restrictions on education expenditure (burke and spaull 2001). there are many different measures that can be used to show how education has expanded, but one of the most common is the participation rate at selected ages. the participation rate measures the proportion of the population at a selected age enrolled in an educational institution. the participation rate at age 16 from 1966 to 2000 is shown in figure 1. there was a dramatic rise in participation at age 16 throughout the 1960s, a plateau in the late 1970s and 1980s which may have been due to the economic crisis, followed again by an increase in the 1980s. in the mid-60s less than half of boys and girls aged 16 attended school. by 1970 this had increased to 55 per cent and 47 per cent respectively, and by 1980 to 56 and 60 per cent, with girls surpassing boys in the mid-1970s. another education indicator often used to study trends over time is apparent retention rate to year 12, which is the final year of high school completed at around age 18. this apparent retention rate is expressed as the number of students of a particular sex enrolled in year 12 as a percentage of the cohort of students of the same sex who first commenced secondary schooling (le and miller 2002). in 1971 the apparent retention rate to year 12 was 27 per cent for females and 34 per cent for males. by 1991 this had increased to 77 per cent for females and 66 per cent for males. from 1976 the percent of female students continuing to year 12 exceeded that of males, although this is partly explained by the greater uptake of trade and apprenticeship courses by males than females after completing year 10 (abs 1993 p. 95). 4 gray & evans australian population studies 3 (2) 2019 figure 1: participation rate in school at age 16 by sex, 1966-2000 source: anderson and vernon 1983; long et al. 1999; abs 2001, 2009. in line with increasing school completion already noted, there has also been an increase in the population with tertiary education. the increase in the total enrolment in higher education is shown in figure 2. part of the increase shown is a consequence of the progressive upgrading to degree status of courses which were at diploma level or lower. this includes many vocational-oriented qualifications in fields such as teaching, accounting, surveying, and nursing, originally provided by the former institutes of technology, teacher’s colleges, colleges of advanced education, and technical colleges (abs 1993). the fields of teaching and nursing are disproportionally female, so these changes had a direct impact on the percentage of women completing degree qualifications. figure 2: total enrolment in higher education by sex, 1966-2000 note: includes government teachers colleges from 1973 onwards. includes non-government teachers colleges from 1974 onwards. figures for years from 1985 to 1993 progressively include state-funded basic nursing students who would previously have been trained in hospitals. source: department of education, training and youth affairs (2001) australian population studies 3 (2) 2019 gray & evans 5 the other change in the education landscape is the changing nature of government funding models. in the early part of this period students paid tuition fees which were largely covered by merit-based commonwealth scholarships. in 1973 the whitlam government abolished fees. the impact of this is evident in a small but steep increase in tertiary enrolment, especially for women. in 1987 the government re-introduced fees through the higher education administrative charge (heac) and in 1989 introduced the higher education contribution scheme (hecs), an interest free incomecontingent loan for students to assist with payments of tuition fees (chapman 2011). these funding changes have clearly impacted total enrolment for men and women during the period of early adulthood for the cohorts covered in this study. 3. data and methods in order to examine the relationship between education and fertility we use data from the 2016 census. the census collects information on the three key aspects needed for this analysis: highest level of education, age, and number of children ever born. as seen in figures 1 and 2, the key period of educational expansion was in the 1960s to 1990s. to capture the experience of women going through secondary school and university during this period, we selected women born in a 19-year period from 1952 to 1976, split into five birth cohorts: 1952-56, 1957-61, 1962-66, 1967-71, 1972-76. at the time of the 2016 census they were aged 60-64, 55-59, 50-54, 45-49 and 40-44 respectively (table 1). table 1: cohort birth years and age at census cohort age at 2016 census year turned 16 1952-56 60-64 1968-72 1957-61 55-59 1973-77 1962-66 50-54 1978-82 1967-71 45-49 1983-87 1972-76 40-44 1988-92 detailed education categories are constructed from two variables: highest level of schooling and highest level of post-school qualifications gained. we were particularly interested in looking at detailed categories of educational attainment to gain a richer insight on the differences between high and low education. our measure of completed fertility is based on the census question asked of all females aged 15 and over: how many babies they have ever given birth to. using this ‘children ever born’ variable we calculate cohort fertility or average completed family size for each education category within each cohort. cohort completed fertility is calculated by multiplying the number children by the total number of women, including those who have no children. for the cohort born in 1972-76 who are aged 40-44, childbearing is not fully complete. however, we note that in 2017, just under 5 per cent of births were to mothers aged 40 or over (abs 2018), and more than half of these births were at 6 gray & evans australian population studies 3 (2) 2019 ages 40 and 41, so the overall impact of right-censoring and loss of information about future births for this group is relatively small. we begin by outlining the educational composition and completed fertility for each cohort. we then decompose the overall change in completed fertility to see how it changed due to two factors: the proportion of women in a particular education category, and the completed fertility of women in that education category. we use das gupta’s (1993) decomposition method developed as a stata program by li (2017). as we are only using one factor (education) the formula to calculate the effect of education and composition, respectively is: 𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸 𝐸𝐸𝐸𝐸𝑐𝑐𝑐𝑐𝐸𝐸𝑐𝑐𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸 𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝐸𝐸𝐸𝐸 = � 𝐸𝐸𝑖𝑖𝑖𝑖 𝐸𝐸𝑖𝑖 + 𝐸𝐸𝑖𝑖𝑖𝑖 𝐸𝐸𝑖𝑖 2 𝑖𝑖 × 𝐶𝐶𝐶𝐶𝐶𝐶𝑖𝑖𝑖𝑖 − � 𝐸𝐸𝑖𝑖𝑖𝑖 𝐸𝐸𝑖𝑖 + 𝐸𝐸𝑖𝑖𝑖𝑖 𝐸𝐸𝑖𝑖 2 𝑖𝑖 × 𝐶𝐶𝐶𝐶𝐶𝐶𝑖𝑖𝑖𝑖 𝐶𝐶𝑒𝑒𝐹𝐹𝐸𝐸𝐸𝐸𝐹𝐹𝐸𝐸𝐸𝐸𝐹𝐹 𝐹𝐹𝐸𝐸𝐸𝐸𝑒𝑒 𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝐸𝐸𝐸𝐸 = � 𝐶𝐶𝐶𝐶𝐶𝐶𝑖𝑖𝑖𝑖 + 𝐶𝐶𝐶𝐶𝐶𝐶𝑖𝑖𝑖𝑖 2 𝑖𝑖 × 𝐸𝐸𝑖𝑖𝑖𝑖 𝐸𝐸𝑖𝑖 − � 𝐶𝐶𝐶𝐶𝐶𝐶𝑖𝑖𝑖𝑖 + 𝐶𝐶𝐶𝐶𝐶𝐶𝑖𝑖𝑖𝑖 2 𝑖𝑖 × 𝐸𝐸𝑖𝑖𝑖𝑖 𝐸𝐸𝑖𝑖 where cfr is the completed fertility rate, n is the population, i is education category, and 𝐸𝐸 and 𝑏𝑏 are the two populations being compared. for any two populations in this simple decomposition the difference between completed fertility rates is attributed to fertility or education composition, with these two components adding up to 100 per cent. 4. results 4.1. change in completed fertility as expected, due to the rise of education for women in the recent past, the educational composition of women in these cohorts is dramatically different (figure 3). the major change in education has been a decline in the proportion of women who only completed year 10 or below, from 30 per cent among the 1952-56 cohort to 10 per cent in the youngest cohort. in turn there has been a corresponding increase in women who have a bachelor’s degree, from 13 per cent in the earliest cohort to 23 in the latest cohort. those with complete high school (year 12) rose slightly, as did all categories of post-school qualification, except basic vocational. completed fertility has also changed across successive cohorts of women. overall, completed fertility has fallen from 2.23 in the 1952-56 cohort to 1.97 in the 1972-76 cohort. figure 4 shows how completed fertility has changed across the cohorts by education level. at higher levels of education completed fertility has declined more or less steadily, whereas for year 11 and year 10 or below, after a decline in the middle cohorts, we see a rise again in fertility for the most recent cohort. for example, for year 10, completed fertility fell from 2.44 among the 1952-56 cohort to 2.34 for women born in 1962-66. it then increased again to 2.43 for the 1972-76 cohort. this change of fertility for women in the lowest education categories are aligned with the dramatic decline of women in these education categories suggesting that the underlying composition of these groups has changed over time. australian population studies 3 (2) 2019 gray & evans 7 graduate or post graduate bachelors diploma or advanced diploma higher certificate basic vocational year 12 year 11 year 10 or below figure 3: percentage distribution of women by highest level of education and birth cohort source: abs 2019 – authors’ calculations graduate or post graduate bachelors diploma or advanced diploma higher certificate basic vocational year 12 year 11 year 10 or below figure 4: completed fertility by highest level of education and birth cohort source: abs 2019 – authors’ calculations 0 5 10 15 20 25 30 52 -5 6 57 -6 1 62 -6 6 67 -7 1 72 -7 6 pe rc en ta ge 52 -5 6 57 -6 1 62 -6 6 67 -7 1 72 -7 6 52 -5 6 57 -6 1 62 -6 6 67 -7 1 72 -7 6 1.4 1.6 1.8 2.0 2.2 2.4 52 -5 6 57 -6 1 62 -6 6 67 -7 1 72 -7 6 co m pl et ed fe rt ili ty 52 -5 6 57 -6 1 62 -6 6 67 -7 1 72 -7 6 52 -5 6 57 -6 1 62 -6 6 67 -7 1 72 -7 6 52 -5 6 57 -6 1 62 -6 6 67 -7 1 72 -7 6 52 -5 6 57 -6 1 62 -6 6 67 -7 1 72 -7 6 52 -5 6 57 -6 1 62 -6 6 67 -7 1 72 -7 6 52 -5 6 57 -6 1 62 -6 6 67 -7 1 72 -7 6 52 -5 6 57 -6 1 62 -6 6 67 -7 1 72 -7 6 52 -5 6 57 -6 1 62 -6 6 67 -7 1 72 -7 6 52 -5 6 57 -6 1 62 -6 6 67 -7 1 72 -7 6 52 -5 6 57 -6 1 62 -6 6 67 -7 1 72 -7 6 52 -5 6 57 -6 1 62 -6 6 67 -7 1 72 -7 6 52 -5 6 57 -6 1 62 -6 6 67 -7 1 72 -7 6 cohort cohort 8 gray & evans australian population studies 3 (2) 2019 the pattern of change in completed total fertility from the 1952-56 cohort to the 1972-76 cohort is shown in figure 5. figure 5 also shows the contribution to completed fertility according to the women’s highest level of education. not only did completed total fertility decline across the cohorts but the contribution by each category also changed. most notably the contribution of completed fertility by women with year 10 or below schooling declined, whereas the contribution to completed fertility by women with a bachelor’s degree increased. the contribution of an education level to completed total fertility can decline if the proportion of women in that category decreases, or their completed fertility rate decreases, or both. in the next section we decompose the change in completed fertility rate to these two factors: compositional change in education categories and actual change in fertility within education categories. figure 5: relative contribution of education categories to overall completed fertility, by cohort source: abs 2019 – authors’ calculations 4.2. decomposing fertility rates and educational composition with this decomposition we can separately identify changes in completed fertility attributable to changes in the educational distribution as well as changes in fertility within each educational group. if, over time, the proportion of women in different educational groups changed but the completed fertility within each group remained the same then changes in total completed fertility would be 100 per cent due to changes in education composition. similarly, if the educational composition stayed the same, but fertility changed within the education groups, then 100 per cent of change in overall fertility could be attributed to fertility rate changes. the decomposition of completed fertility by cohort is shown in table 2. comparing the 1952-56 cohort with the 1957-61 cohort we see a decline of 0.07 children, from 2.23 to 2.16. of this decline, 72 per cent can be attributed to changes in fertility rates and 28 per cent to changes in educational composition. comparing the 1962-66 cohort with the 1957-61 cohort, there is a decline of 0.08 in completed fertility from 2.16 to 2.08. this decline was primarily due to declines in fertility across all education groups, with only 15 per cent attributable to changes in the composition of women across education categories. for the next two cohorts, completed fertility declined by 0.06 and this was 0.00 0.50 1.00 1.50 2.00 1952-1956 1957-61 1962-66 1967-71 1972-76 co m pl et ed fe rt ili ty year 10 or below year 11 year 12 basic vocational higher certificate diploma or advanced diploma bachelors graduate or post graduate australian population studies 3 (2) 2019 gray & evans 9 attributable fairly evenly due to declines in fertility and changes in the composition across education categories. the majority of the decline between the two most recent cohorts is attributable to changes in the composition across education categories. table 2: decomposition of completed fertility change attributable to: completed fertility of cohort completed fertility of previous cohort difference fertility rate effect (%) composition effect (%) 1952-56 2.23 1957-61 2.16 2.23 -0.07 72 28 1962-66 2.08 2.16 -0.08 85 15 1967-71 2.02 2.08 -0.06 53 47 1972-76 1.97 2.02 -0.05 31 70 total (1952-57 to 1972-61) -0.26 56 44 source: abs 2019 – authors’ calculations next we look in more detail at how fertility and education composition contributed to the decline in completed fertility between the cohorts. for each education category, figure 6 shows the effect on fertility of the changing proportion of women in that education category (proportion), and the changing fertility (rate) within each education group. the corresponding table can be found in the appendix. for example, as seen above, between the 1952-57 and 1957-61 cohorts completed cohort fertility declined by 0.07. between these two cohorts the education composition of women changed, and the fertility of women within education categories also changed. for these cohorts, the increase in the proportion of women with a graduate diploma, masters or phd contributed 0.013 to the difference in overall cohort fertility. the decline of fertility among women with a post-graduate degree contributed to -0.006, leading to an overall positive impact of 0.007. if we add up all the positive and negative effects of composition and fertility within each of the education categories it equals -0.07. for the two highest levels of educational attainment, postgraduate and bachelor, between every cohort comparison there was a positive effect on overall completed fertility due to increasing proportion of women with these degrees but a negative effect due to declining fertility in these education categories. however, the negative influence of declining fertility within the education categories was not large enough to offset the positive impact of increasing proportion of women with degrees leading to an overall positive effect on fertility. the increasing proportion of women with a bachelor’s degree had a particularly positive effect on completed fertility for the most recent cohorts, and the offsetting negative effect in the fertility rate among women with a bachelor’s degree was only minor in comparison. as a result, between the 1962-66 and 1967-71 cohort, women with a bachelor’s degree contributed 0.059 to the difference between overall completed fertility. the diploma level had a negligible effect on differences in fertility over time, as did certificate i and iv. for certificate iii and iv initially the increasing composition of women in the oldest cohorts had a positive effect, but in the in later cohorts the effect was minor. 10 gray & evans australian population studies 3 (2) 2019 figure 6: decomposition by cohort and detailed education categories source: abs 2019 – authors’ calculations (full results in appendix) australian population studies 3 (2) 2019 gray & evans 11 for year 12 the increasing composition coupled with very little change in their fertility rate led to an overall positive, though small, contribution to changes in completed fertility. for year 11 an interesting pattern emerges. for the early cohorts there was an increasing proportion of women in year 11 and a small but declining rate of fertility. in the most recent cohorts this switched and the proportion with year 11 declined, but the fertility of women with year 11 education increased. the overall impact on fertility was negative for the most recent cohorts. for year 10 we see the strongest effect of proportion and fertility rate. for each cohort comparison, the declining proportion in this education category had the effect of decreasing completed fertility. among the older cohorts in this group fertility also declined but in the younger cohorts it increased. 5. conclusions for older cohorts of australian women, attaining a degree was a rare occurrence and it was unusual to combine work and family. today the proportion of women with a degree has increased dramatically, and women with a degree are no longer such a selective group. one might therefore expect that the negative gradient between education and fertility would have weakened over time. in australia, we found no evidence of this. instead, as in britain, we find large and widening educational differences. the increasingly large proportion of women with university degrees are still experiencing declining fertility while women with year 11 or below schooling appear to have become an increasingly small and select proportion of the population with increasing fertility rates. these are also women with the lowest investment in human capital and the most precarious attachment to the labour force. in terms of the institutional support for women wanting to combine work and childbearing, australia still lags behind many european and scandinavian countries (heard and arunachalam 2015). the continued barriers to combining motherhood and work for educated women is likely one explanation for why in australia we continue to see educational differences in fertility. in terms of the effect overall, we find that for older cohorts the primary component of change in completed fertility was declining fertility. in the most recent cohort there is evidence that a composition effect due to changing educational trends is now the primary driver of declining fertility. in particular, the impact of the growing group of women with a bachelor’s degree and the shrinking proportion of women who have not completed high school reflects the growth in higher education enrolment shown in figure 2. using detailed education categories also enables us to capture the differences between high and low education. we find large differences in the pattern of fertility between the year 12, year 11 and year 10 or below-groups which are often grouped together in other studies as ‘no post school qualification’. while year 12 followed the same composition and rate trend as the bachelor’s level (increasing composition, declining fertility), the incomplete secondary groups were very different signalling an increasing marginalisation of women with very low levels of education. it is important to note that we use highest completed level of education at the time of the census as our indicator of educational attainment. this anticipatory approach could possibly lead to a bias in the estimates of the educational gradient as the attainment of the highest level of education may change across the reproductive life course and may have occurred after childbearing (hoem and 12 gray & evans australian population studies 3 (2) 2019 kreyenfield 2006). in addition, because we include women up to the age of 64 the results presented may also be biased by differential mortality to the extent that mortality varies by education (hoem and kreyenfield 2006). australia has one of the highest levels of tertiary education completion in the world. nearly 50 per cent of young adults (25-34 year-olds) in australia have attained a tertiary qualification, one of the highest across oecd countries and considerably above the oecd average of 43 per cent. given the significant increase in education, and the continued rising levels of education for women, future national fertility levels will in large part be determined by the childbearing of the increasingly large proportion of women with higher education (heard and arunchalam 2015). it is difficult to predict whether these differences will continue to diverge or whether they will follow european trends and begin to narrow. this will largely depend on the childbearing patterns of the increasingly select group who do not complete high-school, and the fertility of those with university degrees. both can be significantly influenced by supportive family policy. key messages • australia has seen a widespread increase in the level of education of women, and a corresponding decrease in completed fertility. • decomposition analysis allows a separation of the components of fertility decline into two parts: decline due to a change in the composition of education categories, and decline due to changes in fertility rates of cohorts. • for older cohorts, most of the decline is due to declining fertility rates. for the youngest cohort, the decline is due to changing education composition. • focussing on the different education levels provides additional insight into fertility change by level of education. incomplete secondary schooling has a declining effect on completed fertility while tertiary education has a substantial and growing contribution on the change in completed fertility. • overall, the gap between completed fertility by highest level of education is growing. acknowledgements the 2016 australian census data for this paper were made available by the australian bureau of statistics through their tablebuilder product. this research is funded by the australian research council (arc) (dp150104248). we would like to gratefully acknowledge the helpful comments provided by two referees and to thank anna reimondos for research assistance. references adsera a (2017) the future fertility of highly educated women: the role of educational composition shifts and labor market barriers. vienna yearbook of population research 15: 19-25. anderson d s and vernon a e (1983) access to privilege: patterns of participation in australian postsecondary education. canberra: australian national university press. australian population studies 3 (2) 2019 gray & evans 13 andersson g, ronsen m, knudsen l b, lappegård t, neyer g, skrede k, teschner k and vikat a (2009) cohort fertility patterns in the nordic countries. demographic research 20(14): 313-352. australian bureau of statistics (1988) census 86 cross-classified characteristics of persons and dwellings. catalogue number 2102.0. canberra: abs. australian bureau of statistics (1993) women in australia. catalogue no 4113.0. canberra: abs. australian bureau of statistics (2001) year book australia 2001. catalogue no 1301.0. canberra: abs. australian bureau of statistics (2009) schools, australia, 2008. catalogue no 4221.0. canberra: abs. australian bureau of statistics (2018) births, australia, 2017. catalogue no 3301.0. canberra: abs. australian bureau of statistics (2019) 2016 census counting persons, place of usual residence, table builder pro. https://www.abs.gov.au/websitedbs/censushome.nsf/home/tablebuilder. bavel j, klesment m, beaujouan é, brozozowska z, purr a, reher d, requena m, sandström g, sobotka t and zeman k (2018) seeding the gender revolution: women's education and cohort fertility among the baby boom generations. population studies 72(3): 283-304. berrington a, stone j and beaujouan é (2015) educational differences in timing and quantum of childbearing in britain: a study of cohorts born 1940-1960. demographic research 33(26): 733764. brzozowska z (2014) fertility and education in poland during state socialism. demographic research 31(12): 319-336. burke g and spaull a (2001) australian schools: participation and funding 1901-2000. in abs (eds.) year book australia, 2001. catalogue no. 1301.0. canberra: abs; 433-446. caldwell j c (1980) mass education as a determinant of the timing of fertility decline. population and development review 6(2): 225-255. campbell c and proctor h (2014) a history of australian schooling. allen and unwin. chapman b (2011) the australian university student financing system: the rationale for, and experience with, income-contingent loans. in: armstrong s and chapman b (eds.) financing higher education and economic development in east asia. canberra: australian national university epress; 83-105. cochrane s h (1979) fertility and education: what do we really know? baltimore: johns hopkins university press. das gupta p (1993) standardization and decomposition of rates: a user's manual. washington dc: us bureau of the census. department of education, training and youth affairs (2001) higher education students time series tables selected higher education statistics 2000. canberra. fort m, schneeweis n and winter-ebmer r (2016) is education always reducing fertility? evidence from compulsory schooling reforms. the economic journal 126: 1823-1855. heard g and arunachalam, d (2015) fertility differentials. in: heard g and arunachalam d (eds) family formation in 21st century australia. dordrecht: springer; 159-196. heck k e, schoendorf k c, ventura s j and kiely j l (1997) delayed childbearing by education level in the united states, 1969–1994. maternal and child health journal 1(2): 81-88. hoem j m and kreyenfeld m (2006) anticipatory analysis and its alternatives in life-course research. part 1: the role of education in the study of first childbearing. demographic research 15(16): 461-484. jalovaara m, neyer g, andersson g, dahlberg j, dommermuth l, fallesen p and lappegård t (2018) education, gender, and cohort fertility in the nordic countries. european journal of population 35(3): 563-586. kravdal o and rindfuss r r (2008) changing relationships between education and fertility: a study of women and men born 1940 to 1964. american sociological review 73(5): 854-873. le a t and miller p w (2002) the rising education levels of females in australia. education economics 10(1): 1-24. https://www.abs.gov.au/websitedbs/censushome.nsf/home/tablebuilder 14 gray & evans australian population studies 3 (2) 2019 li j (2017) rate decomposition for aggregate data using das gupta's method. the stata journal 17(1): 113. liefbroer a c and corijn m (1999) who, what, where and when? specifying the impact of educational attainment and labour force participation on family formation. european journal of population 15(1): 45-75. long m, carpenter p and hayden m (1999) participation in education and training. camberwell: australian council for educational research. merz e-m and liefbroer a c (2017) cross-national differences in the association between educational attainment and completed fertility. do welfare regimes matter? vienna yearbook of population research 15: 95-120. musick k, england p, edgington s and kangas n (2009) education differences in intended and unintended fertility. social forces 88(2): 543-572. ní bhrolcháin m and beaujouan é (2012) fertility postponement is largely due to rising educational enrolment. population studies 66(3): 311-327. neels k, murphy m, ni brochlain m and beaujouan, é (2017) rising educational participation and the trend to later childbearing. population and development review 43(4): 667-639. rindfuss r r, morgan s p and offutt k (1996) education and the changing age pattern of american fertility: 1963-1989. demography 33(3): 277-290. sobotka t, beaujouan, é and van bavel j (2017) introduction: education and fertility in low-fertility settings. vienna yearbook of population research 15: 1-16. yoo s h (2014) educational differences in cohort fertility during the fertility transition in south korea. demographic research 30(53): 1463-1494. australian population studies 3 (2) 2019 gray & evans 15 appendix table a: detailed decomposition of change in completed fertility due to composition and due to fertility rate, by highest level of education cohort 1952-56 to 1957-61 1957-61 to 1962-66 1962-66 to 1967-71 1967-71 to 1972-76 graduate diploma, masters or phd composition effect 0.013 0.008 0.02 0.022 fertility rate effect -0.006 -0.006 -0.002 -0.005 total effect 0.007 0.002 0.018 0.017 bachelor composition effect 0.028 0.021 0.064 0.089 fertility rate effect -0.008 -0.012 -0.005 -0.01 total effect 0.02 0.009 0.059 0.079 diploma or advanced diploma composition effect 0.025 0.011 0.016 0.001 fertility rate effect -0.005 -0.011 -0.009 -0.009 total effect 0.02 0.00 0.007 -0.008 certificate iii or iv composition effect 0.054 0.036 0.008 -0.002 fertility rate effect -0.004 -0.009 -0.009 -0.008 total effect 0.05 0.027 -0.001 -0.01 certificate i or ii composition effect 0.005 0.001 -0.008 0.00 fertility rate effect -0.003 -0.003 -0.002 0.00 total effect 0.002 -0.002 -0.01 0.00 year 12 composition effect 0.018 0.012 0.023 0.027 fertility rate effect -0.004 -0.004 -0.005 -0.001 total effect 0.014 0.008 0.018 0.026 year 11 or equivalent composition effect 0.01 0.009 -0.02 -0.039 fertility rate effect -0.002 -0.002 0.001 0.005 total effect 0.008 0.007 -0.019 -0.034 year 10 or below composition effect -0.147 -0.091 -0.115 -0.126 fertility rate effect -0.011 -0.014 0.001 0.011 total effect -0.158 -0.105 -0.114 -0.115 not stated composition effect -0.026 -0.019 -0.014 -0.01 fertility rate effect -0.006 -0.007 -0.002 0.00 total effect -0.032 -0.026 -0.016 -0.01 column total -0.07 -0.08 -0.06 -0.05 abstract background aims data and methods the study uses 2016 census data on the number of children ever born of five cohorts of women born between 1952 and 1976. decomposition is used to distinguish the effects of the two components. results the educational composition of women in these cohorts is dramatically different, with an increasing number of women having completed tertiary education in later cohorts. completed fertility has also changed across successive cohorts. we find that for ... conclusions despite tertiary education becoming much more common, fertility within this group remains lower than other education groups. while other countries have seen a narrowing of the gap in fertility rates between education groups, this pattern is not found ... key words education; fertility; fertility decline; children ever born; census; decomposition. 3. data and methods key messages acknowledgements references australian population studies 2017 | volume 1 | issue 1 | pages 73–85 © lomax and smith. published under the creative commons attribution-noncommercial licence 3.0 australia (cc by-nc 3.0 au). journal website: www.australianpopulationstudies.org an introduction to microsimulation for demography nik lomax* university of leeds andrew p. smith university of leeds *corresponding author. email: n.m.lomax@leeds.ac.uk. address: school of geography and leeds institute for data analytics, university of leeds, leeds ls2 9jt, united kingdom paper received 18 august 2017; accepted 18 october 2017; published 20 november 2017 abstract background microsimulation consists of a set of techniques for estimating characteristics and modelling change in populations of individuals. aims to demonstrate how microsimulation can be used by demographers who want to undertake population estimates and projections. data and methods we use data from the 2011 united kingdom (uk) census of population to create a synthetic population by age, sex and ethnic group. static and dynamic microsimulations and the visualisation of results are undertaken using the statistical package r. the code and data used in the static and dynamic microsimulation are available via a github repository. results a synthetic population in 2011 by age, sex and ethnicity was produced for the east london borough of tower hamlets, estimated from two census tables. a population projection was produced for each of these age, sex and ethnicity groups to 2021. we used a projection of the bangladeshi population to visualise population growth by middle-layer super output area (msoa) and to produce a population pyramid of estimates in 2021. conclusions we argue that microsimulation is an adaptable technique which is well suited to demography, for both population estimation and projection. although our example is applied to the east london borough of tower hamlets, the approach could be readily applied in australia, or any other country. key words microsimulation; static; dynamic; r code; demography; data visualisation; population estimation. http://www.australianpopulationstudies.org/ mailto:n.m.lomax@leeds.ac.uk 74 lomax n and smith a p australian population studies 1 (1) 2017 1. introduction this paper outlines how microsimulation can be of use to demographers as a method for the estimation and projection of populations. we outline what microsimulation is and how it has been used in a range of applications, from broader social science to demography specific examples. we provide two examples of how to apply microsimulation (with accompanying r code), firstly creating a synthetic population, and then projecting this population forward in time. while the examples presented in this paper relate to the united kingdom (uk), the methods could be readily applied to australia. we offer some examples of how the data can be visualised and analysed and a discussion of how the examples could be further extended. 2. what is microsimulation? microsimulation is an approach used to estimate the characteristics of a population from a range of attribute-rich data sources. micro refers to individual units (people, households, etc.) while simulation refers to the process of assigning attributes for those units. microsimulation can, for example, be used to allocate characteristics from a sample survey dataset to a larger population enumerated in a census. microsimulation can be used to create static or dynamic models. a static model is a way of synthesising data to produce an entire population estimate of individuals. dynamic microsimulation introduces time into the model and the ability to ‘age’ the static population. it is dynamic microsimulation which is used in population projections and to perform ‘what if’ scenario testing of policy or market interventions. the technique is often referred to as spatial microsimulation. essentially, the spatial element is concerned with allocating some form of geography to your individual or unit. we provide examples of static and dynamic microsimulation later in the paper. both are inherently spatial because they estimate people within specific areas. microsimulation has a wide range of utilities across a number of disciplines. for example, microsimulation has been used to model future elderly health care demand (clark et al. 2017), project educational attainment (nelissen 1991), commuting patterns (lovelace, ballas and watson 2014) and population projections (ballas et al. 2005). many of the applications of microsimulation we see in practice today stem from the work of orcutt (1957) and orcutt et al. (1961). spielauer (2011) provides a good overview of microsimulation applications for the social sciences (including the differences between static and dynamic models); li and o’donoghue (2013) provide a comprehensive discussion of dynamic microsimulation models implemented across a range of settings. for a comprehensive overview of spatial microsimulation see birkin and clarke (2011). ballas et al. (2005, p. 15) contend that microsimulation fundamentally is not that complicated an approach, noting that ‘even calculating simple social statistics such as life expectancy is a method that is similar to microsimulation’. when it comes to more complicated dynamic applications, bélanger and sabourin (2017) observe that microsimulation is an approach which has become increasingly popular with improved computing power and better availability of survey microdata. australian population studies 1 (1) 2017 lomax n and smith a p 75 3. why is microsimulation a useful technique for demographers? as a technique, demographers can make use of microsimulation to estimate and project populations and their attributes. in an age where more information is becoming available from disparate sources (e.g. censuses, surveys, administrative or commercial data), there is a need to synthesise this information if we wish to estimate populations with a range of attributes. the need for this attributerich information stems from the fact that we can produce more accurate models and make better recommendations if we know more about each individual. bélanger and sabourin (2017, p. xix) argue that micro models are superior to macro models because ‘the most powerful theoretical models for explaining human behaviour, such as decisions to have a child or migrate, operate at the level of the individual’. for a balanced assessment of microsimulation vis-a-vis macrosimulation models for population projection see van imhoff and post (1998, p. 134) who conclude that, while there are limitations (largely related to difficulty in implementation) ‘microsimulation can do certain things that macrosimulation cannot’. one of the key advantages they identify is the ability of microsimulation to perform in large state space – ‘the representation of the components of the system of interest’ (van imhoff and post 1998, p. 102) – where a large number of individual attributes would not be handled well by a macro model. wilson and rees (2005), in a review of population projection methods, conclude that microsimulation should be considered when projecting populations with multiple attributes. in the demographic literature, microsimulation has been used for a range of applications. for example, it was used by ruggles (1992) to correct missing data bias in historical demography, and by thomson et al. (2012) to estimate completed fertility in france when union formation and dissolution were varied in the simulation. models have been implemented in population projections: see, for example, smile (developed for ireland) (ballas et al. 2005) and moses (developed for the uk) (wu and birkin 2012). microsimulation models are also used to test the implications of policy change. ballas and clarke (2001) model the local impact of changes to national social policy in the uk; brown and harding (2002) provide an overview of a selection of microsimulation models examining policy decisions in australia. microsimulation is a technique which is useful outside the constraints of academia. increasingly microsimulation is being adopted to undertake official projections by national statistics agencies. mosart is used by statistics norway to project education, labour force and public pension benefits (fredriksen and stolen 2007); statistics canada’s projection model demosim uses microdata from the 2011 national household survey to produce estimates of future ethnocultural composition, aboriginal populations and labour force (bélanger and sabourin 2017); pensim is used by the uk department for work and pensions to estimate the future distribution of pensioner incomes (emmerson, reed and shephard. 2004); and sesim is used by the swedish ministry of finance for similar purposes (sundberg 2007). notwithstanding their utility, there are challenges and limitations associated with implementing microsimulation models which must be considered. first, a certain amount of technical expertise is required, although we explain in the next section that there are a range of options for implementing microsimulation models which are suitable for different levels of expertise. second, microsimulation can be computationally intensive. this was a key problem for early implementations of micro models; 76 lomax n and smith a p australian population studies 1 (1) 2017 however, advances in computing means that this is less of a limitation now than it once was. spielauer (2011, p. 18) outlines a third, more conceptual criticism: ‘randomness caused by accumulated errors and biases’. because microsimulation draws from a range of data sources, there is the potential for errors or biases in those data being extrapolated because of the level of detail required (or expected) from the microsimulation model. this randomness, similarly identified by van imhoff and post (1998) and wilson and rees (2005), is more acute in dynamic models because they are stochastic: results of a projection, for example, are randomly drawn from an expected distribution. brown and harding (2002, p. 1) highlight that microsimulation models are limited by design, assumptions, algorithms and data requirements but the solution is to ‘make these explicit and then interpret the results within the models' limitations and capacities’. this is advice which should be applied to any model, be it micro or macro in design. 4. how can microsimulation be applied? there are a multitude of ways to undertake microsimulation, from freely available custom software packages to writing your own model in a development language. bespoke packages are probably the easiest place to start. one excellent example is a program called the flexible modelling framework (fmf), created specifically to perform spatial microsimulation (harland 2013). the fmf has a graphical user interface (gui) which allows the user to select input files and specify required options and outputs, circumnavigating any requirement for the user to write their own code. the model is, however, only capable of static microsimulation. in the programming language r, there are several packages which have been developed to undertake microsimulation. the micsim package is one example which provides the required functionality to undertake microsimulation. for a comprehensive overview of how to use r to undertake spatial microsimulation, see lovelace and dumont (2016). an interim step between writing a program from scratch and an off-the-shelf gui is offered by statistics canada (bélanger and sabourin 2017). their framework, called modgen (model generator), implements an extension of the c++ language to help researchers build a model to their own specification, but does require some development skills to implement properly. for the examples presented in this paper we have chosen to write the code in r. this is because it offers the flexibility to incorporate data and functions of our choosing. also, because the source code is freely available, it offers researchers the option of adapting the code to their specific needs. a slightly longer initial outlay in terms of time and learning yields a more flexible and adaptable approach. 5. two examples: static and dynamic microsimulation we have produced an r package to demonstrate the techniques of both static and dynamic microsimulation with some example visualisations of the results. this package and associated documentation can be accessed via a github repository (see https://github.com/virgesmith/demographymicrosim). in this section, we give a description of the data and methodology, and provide some example illustrations from the results. detailed instructions on how to install and use the package are given in the readme.md file at the github link. the r package is self-contained in that it contains all the code and input data needed for the microsimulation. https://github.com/virgesmith/demographymicrosim australian population studies 1 (1) 2017 lomax n and smith a p 77 in the first example we use static microsimulation to create a base human population for the east london borough of tower hamlets from 2011 uk census data, estimating the population by geographical location, gender, age and ethnicity. in the second example we take this synthetic population and undertake a projection with dynamic microsimulation using detailed ethnicityspecific fertility and mortality data. we demonstrate ways to visualise the estimates using a map and population pyramid. 5.1 case study area the east london borough of tower hamlets makes an interesting case study area because it is one of the most diverse areas in the uk, both in terms of its ethnic composition and its socioeconomic structure. this part of london has seen successive waves of immigration and onward migration for centuries, including huguenots, jews, irish, bangladeshis and, more recently, migrants from eastern europe, somalia, eritrea and yemen. the borough is also a place of economic extremes: areas of poverty such as shadwell, whitechapel and poplar contrast with the financial district of canary wharf and the high-end ‘docklands’ riverside housing developments which have replaced the old working docks. our estimates are produced at middle-layer super output area (msoa) spatial resolution. msoas are a census geography where zones are aggregations of output areas, the smallest geography at which census data are released, and designed to provide consistent population counts across the country. msoas contain an average of just under 8,000 people. tower hamlets is split into 32 msoas and the borough’s total population is recorded as just over 250,000. the method outlined here could be adapted and applied to any other geography in the uk or elsewhere, so long as the data are available. the australian equivalent to these hierarchical census geographies are statistical areas (sas). sa2s have an average population of about 10,000 so are the nearest equivalent to the uk’s msoa geography. 5.2 input data the input data consist of two datasets, which are used for the static and dynamic microsimulation. we also provide geographical boundaries for the visualisation of results. the specific data used in our example are described in more detail in the following sections. however, in general terms, to undertake a static microsimulation as described in this paper the user will need one or more datasets containing full enumeration of the population of interest (e.g. from a census), disaggregated by some attributes. subsequent tables don’t need to provide full enumeration and some attributes need to match if drawing from a number of tables. to undertake dynamic microsimulation the user will additionally need demographic rates for the population of interest (e.g. from vital registration data). these need not be at the same spatial scale, as shown in our example. to replicate the visualisation of results, our code can be adapted to deal with any spatial system if shapefiles are available. 5.2.1 aggregate population data we use data from the uk 2011 census of population to create the microsimulated synthetic population. the 2011 census provides counts of people by area (in this case msoa), cross-tabulated by a number of characteristics. census data used here are sourced from nomis official labour market statistics (https://www.nomisweb.co.uk/) and contain public sector information licensed under the open government licence v3.0. https://www.nomisweb.co.uk/ 78 lomax n and smith a p australian population studies 1 (1) 2017 we have used the following tables: • dc2101ew: ethnic group by sex by age. after processing we have named this file sexageeth.csv in the github repository. it provides a count of persons by msoa by sex by age band by ethnicity. • dc1117ew: sex by single year of age. after processing we have named this file sexageyear.csv. it provides a count of persons by msoa by sex by single year of age. the data tables contain the following fields, where ‘persons’ is the count of each cross tabulation: • msoa: identification code for the 32 msoas within tower hamlets • sex: m (male) and f (female) • age band: 0-4, 5-7, 8-9, 10-14, 15, 16-17, 18-19, 20-24, 25-29, 30-34, 35-39, 40-44, 45-49, 50-54, 55-59, 60-64, 65-69, 70-74, 75-79, 80-84, 85+ • age: individual years up to 84, then a single 85+ category • ethnicity: ban (bangladeshi), bla (black african), blc (black caribbean), chi (chinese), ind (indian), mix (mixed), oas (other asian), obl (other black), oth (other), pak (pakistani), wbi (white british), who (white other). the categories for ethnicity have been aggregated from the original census categories in order to be consistent with the fertility and mortality data discussed below. figure 1 provides an extract from the two constraint tables. we can see that both tables contain useful information but are not consistent with each other. a b figure 1: extracts from (a) table sexageeth.csv and (b) table sexageyear.csv. 5.2.2 fertility and mortality rate data there is substantial variation in the rates of mortality and fertility for different ethnic groups (rees et al. 2009; norman et al. 2014), and it is important that our microsimulation captures this. we use ethnic-specific fertility and mortality rates produced as part of the newethpop project. newethpop has produced population projections for uk local authorities, disaggregated by age, sex and ethnic group, that capture the variation in demographic components which is mostly missed by other (nonethnic disaggregated) projections. for more information on the project see rees et al. (2017). projection results can be downloaded from www.ethpop.org. https://github.com/virgesmith/demographymicrosim/blob/master/data/sexageeth.csv https://github.com/virgesmith/demographymicrosim/blob/master/data/sexageyear.csv http://www.ethpop.org/ australian population studies 1 (1) 2017 lomax n and smith a p 79 rates by ethnicity and single year of age are available for the entire borough of tower hamlets. these rates are not differentiated at any smaller geographical scale. we use this information in the microsimulated population projection. note that while ethnic-specific migration data are used in the newethpop project, we do not use this in our paper. our aim is to keep the examples we present simple and reproducible. we provide the following files, which provide the ethnic-specific information, in the github repository: • towerhamletsfertility.csv, which is the ethnic and single year of age specific fertility rate for tower hamlets. • towerhamletsmortality.csv, which is the ethnic and single year of age specific mortality rate for tower hamlets. 5.2.3 geographical boundaries the code to map the results for the tower hamlets projection is also provided at github. this includes msoa boundaries made available by the office for national statistics under the open government licence v3.0. background map data are sourced from carto through openstreetmap under the open data commons open database license. 5.3 methodology in the following examples, microsimulation is necessary because we have information from different data sources which are not entirely consistent. while we have counts of people by single year of age by sex in one census table, the ethnic disaggregated counts from the other census table are only available for grouped age bands. the information on fertility and mortality rates is available for single year of age and sex, but is not geographically disaggregated below borough level. the aim here is to use static microsimulation to produce a population of individuals by sex, ethnicity, single year of age and msoa. this base population is then aged on using the ethnic-specific fertility and mortality rates in the dynamic microsimulation. 5.3.1 static microsimulation using microsimulation, we can generate a synthetic population that enumerates both ethnicity and single year of age for every individual. the sum of this population is entirely consistent with the input data: it will add up to both the ethnicity totals (by age band) and the population totals (by single year of age) for each geographical area and sex. for the microsimulation we use the humanleague r package that generates a population using quasirandom sampling of the marginal data (smith et al. 2017). iterative proportional fitting (ipf) achieves a similar goal, minimising statistical significance at the expense of not always producing a whole-number population in each (age, ethnicity) state, and thus requires some form of adjustment. see lomax and norman (2016) for a step-by-step guide to ipf. in our example, in any given area, for each gender and age band, we know the total persons of each ethnicity, and the total persons of each (single year of) age. we fill a table with people in such a way that: (1) there is a whole-number population for each age and ethnicity; (2) the totals for each age and ethnicity are correct; and (3) the population has a low statistical significance, i.e. age and ethnicity are not correlated. https://github.com/virgesmith/demographymicrosim/blob/master/data/towerhamletsfertility.csv https://github.com/virgesmith/demographymicrosim/blob/master/data/towerhamletsmortality.csv 80 lomax n and smith a p australian population studies 1 (1) 2017 the algorithm has three broad functional components: 1. load the input data and compute various data that will be required later (such as the categories, and a match between age band and age). 2. perform microsimulation for each geographical area and insert into the population. 3. perform checks on the synthesised population to ensure it is consistent with the input data. the result is a population of 254,096 individuals who have some combination of age, sex, ethnicity and msoa location. this is the base population which will be used for the dynamic projection. 5.3.2 dynamic microsimulation in this example we project the base population from 2011 to 2021. the projection uses a montecarlo simulation that assign births and deaths to the population based on the ageand ethnicityspecific fertility and mortality rates for the borough of tower hamlets. this type of methodology is described in van imhoff and post (1998). in this example, for each iteration (i.e. year) we draw two uniform independent random variates for each eligible person. we compare these values to the fertility and mortality rates for the appropriate age and ethnicity: if the first random variate is lower than the fertility rate, a birth is assigned to this person; if the second is lower, a death is assigned. only females have nonzero fertility rates, which are strongly age dependent. if the fertility rate is 0.1, this will result in on average one in ten of the population drawing a random lower number and having a birth assigned to them. for a mortality rate of 0.01, only 1 per cent of the population will on average draw a number lower than this. the simulation is discrete in that it operates in one-year intervals, and the user specifies the number of years to run the projection. in this relatively simple example, our population is 254,096 persons. each person exists in one of 66,048 possible states (32 [msoas] × 86 [ages] × 2 [genders] × 12 [ethnicities]). in a macrosimulation we would need to keep track of the number of people in each state (i.e. 66,048 values). however, in a microsimulation, we need to keep track of individuals and the states that they occupy. this requires storing about 1,000,000 values, since there are four separate categories. as more categories are added the number of states grows exponentially, while the storage required for individuals only grows linearly. it would only take the addition of one or two more categories before the number of states would exceed the population, and microsimulation would be a more efficient approach in terms of storage. note that the following assumptions are made in the model, which are reflected in the source code in the ‘microsimulate’ function (see lines 71–82 in microsimulation.r): • only single births occur (i.e. we assume that multiple births are factored into the fertility rate) • newborns have an equally probable chance of being male or female • the ethnicity and msoa of the newborn is the same as their mother’s • births occur before deaths – thus a newborn will survive if a parent dies within the same year • migration is not taken into account. see the final section of this paper for discussion about developing the example further. australian population studies 1 (1) 2017 lomax n and smith a p 81 the algorithm can be described as follows: 1. load the ethnicity-specific fertility and mortality rates. 2. randomly assign births and deaths to members of the population in a manner that is consistent with the fertility and mortality rates. 3. age the population by one year. 4. add newborns (aged zero) and remove the deceased from the population. 5. repeat from step 2 until the target year is reached. 5.4 visualisation figure 2: map of projected percentage population growth, 2011–2021 source: map tiles by carto under cc by 3.0. data by openstreetmap under odbl. note: low growth is blue and higher growth orange. to interpret the results, it is important to be able to visualise them effectively. the package provides functionality to calculate summary measures for visualisation. firstly growth (𝑔𝑇), 𝑔𝑇 = 𝑃𝑇 𝑃0 − 1 where 𝑃0 is the initial population and 𝑃𝑇 the final one. the value will be negative if the population shrinks, zero if unchanged, and positive if the population increases. the value is not annualised. secondly diversity, 𝑑 = 1 − 𝑛. var(𝐩) 82 lomax n and smith a p australian population studies 1 (1) 2017 where 𝑛 is the number of categories (i.e. ethnicities), 𝐩 a vector of populations in each category, and var is the sample variance. while numerous measures for diversity are used to explain population distributions (e.g. rees and butt 2004 use the diversity index), this measure has the useful properties that the value is: zero when the population is all in one category; one (1) when the populations in each category are equal; and essentially independent of the number of categories. we include code to produce maps of these data in the github repository. figure 2 (previous page) provides an example of total population growth between 2011 and 2021. figure 3: 2021 projected bangladeshi population pyramid, 2021 source: authors’ projections. we also provide a function to produce population pyramids of subsets of the base or projected population. figure 3 shows one of these population pyramids for the age and sex structure of the bangladeshi population in 2021. the code can be adapted to produce figures for other ethnic groups. 6. adaption and extension of the example the examples outlined in the previous section provide the building blocks needed for researchers to implement both static and dynamic microsimulation. in order to use the example in different contexts, researchers will want to adapt the data inputs to take in to account their own research interests. for example, fertility and mortality rates for different groups (e.g. stratified by socioeconomic status or health status) could replace ethnicity, and different geographic systems could replace the msoas used here. australian population studies 1 (1) 2017 lomax n and smith a p 83 there are also improvements which would make the microsimulation models more sophisticated. the first obvious extension is to include migration (both internal and international) in the model. this could be done by incorporating migration rates in a similar way to those used for fertility and mortality, but it is important to note that the model complexity and run time would be increased. for an overview of how complex and multifaceted a projection with dynamic microsimulation can be, see the work of holm et al. (2008) who describe the sverige model which not only incorporates migration, but also education, marriage, leaving home, divorce, employment and earnings. an individual simulated population cannot be considered an accurate prediction of the future. it is essentially an extrapolation that is subject to biases and uncertainties in the input data, the model assumptions and the random noise that is inherent in any monte-carlo simulation, which may amplify the initial biases. the second improvement, which would help to negate this problem, would be to perform a number of iterations of the model run and from these compute not a single value but a confidence interval for aspects of the final population. this would go some way towards alleviating the criticisms about randomness discussed in section 3. multiple runs of the model would give the researcher an idea about the variability of results, providing some indication of uncertainty which could be attached to projection estimates. this significantly increases the amount of computation required, but there are various statistical techniques available (outside the scope of this paper) which are more efficient than a brute-force approach. a third improvement is to test the sensitivity of the model to the input data or model assumptions. if there is a suspicion that some value or values in the input data are inaccurate or biased, perturbing these values (i.e. changing them by a small amount) and re-running the microsimulation can establish the sensitivity of the model to these inputs. if the projected population (or some aspect of it) is very similar the model can be considered insensitive to the input and the input’s accuracy may not be a major issue. conversely, if the projected population differs significantly the model is highly sensitive to the input, and perhaps the results cannot be accepted with much confidence. a final important consideration is the validation of model results. limitations in the amount of space available negate a discussion of model validation in this paper but for a good summary see ballas et al. (2005). 7. conclusion this paper has provided an overview of microsimulation as a technique which should be considered by demographers who are interested in estimating individuals within a population (through static models) and undertaking projections of those individuals (through dynamic models). we have argued that microsimulation is a technique which is useful for demographers and one which is becoming increasingly used as data, computing power and user support increases. our example is applied to the east london borough of tower hamlets but the approach could be readily applied in australia, or any other country for that matter, so long as appropriate data are available. useful next steps would be to look at the full r code provided for the examples outlined in this paper and have a go at implementing the model. the code can be adapted to suit the researchers’ requirements. 84 lomax n and smith a p australian population studies 1 (1) 2017 references ballas d and clarke g (2001) modelling the local impacts of national social policies: a spatial microsimulation approach. environment and planning c: government and policy 19(4): 587–606. ballas d, clarke g, dorling d, eyre h, thomas b and rossiter d (2005) simbritain: a spatial microsimulation approach to population dynamics. population, space and place 11(1): 13–34. bélanger a and sabourin p (2017) microsimulation and population dynamics: an introduction to modgen 12. the springer series on demographic methods and population analysis 43. switzerland: springer. birkin m and clarke m (2011) spatial microsimulation models: a review and a glimpse into the future. in: stillwell j and clarke m (eds) population dynamics and projection methods. netherlands: springer; 193–208. brown l and harding a (2002) social modelling and public policy: application of microsimulation modelling in australia. journal of artificial societies and social simulation 5(4), http://jasss.soc.surrey.ac.uk/5/4/6.html. clark s, birkin m, heppenstall a and rees p (2017) using 2011 census data to estimate future elderly heath care demand. in: stillwell j and duke-williams o (eds) the routledge handbook of census resources, methods and applications: unlocking the uk 2011 census. london: routledge; 305–319. emmerson c, reed h and shephard a (2004) an assessment of pensim2. london: ifs working papers, institute for fiscal studies (ifs) no. 04/21, doi: 10.1920/wp.ifs.2004.0421. fredriksen d and stolen n (2007) model 1: mosart (dynamic cross-sectional microsimulation model). in: gupta a and harding a (eds), modelling our future: population ageing, health and aged care. bingley: emerald group publishing limited; 433–437. harland k (2013) microsimulation model user guide: flexible modelling framework. uk: national centre for research methods (ncrm), ncrm working paper no. 06/13. holm e, holme k, lindgren u and mäkilä k. (2008) the sverige spatial microsimulation model. sweden: department of social and economic geography, umeå university, discussion paper no. 595. li j and o’donoghue c (2013) a survey of dynamic microsimulation models: uses, model structure and methodology. international journal of microsimulation 6(2): 3–55. lomax n and norman p (2016) estimating population attribute values in a table: “get me started in” iterative proportional fitting. the professional geographer 68(3): 451–461. lovelace r, ballas d and watson m (2014) a spatial microsimulation approach for the analysis of commuter patterns: from individual to regional levels. journal of transport geography 34: 282–296. lovelace r and dumont m (2016) spatial microsimulation with r. florida: chapman and hall, crc press. nelissen j (1991) household and education projections by means of a microsimulation model. economic modelling 8(4): 480–511. norman p rees p and wohland p (2014) the use of a new indirect method to estimate ethnic-group fertility rates for subnational projections for england. population studies 68(1): 43–64. orcutt g (1957) a new type of socio-economic system. review of economics and statistics 39(2): 116–123. reproduced in international journal of microsimulation 2007 1(1): 3–9. orcutt g, greenberger m, korbel j and rivlin a (1961) microanalysis of socioeconomic systems: a simulation study. new york: harper and row. rees p and butt f (2004) ethnic change and diversity in england, 1981–2001. area 36(2): 174–186. rees p, wohland p and norman p (2009) the estimation of mortality for ethnic groups at local scale within the united kingdom. social science and medicine 69(11): 1592–1607. rees p, wohland p, norman p and clark s (2017) population projections by ethnicity: challenges and solutions for the united kingdom. in: swanson d (ed.) the frontiers of applied demography. switzerland: springer; 384–408. http://jasss.soc.surrey.ac.uk/5/4/6.html http://dx.doi.org/10.1920/wp.ifs.2004.0421 australian population studies 1 (1) 2017 lomax n and smith a p 85 ruggles s (1992) migration, marriage, and mortality: correcting sources of bias in english family reconstitutions. population studies 46(3): 507–522. smith a, lovelace r and birkin m (2017) population synthesis with quasirandom integer sampling. journal of artificial societies and social simulation 20(4): 14, doi: 10.18564/jasss.3550. spielauer m (2011) what is social science microsimulation? social science computer review 29(1): 9–20. sundberg o (2007) model 5: sesim (longitudinal dynamic microsimulation model). in: gupta a and harding a (eds), modelling our future: population ageing, health and aged care. bingley: emerald group publishing limited; 453–460. thomson e, winkler-dworak m, spielauer m and prskawetz a (2012) union instability as an engine of fertility? a microsimulation model for france. demography 49(1): 175–195. van imhoff e and post w (1998) microsimulation methods for population projection. population: an english selection 10(1): 97–138. wilson t and rees p (2005) recent developments in population projection methodology: a review. population, space and place 11(5): 337–360. wu b and birkin m (2012) agent-based extensions to a spatial microsimulation model of demographic change. in: heppenstall a, crooks t, see l and batty m (eds) agent-based models of geographical systems. netherlands: springer; 347–360. australian population studies 2018 | volume 2 | issue 2 | pages 33-58 © dennett 2018. published under the creative commons attribution-noncommercial licence 3.0 australia (cc by-nc 3.0 au). journal website: www.australianpopulationstudies.org an introduction to modelling population flows using spatial interaction models adam dennett* university college london * corresponding author. email: a.dennett@ucl.ac.uk. address: bartlett centre for advanced spatial analysis, university college london, gower street, london, wc1e 6bt, uk paper received 18 april 2018; accepted 10 may 2018; published 12 november 2018 abstract background spatial interaction models have been used for decades to explain and predict flows (of migrants, capital, traffic, trade etc.) between geographic locations. aims this paper will guide users through the process of fitting and calibrating spatial interaction models in order to understand, explain and predict internal migration flows in australia. data and methods internal migration data from the australian 2011 census of population and housing, which records people who have moved between greater capital city statistical areas over 5-year periods, is used to exemplify the modelling process. the r statistical software is used to process and visualise the data as well as run the models. results the full suite of wilson’s family of spatial interaction models is fitted to the internal migration data, revealing that distance and origin/destination populations are some of the most important influencing factors affecting internal migration flows. we see whether constraining the model to known flows about origins and/or destinations will improve the fits and model estimates. conclusions spatial interaction modelling has been a tool in the box of some population geographers for a number of decades. however, recent advances in more forgiving programming languages such as r and python now mean that this powerful modelling methodology is no longer only available to those who also possess advanced computer programming skills. this guide has exemplified the process of fitting and calibrating spatial interaction models on australian internal migration data, but the methods could easily be applied to other flow data sets in other contexts. key words spatial interaction model; gravity model; migration; r; estimation; visualisation; census; australia. http://www.australianpopulationstudies.org/ mailto:a.dennett@ucl.ac.uk 34 dennett australian population studies 2 (2) 2018 1. introduction in this introductory guide, you will learn how we can use spatial interaction models to model population flows for a variety of different purposes such as estimating unknown flows, predicting future patterns, understanding the drivers of those flows, or exploring the differences between the flows of different groups. the empirical example uses migration flows taken from the australian bureau of statistics (abs) 2011 census of population and housing (census), but the method is generic and could be used on any flow data (other population data such as commuting data or economic data such as flows of capital or trade, for example). the examples shown here will use the r software environment, and an accompanying practical walk-through guide will be referred to throughout this paper which can be accessed via the following link: https://rpubs.com/adam_dennett/376877. it can also be followed in its entirety if you would like to learn the code required for any of the models. 1.1 what constitutes a population flow? in fields such as population geography and demography, the population flows of interest are usually low-frequency migration or residential mobility moves. both assume some permanent change of residential address which can be either within a country (internal migration or residential mobility) or between countries (international migration). some scholars make a clear distinction between what they term ‘residential mobility’ (short-distance moves, usually within settlements or regions where individuals may retain the same social groups or job) and ‘internal migration’ (longer distance moves, which may involve changing jobs and social groups). however, in reality this is a continuum with no clear line demarking one or the other. most national censuses will simply refer to any internal move over any distance as ‘internal migration’. depending on where you are in the world, the proportion of a population changing their residential address in a given year is around 10 per cent, with most people only moving a handful of times in their lifetime. these population movements can be contrasted with high-frequency flows that occur on daily or weekly cycles. the commute to work or school or travel to the shops, while of interest to some population geographers, is frequently the domain of transport planners and analysts who are concerned with the impact of these moves on transport infrastructure and systems. 1.2. population flow data data related to these flows can vary. for migration, data captured by censuses or surveys tend to be ‘transitions’ over a period of time (a year, 5 years, 10 years) with a flow recorded if there is a difference between residential address between the start and the end of this period (rees 1977). transition data will not record multiple moves during a period and, as such, are simpler than movement data which record multiple moves. movement data can be found more commonly in population registers that track population changes continuously. commuting data, while frequently transitions derived from census returns, are increasingly being obtained from sources such as mobile phone cell tower connections and apps (erhardt and dennett 2017). in this guide we will use censusbased transition data; however, being aware of these other sources is important, particularly in a world that is looking increasingly to move away from censuses. https://rpubs.com/adam_dennett/376877 australian population studies 2 (2) 2018 dennett 35 1.3. why would we want to model population flows? as hinted in the introduction, there are a number of reasons for wanting to model population flows. in migration studies, in a number of papers by raymer and colleagues (raymer 2007; raymer, abel and smith 2007; raymer and abel 2008) as well abel (2010), willekens (1999) and dennett and wilson (2013), the problem of missing or incomplete data was addressed. pooler (1987) used similar models to tackle the problem of predicting migration, as did fotheringham et al. (2001). implicit in much of this modelling research is the evaluation of factors influencing the observed patterns – a paper by kim and cohen (2010) is more explicit in this regard. 1.4. how can we model population flows? population flows can be conceptualised as interactions between two entities – origins and destinations – which have different properties of emissivity and attractiveness (see lee 1966 for the classic paper on this topic in relation to migration). the strength of interaction is a function of these origin and destination properties and the negative influence of the cost (frequently some measure of distance, but equally could be financial, time or some other cost) that might be associated with travelling between them. this situation is analogous to that observed by newton when he defined the laws of gravity – the larger the entities interacting, the stronger the force of interaction; the further the distance between them, the weaker the interaction. hence the term ‘gravity model’ has been used and the equation applied in studying population flows for a long time. see zipf (1946) for one of the earlier applications of the model to population flows. over the years, various improvements have been made to the basic gravity model. perhaps the most important paper in this respect is by wilson (1971), where he introduces the idea of ‘constraints’ which force the flows or interactions estimated by the model to adhere to known information about the system. for example, there may be data on the total number of people leaving an origin or arriving at a destination (or both). in the basic gravity model, the flows estimated might exceed this known information, which is clearly an issue. by introducing constraints, it is possible to force the modelled flows to correspond to this known information, significantly improving accuracy. wilson called his new family of constrained models ‘spatial interaction models’. 2. modelling population flows in practice the basic theory behind spatial interaction or gravity models is not too difficult to comprehend. where the challenges begin is in running a spatial interaction model in practice. in the 1970s and 1980s, when spatial interaction modelling was being established as part of the tool kit for people working on spatial problems, running a model would require some fairly high-level computer programming knowledge. today, however, software has advanced and more forgiving languages such as python and r mean that spatial interaction models can be used as a tool by many more researchers. in this guide we will be using r, with details of the full implementation referred to in this paper found in an accompanying walk-through guide designed to be worked through while reading the text here. the guide includes all of the data and code you need to run a spatial interaction model, and can be https://rpubs.com/adam_dennett/376877 36 dennett australian population studies 2 (2) 2018 accessed via the link at the beginning of this paper1. if, however, you would like to explore these models in python, then oshan (2016) has written an excellent primer that is worth reading, while dan lewis has translated a similar r walk-through of mine into python using uk data2. for consistency, oshun’s notation is adopted in this paper. 2.1. data to illustrate the modelling exercise, migration data (derived from the answer about previous residence 5 years ago – therefore comprising 5-year transitions) from the 2011 census have been obtained. these data are at the greater capital city statistical area (gccsa) level, which is comprised of 15 zones (figure 1). figure 1: five-year migration flows between gccsas, 2006–2011 source: abs 2011 census. note: line weights indicative of volumes. 1 https://rpubs.com/adam_dennett/376877 2 https://github.com/danlewis85/ucl_casa_urban_simulation https://rpubs.com/adam_dennett/376877 https://github.com/danlewis85/ucl_casa_urban_simulation australian population studies 2 (2) 2018 dennett 37 accompanying the migration flows are variables for each origin and destination relating to: • total population • unemployment rate • median weekly income • percentage of households living in rented accommodation. these variables can be used to try and explain observed migration flows or predict flows if none are available. a table containing origin/destination pairs, the flows between them and these origin and destination specific variables can be observed in table 1 and is downloadable in the accompanying exercise. table 1: sample of pair-wise migration flow data with accompanying data relating to origin and destination characteristics origin destination flow origpop destpop orig unemp dest unemp orig med income dest med income orig % rented dest % rented greater sydney greater sydney 3,395,019 4,391,673 4,391,673 5.74 5.74 780.64 780.64 31.77 31.77 greater sydney rest of nsw 91,043 4,391,673 2,512,952 5.74 6.12 780.64 509.97 31.77 27.20 greater sydney greater melbourne 22,605 4,391,673 3,999,981 5.74 5.47 780.64 407.95 31.77 27.34 greater sydney rest of vic. 4,420 4,391,673 1,345,717 5.74 5.17 780.64 506.58 31.77 24.08 greater sydney greater brisbane 22,874 4,391,673 2,065,998 5.74 5.86 780.64 767.08 31.77 33.19 greater sydney rest of qld 27,447 4,391,673 2,253,723 5.74 6.22 780.64 446.48 31.77 32.57 greater sydney greater adelaide 5,829 4,391,673 1,225,235 5.74 5.78 780.64 445.53 31.77 28.27 greater sydney rest of sa 795 4,391,673 368,260 5.74 5.45 780.64 522.71 31.77 26.17 greater sydney greater perth 10,572 4,391,673 1,728,865 5.74 4.76 780.64 730.84 31.77 27.52 source: abs 2011 census 2.2. the ‘unconstrained’ / ‘total constrained’ spatial interaction model 2.2.1. the multiplicative modelling framework the classic gravity model, which estimates flows/interactions as a function of predictor variables (a model virtually identical to that used by zipf), can be written as follows: 𝑇𝑇𝑖𝑖𝑖𝑖 = 𝑘𝑘 𝑉𝑉𝑖𝑖 𝜇𝜇𝑊𝑊𝑗𝑗 𝛼𝛼 𝑑𝑑𝑖𝑖𝑗𝑗 𝛽𝛽 (1) this model can be rearranged and written in the multiplicative form more familiar from wilson’s 1971 paper: 𝑇𝑇𝑖𝑖𝑖𝑖 = 𝑘𝑘𝑉𝑉𝑖𝑖 𝜇𝜇𝑊𝑊𝑖𝑖 𝛼𝛼𝑑𝑑𝑖𝑖𝑖𝑖−𝛽𝛽 (2) 38 dennett australian population studies 2 (2) 2018 where: 𝑇𝑇𝑖𝑖𝑖𝑖 is the transition/trip or flow, 𝑇𝑇, between origin 𝑖𝑖 (always the rows in a matrix) and destination 𝑗𝑗 (always the columns in a matrix). if you are not overly familiar with matrix notation, the 𝑖𝑖 and 𝑗𝑗 are just generic indexes to allow us to refer to any cell in the matrix. 𝑉𝑉 is a vector (a 1 dimensional matrix – or, if you like, a single line of numbers) of origin attributes which relate to the emissivity of all origins in the dataset, 𝑖𝑖 – this could be any of the origin-related variables in table 1. 𝑊𝑊 is a vector of destination attributes relating to the attractiveness of all destinations in the dataset, 𝑗𝑗 – similarly, this could be any of the destination related variables in table 1. 𝑑𝑑 is a matrix of costs (frequently distances – hence, d) relating to the flows between 𝑖𝑖 and 𝑗𝑗. 𝑘𝑘, 𝜇𝜇, 𝛼𝛼 and 𝛽𝛽 are all model parameters to be estimated. 𝛽𝛽 is assumed to be negative, as with an increase in cost/distance we would expect interaction to decrease. wilson’s term for this basic gravity model is the ‘unconstrained’ model. however, 𝑘𝑘 is a constant of proportionality and forces all flow estimates to add up to the total number of flows observed in a system. this leads to this particular model being more accurately described as a ‘total constrained’ model, where: 𝑘𝑘 = 𝑇𝑇 ∑ ∑ 𝑉𝑉𝑖𝑖 𝜇𝜇𝑗𝑗𝑖𝑖 𝑊𝑊𝑗𝑗 𝛼𝛼𝑑𝑑𝑖𝑖𝑗𝑗 −𝛽𝛽 (3) and 𝑇𝑇 is the sum of our matrix of observed flows or: 𝑇𝑇 = ∑ ∑ 𝑇𝑇𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖 (4) in plain language, this is just the sum of all observed flows divided by the sum of all of the other elements in the model. if we use distance as a basic measure of cost, then the simplest distance to measure is the euclidean distance between the centroids of the zones for which you have data. in the accompanying walkthrough exercise, the process of generating a distance matrix from a set of gccsa boundaries is demonstrated using the spdists() function in r. observing equations 1 and 2 above you would be forgiven for asking how the values for the four parameters are known or estimated. one solution could be to simply insert some arbitrary or expected values as parameters. this becomes more feasible once we understand what the parameters are in reality: they relate to the scaling effect/importance of the variables with which they are associated. most simply, where the effects of origin and destination attributes on flows scale in a linear fashion (i.e. for a 1 unit increase in, say, population at origin, we might expect a 1 unit increase in flows of people from that origin; or for a halving in average salary at destination, we might expect a halving of migrants), then the parameter/scaling factor for our origin variable would equal 1, e.g. μ = 1 and α = 1. in newton’s original gravity equation the negative influence of distance is not linear. rather, it follows a power law. for example, where β = -2 for a 1 unit increase in distance, we have a 1-2 (or 1) australian population studies 2 (2) 2018 dennett 39 unit decrease in interaction/flow. for a 2 unit increase in distance, we have 2-2 (0.25 or 1/4) of the interaction, for a 3 unit increase, 3-2 (0.111) of the interaction and so on. we can check to see if μ = 1 and α = 1 and β = -2 are a good or poor guess by looking at our data and plotting observed flows against each variable and seeing whether the value of these variables raise to these powers (figure 2). reviewing the three graphs at figure 2a, 2b, 2c, it can be observed that -2 looks like a fairly good estimate for β with the red modelled line matching quite closely the observed relationship between migration flows and distances. 1 also looks like a fair estimate of μ for at least some of the relationship between origin population and migration flows; however, it appears that there is little discernible visible relationship between destination median incomes and migration flows, so the value of α may be of little consequence. if we accept that these first parameter estimates are plausible, then they can be inserted into equation 2 to generate a first set of estimates. such a set of estimates, where k=3.28, μ = 1, α=1 and β=-2, are shown in table 2. these can be compared with the observed flows in table 3. manual inspection of the flows reveals that in some cases the estimates are not too far from the observed flows, but in others we can clearly see that the estimates are a long way out. whilst it is ok to ‘eyeball’ small flow matrices like these, when you have much larger matrices, another solution is required to test the so-called ‘goodness-of-fit’. there are a number of ways to do this but two of the most common are to calculate the coefficient of determination (r2) or the square root of mean squared error (rmse). anyone who has run a linear regression model before will have come across r2 but rmse may be less familiar. there are other methods and they all do more-or-less the same thing: compare the modelled estimates with the real data and represent the degree of agreement with a single number. r2 is popular as it is quite intuitive and can be compared across models. rmse is less intuitive, but some argue is better for comparing changes to the same model. guidance on how to quickly calculate r2 and rmse can be found in the accompanying practical guide. in this initial case, the r2 value is 0.18. this tells us that this first model accounts for about 18 per cent of the variation of flows in the system – not brilliant, but a starting point nevertheless. as a result, two immediate questions emerge: can we improve these estimates? can we tell which predictor variables are best? fortunately, the answer to both of these questions is ‘yes’. one way that we can begin to answer both of these questions is through the process of model calibration. 40 dennett australian population studies 2 (2) 2018 (a) distance note: red line is distance or 𝑑𝑑𝑖𝑖𝑖𝑖−𝛽𝛽 estimate where 𝛽𝛽 = -2 (b) origin population note: red line is origin population or 𝑉𝑉𝑖𝑖 𝜇𝜇estimate where 𝜇𝜇 = 1 (c) destination median income note: red line is destination median income or 𝑊𝑊𝑖𝑖 𝛼𝛼estimate where 𝛼𝛼 = 1 figure 2: the relationship between migration flows and three predictor variables australian population studies 2 (2) 2018 dennett 41 table 2: modelled flows from the initial total constrained gravity model with crude estimated parameters origin / destination 1gsyd 1rnsw 2gmel 2rvic 3gbri 3rqld 4gade 4rsau 5gper 5rwau 6ghob 6rtas 7gdar 7rnte 8acte (all) 1gsyd 0 47944 12607 15513 22050 3346 5187 3522 1012 1325 5627 7789 1386 1395 193311 322014 1rnsw 41995 0 8089 12786 11218 3039 5467 3675 721 1008 2566 3703 1021 1138 42659 139085 2gmel 21972 16095 0 373099 5627 2040 13506 4665 1294 1539 15973 29621 1221 1271 62311 550234 2rvic 7325 6893 101083 0 2014 784 6705 1971 465 570 3652 6304 448 481 18995 157690 3gbri 10556 6131 1546 2041 0 3035 1256 1293 397 573 955 1259 791 771 6639 37243 3rqld 3002 3113 1050 1491 5688 0 1508 2718 615 1118 630 844 2101 2759 2873 29510 4gade 2536 3051 3788 6941 1283 821 0 5788 653 858 1300 1973 543 657 4041 34233 4rsau 441 525 335 523 338 379 1483 0 250 462 172 241 275 439 569 6432 5gper 425 346 312 414 349 288 562 839 0 4637 273 356 730 604 535 10670 5rwau 156 136 104 142 141 147 207 434 1299 0 81 107 523 508 190 4175 6ghob 478 249 778 657 169 60 226 116 55 58 0 25438 46 43 876 29249 6rtas 724 393 1579 1240 244 87 375 179 79 84 27825 0 65 62 1416 34352 7gdar 33 28 17 23 39 56 26 52 41 105 13 17 0 326 36 812 7rnte 42 39 22 31 49 93 41 107 44 131 15 20 417 0 47 1098 8acte 13901 3502 2571 2893 997 230 594 327 92 116 742 1096 110 112 0 27283 (all) 103586 88445 133881 417794 50206 14405 37143 25686 7017 12584 59824 78768 9677 10566 334498 1384080 source: abs 2011 census. notes: greater sydney = 1gsyd; rest of new south wales = 1rnsw; greater melbourne = 2gmel; rest of victoria = 2rvic; greater brisbane = 3gbri; rest of queensland = 3rqld; greater adelaide = 4gade; rest of south australia = 4rsau; greater perth = 5gper; rest of western australia = 5rwau; greater hobart = 6ghob; rest of tasmania = 6rtas; greater darwin = 7gdar; rest of northern territories = 7rnte; australian capital territory = 8acte 42 dennett australian population studies 2 (2) 2018 table 3: original flow data for comparison origin / destination 1gsyd 1rnsw 2gmel 2rvic 3gbri 3rqld 4gade 4rsau 5gper 5rwau 6ghob 6rtas 7gdar 7rnte 8acte (all) 1gsyd 0 91043 22605 4420 22874 27447 5829 795 10572 2127 1654 1984 1992 828 10658 204828 1rnsw 53568 0 12418 13072 21289 35191 3613 1587 4999 3295 978 1885 2252 1431 15766 171344 2gmel 15569 11094 0 70264 13055 16164 6017 1292 10111 2570 2126 2553 2029 1004 4727 158575 2rvic 2528 11968 47988 0 4328 10110 3468 2217 3449 2597 667 1428 1548 721 1362 94379 3gbri 12333 16056 13080 4249 0 84649 3044 818 4810 1796 1388 2295 1802 905 3127 150352 3rqld 11629 26699 12284 7566 74412 0 3772 1758 6583 4688 1479 3086 3126 2142 3125 162349 4gade 5415 3517 8803 3188 5449 6178 0 25679 3831 1230 598 872 1843 927 1995 69525 4rsau 477 1490 1154 2439 824 2631 22020 0 1051 1354 148 429 679 484 185 35365 5gper 6523 4064 11721 2931 5086 7019 2625 865 0 41332 1022 1802 1305 416 1675 88386 5rwau 714 2241 1490 1811 1141 4333 808 982 42149 0 277 1161 1093 627 251 59078 6ghob 1221 998 3014 624 1307 1810 532 111 901 365 0 5019 195 113 564 16774 6rtas 1029 1871 2637 1647 1543 2884 658 343 1210 1028 7214 0 272 164 288 22788 7gdar 1237 2185 1957 1481 2763 5107 2111 641 2149 949 239 333 0 1998 824 23974 7rnte 406 1432 700 792 896 3018 1296 961 699 826 96 213 2684 0 229 14248 8acte 7065 16829 5930 1994 5225 6968 2657 1091 2212 1110 466 480 3304 56779 0 112110 (all) 119714 191487 145781 116478 160192 213509 58450 39140 94726 65267 18352 23540 24124 68539 44776 1384075 source: abs 2011 census australian population studies 2 (2) 2018 dennett 43 2.2.2. regression modelling framework calibration is the process of adjusting parameters in the model to try and get the estimates to agree with the observed data as much as possible. adjusting the parameters is the sort of iterative process that computers are particularly good at and the goodness-of-fit statistics can be used to indicate when the optimum solution is found. historically this process required a researcher with the requisite programming skills to write a computer algorithm to iteratively adjust each parameter, check the goodness-of-fit, and then start all over again until the goodness-of-fit statistic was maximised/minimised. there are various well-established routines that can achieve this, such as the newton-raphson algorithm, but without the necessary programming skills this can be a serious barrier and probably why spatial interaction modelling was the preserve of a few specialists for so long. however, since the early days of spatial interaction modelling, a number of useful developments have occurred. perhaps the most important in the context of calibration is the fact that it is possible to turn the multiplicative model in equation 2 into an additive model. taking the logarithms of both sides of equation 2, you end up with the following equation: ln𝑇𝑇𝑖𝑖𝑖𝑖 = 𝑘𝑘 + 𝜇𝜇ln𝑉𝑉𝑖𝑖 + 𝛼𝛼ln𝑊𝑊𝑖𝑖 − 𝛽𝛽ln𝑑𝑑𝑖𝑖𝑖𝑖 (5) what we have now is a regression model. anyone who has been introduced to regression models in introductory statistics classes will be aware that there are various pieces of software available to us to run regressions (such as r) and calibrate the parameters (or ‘estimate the coefficients’ in the language of statistics), so expert programming skills are no longer required. there are some papers that are worth reading at this point if you would like to learn more. perhaps the best is by flowerdew and aitkin (1982). one of the key points that flowerdew and aitkin make is that the model in equation 5 (known as a log-normal model) has various problems associated with it, which mean that the estimates produced might not be reliable. the paper (and also wilson’s 1971 paper) details these issues; however, the salient point is that the way around many of these issues is to re-specify the model, not as a log-normal regression but as a poisson or negative binomial regression model. the flows that spatial interaction models deal with (such as migration or commuting) relate to nonnegative integer counts (you cannot have negative people moving between places and you cannot normally – if they are alive! – have fractions of people moving either). as such, the probability of migrating or commuting is not described by a continuous (normal) probability distribution (the distribution which underpins the error distribution in standard linear regression models), but a discrete probability distribution such as the poisson distribution or the negative binomial distribution (of which the poisson distribution is a special case). there is a family of generalised linear models, but in the analysis of migration and other population flows poisson and negative binomial regression models have been used most frequently (abel 2010; congdon 1993, 1988; crymble, dennett and hitchcock 2017; flowerdew 2010, 1982; flowerdew and aitkin 1982; shen 2017, 2015; willekens 1999). the differences between the two are technical, but some, including congdon (1993), argue that negative binomial models should be used over poisson due to a statistical phenomenon known as overdispersal. for ease of explanation here we will 44 dennett australian population studies 2 (2) 2018 continue with the poisson model, but be aware that in practice a negative binomial model may be better. in the migration modelling literature others, such as raymer (raymer 2007; raymer, abel and smith 2007; raymer and giulietti 2010; rogers and raymer 1998), have used what are described as loglinear models. these are exactly the same as poisson models with the distinction that poisson models will usually contain both continuous and categorical predictor variables, whereas log-linear models will only contain categorical predictors. for a two-dimensional population flow matrix between origins and destinations, these categorical predictors would generally be the origin and destination zones – in effect we would have a two-dimensional contingency table. the analysis of contingency tables is well established in statistics and as such has its own lexicon. in log-linear modelling terminology, these origin and destination zones would be described as the ‘main’ or ‘fixed’ effects and are equivalent to the ‘constraints’ that will be introduced later on in this paper. this discussion of poisson, log-linear and spatial interaction models is included here for the purpose of highlighting that in the migration modelling literature it can be particularly confusing when reading papers by different authors who all use very different terminology and modelling paradigms. iterative proportional fitting (ipf) is another term that may appear when researching papers in this area (lomax and norman 2016). the salient point is that all of the models are essentially doing the same thing, but the papers in which they are outlined ascribe to different definitional conventions. how is the poisson distribution different to a normal distribution? aside from them describing different frequency/probability distributions, they behave differently for different sets of observations. below are two histograms (figure 3). the first is a random variable with a normal distribution, with a mean of 75 and a standard deviation of 5; the second is a histogram of poisson distributed variable with the same mean (poisson distributions have only one parameter – the mean). you will notice that they look broadly similar. however, with a poisson distributed variable, when the mean (𝜆𝜆 lambda) changes, so does the shape of the frequency distribution. as the mean gets smaller, and this is often the case with flow data where small flows are very likely, the distribution starts to look a lot more like a skewed or log-normal continuous distribution. the key point is that it is not a continuous distribution but a discrete (poisson) distribution. figure 4 plots a frequency distribution for a discrete variable with a small mean. the shape of the histogram is very similar to a positively skewed continuous distribution. for any system of population flows between a matrix of origins and destinations, the flows will have a mean value of 𝜆𝜆𝑖𝑖𝑖𝑖 which will normally be quite low and will dictate the distribution. plotting the frequency distribution of migration flows between our australian gccsas (excluding within-area flows – figure 5), reveals a histogram which looks like a skewed normal or, more accurately, a poisson distribution with a small mean. australian population studies 2 (2) 2018 dennett 45 (a) random variable with a normal distribution (b) random variable with a poisson distribution figure 3: normal and poisson distributions figure 4: poisson distribution with a small mean 46 dennett australian population studies 2 (2) 2018 figure 5: frequency distribution of migration flows (excluding intra-zonal flows) between gccsas in all of this discussion about frequency/probability distributions, it can be easy to lose track of the purpose for understanding all of this. perhaps the easiest way to remind ourselves of the purpose is in understanding the basics of what a regression model is trying to do. at its most simple a regression model is nothing more than a line of best fit drawn through a cloud of observations. if we think of a spatial interaction model as partially representing the relationship between volume of flow and cost of interaction (distance), then we would expect to see a straight line between flow volumes and distance. figure 2a shows that when plotting raw flows against distance, the relationship cannot be represented by a straight line. however, if both the flows and the distance are logged, as in equation 5, a plot similar to the one below in figure 6 is produced. it might not be a clear straight-line relationship, but there is certainly a suggestion that the blue regression line does represent the underlying relationship between migration flows and distance. figure 6: relationship between log(distance) and log(migration flows) between gccsas australian population studies 2 (2) 2018 dennett 47 now the discussion in the previous session indicates that the 𝑦𝑦 variable in our model is not logged as in the graph above. however, it can still be modelled using something like the blue line if we assume a poisson distribution. equation 5 can now be re-specified as a poisson regression model. instead of the dependent variable being ln𝑇𝑇𝑖𝑖𝑖𝑖, it is now the mean of the poisson distribution 𝜆𝜆𝑖𝑖𝑖𝑖 and the model becomes: 𝜆𝜆𝑖𝑖𝑖𝑖 = exp(𝑘𝑘 + 𝜇𝜇ln𝑉𝑉𝑖𝑖 + 𝛼𝛼ln𝑊𝑊𝑖𝑖 − 𝛽𝛽ln𝑑𝑑𝑖𝑖𝑖𝑖) (6) what this model says is 𝜆𝜆𝑖𝑖𝑖𝑖 (the dependent variable – the estimate of 𝑇𝑇𝑖𝑖𝑖𝑖) is logarithmically linked to (or modelled by) a linear combination of the logged independent variables in the model. using equation 6, a poisson regression model can be fitted to produce estimates of 𝑘𝑘, 𝜇𝜇, 𝛼𝛼 and 𝛽𝛽 – or put another way, we can use the regression model to calibrate our parameters. it is very straight forward to run a poisson regression model in r using the glm() (generalised linear models) function. the code to do this can be found in the accompanying guide. delving into the depths of the glm() function documentation will reveal that the parameters are calibrated though an ‘iteratively re-weighted least squares’ algorithm. this essentially fits lots of lines similar to that in figure 6 to the data until it finds the best one. it continually adjusts the parameters to minimise the error between the observed and expected (blue line) values using some goodness-of-fit measure, not dissimilar to an r2 or rmse. running the model will produce some output similar to that shown below in box 1. call: glm(formula = flow ~ log(vi1_origpop) + log(wj3_destmedinc) + log(dist), family = poisson(link = "log"), data = mdatasub, na.action = na.exclude) deviance residuals: min 1q median 3q max -177.78 -54.49 -24.50 9.21 470.11 coefficients: estimate std. error z value pr(>|z|) (intercept) 7.1953790 0.0248852 289.14 <0.0000000000000002 *** log(vi1_origpop) 0.5903363 0.0009232 639.42 <0.0000000000000002 *** log(wj3_destmedinc) -0.1671417 0.0033663 -49.65 <0.0000000000000002 *** log(dist) -0.8119316 0.0010157 -799.41 <0.0000000000000002 *** -- signif. codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1 (dispersion parameter for poisson family taken to be 1) null deviance: 2750417 on 209 degrees of freedom residual deviance: 1503573 on 206 degrees of freedom aic: 1505580 number of fisher scoring iterations: 5 box 1: model output including calibrate parameters from r glm() implementation of equation 6 https://rpubs.com/adam_dennett/376877 48 dennett australian population studies 2 (2) 2018 box 1 contains various pieces of information. the ‘call’ section at the top is the code used to run the model in r. then, under the ‘coefficients’ section, are the values for the four calibrated parameters in the model. in our original model, we estimated the four parameters as follows: 𝑘𝑘 = 3.28, 𝜇𝜇 = 1, 𝛼𝛼 = 1 and 𝛽𝛽 = −2 after fitting the poisson model, the values for the parameters can be found in the ‘estimate’ column. they change to: 𝑘𝑘 = 7.195, 𝜇𝜇 = 0.59, 𝛼𝛼 = −0.17 and 𝛽𝛽 = −0.81 the regression model also produces some other useful pieces of output. the p-values in the last column reveal that all variables have a highly (***) statistically significant influence on migration flows, with the z-scores (standardised coefficients) revealing that distance has the most (negative) influence on the model followed by origin population, with destination income only a small influence on the flows, and a counterintuitive one at that, with increases in destination income resulting in decreases in migration flows. in the accompanying practical guide, it is shown how these parameter values can be inserted directly back into equation 6 to produce a new set of estimates similar to those in table 1. the r2 value for this new matrix improves to 0.32, meaning that by simply calibrating the model parameters on observed data, we are able to explain around 14 per cent more of the variation in the migration flows in our system. 2.2. constrained spatial interaction models returning to wilson’s (1971) seminal paper, he introduces a full family of spatial interaction models of which the unconstrained model is just the start. of course, since then, there have been all manner of incremental advances and alternatives (dennett and wilson 2013; fotheringham 1983; pooler 1994; stillwell 1978). however, in this section we will concentrate on wilson’s original family – the production (origin) constrained model; the attraction (destination) constrained model; and the doubly constrained model – but show how the poisson regression framework can be used as with the unconstrained model. recalling the unconstrained/total constrained model above (table 1), while the total flows in the estimates equalled the total observed flows, none of the estimates sum to the observed in-migration and out-migration totals (the margins of the matrix). wilson’s real contribution to the field was in noticing that this unconstrained model was sub-optimal as it did not make use of all of the available information in the system being studied. where there is a full flow matrix to calibrate parameters, then it is possible to incorporate the row (origin) totals, column (destination) totals or both origin and destination totals to constrain flow estimates to these known values. there are various reasons for wanting to do this in different flow modelling contexts, for example: a) if the researcher is interested in flows of money into businesses or customers into shops, then they might have information on the amount of disposable income and shopping habits of the people living in different areas, perhaps from loyalty card data. this is known information about origins and so it would be logical to constrain the estimates from a spatial interaction model to https://rpubs.com/adam_dennett/376877 australian population studies 2 (2) 2018 dennett 49 this known information. other information about the attractiveness of shops and businesses (store size, variety/specialism of goods etc.) can then be used to estimate how much money/customers a new store opening in the area might make/attract or, if a new out-of-town shopping centre opens, how much it might affect the business of shops in the town centre. this is what is known in the literature as the ‘retail model’ and is perhaps the most common example of a production (origin) constrained spatial interaction model. b) other researchers might be interested in understanding the impact of a large new employer in an area on the flows of traffic in the vicinity or on the demand for new worker accommodation nearby. a good example of where this might be the case is with large new infrastructure developments like airports. for example, before the go-ahead for the new third runway at heathrow airport in london, england was given, one option being considered was a new runway in the thames estuary. if a new airport was built here, what would the potential impact on transport flows be in the area and where might workers commute from? this sort of scenario could be tested with an attraction (destination) constrained spatial interaction model where the number of new jobs in a destination is known (as well as jobs in the surrounding area). the model could also be used to estimate where the workers will be drawn from and their likely travel-to-work patterns. these models are known as land use transport interaction (luti) models and have a well-established history in urban planning. c) other researchers might be interested in understanding the changing patterns of commuting or migration over time. data from a census provides an accurate snap-shot of migrating and commuting patterns, but only periodically. in these full data matrices, information about both the numbers of commuters/migrants leaving origins and arriving at destinations and the interactions between them is known. constraining model estimates to this known information at origin and destination allows various things to be examined, including: i. the ways that the patterns of commuting/migration differ from the model predictions – where might there be more migrant/commuter flows than expected? ii. how the model parameters vary over time – for example, how does distance/cost of travel affect flows over time? are people prepared to travel further or less distance than before? 2.2.2. the production constrained model recall the unconstrained model from equation 2. a production constrained model constrains estimates to known information about the origins and so replaces the terms 𝑘𝑘 and 𝑉𝑉𝑖𝑖 𝜇𝜇 to produce the following model: 𝑇𝑇𝑖𝑖𝑖𝑖 = 𝐴𝐴𝑖𝑖𝑂𝑂𝑖𝑖𝑊𝑊𝑖𝑖 𝛼𝛼𝑑𝑑𝑖𝑖𝑖𝑖−𝛽𝛽 (7) where: 𝑂𝑂𝑖𝑖 = ∑ 𝑇𝑇𝑖𝑖𝑖𝑖𝑖𝑖 (8) and: 𝐴𝐴𝑖𝑖 = 1 ∑ 𝑊𝑊𝑗𝑗 𝛼𝛼 𝑗𝑗 𝑑𝑑𝑖𝑖𝑗𝑗 −𝛽𝛽 (9) 50 dennett australian population studies 2 (2) 2018 in the production constrained model, 𝑂𝑂𝑖𝑖 does not have a parameter as it is a known constraint. 𝐴𝐴𝑖𝑖 is known as a balancing factor and is a vector of values which relate to each origin 𝑖𝑖 which do the equivalent job as 𝑘𝑘 in the unconstrained/total constrained model but ensure that flow estimates from each origin sum to the know totals 𝑂𝑂𝑖𝑖 rather than just the overall total. now at this point the 𝑂𝑂𝑖𝑖 and 𝐴𝐴𝑖𝑖 values could be calculated by hand for the sample system and the parameter values for the rest of the model could be guessed. however, the poisson regression framework allows this to be avoided. the production constrained model can be re-specified as a poisson regression model in exactly the same way as before. taking the logs of the right-hand side of the equation, and assuming that these are logarithmically linked to the poisson distributed mean (𝜆𝜆𝑖𝑖𝑖𝑖) of the 𝑇𝑇𝑖𝑖𝑖𝑖 variable, means that equation 7 becomes: 𝜆𝜆𝑖𝑖𝑖𝑖 = 𝑒𝑒𝑒𝑒𝑒𝑒(𝜇𝜇𝑖𝑖 + 𝛼𝛼ln𝑊𝑊𝑖𝑖 − 𝛽𝛽ln𝑑𝑑𝑖𝑖𝑖𝑖) (10) in equation 10 𝜇𝜇𝑖𝑖 is the equivalent of the vector of balancing factors 𝐴𝐴𝑖𝑖. but in regression/log-linear modelling terminology these can also be described as either ‘dummy variables’ or ‘fixed effects’. in practical terms what this means is that in the regression model 𝜇𝜇𝑖𝑖 is modelled as a categorical predictor3, and therefore in the poisson regression model the numeric values of 𝑂𝑂𝑖𝑖 are ignored and replaced by a categorical identifier for the origin. in terms of the origin/destination migration matrix shown in table 3, rather than the flow of 204,828 migrants leaving sydney (row 1) being used as a predictor, simply the code ‘1gsyd’ is used as a dummy variable. in the accompanying practical guide the code for running this model using the glm() function in r is provided. running the model will produce the following output (box 2). there are elements of the model output that should be familiar from the unconstrained model: • the α parameter related to the destination attractiveness (in this case, median weekly income): -0.27 is not much different from the unconstrained model. the z-score indicates that this is not a very important variable in explaining variation in migration behaviours in australia. • the β distance decay parameter: -1.23 has decreased meaning after controlling for origin characteristics, distance becomes more of a deterrent. where the model output differs is that the intercept (𝑘𝑘) parameter has been replaced by the vector of dummy variables/constraints 𝜇𝜇𝑖𝑖 relating to each origin. we can see from the standard outputs from the model that all of the explanatory variables are statistically significant (***); the z-scores indicate that the rest of queensland and greater sydney have greater emissivity properties than all other zones in the model, with factors associated with these zones more important in explaining migration patterns in australia than distance, which while still important, is less important than a number of origin zones. in the accompanying practical guide there is code that allows you to create a new set of estimates by both plugging these values back into a multiplicative model or by using a much easier built-in function in glm(). using either method will produce the set of flows shown in table 4. 3 https://en.wikipedia.org/wiki/categorical_variable https://en.wikipedia.org/wiki/categorical_variable https://en.wikipedia.org/wiki/categorical_variable https://rpubs.com/adam_dennett/376877 https://rpubs.com/adam_dennett/376877 https://en.wikipedia.org/wiki/categorical_variable australian population studies 2 (2) 2018 dennett 51 call: glm(formula = flow ~ orig_code + log(wj3_destmedinc) + log(dist) 1, family = poisson(link = "log"), data = mdatasub, na.action = na.exclude) deviance residuals: min 1q median 3q max -225.71 -54.10 -15.94 20.45 374.27 coefficients: estimate std. error z value pr(>|z|) orig_code1gsyd 19.541851 0.023767 822.22 <0.0000000000000002 *** orig_code1rnsw 19.425497 0.023913 812.35 <0.0000000000000002 *** orig_code2gmel 18.875763 0.023243 812.12 <0.0000000000000002 *** orig_code2rvic 18.335242 0.022996 797.31 <0.0000000000000002 *** orig_code3gbri 19.856564 0.024063 825.20 <0.0000000000000002 *** orig_code3rqld 20.094898 0.024300 826.94 <0.0000000000000002 *** orig_code4gade 18.747938 0.023966 782.28 <0.0000000000000002 *** orig_code4rsau 18.324029 0.024407 750.75 <0.0000000000000002 *** orig_code5gper 20.010551 0.024631 812.43 <0.0000000000000002 *** orig_code5rwau 19.392751 0.024611 787.96 <0.0000000000000002 *** orig_code6ghob 16.802016 0.024282 691.97 <0.0000000000000002 *** orig_code6rtas 17.013981 0.023587 721.33 <0.0000000000000002 *** orig_code7gdar 18.607483 0.025012 743.93 <0.0000000000000002 *** orig_code7rnte 17.798856 0.025704 692.45 <0.0000000000000002 *** orig_code8acte 17.796693 0.023895 744.79 <0.0000000000000002 *** log(wj3_destmedinc) -0.272640 0.003383 -80.59 <0.0000000000000002 *** log(dist) -1.227679 0.001400 -876.71 <0.0000000000000002 *** -- signif. codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1 (dispersion parameter for poisson family taken to be 1) null deviance: 23087017 on 210 degrees of freedom residual deviance: 1207394 on 193 degrees of freedom aic: 1209427 number of fisher scoring iterations: 6 box 2: outputs from the production constrained model specified in equation 10 comparing table 4 with table 3, it is very easy to see the origin constraints working. the sum across all destinations for each origin in the estimated matrix (table 4) is exactly the same (give or take the odd rounding error) as the same sum across the observed matrix (table 3): ∑ 𝑇𝑇𝑖𝑖𝑖𝑖𝑖𝑖 = ∑ 𝜆𝜆𝑖𝑖𝑖𝑖𝑖𝑖 = 𝑂𝑂𝑖𝑖. but clearly the same is not true when you sum across all origins for each destination: ∑ 𝑇𝑇𝑖𝑖𝑖𝑖𝑖𝑖 ≠ ∑ 𝜆𝜆𝑖𝑖𝑖𝑖𝑖𝑖 ≠ 𝐷𝐷𝑖𝑖. calculating the r2 value, the fit of the model has improved quite considerably: from around 0.32 in the unconstrained model to around 0.43 in this model. the rmse has also dropped quite noticeably. one of the advantages of singly constrained models is that once initial parameters have been calibrated on existing data, then changes can be made to destination variables (for origin/production constrained models) or origin variables (for destination/attraction constrained models) and the impact on flow estimates explored. for example, what would happen if average wages suddenly increased in an area? how would this impact migration flows to and from that area? the accompanying worked example demonstrates how this can be explored. https://rpubs.com/adam_dennett/376877 52 dennett australian population studies 2 (2) 2018 table 4: modelled flows from a production constrained spatial interaction model with parameters calibrated by a poisson regression model origin / destination 1gsyd 1rnsw 2gmel 2rvic 3gbri 3rqld 4gade 4rsau 5gper 5rwau 6ghob 6rtas 7gdar 7rnte 8acte (all) 1gsyd 0 36794 19752 18516 15905 8076 10591 7248 2504 2860 11192 11454 2519 4105 53308 204824 1rnsw 29163 0 18862 20620 13173 9548 13715 9329 2549 3032 8667 9100 2619 4543 26439 171359 2gmel 8501 10243 0 70950 3742 3243 10367 4685 1584 1705 11552 14147 1268 2109 14474 158570 2rvic 4924 6918 43838 0 2263 2050 7667 3139 961 1053 5309 6221 779 1320 7935 94377 3gbri 21684 22658 11852 11604 0 16555 9653 8526 3069 3722 8200 8144 3886 6207 14647 150407 3rqld 12057 17984 11248 11511 18128 0 12989 16188 4832 6746 7639 7664 8515 16335 10539 162375 4gade 4109 6714 9345 11186 2747 3376 0 9731 1895 2167 4506 4879 1403 2558 4912 69528 4rsau 1922 3122 2887 3130 1659 2876 6653 0 1438 2028 1780 1840 1264 2736 2017 35352 5gper 3930 5048 5777 5673 3533 5080 7666 8507 0 17470 4952 4882 4812 6954 4064 88348 5rwau 2445 3269 3387 3386 2333 3862 4775 6535 9514 0 2696 2679 4515 7196 2476 59068 6ghob 619 605 1485 1105 333 283 643 371 175 175 0 9840 129 201 807 16771 6rtas 827 829 2374 1689 431 371 908 501 225 226 12842 0 166 261 1121 22771 7gdar 1030 1350 1204 1198 1165 2331 1478 1948 1253 2159 950 937 0 6000 981 23984 7rnte 644 899 769 779 714 1716 1034 1618 695 1321 569 568 2303 0 618 14247 8acte 9622 6021 6070 5386 1939 1274 2285 1373 467 523 2631 2802 433 712 0 41538 (all) 101477 122454 138850 166733 68065 60641 90424 79699 31161 45187 83485 85157 34611 61237 144338 1313519 australian population studies 2 (2) 2018 dennett 53 2.2.3. the attraction constrained model the attraction constrained model is virtually the same as the production constrained model: 𝑇𝑇𝑖𝑖𝑖𝑖 = 𝐷𝐷𝑖𝑖𝐵𝐵𝑖𝑖𝑉𝑉𝑖𝑖 𝜇𝜇𝑑𝑑𝑖𝑖𝑖𝑖−𝛽𝛽 (11) where: 𝐷𝐷𝑖𝑖 = ∑ 𝑇𝑇𝑖𝑖𝑖𝑖𝑖𝑖 (12) and: 𝐵𝐵𝑖𝑖 = 1 ∑ 𝑉𝑉𝑖𝑖 𝜇𝜇 𝑖𝑖 𝑑𝑑𝑖𝑖𝑗𝑗 −𝛽𝛽 (13) the poisson model equation for the attraction constrained model would be: 𝜆𝜆𝑖𝑖𝑖𝑖 = 𝑒𝑒𝑒𝑒𝑒𝑒(𝜇𝜇ln𝑉𝑉𝑖𝑖 + 𝛼𝛼𝑖𝑖 − 𝛽𝛽ln𝑑𝑑𝑖𝑖𝑖𝑖) (14) its implementation in r is virtually identical to the production constrained model. see the accompanying walk-through exercise for full details of how to run these models with the sample dataset. because of the similarities to the production constrained model, the attraction constrained model will not be dwelt upon here. 2.2.3. the doubly constrained model the final model in the wilson (1971) family is the doubly constrained model. let’s begin with the formula: 𝑇𝑇𝑖𝑖𝑖𝑖 = 𝐴𝐴𝑖𝑖𝑂𝑂𝑖𝑖𝐵𝐵𝑖𝑖𝐷𝐷𝑖𝑖𝑑𝑑𝑖𝑖𝑖𝑖−𝛽𝛽 (15) where: 𝑂𝑂𝑖𝑖 = ∑ 𝑇𝑇𝑖𝑖𝑖𝑖𝑖𝑖 (16) 𝐷𝐷𝑖𝑖 = ∑ 𝑇𝑇𝑖𝑖𝑖𝑖𝑖𝑖 (17) and: 𝐴𝐴𝑖𝑖 = 1 ∑ 𝐵𝐵𝑗𝑗𝑗𝑗 𝐷𝐷𝑗𝑗𝑑𝑑𝑖𝑖𝑗𝑗 −𝛽𝛽 (19) 𝐵𝐵𝑖𝑖 = 1 ∑ 𝐴𝐴𝑖𝑖𝑖𝑖 𝑂𝑂𝑖𝑖𝑑𝑑𝑖𝑖𝑗𝑗 −𝛽𝛽 (20) astute readers will have noticed that the calculation of 𝐴𝐴𝑖𝑖 relies on knowing 𝐵𝐵𝑖𝑖 and the calculation of 𝐵𝐵𝑖𝑖 relies on knowing 𝐴𝐴𝑖𝑖 – something of a conundrum to which the solution is elegantly described by senior (1979), who sketches out a very useful algorithm for iteratively arriving at values for 𝐴𝐴𝑖𝑖 and 𝐵𝐵𝑖𝑖 by setting each to equal 1 initially and then continuing to calculate each in turn until the difference between successive iterations of the 𝐴𝐴𝑖𝑖 and 𝐵𝐵𝑖𝑖 values is small enough not to matter. in the accompanying practical guide an algorithm to achieve this using this multiplicative framework is provided. however, as you will have probably guessed by now, the poisson regression framework allows for the doubly constrained model to be fitted very easily. https://rpubs.com/adam_dennett/376877 https://rpubs.com/adam_dennett/376877 54 dennett australian population studies 2 (2) 2018 the poisson doubly constrained model takes the form: 𝜆𝜆𝑖𝑖𝑖𝑖 = 𝑒𝑒𝑒𝑒𝑒𝑒(𝜇𝜇𝑖𝑖 + 𝛼𝛼𝑖𝑖 − 𝛽𝛽ln𝑑𝑑𝑖𝑖𝑖𝑖) (21) when run in r and applied to the australian migration data we have been using, this model will produce the outputs shown in box 3. the coefficients in this version of the doubly constrained model will look a little different. this is explained in the accompanying walk-through exercise and is because an intercept has been added along with reference categories for the categorical variables, as two factor levels are used in this model. importantly the model estimates are not altered in any way by this. the reference level means that the origin and destination coefficients need to be interpreted in relation to a reference category. in this example, the first zone in the system is used (sydney), with the direction and size of the coefficients referring to whether another origin or destination zone has a greater or lesser positive or negative effect on migration flows in the system when compared to sydney. the estimates produced by the doubly constrained model are the most accurate in the wilson family of models (in this example, an r2 value of 0.87). however, there is a loss of flexibility when compared to the singly constrained models, as only alternatives to origin/destination interaction explanatory variables such as historic flows or something other than distance can be experimented with. the double constraints mean that originand destination-specific explanatory variables cannot be used. 2.2.4. further experimentation all of the way through this paper there has been an assumption that the distance decay parameter follows a negative power law. this does not have to be the case and empirically might not necessarily be so. in his original paper, wilson (1971) generalised the distance decay parameter to: 𝑓𝑓(𝑑𝑑𝑖𝑖𝑖𝑖) (22) where 𝑓𝑓 represents some function of distance describing the rate at which the flow interactions change as distance increases. lots of people have written about this, including taylor (1983) and more recently lovelace (2015) in a transport context. the inverse power law that has been used to this point is one possible function of distance; the other common one that is used is the negative exponential function: 𝑒𝑒𝑒𝑒𝑒𝑒(−𝛽𝛽𝑑𝑑𝑖𝑖𝑖𝑖) (23) the exact effect that the different function has on the rate of distance decay will depend on the value of 𝛽𝛽 as well as the function. however, figure 7 shows how the different functions and values of 𝛽𝛽 can combine to affect distance decay. https://rpubs.com/adam_dennett/376877 australian population studies 2 (2) 2018 dennett 55 call: glm(formula = flow ~ orig_code + dest_code + log(dist), family = poisson(link = "log"), data = mdatasub, na.action = na.exclude) deviance residuals: min 1q median 3q max -93.018 -26.703 0.021 19.046 184.179 coefficients: estimate std. error z value pr(>|z|) (intercept) 20.208178 0.011308 1786.999 <0.0000000000000002 *** orig_code1rnsw -0.122417 0.003463 -35.353 <0.0000000000000002 *** orig_code2gmel -0.455872 0.003741 -121.852 <0.0000000000000002 *** orig_code2rvic -1.434386 0.004511 -317.969 <0.0000000000000002 *** orig_code3gbri 0.241303 0.003597 67.091 <0.0000000000000002 *** orig_code3rqld 0.772753 0.003599 214.700 <0.0000000000000002 *** orig_code4gade -0.674261 0.004527 -148.936 <0.0000000000000002 *** orig_code4rsau -1.248974 0.005889 -212.091 <0.0000000000000002 *** orig_code5gper 0.742687 0.004668 159.118 <0.0000000000000002 *** orig_code5rwau -0.317806 0.005131 -61.943 <0.0000000000000002 *** orig_code6ghob -2.270736 0.008576 -264.767 <0.0000000000000002 *** orig_code6rtas -1.988784 0.007477 -265.981 <0.0000000000000002 *** orig_code7gdar -0.797620 0.007089 -112.513 <0.0000000000000002 *** orig_code7rnte -1.893522 0.008806 -215.022 <0.0000000000000002 *** orig_code8acte -1.921309 0.005511 -348.631 <0.0000000000000002 *** dest_code1rnsw 0.389478 0.003899 99.894 <0.0000000000000002 *** dest_code2gmel -0.007616 0.004244 -1.794 0.0727 . dest_code2rvic -0.781258 0.004654 -167.854 <0.0000000000000002 *** dest_code3gbri 0.795909 0.004037 197.178 <0.0000000000000002 *** dest_code3rqld 1.516186 0.003918 386.955 <0.0000000000000002 *** dest_code4gade -0.331189 0.005232 -63.304 <0.0000000000000002 *** dest_code4rsau -0.627202 0.006032 -103.980 <0.0000000000000002 *** dest_code5gper 1.390114 0.005022 276.811 <0.0000000000000002 *** dest_code5rwau 0.367314 0.005362 68.509 <0.0000000000000002 *** dest_code6ghob -1.685934 0.008478 -198.859 <0.0000000000000002 *** dest_code6rtas -1.454819 0.007612 -191.112 <0.0000000000000002 *** dest_code7gdar -0.308516 0.007716 -39.986 <0.0000000000000002 *** dest_code7rnte -1.462020 0.009743 -150.060 <0.0000000000000002 *** dest_code8acte -1.506283 0.005709 -263.866 <0.0000000000000002 *** log(dist) -1.589102 0.001685 -942.842 <0.0000000000000002 *** -- signif. codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1 (dispersion parameter for poisson family taken to be 1) null deviance: 2750417 on 209 degrees of freedom residual deviance: 335759 on 180 degrees of freedom aic: 337818 number of fisher scoring iterations: 6 box 3: outputs from the doubly constrained model specified in equation 21 56 dennett australian population studies 2 (2) 2018 figure 7: alternative distance decay curves for alternative values of 𝜷𝜷 and different functions in this particular example with these parameters and 𝛽𝛽 values, the inverse power function has a far more rapid distance decay effect than the negative exponential function. in real life, what this means is that if the observed interactions drop off very rapidly with distance, then they might be more likely to follow an inverse power law. this might be the case when looking at trips to the local convenience store by walking, for example. on the other hand, if the effect of distance is less severe – for example, migration across the country for a new job – then the negative exponential function with a small value of 𝛽𝛽 function might be more appropriate. there is no hard and fast rule as to which function to pick. it will just come down to which fits the data better. following oshan’s (2016) example, the accompanying walk-through exercise will allow you to explore the effect on model fits and predictions of fitting different distance decay functions to the distance variable. the final point to note is that the regression modelling framework means that adding additional explanatory variables into the spatial interaction model is very easy compared with the multiplicative framework. in the accompanying walk-through guide, in addition to variables relating to median income, there are variables on unemployment rate and the percentage of households living in rented accommodation. experiment with these variables for origins and destinations to see whether the singly constrained models can be improved in any way. 3. conclusions spatial interaction modelling is one of the key tools in the population geographer’s tool kit, but for too long has been inaccessible to researchers new to the field or without computer programming expertise. recent advances in more forgiving software environments like r and python now mean that with much less (although admittedly still some) effort, this powerful modelling tool can be accessed by more people. this guide has been designed to introduce researchers to spatial interaction modelling in, hopefully, an accessible way through exemplification of two modelling frameworks – the wilson-esque multiplicative framework and the poisson regression additive modelling framework – and an accompanying walk-through guide. https://rpubs.com/adam_dennett/376877 http://rpubs.com/adam_dennett/376877 http://rpubs.com/adam_dennett/376877 australian population studies 2 (2) 2018 dennett 57 acknowledgements thanks to hadrien salat in casa who helped with some earlier technical errors in the modelling. references abel g j (2010) estimation of international migration flow tables in europe: international migration flow tables. journal of the royal statistical society. series a, statistics in society 173(4): 797– 825. congdon p (1993) approaches to modelling overdispersion in the analysis of migration. environment and planning a: economy and space 25(1): 1481–1510. congdon p (1988) modelling migration flows between areas: an analysis for london using the census and opcs longitudinal study. regional studies 23(2): 87–103. crymble a, dennett a and hitchcock t (2017) modelling regional imbalances in english plebeian migration to late eighteenth-century london. the economic history review 71(3): 747-771. dennett a and wilson a (2013) a multi-level spatial interaction modelling framework for estimating inter-regional migration in europe. environment and planning a: economy and space 45(6): 1491–1507. erhardt g d and dennett a (2017) understanding the role and relevance of the census in a changing transportation data landscape. paper presented at the transportation research board conference on applying census data for transportation, kansas city, missouri. flowerdew r (2010) modelling migration with poisson regression. in: stillwell j, duke-williams o and dennett a (eds) technologies for migration and commuting analysis: spatial interaction data applications. hershey pa: igi global. flowerdew r (1982) fitting the lognormal gravity model to heteroscedastic data. geographical analysis 14(3): 263–267. flowerdew r and aitkin m (1982) a method of fitting the gravity model based on the poisson distribution. journal of regional science 22(2): 191–202. fotheringham a s (1983) a new set of spatial-interaction models: the theory of competing destinations. environment and planning a: economy and space 15(1): 15–36. fotheringham a s, nakaya t, yano k, openshaw s and ishikawa y (2001) hierarchical destination choice and spatial interaction modelling: a simulation experiment. environment and planning a: economy and space 33(5): 901–920. kim k and cohen j e (2010) determinants of international migration flows to and from industrialized countries: a panel data approach beyond gravity. international migration review 44(4): 899– 932. lee e s (1966) a theory of migration. demography 3(1): 47–57. lomax n and norman p (2016) estimating population attribute values in a table: “get me started in” iterative proportional fitting. the professional geographer 68(3): 451–461. lovelace r (2015) estimating distance decay for the national propensity to cycle tool. https://www.slideshare.net/itsleeds/estimating-distance-decay-for-the-national-propensityto-cycle-tool. oshan t m (2016) a primer for working with the spatial interaction modeling (spint) module in the python spatial analysis library (pysal). region 3: r11–r23. pooler j (1994) an extended family of spatial interaction models. progress in human geography 18(1): 17–39. https://www.slideshare.net/itsleeds/estimating-distance-decay-for-the-national-propensity-to-cycle-tool https://www.slideshare.net/itsleeds/estimating-distance-decay-for-the-national-propensity-to-cycle-tool 58 dennett australian population studies 2 (2) 2018 pooler j (1987) modeling interprovincial migration using entropy-maximizing methods. the canadian geographer 31(1): 57–64. raymer j (2007) the estimation of international migration flows: a general technique focused on the origin–destination association structure. environment and planning a: economy and space 39(4): 985–995. raymer j and abel g (2008) methods to improve estimates of migration flows – the mimosa model for estimating international migration flows in the european union. unece/eurostat work session on migration statistics, working paper 8, geneva. raymer j, abel g and smith p w (2007) combining census and registration data to estimate detailed elderly migration flows in england and wales. journal of the royal statistical society. series a, statistics in society 170(4): 891–908. raymer j and giulietti c (2010) analysing structures of interregional migration in england. in: stillwell j, duke-williams o and dennett a (eds) technologies for migration and commuting analysis: spatial interaction data applications. hershey pa: igi global. rees p (1977) the measurement of migration from census data and other sources. environment and planning a: economy and space 9(3): 257–280. rogers a and raymer j (1998) the spatial focus of us interstate migration flows. population, space and place 4(1): 63–80. senior m l (1979) from gravity modelling to entropy maximizing: a pedagogic guide. progress in human geography 3(2): 175–210. shen j (2017) modelling interregional migration in china in 2005–2010: the roles of regional attributes and spatial interaction effects in modelling error. population, space and place 23(3): e2014. shen j (2015) explaining interregional migration changes in china, 1985–2000, using a decomposition approach. regional studies 49(7): 1176–1192. stillwell j (1978) interzonal migration: some historical tests of spatial-interaction models. environment and planning a: economy and space 10(1): 1187–1200. taylor p j (1983) distance decay in spatial interactions. norwich: geo books. willekens f (1999) modeling approaches to the indirect estimation of migration flows: from entropy to em. mathematical population studies 7(3): 239–278. wilson a (1971) a family of spatial interaction models, and associated developments. environment and planning a: economy and space 3(1): 1–32. zipf g k (1946) the p1 p2 / d hypothesis: on the intercity movement of persons. american sociological review 11(6): 677–686. australian population studies 2019 | volume 3 | issue 2 | pages 34-36 © lazzari 2019. published under the creative commons attribution-noncommercial licence 3.0 australia (cc by-nc 3.0 au). journal website: www.australianpopulationstudies.org socio-economic changes in the age-patterns of childbearing in australia ester lazzari* australian national university * corresponding author. email ester.lazzari@anu.edu.au. address: school of demography, anu college of arts and social sciences, the australian national university, 9 fellows road, acton, act 2601, australia paper received 1 august 2019; accepted 17 october 2019; published 18 november 2019 over the past three decades the median age of mothers in australia has continued to rise, although age-patterns of childbearing vary widely between poorer and wealthier areas. using data on the number of children ever born to women aged 15 years and over sourced from the 2006 and 2016 censuses, this demographic examines changes in the timing and quantum of childbearing of australian mothers across different socio-economic statuses. in developed countries, fertility has been progressively decreasing at younger ages and increasing at older ages, a phenomenon known as postponement (kohler et al. 2002). despite the overall shift towards later childbearing, significant variation in the median age of mothers can be found across different socio-economic groups. there is a vast literature showing that the trend towards postponed childbearing primarily occurs among women with high educational attainment (heck et al. 1997). not only are highly educated women likely to start their families later because of their pursuit of education and career (kaizer et al. 2007), but they also tend to have fewer children due to the natural decline in fertility with age (steiner and jukic 2016). the 2016 census was used to quantify the number of babies ever born to women aged 15 years and over by the statistical areal level 2 (sa2) of usual residence, as defined by the australian statistical geography standard (asgs) of the abs. to analyse whether socio-economic circumstances have an effect on the timing of childbearing, i used the index of relative socio-economic advantage and disadvantage (irsad) developed by the abs, which ranks areas in australia according to a set of variables including relative educational and occupational level (abs 2018). the index is used to group the population into five socio-economic statuses: ses1 (most disadvantaged), ses2, ses3, ses4, and ses5 (most advantaged). the index is based on census data and assigns summary values of socioeconomic conditions to each geographic area. between 2006 and 2016, the average age of all women who gave birth increased from 29.8 to 30.5 years old (aihw 2016). figure 1 displays the changes over the ten year intercensal period in the proportion of women with one, two and three children by age and socio-economic status (ses) as defined by the 2016 irsad index values. despite a general increase in the median age of australian mothers across parities, a variation by socio-economic status can be observed, with mothers from wealthier areas experiencing substantially later ages at first, second and third birth compared to d em og ra ph ic http://www.australianpopulationstudies.org/ mailto:ester.lazzari@anu.edu.au australian population studies 3 (2) 2019 lazzari 35 figure 1: cumulative difference in the proportion of children ever born between 2006 and 2016 by socioeconomic status and parity source: calculated by the author using data extracted from the 2006 and 2016 censuses using tablebuilder women from poorer areas. at the ses5 level, the gap in the proportion of first-time mothers between 2006 and 2016 was at its greatest at the age of 26 (9.3 per cent). a recuperation trend can be observed after age 26, with the proportion of mothers slightly increasing until age 36, when 4.5 per cent more women were first-time mothers than in 2006. similar patterns can be observed at parity 2 and 3, even though recuperation at parity 3 is rather modest and the proportion of third-time mothers never surpasses that of 2006. by contrast, at the ses1 level, the peak of first-time births was reached at the age of 22 at parity 1, 25 at parity 2 and 30 at parity 3, showing an anticipation rather than a postponement in childbearing compared to the benchmark year. 36 lazzari australian population studies 3 (2) 2019 socio-economic circumstances seem to have different impacts on the probability of entry into motherhood in different periods of life. the trend of childbearing postponement differs across socioeconomic groups and the increase in the age at birth of australian mothers is driven by women with high socio-economic status. low socio-economic status was associated with entry into motherhood at younger ages, while high socio-economic status was associated with childbearing postponement and increased fertility at older ages. interestingly, fertility was lower among the most disadvantaged groups, suggesting that socio-economic circumstances affect both timing and quantum of fertility. by the end of the reproductive years, there were 2.5 per cent more first-time mothers and 9.2 more second-time mothers in the ses5 than in the ses1, while only a slightly higher proportion of women in the ses1 gave birth to a third child by the age of 49 (1.8 per cent). the decrease in the total fertility rate from 1.87 in 2006 to 1.79 in 2016 (abs 2017) seems to be driven by women with low socioeconomic status, which despite having their first child relatively young, have fewer children overall than women in higher ses groups. family policies tend to mainly favour working women and, hence, it has been argued that women in higher socio-economic statuses account for these observed fertility patterns (mcdonald and moyle 2019). references abs (2017) births, australia, 2016. catalogue no. 3301.0. canberra: abs. abs (2018) census of population and housing: socio-economic indexes for areas (seifa), australia, 2016. catalogue no. 2033.0.55.001. canberra: abs. australian institute of health and welfare (2019) australia’s mothers and babies 2017 in brief. perinatal statistics series 35. canberra: aihw. heck k, schoendorf k c, ventura s j and kiely j (1997) delayed childbearing by education level in the united states, 1969–1994. maternal and child health journal 1 (2): 81-88. kaizer r, dykstra p a and jansen m d (2007) pathways into childlessness: evidence of gendered life course dynamics. journal of biosocial science 40 (6): 863-878. kohler h-p, billari f c and ortega j a (2002) the emergence of lowest-low fertility in europe during the 1990s. population and development review 28(4): 641-680. mcdonald p and moyle h (2019) in australia fertility is falling only for low educated women. n-iussp. http://www.niussp.org/article/in-australia-fertility-is-falling-only-for-low-educated-womenenaustralie-la-fecondite-baisse-uniquement-chez-les-femmes-peu-scolarisees/. accessed on 4 july 2019. steiner a z and jukic a m z (2016) impact of female age and nulligravidity on fecundity in older reproductive age cohort. fertility and sterility 105(6): 1584-1588. http://www.niussp.org/article/in-australia-fertility-is-falling-only-for-low-educated-womenen-australie-la-fecondite-baisse-uniquement-chez-les-femmes-peu-scolarisees/ http://www.niussp.org/article/in-australia-fertility-is-falling-only-for-low-educated-womenen-australie-la-fecondite-baisse-uniquement-chez-les-femmes-peu-scolarisees/ australian population studies 2021 | volume 5 | issue 1 | pages 3-8 © charles-edwards 2021. published under the creative commons attribution-noncommercial licence 3.0 australia (cc by-nc 3.0 au). journal website: www.australianpopulationstudies.org commentary living in more than one place: capturing dual-local lives in the 2026 census elin charles-edwards the university of queensland email: e.charles-edwards@uq.edu.au. address: queensland centre for population research, school of earth and environmental sciences, the university of queensland, brisbane, australia 4072. paper received 5 march 2021; accepted 20 april 2021; published 31 may 2021 1. introduction the concept of “usual residence” is central to the collection of population data in the australian census of population and housing and the basis of australia’s official population statistic, the estimated resident population. usual residence is defined in the census as “...the address at which a person lives or intends to live for six months or more” (abs 2017). growth in spatial mobility and a “pluralisation” of living arrangements makes that the notion of a single usual residence problematic for a growing share of the population (schier et al. 2015). this includes second (vacation) homeowners (atkinson et al. 2009), long-distance commuters (including fly-in fly-out workers), children living in shared custody arrangements (stjernström and strömgren 2012), and couples “living apart together” (reimondos, evans, and gray 2011). few data sources capture these populations in australia, and available data lack detailed socioeconomic attributes and reliable spatial information which are critical for policy and planning. there are myriad applications of such data, ranging from the development of service population estimates (mckenzie, martin, and paris 2008) and emergency preparedness, to better understanding of residential multi-locality (schier 2016) and regional population dynamics (adamiak et al. 2017). there are longstanding calls for the inclusion of a census question on dual or second residence (abs 2007). a question on second residence was researched and tested for 2016 but not selected as it was “…difficult for respondents to understand” (abs 2018) as well as having a significant cost impost. despite renewed calls for inclusion in the 2021 census, a question on second residence was again rejected. public submissions for the 2021 census also called for information on children living in multiple residences as part of shared care arrangements (abs 2018), a form of dual-local living, but was not selected for inclusion. while there are certainly barriers to the inclusion of questions on dual residence including respondent burden, recall issues and cost, calls for better information on dual-local individuals and households are likely to persist and grow over time. http://www.australianpopulationstudies.org/ mailto:e.charles-edwards@uq.edu.au 4 charles-edwards australian population studies 5 (1) 2021 2. proposed census questions given the reticence of the abs to include a question on dual residence in the census, it is illustrative to turn to other jurisdictions. second or dual residence data can be collected directly via a targeted census question or indirectly as a by-product of the census enumeration strategy. the england and wales censuses have included questions on dual residence since 2011. respondents are first asked, “do you stay at another address for more than 30 days a year?”. if respondents answered “yes” they were prompted to enter the full uk residential address or the country if the second residence was abroad. this was followed by a question asking respondents to nominate the type of residence from the following options: armed forces base address; another address when working away from home; student’s home address; student’s term-time address; another parent or guardian’s address; holiday home; other. the question has been retained for the 2021 census with only the addition of “partner’s house” as a response for the type of residence. the formulation used by the england and wales censuses allows the identification of a range of dual-local arrangements including second homes, long-distance commuting, couples living apart together and children in shared custody arrangements. a question on dual residence was included in the 2001 italian census. respondents were asked to “indicate whether during the past 12 months (21 october 2000 21 october 2001) the person lived in accommodation or institutional household (e.g., relatives or friends house, barracks, hospital) other than the present”. respondents were then asked to nominate the number of days from a closed-response set. persons living at another address for more than 90 days were asked to indicate if the person was currently absent, followed by the location of the accommodation (in this municipality; in another municipality; and abroad). a final question asked the reasons for absence. pre-coded responses included: work; study; vacation; presence of relatives; vacation; previous usual residence; other. the formulation used in the 2001 italian census produced data on dual-local living arrangements tied to work, study, family and leisure. this question was not retained in the 2011 italian census. the 2020 united states census collected information on second homes (indirectly) as a by-product of its enumeration strategy. individuals were directed to complete the census at their primary residence; however, homeowners were asked to complete a form for all properties owned and had the option of nominating these properties as seasonal homes. us census data on second homes are available back to the 1940s but are limited to the number and distribution of second homes across us states (us census bureau 2017). information on the utilisation of seasonal homes, including the duration and purpose of use, is not available. following the example of the england and wales census and the 2011 italian census, the following suite of questions is proposed for the 2026 census in australia (figure 1). the proposed questions follow the same format as the england and wales censuses. this is due to their simplicity and low respondent burden compared with the italian formulation. an indirect data item, such as collected in the us, is rejected due to its inability to enumerate dual-local households, such as couples living apart together and fly-in fly-out arrangements. it is proposed that the 30-day threshold is replaced by the term “regularly” to reduce recall burden for respondents. supplementary information is, however, required to provide respondents with the definition of “regular”, here proposed as australian population studies 5 (1) 2021 charles-edwards 5 “recurring at uniform or fixed intervals e.g., weekly, monthly or seasonally”. respondents regularly living at more than two addresses should be instructed to select only one. the response set for question 2 is shorter than the england and wales equivalent, and aims to capture the main forms of dual residence living in australia including long distance commuting, children in shared custody, couples living apart together and second homeowners. [1] in the past year, have you regularly stayed at another address? [ ] no [ ] yes, write in other australian address below (address write-in) or [ ] yes, outside australia , write in country [2] what is that address? [ ] another address when working away from home [ ] parent or guardian's address [ ] a partner’s address [ ] holiday home [ ] other figure 1: proposed census questions on dual residence source: author 3. discussion census statistics on dual residence have several potential applications. in this section, three are described alongside a brief discussion of past research and existing data sources. these are: understanding the impacts of second home communities; the intensity, pattern and composition of long-distance commuting (ldc); and the distribution and characteristics of couples living apart together (lat). second or vacation homes are a significant feature of regional landscapes in australia and overseas. while there is extensive literature on second homes, the vast majority of papers are focused on europe and north america. this literature points to the varied social (gallent 2014), economic (hilber and schöni 2020; velvin et al. 2013) and environmental (hiltunen 2007; næss et al. 2019) impacts of second homes. besides improving our understanding of these varied impacts, accurate statistics on second homes would be useful for planning and provision of goods and services and emergency preparedness as part of service population estimates (mckenzie, martin, and paris 2008; charlesedwards, bell, and brown 2008). several australian sources provide partial insight into the number and distribution of second homes in australia, including census counts of unoccupied dwellings, the national visitor survey (nvs), and the household, income and labour dynamics in australia (hilda) survey. at the 2016 census, 11 per cent of private dwellings were unoccupied, up from 10 per cent in 2011. over fifty per cent of private dwellings were unoccupied on census night in the local government areas of central highlands, robe, queenscliffe, tasman, dandaraga, and glamorgan/spring bay (table 1). 6 charles-edwards australian population studies 5 (1) 2021 table 1: percentage of unoccupied private dwellings, selected local government areas, 2016 census local government area state % unoccupied private dwelling central highlands tasmania 61 robe south australia 58 queenscliffe victoria 54 tasma tasmania 52 dandaragan western australia 51 glamorgan/spring bay tasmania 50 source: abs tablebuilder, 2016 census this statistic includes homes for sale, new homes yet to be occupied, homes in which the residents were absent on census night as well as second homes. no information on individuals using second homes is collected and thus the utility of census data is limited to identifying (potential) second home locations. data have, however, been used for this purpose to good effect (see e.g., frost 2004). the nvs collects data on trips and visitors to second homes (recorded as “own property”) by australians aged 15 years and over and includes information on trip activities as well as limited demographic data. in 2019, the nvs recorded almost 5 million overnight trips to second homes of a total of 117 million overnight trips (tourism research australia 2021). sampling variability is large at the local area level which limits the utility of these data for understanding second home impacts. hilda collects data on second home ownership, along with a rich array of sociodemographic data, but does not collect any information on the location of second homes, again limiting their utility for studying impacts at the local area level. the proposed census questions on dual residence overcome the main shortcomings of the extant data sets providing reliable information on both the spatial distribution of second homes at the small area level as well as detailed information on the attributes of individuals visiting them. fly-in fly-out (fifo) and drive-in drive-out (dido) arrangements in the mining and construction sectors has disrupted the traditional relationship between place of usual residence and place of work in many parts of australia (see the article by haslam mckenzie in this issue). fifo arrangements are just one form of long-distance commuting (ldc), an arrangement whereby people live beyond daily commuting distance from their workplace, necessitating several nights stay near their place of work before returning “home”. several studies have attempted to estimate the intensity (kpmg for the minerals council of australia 2013), geography (nicholas and welters 2016), and characteristics (de silva, johnson, and wade 2011) of ldc by cross-classifying a combination of census data on place of usual residence, place of enumeration and place of work and then applying some distance threshold that precludes daily commuting. estimates of the extent of ldc vary depending on the classification method and distance thresholds used (nicholas and welters 2016). more accurate data on ldc is important to understand the social and economic impacts of ldc on individuals, families and communities. this may become increasingly important in the post-covid-19 era with increased acceptance of working from home potentially facilitating more ldc arrangements. a final application of the proposed question on dual residences relates to our understanding of couples living apart together (lat). in recent decades, lat has emerged as an important new household form, with estimates suggesting it involves around 10 per cent of the adult population in australian population studies 5 (1) 2021 charles-edwards 7 western europe, north america and australasia (duncan et al. 2014). relatively little research on lat has been undertaken in australia (for exceptions see tai et al. 2014 and reimondos et al. 2011). research from overseas suggests that the prevalence of these relationships is likely to increase over time, particularly among older adults (benson and coleman 2016). growth in lat relationships has the potential to impact average household size and dwelling demand (reuschke 2010), as well as residential mobility (wagner and mulder 2015). while there are some survey data on lat in australia (e.g., hilda), little remains known about the geographic distribution of these relationship and their impacts on urban form and spatial mobility. the three applications described above highlight the growing complexity of living arrangements in australia as well as the sustained efforts of researchers and policymakers to understand the intensity, composition and geography of dual-local arrangements and their impacts. the increase in dual-local individuals presents a challenge to traditional population statistics which are founded on the notion of a single place of usual residence. the growth in personal mobility, improvements in telecommunication technology and continued dismantling of traditional societal norms has the potential to increase the prevalence of such arrangements. statistics on dual resident individuals and households are therefore likely to become more important over time. references abs (2007) 2007.0 information paper: census of population and housing: abs views on content and procedures, 2011. https://www.abs.gov.au/ausstats/abs@.nsf/latestproducts/53a690e9f185e186ca25737f00166a 7f?opendocument. accessed on 28 february 2020. abs (2017) 2900.0 census of population and housing: understanding the census and census data, australia, 2016. https://www.abs.gov.au/ausstats/abs@.nsf/lookup/2900.0main+features62016. accessed on 28 february 2020. abs (2018) 2007.0.55.001 census of population and housing: topic directions, 2021. https://www.abs.gov.au/ausstats/abs@.nsf/lookup/by subject/2007.0.55.001~2021~main features~topic directions~6. accessed on 28 february 2020. adamiak c, pitkänen k and lehtonen o (2017) seasonal residence and counterurbanization: the role of second homes in population redistribution in finland. geojournal 82: 1035-50. https://doi.org/10.1007/s10708-016-9727-x atkinson r, picken f and tranter b (2009) home and away from home: the urban-regional dynamics of second home ownership in australia. urban research & practice 2: 1-17. https://doi.org/10.1080/17535060902727009 benson j and coleman m (2016) older adults developing a preference for living apart together. journal of marriage and family 78: 797-812. https://doi.org/10.1111/jomf.12292 charles-edwards e, bell m and brown d (2008) where people move and when: temporary population mobility in australia. people and place 16(1):21-30. de silva h, johnson l and wade k (2011) long distance commuters in australia: a socio-economic and demographic profile. australasian transport research forum 201, adelaide. https://www.bitre.gov.au/sites/default/files/sp_005_desilva_johnson_wade.pdf duncan s, phillips m, carter j, roseneil s and stoilova m (2014) practices and perceptions of living apart together. family science 5: 1-10. https://doi.org/10.1080/19424620.2014.927382 https://www.abs.gov.au/ausstats/abs@.nsf/latestproducts/53a690e9f185e186ca25737f00166a7f?opendocument https://www.abs.gov.au/ausstats/abs@.nsf/latestproducts/53a690e9f185e186ca25737f00166a7f?opendocument https://www.abs.gov.au/ausstats/abs@.nsf/lookup/2900.0main+features62016 https://www.abs.gov.au/ausstats/abs@.nsf/lookup/by%20subject/2007.0.55.001~2021~main%20features~topic%20directions~6 https://www.abs.gov.au/ausstats/abs@.nsf/lookup/by%20subject/2007.0.55.001~2021~main%20features~topic%20directions~6 https://doi.org/10.1007/s10708-016-9727-x https://doi.org/10.1080/17535060902727009 https://doi.org/10.1111/jomf.12292 https://www.bitre.gov.au/sites/default/files/sp_005_desilva_johnson_wade.pdf https://doi.org/10.1080/19424620.2014.927382 8 charles-edwards australian population studies 5 (1) 2021 frost w (2004) a hidden giant: second homes and coastal tourism in south-eastern australia. in hall cm and müller d (eds.) tourism, mobility and second homes: between elite landscape and common ground. bristol: channel view publications; 162-173. gallent n (2014) the social value of second homes in rural communities. housing, theory and society 31: 174-191. https://doi.org/10.1080/14036096.2013.830986 hilber c and schöni o (2020) on the economic impacts of constraining second home investments. journal of urban economics 118: 103266. https://doi.org/10.1016/j.jue.2020.103266 hiltunen m (2007) environmental impacts of rural second home tourism – case lake district in finland. scandinavian journal of hospitality and tourism 7: 243-65. https://doi.org/10.1080/15022250701312335 kpmg for the minerals council of australia (2013) analysis of the long distance commuter workforce across australia. canberra: minerals council of australia. mckenzie f, martin j, and paris c (2008) fiscal policy and mobility: the impact of multiple residences on the provision of place-based service funding. australasian journal of regional studies 14: 53-71. https://www.anzrsai.org/assets/uploads/publicationchapter/245-mckenzieetal.pdf næss p, xue j, stefansdottir h, steffansen r and richardson t (2019) second home mobility, climate impacts and travel modes: can sustainability obstacles be overcome? journal of transport geography 79: 102468. https://doi.org/10.1016/j.jtrangeo.2019.102468 nicholas c and welters r (2016) exploring determinants of the extent of long distance commuting in australia: accounting for space. australian geographer 47: 103-120. https://doi.org/10.1080/00049182.2015.1090300 reimondos a, evans a and gray e (2011) living-apart-together (lat) relationships in australia. family matters 87: 43-55. https://aifs.gov.au/publications/family-matters/issue-87/living-aparttogether-lat-relationships-australia reuschke d (2010) living apart together over long distances — time-space patterns and consequences of a late-modern living arrangement. erdkunde 64: 215-226. https://doi.org/10.3112/erdkunde.2010.03.01 schier m, hilti n, schad h, tippel c, dittrich-wesbuer a and monz a (2015) residential multi-locality studies the added value for research on families and second homes. tijdschrift voor economische en sociale geografie 106: 439-452. https://doi.org/10.1111/tesg.12155 stjernström o and strömgren m (2012) geographical distance between children and absent parents in separated families. geografiska annaler. series b, human geography 94: 239-253. https://doi.org/10.1111/j.1468-0467.2012.00412.x tai t, baxter j and hewitt b (2014) do co-residence and intentions make a difference? relationship satisfaction in married, cohabiting, and living apart together couples in four countries. demographic research 31: 71-104. https://doi.org/10.4054/demres.2014.31.3 tourism research australia (2021) tra online. https://www.tra.gov.au/tra-online/tra-online. accessed on 16 february 2020. us census bureau (2017) historical census of housing tables: vacation homes https://www.census.gov/data/tables/time-series/dec/coh-vacation.html. accessed on 16 february 2020. velvin j, martin kvikstad t, drag e and krogh e (2013) the impact of second home tourism on local economic development in rural areas in norway. tourism economics 19(3): 689-705. https://doi.org/10.5367/te.2013.0216 wagner m and mulder c (2015) spatial mobility, family dynamics, and housing transitions. kzfss kölner zeitschrift für soziologie und sozialpsychologie 67: 111-135. http://doi.org/10.1007/s11577-0150327-4 https://doi.org/10.1080/14036096.2013.830986 https://doi.org/10.1016/j.jue.2020.103266 https://doi.org/10.1080/15022250701312335 https://www.anzrsai.org/assets/uploads/publicationchapter/245-mckenzieetal.pdf https://doi.org/10.1016/j.jtrangeo.2019.102468 https://doi.org/10.1080/00049182.2015.1090300 https://aifs.gov.au/publications/family-matters/issue-87/living-apart-together-lat-relationships-australia https://aifs.gov.au/publications/family-matters/issue-87/living-apart-together-lat-relationships-australia https://doi.org/10.3112/erdkunde.2010.03.01 https://doi.org/10.1111/tesg.12155 https://doi.org/10.1111/j.1468-0467.2012.00412.x https://doi.org/10.4054/demres.2014.31.3 https://www.tra.gov.au/tra-online/tra-online https://www.census.gov/data/tables/time-series/dec/coh-vacation.html https://doi.org/10.5367/te.2013.0216 http://doi.org/10.1007/s11577-015-0327-4 http://doi.org/10.1007/s11577-015-0327-4 australian population studies 2019 | volume 3 | issue 1 | pages 40-56 © wilson 2019. published under the creative commons attribution-noncommercial licence 3.0 australia (cc by-nc 3.0 au). journal website: www.australianpopulationstudies.org introductory guide an introduction to population projections for australia tom wilson* charles darwin university * email: tom.wilson@cdu.edu.au. address: northern institute, charles darwin university, darwin, nt 0909 paper received 30 january 2019; accepted 19 april 2019; published 27 may 2019 abstract background population projections for australia are produced by many organisations. they differ in projected population numbers, methods used, level of output detail, temporal extent, frequency of revision, quality and purpose, and they are not always easy to find. aims this paper provides a brief guide to many of the population projections prepared for australia in recent years. it gives an overview of projection methods and selected results, a brief commentary on key aspects of the projections, and shows readers where to find more data and information. data and methods projections data were obtained from the various organisations producing projections. they are presented in order of spatial detail: national scale, states and territories; large sub-state regions; and then local and small areas. results the abs and state and territory governments are the main producers of population projections and forecasts in australia, and generally these projections are good quality. they cover a wide variety of spatial scales from the national level to local areas, such as sa2s. a great deal of projections data and information is now freely available online. conclusions population projections and forecasts can be very useful data for a wide variety of planning, policy and research purposes. but it is important to be aware of their limitations. key words population projections; population forecasts; projections data; australia http://www.australianpopulationstudies.org/ mailto:tom.wilson@cdu.edu.au australian population studies 3 (1) 2019 wilson 41 1. introduction population projections are calculations of population beyond the latest population estimate based on assumptions about the future of fertility, mortality and migration. projected populations, and sometimes also projected births, deaths and migration, are used in a wide variety of planning, policy development, service delivery, market assessment, and research activities, as well as in public debates about australia’s future. for example:  state and territory governments formulate urban and regional plans which take into account the future demand for housing, schools, hospitals, public transport, and so on;  utility companies wish to plan for future electricity, water and sewage treatment needs;  the commonwealth government uses projections to assess the likely costs of future health provision and aged pensions (as shown in its intergenerational reports, e.g. australian treasury 2015);  the australian electoral commission uses short-term projections of enrolments to vote to adjust electoral boundaries to ensure constituencies are likely to contain roughly equal numbers of electors (e.g. australian electoral commission 2017). projections are also sometimes used as interim population estimates for the present because official estimated resident populations (erps) are published some time after their reference dates (between around 6 and 15 months depending on geographical scale and level of demographic detail). population projections for australia are produced by many organisations. they differ in projected population numbers, methods used, level of output detail, temporal extent, frequency of revision, quality, and purpose. and they are not always easy to find on the web. the aim of this paper is to provide a brief guide to many of the population projections prepared for australia in recent years. it gives an overview of methods employed, data outputs, where projections data and accompanying information can be obtained, along with a few comments about the projections. national projections are covered first, followed by those for states and territories, major sub-state regions, and then local and small areas. details of the projections are not presented and interested readers are encouraged to look up the web resources listed in the paper to find out more. projections of other demographic variables, such as households, living arrangements, dwellings, populations by indigenous status, school enrolments and so on, are outside the scope of the paper. be aware that all projections mentioned in the paper (along with urls) refer to those publically available as of 30th april 2019. projections are revised regularly and may have been updated since the time this article was finalised. before proceeding any further i define the meanings of some common projections terms such as projection horizon, jump-off year, and the difference between a projection and a forecast. these are listed in box 1. 42 wilson australian population studies 3 (1) 2019 box 1: projections terminology jump-off year: the starting year of population projections; the year in which they ‘jump off’ into the demographic future. it is sometimes also known as the launch year. population estimate: the population for a past or current point in time derived from demographic information available for that time. population projection: a calculation of population beyond the jump-off year based on certain assumptions about the future of fertility, mortality and migration (and sometimes other variables). projection series: a projection with a specified combination of fertility, mortality and migration assumptions (often labelled low, medium, or high; or sometimes labelled with letters and numbers). projection scenario: a projection based on a vision, story, or set of particular circumstances (e.g. restrictive immigration policy scenario; major new employer scenario; mine closure scenario). population forecast: a population projection deemed to be the most likely future population; a prediction. (therefore all forecasts are projections, but many projections are not forecasts). projection/forecast horizon: the time period between the jump-off year and the final year of the projections/forecasts. projection assumptions: data inputs to the projection calculations which usually include the assumed future of fertility, mortality, and migration (and sometimes other input variables, e.g. housing developments). if more broadly defined, projection assumptions also implicitly or explicitly include qualitative elements, such as expectations about future immigration policy, and the absence of major disasters. probabilistic projections/forecasts: projections/forecasts which are expressed as a range of future populations, often with prediction intervals. prediction intervals consist of a range with a probability of future population lying within that range. deterministic projections/forecasts: projections/forecasts which are expressed as a single set of values; most national projections and almost all subnational projections are of this type. net migration: the value of inward migration minus outward migration, such as net interstate migration (nim) which is interstate in-migration minus interstate out-migration, or net overseas migration (nom) which is immigration (which the australian bureau of statistics (abs) terms nom arrivals) minus emigration (nom departures). directional migration: flows of migration from one country or region to another; for example, migration from overseas to australia (immigration), or from new south wales to other jurisdictions within australia (interstate out-migration). source: the author australian population studies 3 (1) 2019 wilson 43 2. population projections for australia 2.1. national population projections the best known population projections for australia are those produced by the australian bureau of statistics (abs) in their population projections, australia publication. updated every five years, the most recent abs projections were released in november 2018 (abs 2018a). these projections start from a jump-off year of 2017 and cover a projection horizon of almost 50 years, ending in 2066. the abs applies a standard cohort-component model1 incorporating directional migration flows (immigration and emigration) constrained to total net overseas migration (nom) assumptions (abs 1999). most emphasis is placed on three main projection series – a, b and c – which are often interpreted as high, medium and low series. a further 69 projection series with differing fertility, mortality and migration assumptions are also prepared and are available for download from the abs website. series b is generally interpreted by users to be a forecast, though abs deliberately avoids the term ‘forecast’ and emphasises that all its future population numbers are merely projections. the series b projection sees australia’s population increasing from 24.6 million in 2017 to 42.6 million by 2066, the result of a long-run total fertility rate of 1.80, net overseas migration of 225,000 per annum, and modest and decelerating increases in life expectancy at birth (86.0 years for females and 83.0 years for males by the end of the projection horizon). the projected population growth rate gradually declines into the future because population ageing results in the number of deaths rising slightly faster than births, and because a fixed level of net overseas migration in the context of an increasing population represents a falling crude rate of net migration. the projection assumptions are based primarily on recent levels of fertility and net overseas migration observed over the last few years, and the small increases in life expectancy observed in recent years. these trends are assumed to continue. interestingly, life expectancy at birth is expected to rise by very little and is less optimistic than in the previous set of abs projections (abs 2013), and by the 2060s is over 2 years of life lower. some users may wish to download projection series 26 which incorporates the abs high life expectancy assumptions (increasing to 87.7 years and 89.2 years for males and females respectively by 2065-66). the series 26 projection is 1.1 million higher than series b by 2066, nearly all of which is in the 65+ age group. figure 1 illustrates the latest abs series b projection alongside several others for australia, while figure 2 presents some of the assumptions on which those projections were based. box 2 includes urls which will take readers to these and other projections mentioned in the paper. the united nations population division updates population projections for australia every two years as part of its world population prospects publication (un 2017a). this publication contains projections for all countries of the world (except the very smallest) along with projections for the world as a whole. the projections in the 2017 revision of world population prospects start from a jump-off year of 2015 and extend all the way out to 2100. the un uses a cohort-component model with five year age groups which proceeds forward in five year intervals; overseas migration is generally projected as net migration numbers due to the lack of reliable data on immigration and 1 for basic introductions to the cohort-component model see rowland (2003) chapter 12, or smith et al. (2013) chapters 3 to 7. for a description of subnational and multiregional cohort-component models see rees (1997) and the review in rees et al. (2015). 44 wilson australian population studies 3 (1) 2019 emigration flows for all countries (un 2017b). the demographic transition model (dyson 2010) provides the over-arching theoretical framework for setting fertility and mortality assumptions, so that all countries are assumed to progress through the transition from high to low fertility and mortality rates. net overseas migration is assumed to remain constant until 2045-50 if recent numbers have been stable, and then decline to half that number by 2095-2100. this approach was taken because it “represents a compromise between the difficulty of predicting the levels of immigration or emigration for each country of the world over such a far horizon, and the recognition that net migration is unlikely to reach zero in individual countries” (un 2017b p. 30). the un’s projection assumptions for australia consist of a long-run tfr varying slightly between 1.76 and 1.80, life expectancy reaching 87.3 years for males and 90.5 years for females by 2060-65 (and 91.5 and 94.7 years respectively by 2095-2100), and net overseas migration set to an annual average of 150,000 from 2020 to 2050, declining gradually to 75,000 by 2095-2100. the un assumptions for mortality and migration differ considerably from those of the abs series b projection, with the un selecting much higher life expectancy and much lower net overseas migration levels. according to the un’s main series projection, australia’s population is projected to increase from 23.8 million in 2015 to 41.8 million by 2100. by 2066 it is expected pass 36.6 million (7 million lower than the abs series b projection). differences with abs projections are largely due to nom assumptions and are particularly noticeable in the 0-14 and 15-64 age groups (figure 1b). the un projections comprise nine variants as well as probabilistic projections created from probabilistic fertility and mortality inputs (though, interestingly, not probabilistic overseas migration inputs). outputs from, and information about, these alternative projections can be found at the webpage listed in box 2. intergenerational reports are prepared by the australian treasury every few years, with the most recent being the 2015 report (australian treasury 2015). the purpose of these reports is to “assess the long-term sustainability of current government policies and how changes to australia’s population size and age profile may impact economic growth, workforce and public finances” (australian treasury 2015 p. xxiii). the assessment is underpinned by a series of demographic and economic projections looking ahead over the next four decades. the population projections are prepared with a standard cohort-component model. treasury’s projection for australia’s population in 2055 is 39.7 million, the result of a total fertility rate of 1.90, life expectancy at birth increasing to 87.5 years for males and 90.1 years for females by 2050, and long-run net overseas migration levels of 215,000 per annum. population projections for australia are also produced by other organisations and individuals as occasional or one-off exercises. three examples are mentioned here. the productivity commission created population projections as part of the report an ageing australia (productivity commission 2013). their projections from 2012 to 2060 put the national population at 38.3 million by the end of the projection horizon. lutz et al. (2018) created projections for most countries of the world by educational status from 2015 to 2100. their medium scenario has australia’s total population at 32.5 million by 2050 and 43.6 million by 2100. bell et al. (2011) prepared probabilistic population forecasts for australia from 2010 to 2051. their forecasts included a 95% prediction interval for australia’s total population in 2051 spanning 29.4 to 43.0 million, with the median of the distribution at 36.1 million. australian population studies 3 (1) 2019 wilson 45 (a) projections of australia’s total population (b) projections of australia’s population aged 0-14, 15-64 and 65 years and over figure 1: some recent projections of australia’s population notes: erp = estimated resident population sources: abs (2018a; 2018b), un (2017a), australian treasury (2015) 46 wilson australian population studies 3 (1) 2019 (a) net overseas migration assumptions (b) life expectancy at birth assumptions (c) total fertility rate assumptions figure 2: assumptions behind some recent projections of australia’s population notes: tfr = total fertility rate; e0 = life expectancy at birth sources: abs (2014; 2018a; 2018b), human mortality database (2017), un (2017a), australian treasury (2015) australian population studies 3 (1) 2019 wilson 47 2.2. projections at the state and territory scale the abs and state and territory governments all produce population projections at the state and territory scale every few years. abs state and territory projections are published as part of its population projections, australia publication (abs 2018a). these projections have a high profile because they are published by the national statistical office. because they are produced using a consistent method, and have net interstate migration summing to the logical value of zero across all jurisdictions, the projections are useful for analyses involving all states and territories across australia. state and territory governments produce projections for their own state/territory as a whole, plus projections for a variety of sub-state geographies, with updates mostly prepared every 2 to 5 years. these projections are important because government departments within that state or territory are required to use them in their analysis, planning and budgeting. data and information on all these projections can be obtained from the urls listed in box 2. the abs projects state and territory populations with a cohort-component model which is similar to the national-scale model but with the addition of interstate migration. interstate inand outmigration flows are handled as directional (rather than net) migration flows but are constrained to be consistent with assumed net interstate migration (nim) totals. the model is not a fully multiregional model with flows between each state/territory and every other, but a bi-regional type which handles inand out-migration flows between each state/territory and the remainder of the country (abs 1999; wilson and bell 2004). in the latest set of abs projections medium variant nim assumptions are based on the average nim values of the last 10 years. series b state and territory population projections are illustrated in figure 3 alongside those produced by state and territory governments. a log scale is used in the graph to clearly show the projected trends of the less populous states and territories. all state and territory governments make use of cohort-component projection models of various types. most projection models use directional (rather than net) migration, which is conceptually better and avoids implausible outputs. some models work with five year age groups and projection intervals, though most use single year ages and intervals; some state and territory governments produce only one set of projections while others produce multiple series. medium series projections often use assumptions which deliver similar results to the abs series b projections in terms of total population, which is probably a mix of similar thinking and a desire to avoid having to explain to users why their projections are so different. the new south wales department of planning and environment produces official nsw government population projections for nsw. high, medium and low series are produced. the latest projection covers a horizon of 2011 to 2041 and the medium series puts the nsw population at 10.46 million by the end of that period (nsw department of planning and environment 2016). these projections are fractionally lower than those of the abs. the department of environment, land, water and planning in victoria produces population projections for its victoria in future publication. only one projection series is published. the latest ‘victoria in future 2016’ projections cover the period 2011 to 2051 at the state level and show victoria’s population increasing from 5.54 million to 10.09 million over those four decades (victorian 48 wilson australian population studies 3 (1) 2019 department of environment, land, water and planning 2016). these are a little lower than the abs series b projections. the queensland government statistician’s office prepares the official population projections for that state. high, medium and low series are published. its latest projections were published in 2018 and extend from 2016 to 2066. the medium series projections indicate growth from 4.85 million to 9.51 million over the projection horizon (queensland government statistician’s office 2018), showing marginally higher growth than the abs series b projection. western australia tomorrow is the publication produced by the department of planning, lands and heritage containing population forecasts for western australia (western australian planning commission 2018). these projections are unusual in that they are formally labelled ‘forecasts’ and are produced by a form of probabilistic cohort-component model. within the probabilistic forecasts five alternative projection trajectories are distinguished, labelled bands a, b, c, d and e. the latest band c (medium series) forecasts envision wa’s population growing from 2.35 million in 2011 to 3.25 million by the end of the projection horizon in 2031 (western australian planning commission 2018), which is slightly higher than abs series b projection. tasmania’s latest population projections were published recently and include high, medium and low series. the medium series show the state’s population growing from 522,000 in 2017 to 577,000 by 2067 (tasmanian department of treasury and finance 2019). the projected trajectory of tasmania’s total population is close to the latest abs series b projection. the northern territory’s projected population is calculated as the sum of separate projections for its indigenous and non-indigenous residents, a feature which is unique amongst the state and territory government population models. the cohort-component model incorporates overseas and interstate migration flows, permits changes to reported indigenous status over time, and allows the indigenous status of newly-born babies to differ from those of their mothers. the four projection series are labelled main, current, high and low. the main series from the latest projections has the population growing from 246,000 in 2016 to 352,000 by 2046; the indigenous population is expected to increase from 75,000 in 2016 to 104,000 by 2046 while the non-indigenous population is projected to grow from 171,000 to 247,000 over the same projection horizon (nt department of treasury and finance 2019). the nt’s total population is a little under the abs series b projection until the 2040s. the current australian capital territory government population projections start from a jump-off year of 2017 and extend out to 2058 (act government 2019). only one projections series was published. the projections indicate that the population will increase from 412,000 in 2017 to 703,000 by 2058, which is a very similar growth trajectory to the latest abs series b projections. australian population studies 3 (1) 2019 wilson 49 figure 3: some recent projections of state and territory populations notes: state and territory government medium/main series projections are shown. sources: abs (2018a), nsw department of planning and environment (2016), victorian department of environment, land, water and planning (2016), queensland government statistician’s office (2018), western australian planning commission (2018), tasmanian department of treasury and finance (2019), nt department of treasury and finance (2019), act government (2019). 2.3. projections for large sub-state regions the abs projections in population projections, australia include projections for greater capital city statistical areas and balance of state/territory regions. they are produced using a cohortcomponent model projecting internal migration flows in and out of each region constrained to a total net internal migration assumption. the latest series b projections suggest an increasing concentration of population in australia’s metropolitan regions over the coming decades (abs 2018a). the combined population of the capital cities is expected to grow from 16.6 million in 2017 (67.3% of the national population) to 32.1 million by 2066 (75.3%). by the end of the projection horizon sydney and melbourne are projected to be 9.7 and 10.2 million, respectively, with brisbane the next largest at 4.8 million. the greatest proportional increase, however, is projected for greater darwin (149,000 in 2017 and 333,000 in 2066). projections of metropolitan growth such as these tend to gain lots of media attention, but the demographic outcomes for other regions can look very different. the rest of south australia and rest of tasmania are projected to decline in population and experience considerable amounts of population ageing. 50 wilson australian population studies 3 (1) 2019 some state and territory government projections are published for greater capital city statistical areas as well as for other large sub-state regions. in some cases these projections are created directly, but in others they are created by building up from smaller geographical areas. table 1 lists the large sub-state regions for which projections are published in the current sets of projections. in queensland, for example, projections are published for brisbane greater capital city statistical area and all other sa4 regions of the state (queensland government statistician’s office 2018). under the medium series projections brisbane grows from 2.36 million in 2016 to 3.67 million by 2041 (which is very close to the latest abs series b projection). population growth is anticipated for all other regions of queensland with the exception of queensland – outback where the total population remains largely unchanged. table 1: state and territory government projections for large sub-state regions state/territory sub-state regions nsw projection regions, defined as aggregations of local government areas; local health districts vic major regions, defined as melbourne greater capital city statistical area and sa4 areas across the rest of the state qld brisbane greater capital city statistical area and sa4 regions across the rest of the state sa statistical divisions wa no large sub-state regions in the current projections output tas no large sub-state regions in the current projections output nt regions, defined as sa3 areas; sa4 regions act districts, defined as sa3 areas sources: see figure 3. 2.4. projections for local and small areas for many users, local and small area population projections are the most useful. they inform planning, budgeting, and service delivery at the local scale, and small area projections can also be aggregated up to a variety of custom-defined regions. in some cases they can generate strong local reactions, such as objections from a mayor when projections show population decline or little growth, or protests from local community groups about anticipated rapid growth viewed as likely to adversely affect the character and quality of life of a local neighbourhood. the abs does not produce local or small area projections as part of its official projections output (though it does prepare them as a consultancy on behalf of other organisations). the key producers of local and small area projections are state and territory governments who tend to publish local government area (lga) and/or sa2 area projections, as shown in table 2. see box 2 for urls. often local and small area projections in metropolitan regions are derived from cohort-component models linked to housing unit models (foss 2002; wilson 2011), which are based on assumptions about expected future dwelling growth. the incorporation of dwelling data in these projections can result in substantially different outputs compared to trend-based projections. for example, the populations of urban fringe areas experiencing considerable increases in the amount of residential land becoming available will not be projected accurately from trend-based models. but reasonable projections may be obtained via housing-unit models if the land development and dwelling forecasts are about right. australian population studies 3 (1) 2019 wilson 51 table 2: state and territory government projections for local and small areas state/territory areas nsw local government areas vic local government areas; victoria in future small areas (aggregations and modifications of sa2 areas which nest into lgas) qld local government areas; sa2 areas sa local government areas; statistical local areas wa sa2 areas tas local government areas nt no local or small areas in the current projections output act sa2 areas sources: see figure 3 a variety of other organisations also prepare local and small area projections from time to time. the commonwealth department of social services (2014) has published sa2 area projections by age and sex for the whole of australia. they were prepared by the abs as a consultancy and cover a projection horizon of 2012 to 2027, so are a little old now. their main advantage is the use of a common method and data inputs across the whole country. however, the migration assumptions are purely trend-based so projections for sa2 areas within metropolitan regions should be used with caution beyond just a few years into the projection horizon. this also applies to certain nonmetropolitan local areas, such as mining towns (e.g. taylor et al. 2014). the australian electoral commission obtains short-term sa1 area population projections from time to time (usually as a consultancy from the abs). they are used to calculate likely numbers of enrolled electors over the next few years for the purpose of electoral boundary redistributions. the enrolment projections are generally made available, but not the population projections on which they are based (e.g. australian electoral commission 2017). several consulting firms also produce local and small area projections. for example, id prepares projections for many local government areas which are freely available if councils place them on their websites. id also creates smaller geographical areas in its small area forecast information (safi) product (id 2018). pitney bowes produces sa1 area population projections (pitney bowes 2018). the small area projections are prepared on a consulting basis and must be paid for. 3. concluding remarks population projections and forecasts can be very useful inputs to decision-making, policy formulation, budgeting, and planning. but it is important to be aware of their limitations. strictly, population projections, as statements about future population under certain assumptions, are correct provided they have been calculated correctly. this is the case most of the time – but not always. however, population forecasts will almost certainly turn out to be in error to some extent. this can be due to a suboptimal choice of projection model, subsequent revisions to the jump-off population and historical demographic components, random noise, and deviations in actual fertility, mortality and migration from the projection assumptions. as a general rule:  the further a forecast extends into the future the greater the error;  the smaller the population the greater the error; and 52 wilson australian population studies 3 (1) 2019 box 2: selected population projections for australia and where to obtain them population projections produced by the commonwealth government population projections, australia – abs http://www.abs.gov.au/ausstats/abs@.nsf/mf/3222.0 intergenerational report – australian treasury https://treasury.gov.au/intergenerational-report/ an ageing australia – productivity commission https://www.pc.gov.au/research/completed/ageing-australia sa2 population projections – department of social services https://www.gen-agedcaredata.gov.au/resources/access-data/2014/august/population-projections,2012-(base)-to-2027-for-al population projections produced by state and territory governments nsw: population projections – department of planning & environment https://www.planning.nsw.gov.au/projections vic: victoria in future – department of environment, land, water & planning https://www.planning.vic.gov.au/land-use-and-population-research/victoria-in-future-2016 qld: population projections – queensland government statistician’s office http://www.qgso.qld.gov.au/subjects/demography/population-projections/index.php sa: department of planning, transport and infrastructure https://www.saplanningportal.sa.gov.au/data_and_research/population_projections_and_demographics wa: wa tomorrow – department of planning, lands and heritage https://www.dplh.wa.gov.au/information-and-services/land-supply-and-demography/westernaustralia-tomorrow-population-forecasts tas: population projections – department of treasury & finance https://www.treasury.tas.gov.au/economy/economic-data/2019-population-projections-for-tasmaniaand-its-local-government-areas nt: population projections – department of treasury & finance https://treasury.nt.gov.au/dtf/economic-group/population-projections act: population projections – treasury and economic development directorate https://apps.treasury.act.gov.au/snapshot/demography/act population projections produced by overseas organisations world population prospects – united nations population division https://population.un.org/wpp/ population projections by educational status – wittgenstein centre http://dataexplorer.wittgensteincentre.org/wcde-v2/ population projections produced by consulting firms many local and small area projections are produced by consulting firms. summary data may sometimes be freely available but generally these projections must be paid for. examples include: small area forecast information (safi) projections – id https://home.id.com.au/services/population-forecasting/small-area-population-forecasts-safi/ sa1 area projections – pitney bowes https://www.pitneybowes.com/au/data/demographic-data/estimates-projections-australia.html note: urls correct as of 30th april 2019. http://www.abs.gov.au/ausstats/abs@.nsf/mf/3222.0 https://treasury.gov.au/intergenerational-report/ https://www.pc.gov.au/research/completed/ageing-australia https://www.gen-agedcaredata.gov.au/resources/access-data/2014/august/population-projections,-2012-(base)-to-2027-for-al https://www.gen-agedcaredata.gov.au/resources/access-data/2014/august/population-projections,-2012-(base)-to-2027-for-al https://www.planning.nsw.gov.au/projections https://www.planning.vic.gov.au/land-use-and-population-research/victoria-in-future-2016 http://www.qgso.qld.gov.au/subjects/demography/population-projections/index.php https://www.saplanningportal.sa.gov.au/data_and_research/population_projections_and_demographics https://www.dplh.wa.gov.au/information-and-services/land-supply-and-demography/western-australia-tomorrow-population-forecasts https://www.dplh.wa.gov.au/information-and-services/land-supply-and-demography/western-australia-tomorrow-population-forecasts https://www.treasury.tas.gov.au/economy/economic-data/2019-population-projections-for-tasmania-and-its-local-government-areas https://www.treasury.tas.gov.au/economy/economic-data/2019-population-projections-for-tasmania-and-its-local-government-areas https://treasury.nt.gov.au/dtf/economic-group/population-projections https://apps.treasury.act.gov.au/snapshot/demography/act https://population.un.org/wpp/ http://dataexplorer.wittgensteincentre.org/wcde-v2/ https://home.id.com.au/services/population-forecasting/small-area-population-forecasts-safi/ https://www.pitneybowes.com/au/data/demographic-data/estimates-projections-australia.html australian population studies 3 (1) 2019 wilson 53  the greater the volatility of migration the greater the error (smith et al. 2013; wilson et al. 2018). forecast errors are often larger than many users realise. projections and forecasts can also be subject to another form of error – misinterpretation and misreporting. net overseas migration projections are regularly misinterpreted as immigration (which is only correct in the absence of emigration!). sometimes population projections are interpreted as population targets or policies (which would only be correct if a projection had been expressly created as such). for example, the reporting of australia’s population passing the 25 million milestone in 2018 was reported by some news outlets (deliberately not cited) as “33 years ahead of schedule” based on abs projections from the late 1990s, as if those abs projections were part of a national population policy. population projections are highly sensitive to input assumptions, and these can vary considerably from one projection to another (as figure 2 shows). generally assumptions do not attempt to predict trend changes or cyclical paths, but instead aim for a smooth trajectory through the middle of an undulating pattern. assumptions are sometimes based on statistical models, and sometimes on judgement. in reality they often involve a mix of art and science. in australia, overseas migration contributes the bulk of error in national population forecasts while at the state and regional scales both internal and overseas migration tend to contribute most of the error (wilson 2007, 2012). the variability of overseas migration makes it especially challenging to forecast and there is no robust mathematical model available which has proven to give accurate migration forecasts, at least in the medium and long-term (disney et al. 2015). the history of fertility forecasting is also not an unqualified success despite a large literature attempting to understand and forecast period and cohort fertility (bohk-ewald et al. 2018). mortality forecasting at the national scale has been subject to a great deal of research attention and there are an ever increasing number of sophisticated models to choose from (booth and tickle 2008; janssen 2018). generally, errors in mortality forecasts are lower than those of fertility and migration. nonetheless, mortality models are almost all extrapolative models. should a major change in mortality trends occur, as appears to be the case in the us and uk at the moment, errors could be larger than in the past. before using a set of projections, it is valuable to ask who prepared them and why. that may go some way to explaining the choice of projection assumptions and the way in which projection outputs are presented. no set of projections is ever completely objective and free of analysts’ judgments. for example, given the sensitivity about population decline in australia, projections for local areas currently experiencing population decline sometimes err on the ‘optimistic’ side by showing much less decline, or even no decline, in the future. state/territory projections may be influenced by the incumbent government’s view on whether recent population growth is too high or too low, or an explicit population policy. projections produced by consulting firms will sometimes reflect the assumptions, or aspirations, of the client. it is also sensible to conduct a quick assessment of the plausibility of any set of projections you are considering using (see also smith et al. 2013 pp. 385-390; wilson 2017). while most projections have been generated by good models and sensible assumptions, this is not universally the case. treat projections which have been produced using models incorporating net migration numbers or net migration rates with caution (rogers 1990; wilson 2016). they have the potential to generate 54 wilson australian population studies 3 (1) 2019 implausible or impossible outcomes. at the local level in growing urban regions, it is also worth checking that projections are based on anticipated housing developments. be wary of purely trendbased local area projections in such regions. projected age-sex structures should generally evolve slowly and will often maintain characteristic peaks and troughs across the peak migration age groups (late teens to late 30s), as well as continued population ageing. for example, if a local population age structure has a large peak across the late teens and early twenties due to a higher education institution then you would expect that peak to be maintained in the projections. in addition, projected total population, births, deaths and migration flows should follow on smoothly from observed trends – unless there are good reasons for such changes. the projected age pattern of sex ratios should also change gradually in the projections. in summary, this paper has provided a brief guide to australian population projections data at various geographical scales. it has introduced many (though not all) of the projections produced for australia in recent years, and included some brief comments on the limitations of projections. it is hoped that the guide proves a useful starting point for exploring population projections data and understanding the context and broader processes involved in population projections preparation. key messages  population projections are calculations of population beyond the latest population estimate based on assumptions about the future of fertility, mortality and migration.  population projections are used in planning, policy development, service delivery, market assessment, research activities, and public debate.  the abs and state and territory governments are the main producers of population projections and forecasts in australia.  a great deal of projections data and information is now freely available online (box 2).  be aware of the limitations of projections and forecasts. acknowledgements i am grateful for the helpful comments on an earlier draft of this paper by alison taylor, hitesh khanna, abs demography staff, and the anonymous reviewers. references abs (1999) population estimates: concepts, sources and methods, 2009. catalogue no. 3228.0.55.001. canberra: abs. http://www.abs.gov.au/ausstats/abs@.nsf/mf/3228.0.55.001 abs (2013) population projections, australia, 2012 (base) to 2101. catalogue no. 3222.0. canberra: abs. http://www.abs.gov.au/ausstats/abs@.nsf/mf/3222.0 abs (2014) australian historical population statistics, 2014. catalogue no. 3105.0.65.001. canberra: abs. http://www.abs.gov.au/ausstats/abs@.nsf/mf/3105.0.65.001 abs (2018a) population projections, australia, 2017 (base) – 2066. catalogue no. 3222.0. canberra: abs. http://www.abs.gov.au/ausstats/abs@.nsf/mf/3222.0 abs (2018b) australian demographic statistics, jun 2018. catalogue no. 3101.0. canberra: abs. http://www.abs.gov.au/ausstats/abs@.nsf/mf/3101.0 http://www.abs.gov.au/ausstats/abs@.nsf/mf/3228.0.55.001 http://www.abs.gov.au/ausstats/abs@.nsf/mf/3222.0 http://www.abs.gov.au/ausstats/abs@.nsf/mf/3105.0.65.001 http://www.abs.gov.au/ausstats/abs@.nsf/mf/3222.0 http://www.abs.gov.au/ausstats/abs@.nsf/mf/3101.0 australian population studies 3 (1) 2019 wilson 55 act government (2019) projections of the resident population 2019. https://apps.treasury.act.gov.au/snapshot/demography/act accessed 16 january 2019. australian electoral commission (2017) enrolment projections queensland federal redistribution. https://www.aec.gov.au/electorates/redistributions/2017/qld/enrolment-projections.htm accessed 6 jan 2019. australian treasury (2015) 2015 intergenerational report: australia in 2055. canberra: australian treasury. https://treasury.gov.au/publication/2015-intergenerational-report/. accessed 5 march 2015. bell m, wilson t and charles-edwards e (2011) australia’s population future: probabilistic forecasts incorporating expert judgement. geographical research 49(3): 261-275. bohk-ewald c, li p, and myrskyla m (2018) forecast accuracy hardly improves with method complexity when completing cohort fertility. proceedings of the national academy of sciences 115(37): 91879192. https://www.pnas.org/content/115/37/9187. accessed 30 august 2018. booth h and tickle l (2008) mortality modelling and forecasting: a review of methods. annals of actuarial science 3(12): 3-43. commonwealth department of social services (2014) population projections, 2012 (base) to 2027 for all states and territories at statistical area level 2 (sa2) by sex and age. https://www.genagedcaredata.gov.au/resources/access-data/2014/august/population-projections,-2012-(base)-to2027-for-al. accessed 29 december 2018. disney g, wiśniowski a, forster j j, smith p w f, and bijak j (2015) evaluation of existing migration forecasting methods and models. esrc centre for population change, university of southampton. https://www.gov.uk/government/publications/evaluation-of-existing-migration-forecastingmethods-and-models. accessed 6 december 2015. dyson t (2010) population and development: the demographic transition. london: zed books. foss w (2002) small area population forecasting. the appraisal journal 70(2): 163-172. human mortality database (2017) university of california, berkeley (usa), and max planck institute for demographic research (germany). www.mortality.org. accessed 11 december 2017. id (2018) hyperlocal population forecasts. https://home.id.com.au/services/population-forecasting/smallarea-population-forecasts-safi/. accessed 29 december 2018. janssen f (2018) advances in mortality forecasting: introduction. genus 74: 21. https://doi.org/10.1186/s41118-018-0045-7. accessed 17 march 2019. lutz w, goujon a, kc s, stonawski m, and stilianakis n (eds) (2018) demographic and human capital scenarios for the 21st century: 2018 assessment for 201 countries. luxembourg: publications office of the european union. https://ec.europa.eu/jrc/en/publication/demographic-and-human-capitalscenarios-21st-century-2018-assessment-201-countries. accessed 15 april 2019. nsw department of planning and environment (2016) 2016 nsw population and household projections. https://www.planning.nsw.gov.au/projections. accessed 29 december 2018. nt department of treasury and finance (2019) northern territory population projections (2019 release). https://treasury.nt.gov.au/dtf/economic-group/population-projections. accessed 30 april 2019. pitney bowes (2018) demographic estimates and projections australia. https://www.pitneybowes.com/au/data/demographic-data/estimates-projections-australia.html. accessed 29 december 2018. productivity commission (2013) an ageing australia: preparing for the future. melbourne: productivity commission. https://www.pc.gov.au/research/completed/ageing-australia. accessed 22 november 2013. queensland government statistician’s office (2018) queensland government population projections, 2018 edition. http://www.qgso.qld.gov.au/subjects/demography/populationprojections/reports/qld-govt-pop-proj/index.php. accessed 29 december 2018. https://apps.treasury.act.gov.au/snapshot/demography/act https://www.aec.gov.au/electorates/redistributions/2017/qld/enrolment-projections.htm https://treasury.gov.au/publication/2015-intergenerational-report/ https://www.pnas.org/content/115/37/9187 https://www.gen-agedcaredata.gov.au/resources/access-data/2014/august/population-projections,-2012-(base)-to-2027-for-al https://www.gen-agedcaredata.gov.au/resources/access-data/2014/august/population-projections,-2012-(base)-to-2027-for-al https://www.gen-agedcaredata.gov.au/resources/access-data/2014/august/population-projections,-2012-(base)-to-2027-for-al https://www.gov.uk/government/publications/evaluation-of-existing-migration-forecasting-methods-and-models https://www.gov.uk/government/publications/evaluation-of-existing-migration-forecasting-methods-and-models http://www.mortality.org/ https://home.id.com.au/services/population-forecasting/small-area-population-forecasts-safi/ https://home.id.com.au/services/population-forecasting/small-area-population-forecasts-safi/ https://doi.org/10.1186/s41118-018-0045-7 https://ec.europa.eu/jrc/en/publication/demographic-and-human-capital-scenarios-21st-century-2018-assessment-201-countries https://ec.europa.eu/jrc/en/publication/demographic-and-human-capital-scenarios-21st-century-2018-assessment-201-countries https://www.planning.nsw.gov.au/projections https://treasury.nt.gov.au/dtf/economic-group/population-projections https://www.pitneybowes.com/au/data/demographic-data/estimates-projections-australia.html https://www.pc.gov.au/research/completed/ageing-australia http://www.qgso.qld.gov.au/subjects/demography/population-projections/reports/qld-govt-pop-proj/index.php http://www.qgso.qld.gov.au/subjects/demography/population-projections/reports/qld-govt-pop-proj/index.php 56 wilson australian population studies 3 (1) 2019 rees p (1997) problems and solutions in forecasting geographical populations. journal of the australian population association 14(2): 145-166. rees p, wohland p, norman p and lomax n (2015) subnational projection methods for scotland and scottish areas: a review and recommendations. school of geography, university of leeds. http://eprints.whiterose.ac.uk/90125/. accessed 18 march 2018. rogers a (1990) requiem for the net migrant. geographical analysis 22(4): 283-300. rowland d (2003) demographic methods and concepts. oxford: oxford university press. smith s k, tayman j and swanson d a (2013) a practitioner’s guide to state and local population projections. dordrecht: springer. tasmanian department of treasury and finance (2019) 2019 population projections for tasmania and its local government areas. https://www.treasury.tas.gov.au/economy/economic-data/2019population-projections-for-tasmania-and-its-local-government-areas. accessed 30 april 2019. taylor a, carson d and carson d (2014) demographic modelling of the town of nhulunbuy. northern institute, charles darwin university, darwin. un (2017a) world population prospects 2017. https://population.un.org/wpp/. accessed 29 december 2018. un (2017b) world population prospects: the 2017 revision: methodology of the united nations population estimates and projections. new york: united nations. https://population.un.org/wpp/publications/files/wpp2017_methodology.pdf. accessed 29 december 2018. victorian department of environment, land, water and planning (2016) victoria in future 2016. https://www.planning.vic.gov.au/land-use-and-population-research/victoria-in-future-2016. accessed 29 december 2018. western australian planning commission (2018) western australia tomorrow population forecasts. https://www.dplh.wa.gov.au/information-and-services/land-supply-and-demography/westernaustralia-tomorrow-population-forecasts. accessed 29 december 2018. wilson t (2007) the forecast accuracy of australian bureau of statistics national population projections. journal of population research 24(1): 1-27. wilson t (2011) a review of sub-regional population projection methods. report to the office of economic and statistical research. queensland centre for population research, school of geography, planning and environmental management, the university of queensland, brisbane. wilson t (2012) forecast accuracy and uncertainty of australian bureau of statistics state and territory population projections. international journal of population research. volume 2012, article id 419824. http://dx.doi.org/10.1155/2012/419824. wilson t (2016) evaluation of alternative cohort-component models for local area population forecasts. population research & policy review 35(2): 241-261. wilson t (2017) a checklist for reviewing draft population projections. northern institute, charles darwin university. https://www.cdu.edu.au/sites/default/files/the-northern-institute/checklist.pdf. accessed 15 january 2019. wilson t and bell m (2004) comparative empirical evaluations of internal migration models in subnational population projections. journal of population research 21(2): 127-160. wilson t, brokensha h, rowe f, and simpson l (2018) insights from the evaluation of past local area population forecasts. population research and policy review 37(1): 137-155. http://eprints.whiterose.ac.uk/90125/ https://www.treasury.tas.gov.au/economy/economic-data/2019-population-projections-for-tasmania-and-its-local-government-areas https://www.treasury.tas.gov.au/economy/economic-data/2019-population-projections-for-tasmania-and-its-local-government-areas https://population.un.org/wpp/ https://population.un.org/wpp/publications/files/wpp2017_methodology.pdf https://www.planning.vic.gov.au/land-use-and-population-research/victoria-in-future-2016 https://www.dplh.wa.gov.au/information-and-services/land-supply-and-demography/western-australia-tomorrow-population-forecasts https://www.dplh.wa.gov.au/information-and-services/land-supply-and-demography/western-australia-tomorrow-population-forecasts http://dx.doi.org/10.1155/2012/419824 https://www.cdu.edu.au/sites/default/files/the-northern-institute/checklist.pdf australian population studies 2020 | volume 4 | issue 2 | pages 33-38 © kippen 2020. published under the creative commons attribution-noncommercial licence 3.0 australia (cc by-nc 3.0 au). journal website: www.australianpopulationstudies.org australian age-sex-specific covid-19 mortality in international comparative perspective, to june 2020 rebecca kippen* monash university * email: rebecca.kippen@monash.edu. address: school of rural health, monash university, 26 mercy street, bendigo, victoria, australia. paper received 28 july 2020; accepted 9 november 2020; published 16 november 2020 introduction by the end of june 2020, a combination of good luck and good management had resulted in australia having one of the lowest covid-19 death rates in the developed world at that time. australia’s success was heralded nationally and internationally (duckett and stobart 2020; bremmer 2020). the second wave of covid-19 deaths that occurred in australia from july to october 2020 was, in june 2020, only a possible threat. at 30 june 2020, australia had recorded 103 deaths from covid-19, with only 1 death reported in june. this translates to 4 deaths per million population. developed countries with similarly low covid-19 death rates (under 10 deaths per million population), january-june 2020, included japan (8), brunei (7), south korea (6), slovakia (5), singapore (4), and new zealand (4) (calculated from who 2020; united nations 2019). in contrast, the following 11 developed countries had recorded covid-19 death rates more than 50 times higher than that of australia (above 200 deaths per million population), as at 30 june 2020: belgium (842), united kingdom (644), spain (606), italy (575), sweden (528), france (456), united states (382), netherlands (357), ireland (352), chile (298), and canada (227) (calculated from who 2020; united nations 2019). deaths and death rates were still rising strongly in several of these countries in mid-2020. despite vastly differing levels of mortality, age-sex patterns of covid-19 deaths appear to be consistent internationally (at least for those countries with available data). death rates are higher for males than for females, low in childhood and young adulthood, and increase markedly through middle and old age. data and methods this study is based on age-sex-specific covid-19 death rates for the first half of 2020 for 7 developed countries with particularly low or high covid-19 mortality and for which consistent age-sex death data could be sourced. these 7 countries are australia, japan, south korea, belgium, england & wales (constituting 89% of the united kingdom’s population), italy, and the netherlands. d e m o g ra p h ic http://www.australianpopulationstudies.org/ mailto:rebecca.kippen@monash.edu 34 kippen australian population studies 4 (2) 2020 counts of deaths by age and sex were sourced from national governments for australia (agdh 2020) and england and wales (ons 2020), and from coverage-db (riffe et al. 2020) for the remaining 5 countries. population estimates by age and sex for 2020 were sourced from world population prospects (united nations 2019). death counts and population estimates were used to calculate covid-19 death rates for each population by sex for age groups 0-49 (a broad grouping since covid19 mortality under the age of 50 years was relatively low for all considered populations), 50-59, 60-69, 70-79, 80-89, and 90+ years. to show the age-sex covid-19 mortality pattern for each population, taking into account the enormously disparate death-rate levels, rate ratios were used. for each country, the ratio was calculated of the mortality rate for each age-sex category to the mortality rate for males aged 90+ years (the highest age-sex rate for every country considered). these ratios are shown in figure 1. age-standardised mortality rates for each sex for each country were calculated using australia’s 2020 population by age as the standard. these were used to calculate the male-to-female ratio of agestandardised covid-19 mortality for each country, as a summary measure of excess male mortality, shown in figure 2. key features figure 1 shows the age-sex pattern of covid-19 mortality for each country. figure 1a shows 4 populations that have remarkably similar patterns, despite the fact that the overall mortality of 2 of the populations (belgium, and england & wales) was 100-200 times higher than that of the other 2 countries (australia and south korea). for all 4 populations, female covid-19 mortality at age 90+ years was 64-72% that of males of the same age. male and female mortality at age 80-89 years was respectively 34-41% and 23-27% of male mortality at age 90+ years. younger age groups show similar clustered differentials at lower levels. figure 1b displays the remaining 3 populations – japan, italy, and the netherlands (with australia as a comparator) – with dissimilar age-sex patterns of covid-19 mortality, although they all exhibit higher male mortality and a positive correlation of mortality with age. japanese and italian male mortality for age groups under 90 years is relatively high compared to the patterns seen in figure 1a, while japanese female mortality at age 90+ years is remarkably low relative to japanese male mortality at the same age. female mortality at older ages for italy and the netherlands is relatively high. ratios of male-to-female age-standardised covid-19 death rates for each country are shown in figure 2 in order of magnitude. in belgium, australia, netherlands, england & wales, and south korea, age-standardised covid-19 mortality for males was 61-73% higher than for females. in contrast, covid-19 mortality for italian males was more than twice that for females, while japanese male mortality was almost 3 times female mortality. these 7 country-level sex ratios are much higher than those seen for the most similar cause-of-death groupings. for example, for australia in 2019, age-standardised male mortality for certain infectious and parasitic diseases was only 16% higher than female mortality, and for influenza and pneumonia 11% higher (abs 2020). australian population studies 4 (2) 2020 kippen 35 figure 1: similar (1a) and dissimilar (1b) age-sex patterns of covid-19 mortality (rate ratio to mortality for males age 90+ years), selected developed countries, january-june 2020 36 kippen australian population studies 4 (2) 2020 figure 2: male-to-female ratio of age-standardised covid-19 mortality, selected developed countries, january-june 2020 what explains this disparity in sex-specific covid-19 mortality? early research indicates that females produce a more effective immune response to sars-cov-2 (the coronavirus that causes covid-19) than do males, and that this response declines across age for both sexes (takahashi et al. 2020). in addition, co-morbidities (such as hypertension) and risk behaviours (such as smoking and drinking) that are more prevalent in males may play a role in higher male covid-19 mortality (gebhard et al. 2020). areas for future research include the factors behind non-standard age-sex patterns of covid-19 death rates, such as those seen for japan and italy, and reasons for the much higher male-to-female mortality ratios for these countries. in addition, future research will reveal whether the second wave of covid-19 deaths in australia and other countries results in similar age-sex mortality patterns to those seen in the first half of 2020, or whether and why there are substantive differences. limitations although death registration in each country considered is complete, or near complete, covid-19 death rates may be inaccurate because of (a) mis-estimates of the 2020 denominator population, or, more importantly (b) mis-recording of covid-19 as a cause of death. ‘excess deaths’ in 2020 in many countries greatly exceed reported covid-19 deaths (felix-cardoso et al. 2020; the economist 2020). this indicates that covid-19 as a cause of death is under-recorded, and that therefore the death rates discussed above, are, if anything, underestimates. this could also be the case in australia (bennett 2020). the extent of underestimation may vary by age and sex, affecting the age-sex patterns described above. australian population studies 4 (2) 2020 kippen 37 ethics approval this study was approved by the monash university human research ethics committee, project id 25067. acknowledgements the author thanks the editor and two anonymous reviewers whose helpful suggestions improved this work. most data are derived from coverage-db: a database of covid-19 deaths by age (led by researchers at the max planck institute for demographic research), world population prospects (united nations) and coronavirus disease (covid-19) situation reports (world health organization). the author thanks the respective research teams for assembling these international data and making them publicly available under creative commons licenses. references australian bureau of statistics (abs) (2020) causes of death, australia, 2019, catalogue number 3303.0. https://www.abs.gov.au/statistics/health/causes-death/causes-deathaustralia/2019/3303_1%20underlying%20causes%20of%20death%20%28australia%29.xlsx. accessed 8 november 2020. australian government department of health [agdh] (2020) covid-19 deaths by age group and sex, 1 july 2020. https://www.health.gov.au/resources/covid-19-deaths-by-age-group-and-sex. accessed 2 july 2020. bennett c (2020) have there been uncounted coronavirus deaths in australia? we can’t say for sure, but the latest abs data holds clues. the conversation, 26 june 2020. https://theconversation.com/have-there-been-uncounted-coronavirus-deaths-in-australia-wecant-say-for-sure-but-the-latest-abs-data-holds-clues-141363. accessed 2 july 2020. bremmer i (2020) the best global responses to covid-19 pandemic. time, 12 june 2020. https://time.com/5851633/best-global-responses-covid-19/. accessed 8 november 2020. duckett s and stobart a (2020) 4 ways australia's coronavirus response was a triumph, and 4 ways it fell short. the conversation, 4 june 2020. https://theconversation.com/4-ways-australiascoronavirus-response-was-a-triumph-and-4-ways-it-fell-short-139845. accessed 8 november 2020. felix-cardoso j, vasconcelos h, rodrigues p, and cruz-correia r (2020) excess mortality during covid-19 in five european countries and a critique of mortality analysis data. medrxiv. https://www.medrxiv.org/content/10.1101/2020.04.28.20083147v1. accessed 3 july 2020. gebhard c, regitz-zagrosek v, neuhauser h k, morgan r, and klein s l (2020) impact of sex and gender on covid-19 outcomes in europe. biology of sex differences 11(2): 1–13. office for national statistics (ons) (2020) number of deaths registered each month in england and wales, including deaths involving the coronavirus (covid-19), by age, sex and country. https://www.ons.gov.uk/peoplepopulationandcommunity/birthsdeathsandmarriages/deaths/dat asets/deathsinvolvingcovid19englandandwales. accessed 1 november 2020. riffe t, acosta t, et al. (2020) coverage-db: a database of covid-19 deaths by age. https://osf.io/mpwjq/. accessed 7 november 2020. takahashi t, ellingson m k, wong p, israelow b et al. (2020) sex differences in immune responses that underlie covid-19 disease outcomes. nature. https://doi.org/10.1038/s41586-020-2700-3. accessed 8 november 2020. https://www.abs.gov.au/statistics/health/causes-death/causes-death-australia/2019/3303_1%20underlying%20causes%20of%20death%20%28australia%29.xlsx https://www.abs.gov.au/statistics/health/causes-death/causes-death-australia/2019/3303_1%20underlying%20causes%20of%20death%20%28australia%29.xlsx https://www.health.gov.au/resources/covid-19-deaths-by-age-group-and-sex https://theconversation.com/have-there-been-uncounted-coronavirus-deaths-in-australia-we-cant-say-for-sure-but-the-latest-abs-data-holds-clues-141363 https://theconversation.com/have-there-been-uncounted-coronavirus-deaths-in-australia-we-cant-say-for-sure-but-the-latest-abs-data-holds-clues-141363 https://time.com/5851633/best-global-responses-covid-19/ https://theconversation.com/4-ways-australias-coronavirus-response-was-a-triumph-and-4-ways-it-fell-short-139845 https://theconversation.com/4-ways-australias-coronavirus-response-was-a-triumph-and-4-ways-it-fell-short-139845 https://www.medrxiv.org/content/10.1101/2020.04.28.20083147v1 https://www.ons.gov.uk/peoplepopulationandcommunity/birthsdeathsandmarriages/deaths/datasets/deathsinvolvingcovid19englandandwales https://www.ons.gov.uk/peoplepopulationandcommunity/birthsdeathsandmarriages/deaths/datasets/deathsinvolvingcovid19englandandwales https://osf.io/mpwjq/ https://doi.org/10.1038/s41586-020-2700-3 38 kippen australian population studies 4 (2) 2020 the economist (2020) tracking covid-19 excess deaths across countries. https://www.economist.com/graphic-detail/2020/04/16/tracking-covid-19-excess-deaths-acrosscountries. accessed 2 july 2020. united nations (2019) country-level populations, totals, and by age and sex. world population prospects. https://population.un.org/wpp/. accessed 2 july 2020. world health organization (who) (2020) coronavirus disease (covid-19) situation report 163, 1 july 2020. https://www.who.int/emergencies/diseases/novel-coronavirus-2019/situation-reports. accessed 2 july 2020. https://www.economist.com/graphic-detail/2020/04/16/tracking-covid-19-excess-deaths-across-countries https://www.economist.com/graphic-detail/2020/04/16/tracking-covid-19-excess-deaths-across-countries https://population.un.org/wpp/ https://www.who.int/emergencies/diseases/novel-coronavirus-2019/situation-reports australian population studies 2021 | volume 5 | issue 1 | pages 1–2 © wilson et al. 2021. published under the creative commons attribution-noncommercial licence 3.0 australia (cc by-nc 3.0 au). journal website: www.australianpopulationstudies.org editorial introduction to the census questions special issue tom wilson* the university of melbourne elin charles-edwards the university of queensland jonathan corcoran the university of queensland julia loginova the university of queensland kirsten martinus the university of western australia * corresponding author. email: wilson.t1@unimelb.edu.au. address: melbourne school of population and global health, the university of melbourne, 207 bouverie st, melbourne, vic 3010, australia. published on 31 may 2021 in june 2020, the australian bureau of statistics announced which new questions would be included in the 2021 census of population and housing, as well as which existing questions would be dropped (abs 2020). there will be new questions about defence force service and long-term health conditions, while the internet access question will no longer be asked given how widespread internet access has become. the process of adding questions to the census questionnaire is, understandably, a lengthy one. proposed new topics and questions require careful consideration and debate. there needs to be a compelling case for asking the question, insufficient data on the topic from existing data sources, clear public acceptability of the question, limited respondent burden, an ability for the question to be easily interpretable, and a strong likelihood of it yielding accurate data, among other considerations (abs 2018a). any potential new question needs to be balanced against the competing demands of both existing as well as other proposed new questions, as well as the value of consistency across censuses for time series analyses. all potential new questions require stakeholder consultation (abs 2018b), detailed abs assessment, and extensive testing. any new questions which successfully make it through this rigorous assessment process into the final abs assessment are then recommended to parliament to be included in the next census. given we are currently in a census year, australian population studies sought to place a spotlight on all things census and put out a special call for short commentary papers on proposed new census topics and questions. this special issue draws together a select number of papers that answered this call. the aim was to broaden the conversation and stimulate useful discussion about potential new topics and questions that might be considered for the 2026 census. contained in this special issue are commentary pieces that cover a range of topics, including (1) measuring multi-locational living; (2) highly mobile populations; (3) long-distance commuting; (4) commuting to places of education; http://www.australianpopulationstudies.org/ mailto:wilson.t1@unimelb.edu.au 2 editorial: wilson et al. australian population studies 5 (1) 2021 (5) financial wellbeing; (6) gender identity and sexual orientation; and (7) ethnicity and ancestry. together we hope readers find the collection of papers thought-provoking and seek to draw on in this special issue as a new consolidated resource for scholars and practitioners interested in the census. we note that some papers did not quite make the timing to appear in this special issue of australian population studies and will instead be published in a special section in the forthcoming november issue of the journal. finally, we would like to thank each of the contributing authors who submitted these interesting commentary pieces, and to the referees who gave their time and provided useful comments on the submitted papers. references abs (2018a) census of population and housing: topic directions, 2021. https://www.abs.gov.au/ausstats/abs@.nsf/lookup/2007.0.55.001main+features12021. accessed 2 may 2021. abs (2018b) census of population and housing: consultation on topics, 2021. https://www.abs.gov.au/ausstats/abs@.nsf/lookup/2007.0main+features12021?opendocum ent. accessed 2 may 2021. abs (2020) 2021 census questions and date announced. https://www.abs.gov.au/ausstats/abs@.nsf/mediareleasesbyreleasedate/9a2c37e441ce868e ca25858b0019a7f0?opendocument. accessed 2 may 2021. https://www.abs.gov.au/ausstats/abs@.nsf/lookup/2007.0.55.001main+features12021 https://www.abs.gov.au/ausstats/abs@.nsf/lookup/2007.0main+features12021?opendocument https://www.abs.gov.au/ausstats/abs@.nsf/lookup/2007.0main+features12021?opendocument https://www.abs.gov.au/ausstats/abs@.nsf/mediareleasesbyreleasedate/9a2c37e441ce868eca25858b0019a7f0?opendocument https://www.abs.gov.au/ausstats/abs@.nsf/mediareleasesbyreleasedate/9a2c37e441ce868eca25858b0019a7f0?opendocument australian population studies 2019 | volume 3 | issue 2 | pages 41-44 © arringer, sigler and charles-edwards 2019. published under the creative commons attribution-noncommercial licence 3.0 australia (cc by-nc 3.0 au). journal website: www.australianpopulationstudies.org residential concentration patterns of immigrant groups in australia’s major cities renee arringer* the university of queensland thomas sigler the university of queensland elin charles-edwards the university of queensland * corresponding author. email: renee.arringer@uqconnect.edu.au. address: school of earth and environmental sciences, chamberlain building, the university of queensland, st lucia 4072, queensland, australia paper received 7 september 2019; accepted 17 october 2019; published 18 november 2019 immigration is a global process that stems from a range of personal motivations including the prospect of higher income, better access to education and health, and improved living conditions (undp 2009). this is reflected in australia’s migrant profile where, as of the 2016 census, almost half of the population were either born overseas or had at least one parent born overseas (abs 2017a). ethnic diversity is continuously rising in australia; however, migrants remain overwhelmingly concentrated in the major australian cities of sydney, melbourne, brisbane, and perth. this concentration extends to the suburb scale, as specific neighbourhoods are home to large numbers of co-ethnic communities. an extensive literature (e.g. burnley 2005; edgar 2014; forrest and sheshkin 2015; portes and zhou 1993) indicates that concentration can be involuntary (e.g. ghetto formation) or voluntary, where community members benefit from proximity to others speaking their language or sharing their cultural and social values. this demographic shows the changing residential concentration of the largest immigrant groups in australia’s four most populous cities between 2006 and 2016. it includes both migrants settled in australia and those who immigrated during the five years prior to the 2006, 2011, and 2016 censuses. the immigrant groups were selected as those comprising australia’s ten largest migrant stocks in 2016 (concerned with the birthplace of current immigrants living in australia) and those with the ten largest migrant flows in 2016 (the number of immigrants entering australia). stocks were based on abs estimated resident populations by country of birth in 2016 (abs 2017b) and flows were based on data from the department of home affairs (department of home affairs 2018). when combined, these yielded a total of 13 selected countries of birth: china, england, india, ireland, italy, malaysia, nepal, new zealand, pakistan, philippines, scotland, south africa, and vietnam. country of birth data were extracted from the 2006, 2011, and 2016 censuses based on a harmonised geography and extracted by statistical areas level 2 (sa2) for each greater capital city statistical area (gccsa). these data were used to calculate the index of dissimilarity for each d e m o g ra p h ic http://www.australianpopulationstudies.org/ mailto:renee.arringer@uqconnect.edu.au 42 arringer, sigler & charles-edwards australian population studies 3 (2) 2019 immigrant group in comparison to the australian-born population. the index of dissimilarity is defined as: 𝐼𝐷 = 1 2 ∑ |𝑥𝑖 − 𝑦𝑖| 𝑖 where 𝑥 is the percentage of australian-born population in category 𝑖 (sa2) and 𝑦 is the percentage of immigrant-group population in category 𝑖 (sa2). values close to zero indicate that little dissimilarity exists while high values denote high levels of dissimilarity. plotting dissimilarity against the absolute population of each group reveals the progression of residential concentrations between groups the demographic in figure 1 represents the progression of each immigrant group’s residential concentration as a series of plotted lines. the colour of each line represents the immigrant group, and the shape of the first data point represents the city (circle = brisbane; triangle = melbourne; diamond = perth; square = sydney). the arrows signal the directionality of absolute population change for each group, as well as the degree of relative concentration, as measured by the index of dissimilarity. all groups, except for italy-born and the scotland-born (melbourne and sydney) grew between the 2006 and 2016 census. over this same period, most groups became more dissimilar, with only the italy-born (brisbane and sydney); south africa-born (brisbane and melbourne), vietnam-born (brisbane, perth and sydney) and nepal-born (brisbane), becoming less concentrated. the latter reflects rapid population growth from a low base. from figure 1, it appears that immigrant groups disperse into two modes of settlement, with one converging towards dissimilarity of around 10-20 (england, ireland, new zealand, scotland, and south africa) and the other converging towards dissimilarity of around 50 (china, india, malaysia, nepal, pakistan, philippines, and vietnam). italian immigrant groups depart from this trend, fluctuating between dissimilarities of 30 and 40 for all cities, with group size remaining stagnant or decreasing. for groups converging around dissimilarity of 10-20, there is an observable increase in dissimilarity over the three census years as group size increases (excluding scotland). groups located in sydney, melbourne, and perth increase in dissimilarity as they increase in size until 2011. after this point they continue to increase in size, but their dissimilarity decreases. brisbane is an exception to this pattern, with the majority of groups maintaining a steady directionality. the results show that with the exception of very small groups (e.g. nepal-born in perth, vietnam-born in brisbane and perth), the residential patterns of the largest non-australian born groups are becoming more dissimilar over time. when other research is consulted (burnley 2005; edgar 2014; forrest and dandy 2018), these results suggest that factors surrounding ethnicity – particularly english language proficiency – can influence the level of segregation experienced by immigrant groups in australia. it also suggests immigrant groups will not necessarily become completely ‘assimilated’ into australian society (i.e. dissimilarity = 0) after a certain period, as shown by early-immigrant groups such as those born in italy. as the demographic shows, dissimilarity can increase as group size grows, suggesting that group size may be one variable affecting co-ethnic residential preference. the results presented suggest the need for new models of urban residential settlement to better reflect australia’s ever-changing immigration profile. understanding how group-level characteristics australian population studies 3 (2) 2019 arringer, sigler & charles-edwards 43 (such as size) determine residential concentration is critical as australian cities become increasingly diverse in numbers and origins of immigrant groups. figure 1: residential concentration patterns of the largest immigrant groups in australia’s major cities, 2006-16 source: australian bureau of statistics 2006, 2011, and 2016 census data notes: ‘dissimilarity’ values represent the index of dissimilarity for each immigrant group. ‘group size’ is the total populations of immigrant groups represented on a logarithmic scale. country abbreviations: china (chn), england (eng), india (ind), ireland (irl), italy (ita), malaysia (mys), nepal (npl), new zealand (nzl), pakistan (pak), philippines (phl), scotland (sct), south africa (zaf), vietnam (vnm). the letter preceding the country abbreviation corresponds to brisbane (b), melbourne (m), perth (p), or sydney (s). references abs (2017a) census of population and housing: australia revealed, 2016. catalogue no. 2024.0. canberra: abs. abs (2017b) migration, australia, 2015-16. catalogue no. 3412.0. canberra: abs. alba r d and nee v (2003) remaking the american mainstream: assimilation and contemporary immigration. cambridge: harvard university press. burnley i (2005) generations, mobility and community: geographies of three generations of greek and italian ancestry in sydney. geographical research 29(4): 379-392. 44 arringer, sigler & charles-edwards australian population studies 3 (2) 2019 department of home affairs (2018) country profiles. canberra: australian government. https://www.homeaffairs.gov.au/research-and-statistics/statistics/country-profiles/profiles. accessed 21 february 2018. edgar b (2014) an intergenerational model of spatial assimilation in sydney and melbourne, australia. journal of ethnic and migration studies 40(3): 363-383. forrest j and dandy j (2018) proficiency in english, linguistic shift and ethnic capital: an intergenerational analysis of non-english speaking background immigrant groups in sydney, australia. journal of multilingual and multicultural development 39(2): 111-123. forrest j and sheshkin i (2015) strands of diaspora: the resettlement experience of jewish immigrants to australia. journal of international migration and integration 16(4): 911-927. portes a and zhou m (1993) the new second generation: segmented assimilation and its variants. the annals of the american academy of political and social science 530: 74-96. undp (2009) human development report 2009: overcoming barriers: human mobility and development. new york: undp. https://www.homeaffairs.gov.au/research-and-statistics/statistics/country-profiles/profiles australian population studies 2020 | volume 4 | issue 2 | pages 39-47 © loginova & wohland 2020. published under the creative commons attribution-noncommercial licence 3.0 australia (cc by-nc 3.0 au). journal website: www.australianpopulationstudies.org introductory guide how to create an interactive dashboard using r: the example of the queensland covid-19 tracker julia loginova* the university of queensland pia wohland the university of queensland * corresponding author. email: j.loginova@uq.edu.au. queensland centre for population research, school of earth and environmental sciences, chamberlain building, the university of queensland, st lucia, qld 4072, australia paper received 15 september 2020; accepted 2 november 2020; published 16 november 2020 abstract background interactive tools like data dashboards enable users both to view and interact with data. in today’s data-driven environment it is a priority for researchers and practitioners alike to be able to develop interactive data visualisation tools easily and where possible at a low cost. aims here, we provide a guide on how to develop and create an interactive online data dashboard in r, using the covid-19 tracker for health and hospital regions in queensland, australia as an example. we detail a series of steps and explain choices made to design, develop, and easily maintain the dashboard and publish it online. data and methods the dashboard visualises publicly available data from the queensland health web page. we used the programming language r and its free software environment. the dashboard webpage is hosted publicly on github pages updated via github desktop. results our interactive dashboard is available at https://qcpr.github.io/. conclusions interactive dashboards have many applications such as dissemination of research and other data. this guide and the supplementary material can be adjusted to develop a new dashboard for a different set of data and needs. key words interactive dashboards; r; data visualisation; covid-19; queensland http://www.australianpopulationstudies.org/ mailto:j.loginova@uq.edu.au https://qcpr.github.io/ 40 loginova & wohland australian population studies 4 (2) 2020 1. introduction researchers and practitioners increasingly rely on interactive visualisation tools to draw insights from their data, improve strategic and operational decision-making and to disseminate information to stakeholders and the wider public. effective interaction with data can substantially enhance the understanding of underlying trends and patterns. one way to achieve this is by presenting data in a dashboard. dashboards are “a visual display of the most important information needed to achieve one or more objectives, consolidated and arranged on a single screen so the information can be monitored at a glance.” (few 2006, p. 12). dashboards are attractive to users as they amplify cognition and capitalise on human perceptual capabilities. they can be broadly categorised as interactive or static. interactive dashboards represent a new trend of data intelligence that enable not only viewing but also interacting with data through tables, charts, maps and text. these are different from static dashboards that are “read-only” (yigitbasioglu and velcu 2012). interactivity is especially useful in presenting and interpreting large complex data, as it assists in summarising data and alleviating information overload. different techniques and software can be used to present information and provide interactivity (smith 2013). the possible options are arcgis dashboards, tableau, as well as numerous tools developed for the purposes of data intelligence. although providing numerous advantages and appealing aesthetics, they might not be well integrated with the process and outputs of research which is performed in r, a widely used software environment for statistical computing, data science, and graphics. r is often used in conjunction with the integrated development environment rstudio. both are available as free and open-source editions with a range of packages and exemplary support. the following section explains the process of development of the queensland covid-19 tracker in r. 2. queensland covid-19 tracker in r as the covid-19 pandemic started to spread around the world and australia, dashboards that visualise a large amount of rapidly changing information have become useful for the general public and government institutions. examples include the dashboard by the world health organisation (who 2020) and covid-19 map coordinated by johns hopkins coronavirus resource center (dong et al. 2020). numerous interactive web-dashboards have been developed to track the spread of covid19 in australia, for example, covid-19 dashboard developed by government of south australia (2020), new south wales covid-19 cases and community profile by the university of sydney (2020), charting the covid-19 spread in australia by the australian broadcasting corporation (abc 2020), and covid live (2020) developed by volunteers. covering data at the level of states and territories, these dashboards initially did not provide information related to the spread of covid-19 on a smaller geography level. we aimed at filling this gap through the development of the covid-19 tracker focused on queensland regions. to this end, we wanted to effectively utilise publically available information from queensland health (2020a), a ministerial department of the queensland government, and provide interactive analytical features to track the spread of covid-19 throughout time and space. australian population studies 4 (2) 2020 loginova & wohland 41 2.1. setup several r packages provide opportunities for the development of interactive dashboards. the queensland covid-19 tracker has been developed using the r package ‘flexdashboard’ (iannone et al. 2020). it provides an effective solution for developing and publishing a group of related data visualisations as a dashboard. another package that could be used is ‘shinydashboard’ (chang and borges 2018). it is more suitable for developing dynamic dashboards where input interacts with the output requiring a server to execute r code on user input. ‘flexdashboard’ operates as an r markdown framework allowing the dashboard to be ‘knitted’ into a dynamic but independent html. the complete and annotated code in r markdown document “queenslandcovid19tracker.rmd” and associated files to create the queensland covid-19 tracker are published online (loginova and wohland 2020). as demonstrated in this document, the first step in setting up the dashboard is to add the yaml code as the header of the dashboard, including title and type of output file which in this case is html (box 1). the rest of the document contains code to specify the outline of the dashboard and numerous code chunks. a code chunk is a piece of r code that start and end with ``` and can be named. in the code chunk ‘setup’, we assign main directory (the folder where “queenslandcovid19tracker.rmd” and data files were downloaded), install and load packages, and set global options that affect how r displays the results. it is important to stress that versions of r, r studio and r packages may affect the execution of the code. therefore, control of r and packages is necessary to ensure the stability of the dashboard over time. to this end, we indicated that the code for the queensland covid-19 tracker was developed and works in r version 4.0.0 and versions of each package are specified in “queenslandcovid19tracker.rmd”. -- title: "covid-19 in queensland, australia" output: flexdashboard::flex_dashboard -- box 1: the yaml code specifying the header of the dashboard 2.2. data updates and calculations queensland health has published summary data for 16 queensland health and hospital regions on a daily basis since the first cases of covid-19 were identified in queensland on january 21st 2020 (queensland health 2020b). we started gathering data from march 16th 2020, at the point when the number of new cases had begun to increase rapidly. as this paper went to press in november 2020, there were 1,177 confirmed cases of covid-19 in queensland. the data for confirmed, new confirmed, active and recovered cases as well as deaths were downloaded daily from the 42 loginova & wohland australian population studies 4 (2) 2020 queensland health website (queensland health 2020a) and combined in a .csv (comma-separated values) file. as the situation evolved, additional information has been included in the reporting, including the number of active self-quarantine notices. data updates, preparation, and calculations are performed in a single r markdown file. in the code chunk ‘updates’, we load updated data into the r environment, specify values to update, and add updated daily data to existing data files. in ‘data preparation’, we format the data to be used in maps and graphs. in ‘calculations’, we obtain values to be displayed in the dashboard for queensland totals and doubling time (the time it takes for confirmed cases to double in value). it is important to stress that these three code chunks contain ‘include = false’ condition to prevent r from showing the code and its results in the dashboard 2.3. layout, design and components the layout of the dashboard is specified in the r markdown document using ‘flexdashboard’ syntax, as shown in box 2. the layout of the queensland covid-19 trackers includes four pages organised according to themes, including ‘situation today’, ‘change over time’, ‘individual regions’, and ‘about’. the first three pages contain a collection of tabsets to visualise information relevant to each page (figure 1). the layout of each page can be column-based or row-based, allowing control of the width of columns. the colour of the main panel (dark purple) has been assigned at the beginning of the r markdown document using css (cascading style sheets) easily understood by html (see code for <style> in “queenslandcovid19tracker.rmd”). each tabset consists of components including charts, maps, tables, text, images, and value boxes. our dashboard contains the following components that summarise covid-19 cases in queensland, their spatial distribution and temporal change: • 8 value boxes; • geographical map with 6 layers using ‘tmap’ and ‘leaflet’ packages and 1 animated map using ‘tmap’ and ‘imagemagick’; • 3 summary tables using ‘dt’ package; • 18 graphs using ‘ggplot2’ and ‘plotly’ packages; • 2 texts tabs. each of these components was produced in individual code chunks, as outlined in “queenslandcovid19tracker.rmd”. australian population studies 4 (2) 2020 loginova & wohland 43 `r sit.text` {} # page 1 ===================================== # page 1 column {data-width=175} # column for value boxes --------------------------------- column {.tabset} # tabset for geographical map, summary table, and instructions --------------------------------- change over time {} # page 2 ===================================== # page 2 column {.tabset} ---------------------------------# tabset for queensland, hhs regions, active cases, recovered cases column {data-width=250} ---------------------------------# column for animated map individual regions {} # page 3 ===================================== # page 3 column {.tabset} ---------------------------------# tabset for 15 individual regions about {} # page 4 ===================================== # page 4 box 2: the layout of the components of the queensland covid-19 tracker 44 loginova & wohland australian population studies 4 (2) 2020 figure 1: the layout of the dashboard and snapshots of three main pages source: https://qcpr.github.io/ https://qcpr.github.io/ australian population studies 4 (2) 2020 loginova & wohland 45 2.4. knitting and hosting the finished .rmd file needs to be ‘knitted’ into an html document. rendering executes all of the r code chunks and creates the interactive dashboard. the final step is deploying it online. the decision in relation to publishing a dashboard online is required early in the dashboard development process. github pages provides a suitable environment for continuous integration and serverless deployment of dashboards through github, a repository hosting service (utomo and falahah 2020). web hosting on github pages has several advantages as it is free, has collaborative support from the community of users, and is easy to use (see hanson and seeger 2017). the deployment requires an account and a repository on github and an installed git component on a computer, such as github desktop. however, it is not necessary to have data and .rmd file included the repository, they can all be kept locally as done in this case. figure 2 summarises the process from the dashboard development to hosting it online. figure 2: the process of building a dashboard in r and its deployment of github pages notes: 1. the html is generated using r markdown framework in r. 2. the html website is uploaded into github using github desktop. 3. the html is committed to github pages. 4. users can access the interactive dashboard hosted in github pages. 3. conclusions the aim of this introductory guide was to make it easier for researchers and other practitioners to incorporate data dashboards as a visual data communication tool into their work. today, tools are available in r to develop dashboards and deploy them online easily. interactive dashboards allow presenting data in a way that is informative and visually engaging. furthermore, dashboards can be used for data collections, validation of research results, interactive educational activities, and research collaborations. underlying r codes can be easily shared supporting reproducibility and accountability in research and other areas. however, developing effective and stable dashboards requires building visualisation competence and further research is needed about benefits that justify their adoption. 46 loginova & wohland australian population studies 4 (2) 2020 4. resources the complete code and associated files to create the dashboard can be accessed at https://doi.org/10.6084/m9.figshare.12957065.v1. key messages • interactive dashboards are great data visualisation and communication tools. • researchers and practitioners can easily develop interactive dashboards using r, a widely used software environment for statistical computing. • github pages enable continuous integration and deployment of dashboards through github, a repository hosting service. • in this guide, we provide the complete r code that can be easily adjusted to suit specific needs. acknowledgements we thank editors and reviewers for their valuable feedback. we also acknowledge input from seraphim alvanides, alexander smith, paul dennis and qcpr’s colleagues in relation to the initial design of the dashboard. references abc (2020) charting the covid-19 spread in australia. available at: https://www.abc.net.au/news/2020-03-17/coronavirus-cases-data-reveals-how-covid-19spreads-in-australia/12060704?nw=0. accessed on 30 october 2020. chang w and borges b (2018) shinydashboard: create dashboards with ‘shiny’. https://rstudio.github.io/shinydashboard. accessed on 30 october 2020. covid live (2020) available at: https://covidlive.com.au/ .accessed on 30 october 2020. dong e, du h and gardner l (2020) an interactive web-based dashboard to track covid-19 in real time. the lancet infectious diseases 20(5): 533-534. few s (2006) information dashboard design: the effective visual communication of data. sebastopol, ca: o’reilly media. government of south australia (2020) covid-19 dashboard. available at: https://www.covid19.sa.gov.au/home/dashboard. accessed on 30 october 2020. hanson b and seeger c j (2017) creating github pages. gis and mapping task sheets. 4. https://lib.dr.iastate.edu/gis_tasksheets/4. accessed on 30 october 2020. iannone r, allaire j j, and borges b (2020) flexdashboard: r markdown format for flexible dashboards. http://rmarkdown.rstudio.com/flexdashboard. accessed on 30 october 2020. loginova j and wohland p (2020) queensland covid19 tracker. figshare. https://doi.org/10.6084/m9.figshare.12957065.v1. accessed on 30 october 2020. pauwels k, ambler t, clark b h, lapointe p, reibstein d, skiera b, wierenga b, and wiesel t (2009) dashboards as a service: why, what, how, and what research is needed? journal of service research 12(2): 175-189. queensland health (2020a) queensland covid-19 statistics. available at: https://www.covid19.qld.gov.au/. accessed on 30 october 2020. https://doi.org/10.6084/m9.figshare.12957065.v1 https://www.abc.net.au/news/2020-03-17/coronavirus-cases-data-reveals-how-covid-19-spreads-in-australia/12060704?nw=0 https://www.abc.net.au/news/2020-03-17/coronavirus-cases-data-reveals-how-covid-19-spreads-in-australia/12060704?nw=0 https://rstudio.github.io/shinydashboard https://covidlive.com.au/ https://www.covid-19.sa.gov.au/home/dashboard https://www.covid-19.sa.gov.au/home/dashboard https://lib.dr.iastate.edu/gis_tasksheets/4 http://rmarkdown.rstudio.com/flexdashboard https://doi.org/10.6084/m9.figshare.12957065.v1 https://www.covid19.qld.gov.au/ australian population studies 4 (2) 2020 loginova & wohland 47 queensland health (2020b) queensland coronavirus update. available at: https://www.health.qld.gov.au/news-events/doh-media-releases/releases/queenslandcoronavirus-update-290120. accessed on 30 october 2020. smith v s (2013) data dashboard as evaluation and research communication tool. new directions for evaluation 140: 21-45. university of sydney (2020) nsw covid-19 cases and community profile. available at: https://covid19-data.sydney.edu.au/ utomo p and falahah s (2020) building serverless website on github pages. in iop conference series: materials science and engineering 879(1): 012077. iop publishing. who (2020) coronavirus disease (covid-19) dashboard. available at: https://covid19.who.int/. accessed on 30 october 2020. yigitbasioglu o m and velcu o (2012) a review of dashboards in performance management: implications for design and research. international journal of accounting information systems 13(1): 41-59. https://www.health.qld.gov.au/news-events/doh-media-releases/releases/queensland-coronavirus-update-290120 https://www.health.qld.gov.au/news-events/doh-media-releases/releases/queensland-coronavirus-update-290120 https://covid19-data.sydney.edu.au/ https://covid19.who.int/ australian population studies 2021 | volume 5 | issue 1 | pages 18-25 © haslam mckenzie 2021. published under the creative commons attribution-noncommercial licence 3.0 australia (cc by-nc 3.0 au). journal website: www.australianpopulationstudies.org commentary why recording workforce mobility matters fiona haslam mckenzie the university of western australia email: fiona.haslam-mckenzie@uwa.edu.au. address: department of geography, school of social sciences, the university of western australia, 35 stirling highway, crawley, western australia 6009, australia paper received 12 april 2021; accepted 3 may 2021; published 31 may 2021 1. introduction there is considerable literature documenting the peripatetic preferences of australians for leisure, lifestyle and work purposes, and the paucity of reliable data that would provide insights to why people have multiple bases and the impact on local markets and service delivery (mckenzie et al. 2008 p.4; nicholas and welters 2017; nicholas and welters 2016). there have been numerous requests and submissions to the australian bureau of statistics (productivity commission 2014; productivity commission 2019; abs 2007; mckenzie et al. 2008) to systematically record population mobility, and to not limit the place of usual residence to just one location in the quinquennial census. to date, these have been unsuccessful. this paper focuses specifically on long distance commuting (ldc) for work purposes in rural, regional and remote areas, but the principles and benefits of recording labour force mobility (requiring multiple nights away from the place of usual residence), apply equally for other forms of population mobility such as for education, lifestyle and other seasonal vacation work. specifically, this paper shows how the lack of data about workforce mobility undermines prescient planning and informed regional economic development for local government authorities whose capacity to raise revenue is limited. this is particularly the case in mining regions where local governments and other planning authorities are constrained by state agreements which absolve large multinational mining companies from paying local government rates and other taxes for local infrastructure and services (horsley 2013). while most of the research and media attention regarding (ldc) has focused on the resources industries, it has been reported that it is not the only sector that broadly utilises ldc. greer and stokes (2011) showed that a range of industry sectors utilise ldc, with government, especially the health and education workforce, dominating mobile workers based on per head of population commuting over long distances and staying away from home for consecutive nights. similar findings by the productivity commission (2019, 2014) reinforce the scale and breadth of labour ldc across a range of sectors. the practice is now widespread across both public and private sector service delivery, including health, education, telecommunications, transport services and construction in rural, regional and remote communities (hussain et al. 2015; house of representatives standing committee on regional australia 2013; carson et al. 2011). however, it is the resources industries’ ldc practices that have attracted http://www.australianpopulationstudies.org/ mailto:fiona.haslam-mckenzie@uwa.edu.au australian population studies 5 (1) 2021 haslam mckenzie 19 most concern and where the impacts are especially acute, especially in remote locations with limited accommodation and services (ellem and tonts 2017; haslam mckenzie 2016b). for example, in 2011 the pilbara development commission (2012) calculated there were 50,388 transit worker accommodation (twa) beds in the region while the total estimated residential population of the pilbara was 48,610 people (abs 2012). “by 2015, the western australian legislative assembly (2015) estimated that more than 67,000 mining and construction workers commuted to western australianbased mine sites, more than the entire population of the pilbara region” (haslam mckenzie 2020a p.4). the pilbara region in north west western australia, like most of australia’s resource regions, is located long distances from regional and urban centres where the labour and the workforce is generally concentrated. western australian, the largest of the six states, is one of the most economically productive, with rich deposits or mineral resources contributing 15% of gross national product (gnp) in 2019-20 (department of jobs tourism science and innovation 2021), almost 50% higher than its population share of 10.4% (abs 2020). the mining sector’s contribution to gross state product (gsp) dwarfs every other industry sector, exceeding all the other sectors combined (department of jobs tourism science and innovation 2021). mining activity is almost as important in queensland, contributing a further 11% to gnp (queensland treasury 2020). however, despite more and bigger regional centres in queensland, workforce mobility exerts ongoing pressures (cheshire 2010; fleming and measham 2015). this is particularly so in the more remote bowen basin where non-resident workers in the coal industry and industry support services constitute approximately 20% of the population, which in 2020 equated to 28,758 twa beds (queensland government statistician’s office 2020). 2. capturing long distance commuting data in the census for a variety of reasons (nicholas and welters 2016; nicholas and welters 2017; haslam mckenzie, 2020b), the workforce servicing the resource industries has favoured ldc whereby workers leave their resident or source community and live away from home in a host community, (usually considerable distance away), returning (typically, several days or weeks later) for furlough. while at work, the workforce lives in company-provided accommodation with meals, entertainment, and transport to and from the mine site, and have limited, if any, interaction with the host community. the length of the rosters (‘swings’) is often determined by the particular resource being extracted, and whether the employee works fulltime on an ongoing operation or as a contractor with shorter time commitments on a particular site. despite the media attention and more recent public policy focus (productivity commission 2014; house of representatives standing committee on regional australia 2013; house of representatives select committee on regional development and decentralisation 2018), it is difficult to know the scale of the ldc workforce and the impacts of worker mobility. this is because workforce mobility is not accurately counted or measured. consistent and regular accounting of non-resident populations would provide answers to questions such as: whether or not there is a problem with unaccounted ldc; if the practice is the cause of deleterious impacts on host communities, particularly resourcesbased mono-economy communities; or conversely, if repatriated salaries from the resources industries have a positive multiplier effect when they are spent in other places, particularly rural, regional and remote locations where job opportunities are constrained. 20 haslam mckenzie australian population studies 5 (1) 2021 the australian productivity commission (2014) reported that ldc constitutes only two per cent of the australian workforce based on work undertaken by kpmg (2013). however, there are many researchers who dispute that conservative figure (mcintosh 2012; western australian legislative assembly education and health standing committee 2015; storey 2016). definitive data on the residential distribution and mobility of the ldc workforce is difficult to obtain because, to date, the australian bureau of statistics (abs) does not collect information specific to the ldc workforce. there are also obstacles to obtaining sufficiently comprehensive workforce data from all companies sourcing labour from a given area. nonetheless, indicative estimates of ldc populations are developed by calculating the differences between place of usual residence and place of work census counts (abs 2012; kpmg for the minerals council of australia 2013) for specific abs spatial units at given points in time. accuracy is diminished by the lack of clarity regarding the description of particular job classifications. while it is generally accepted that the operational workforce of a resource project who regularly commute on a fixed roster are ldc workers, there are many contractors who also commute but their tenure at a particular location or workplace is highly variable and therefore difficult to accurately capture in the data (haslam mckenzie 2016a). it is likely that over the last two decades, which have witnessed a prolonged mining boom with considerable spill-over effects into the broader western australian economy, many construction workers, exploration staff, engineers, transport workers and work camp employees were all contractors on mine sites under construction, but their census data was unlikely to be recorded as linked to the mining industry (mcintosh 2012). furthermore, the scale of the ldc workforce is potentially obscured by counting only those in the work cycle of rosters and excluding those workers on furlough (mcintosh 2012). contractors are more likely to work intensely while the work is available and then spend time at home between jobs or contracts (haslam mckenzie et al. 2013). furthermore, their commuting work pattern is often highly erratic depending upon work availability and access to transport and accommodation (see hoath and pavez 2012). the census could be modified, as shown in figure 1, to capture information about people who occupy multiple places for a variety of reasons by simply asking for locality information and what proportion of ‘overnight’ time is spent at the respective locations and for what reason (leisure, lifestyle or work purposes). it would also be useful to know how people move between different locations, providing information that would guide timely transport infrastructure investment and future planning. while this paper has focused on labour mobility, the usefulness of the proposed questions is not limited to non-metropolitan locations or communities overwhelmed by non-resident workers. understanding the scale of second home ownership and the duration of stays in non-resident communities would assist local government authorities that experience seasonal population growth regardless of the reason (including seasonal agricultural labour demand, holiday peaks in coastal and other high amenity areas, and ongoing worker long distance commuting). australian population studies 5 (1) 2021 haslam mckenzie 21 [1] are you regularly away from home requiring you to stay away multiple nights? [ ] no [ ] yes [2] what is the reason for you regularly staying away from your usual residence? [ ] lifestyle (second home) [ ] long distance commuting for work purposes [ ] education and training purposes [ ] health and wellbeing [ ] leisure [ ] other reasons [3] do you work away in blocks of time? [ ] no [ ] yes (write in regular blocks of time in number of nights spent away from home) [ ] yes (but blocks of time away are irregular/contract work) [4] do you have a second residence? [ ] no [ ] yes (the location) [5] how many nights per year do you spend at the second residence? [6] which locality (in australia) do you spend most time when not living at home? [7] how do you travel to the other residence/or workplace away from home? [ ] aeroplane/helicopter [ ] vehicle [ ] boat figure 1: proposed census questions on multiple places of residence source: author 3. why accurate data matters there has been considerable interest in how ldc impacts local socio-economic performance for both source and host communities (house of representatives standing committee on regional australia 2013; lawrie et al. 2011; haslam mckenzie 2016a, 2020b; productivity commission 2014) and from time to time, the media focus on the ill effects of unplanned rapid growth. inevitably, the experience of ldc is not homogenous with vigorous debates regarding who benefits and who loses through long distance work arrangements varies. since the 1960s there have been several mining booms, during which the lack of housing, inadequate infrastructure and poor service delivery were blamed for industry inefficiencies, dysfunctional communities and disjointed economic development. the most recent, prolonged boom has amplified the pressures on housing, infrastructure, human resources and public service provision (senate 22 haslam mckenzie australian population studies 5 (1) 2021 select committee on housing affordability in australia 2008; haslam mckenzie and rowley 2013; nicholas and welters 2016). ldc is an easy antidote to the accommodation, infrastructure and service shortages experienced in many regional areas, but it does not solve the problem of inadequate infrastructure or poor planning for population and industry booms. in western australia after more than two decades of a prolonged mining boom with all the accompanying accommodation and infrastructure shortages (haslam mckenzie 2007; haslam mckenzie and hoath 2014), there are still housing shortages and service shortfalls while ldc continues to by-pass underresourced towns (haslam mckenzie 2020b). census questions that more accurately record time spent in blocks of time at different locations and for what purpose would greatly assist planning for a range of services and infrastructure, and enable communities to better prepare for the market cycles of the resources industries. there has been some research undertaken on second home ownership, particularly as it relates to sea/tree change communities (paris 2013; hugo and harris 2013) and some of the shortcomings of inadequate or inaccurate data highlighted in this work certainly applies to ldc host communities. however, the latter communities’ concerns and frustrations are not the same as communities with a large proportion of second homes, many of which are under-utilised for much of the year. rather than episodic demands, ldc host communities have prolonged over-utilisation and/or shortages of infrastructure and services. nonetheless, as noted by mckenzie et al. (2008 p. 54) “funding processes for the provision of local infrastructure and services are typically based on the size of local populations. in most cases a measure of resident – as distinct to mobile – population is used”. importantly for community development and in line with the principle of horizontal fiscal equalisation (ensuring that average standard of living is achieved regardless of location across the nation), funding from central government to enable basic infrastructure and service provision is allocated via untied grants. these grants are based on population and therefore demographic data originally sourced from the census. it is therefore clear that communities with a high proportion of second home owners, ldc workers and transient worker accommodation are not receiving an equitable share of grants commission funding, despite usually having to provide many of the services and infrastructure commensurate with the overall population resident throughout the year. the inevitable shortfall puts the host communities at a significant disadvantage, carrying the costs of worker mobility with limited opportunities to disrupt the cycle (house of representatives standing committee on regional australia 2013). the conundrum is aptly captured by mckenzie et al. (2008 p. 63): if people are assumed to be connected to a single location and the infrastructure and service funding is allocated on a place-based per capita measure, then people living across two locations will receive less than a full service if they had been living in one location and counted as such. as they are absent, from time to time, from the location in which they are deemed to be living all of the time, then they cannot realise the services provided for them. in the second location in which they habitate, a location where they were not counted as residing and therefore were not provided infrastructure and service funding, there will be crowding and they and the locals (those counted as residents) will complete for fewer resources as there are more people wanting to access these resources. australian population studies 5 (1) 2021 haslam mckenzie 23 from an individual perspective, research consistently identifies the range of individual circumstances guiding work choices and decisions about where to live. these decisions are diverse and include specificity and level of experience and qualifications, time of life and family demands, the economic robustness of the local place of residence, availability and range of local employment options, and priority given to residing in a specific locality for natural and lifestyle amenity. the data collected about individuals’ socio-economic profiles, spending and investment patterns are usually collected via focus groups, interviews and small-scale surveys, with the results often extrapolated to the broader ldc population. however, a reliable and systematic accounting of population and workforce mobility would ensure accuracy of demographic data for policy and planning, enabling all spheres of government and community development practitioners to more accurately predict demand for future services and infrastructure. it would also provide useful insights to policy makers regarding how different interventions and policies could influence labour supply, housing and accommodation needs, demand for services and infrastructure, thus ensuring a more equitable and fair means of distributing general revenue to host communities bearing the brunt of the costs associated with ldc and mobility more generally. references abs (2007) census of population and housing: abs views on content and procedures, 2011. canberra: abs. https://www.abs.gov.au/ausstats/abs@.nsf/lookup/2007.0main+features12011?opendocum ent abs (2012) census of population and housing. canberra: abs. https://www.abs.gov.au/websitedbs/censushome.nsf/home/historicaldata2011?opendocument &navpos=280 abs (2020) national, state and territory population. march 2020. canberra: abs. https://www.abs.gov.au/statistics/people/population/national-state-and-territorypopulation/mar-2020 carson d, ensign p, rasmussen r, & taylor a (2011) perspectives on ‘demography at the edge’. in: carson d, rasmussen r, ensign p, huskey l & taylor a (eds.) demography on the edge: remote human populations in developed nations. farnham: ashgate; 3-20. cheshire l (2010) a corporate responsibility? the constitution of fly-in, fly-out mining companies as governance partners in remote, mine affected localities. journal of rural studies 26(1): 12-20. https://doi.org/10.1016/j.jrurstud.2009.06.005 department of jobs tourism science and innovation (2021) western australia: economic profile. perth: government of western australia. ellem b & tonts m (2017) the global commodities boom and the reshaping of regional economies: the australian experience. australian geographer 49(3): 383-395. https://doi.org/10.1080/00049182.2017.1388761 fleming d & measham t (2015) local economic impacts of an unconventional energy boom: the coal seam gas industry in australia. the australian journal of agricultural and resource economics 59(1): 78-94. https://doi.org/10.1111/1467-8489.12043 greer l & stokes k (2011) divorce and separation in the australian mining sector: is it what we expect? the annual conference of the australian sociological association. rockhampton. https://hdl.handle.net/10018/918860 https://www.abs.gov.au/ausstats/abs@.nsf/lookup/2007.0main+features12011?opendocument https://www.abs.gov.au/ausstats/abs@.nsf/lookup/2007.0main+features12011?opendocument https://www.abs.gov.au/websitedbs/censushome.nsf/home/historicaldata2011?opendocument&navpos=280 https://www.abs.gov.au/websitedbs/censushome.nsf/home/historicaldata2011?opendocument&navpos=280 https://www.abs.gov.au/statistics/people/population/national-state-and-territory-population/mar-2020 https://www.abs.gov.au/statistics/people/population/national-state-and-territory-population/mar-2020 https://doi.org/10.1016/j.jrurstud.2009.06.005 https://doi.org/10.1080/00049182.2017.1388761 https://doi.org/10.1111/1467-8489.12043 https://hdl.handle.net/10018/918860 24 haslam mckenzie australian population studies 5 (1) 2021 haslam mckenzie f & hoath a (2014) the socio-economic impact of mine industry commuting labour force on source communities. resources policy 42: 45-52. https://doi.org/10.1016/j.resourpol.2014.09.002 haslam mckenzie f & rowley s (2013) housing market failure in a booming economy. housing studies 28(3): 373-388. https://doi.org/10.1080/02673037.2013.759177 haslam mckenzie f (2007) attracting and retaining skilled and professional staff in remote locations. alice springs: desert knowledge co-operative research centre. http://www.nintione.com.au/resource/dkcrc-report-21-staff-attraction-and-retention.pdf haslam mckenzie f (2016a) long distance commuting in australia. in: haslam mckenzie f (ed.) labour force mobility in the australian resources industry: socio-economic and regional impacts. singapore: springer; 1-10. https://doi.org/10.1007/978-981-10-2018-6 haslam mckenzie f (2016b) the socio-economic impacts of long-distance commuting on people and communities. in: haslam mckenzie f (ed.) labour force mobility in the australian resources industry. singapore: springer; 11-28. https://doi.org/10.1007/978-981-10-2018-6 haslam mckenzie f, rolfe j, hoath a, buckley a & greer l (2013) regions in transition: uneasy transitions to a diversified economy involving agriculture and mining. curtin graduate school of business, curtin university, perth and cquniversity, rockhampton. final report prepared for csiro minerals down under flagship, mineral futures collaboration cluster. horsley j (2013) conceptualising the state, governance and development in a semi-peripheral resource economy: the evolution of state agreements in western australia. australian geographer 44(3): 283-303. https://doi.org/10.1080/00049182.2013.817038 house of representatives select committee on regional development and decentralisation (2018) regions at the ready: investing in australia’s future. canberra: parliament of the commonwealth of australia. https://www.aph.gov.au/parliamentary_business/committees/house/former_committees/regi onal_development_and_decentralisation/rdd/final_report house of representatives standing committee on regional australia (2013) cancer of the bush or salvation of our cities. canberra: parliament of the commonwealth of australia. https://www.aph.gov.au/parliamentary_business/committees/house_of_representatives_commi ttees?url=ra/fifodido/report.htm hugo g & harris k (2013) time and tide: moving towards an understanding of temporal population changes in coastal australia. adelaide: australian population and migration research centre at the university of adelaide for the national sea change taskforce. hussain r, maple m, hunter s, mapedzahama v & reddy p p (2015) the fly-in fly-out and drive-in driveout model of health care service provision for rural and remote australia: benefits and disadvantages. rural and remote health 15: 3068. https://doi.org/10.22605/rrh3068 kpmg for the minerals council of australia (2013) analysis of the long distance commuter workforce across australia. canberra: minerals council of australia. lawrie m, tonts m, & plummer p (2011) boom towns, resource dependence and socio-economic wellbeing. australian geographer 42(2): 139-164. https://doi.org/10.1080/00049182.2011.569985 mcintosh a (2012) thinking space: ten truths about australia’s rush to mine and the mining workforce. australian geographer 43(4): 331-337. https://doi.org/10.1080/00049182.2012.731297 mckenzie f, haslam mckenzie f & hoath a (2014) fly-in/fly-out, flexibility and the future: does becoming a regional fifo source community present opportunity or burden? geographical research 52(4): 430-441. https://doi.org/10.1111/1745-5871.12080. mckenzie f, martin j & paris c (2008) fiscal policy and mobility: the impact of multiple residences on the provision of place-based service funding. australasian journal of regional studies 14(1): 53-71. https://www.anzrsai.org/assets/uploads/publicationchapter/245-mckenzieetal.pdf. https://doi.org/10.1016/j.resourpol.2014.09.002 https://doi.org/10.1080/02673037.2013.759177 http://www.nintione.com.au/resource/dkcrc-report-21-staff-attraction-and-retention.pdf https://doi.org/10.1007/978-981-10-2018-6 https://doi.org/10.1007/978-981-10-2018-6 https://doi.org/10.1080/00049182.2013.817038 https://www.aph.gov.au/parliamentary_business/committees/house/former_committees/regional_development_and_decentralisation/rdd/final_report https://www.aph.gov.au/parliamentary_business/committees/house/former_committees/regional_development_and_decentralisation/rdd/final_report https://www.aph.gov.au/parliamentary_business/committees/house_of_representatives_committees?url=ra/fifodido/report.htm https://www.aph.gov.au/parliamentary_business/committees/house_of_representatives_committees?url=ra/fifodido/report.htm https://doi.org/10.22605/rrh3068 https://doi.org/10.1080/00049182.2011.569985 https://doi.org/10.1080/00049182.2012.731297 https://doi.org/10.1111/1745-5871.12080 https://www.anzrsai.org/assets/uploads/publicationchapter/245-mckenzieetal.pdf australian population studies 5 (1) 2021 haslam mckenzie 25 nicholas c & welters r (2016) exploring determinants of the extent of long distance commuting in australia: accounting for space. australian geographer 47(1): 103-120. https://doi.org/10.1080/00049182.2015.1090300. nicholas c & welters r (2017) what drives long distance commuting into australian regions? a spatial panel model approach. journal of rural studies 49: 140-150. https://doi.org/10.1016/j.jrurstud.2016.11.016. paris c (2013) is anybody listening? the impact of second home ownership on local governments in australia and elsewhere. proceedings of the 3rd national local government researchers’ forum. adelaide: australian centre of excellence for local government. pilbara development commission (2012) transient worker accommodation in the pilbara. karratha: government of western australia. productivity commission (2014) geographic labour mobility. canberra: australian government productivity commission. https://www.pc.gov.au/inquiries/completed/labour-mobility/report. productivity commission (2019) remote area tax concessions and payments: issues paper. canberra: productivity commission. https://www.pc.gov.au/inquiries/completed/remote-tax/issues. queensland government statistician’s office (2020) bowen basin population report 2020. brisbane: qgso. https://www.qgso.qld.gov.au/issues/3366/bowen-basin-population-report-2020.pdf. queensland treasury (2020) about the queensland economy. brisbane: queensland government. https://www.treasury.qld.gov.au/queenslands-economy/about-the-queensland-economy/ senate select committee on housing affordability in australia (2008) a good house is hard to find: housing affordability in australia. canberra: commonwealth of australia. https://www.aph.gov.au/parliamentary_business/committees/senate/former_committees/hsaf /report/index. storey k (2016) the evolution of commute work in the resource sectors in canada and australia. the extractive industries and society 3(3): 584-593. https://doi.org/10.1016/j.exis.2016.02.009. western australian legislative assembly education and health standing committee (2015) inquiry into mental health impacts of fifo work arrangements. perth: western australian legislative assembly. http://resources.news.com.au/files/2015/06/19/1227405/202450-fiforeport2.pdf. https://doi.org/10.1080/00049182.2015.1090300 https://doi.org/10.1016/j.jrurstud.2016.11.016 https://www.pc.gov.au/inquiries/completed/labour-mobility/report https://www.pc.gov.au/inquiries/completed/remote-tax/issues https://www.qgso.qld.gov.au/issues/3366/bowen-basin-population-report-2020.pdf https://www.treasury.qld.gov.au/queenslands-economy/about-the-queensland-economy/ https://www.aph.gov.au/parliamentary_business/committees/senate/former_committees/hsaf/report/index https://www.aph.gov.au/parliamentary_business/committees/senate/former_committees/hsaf/report/index https://doi.org/10.1016/j.exis.2016.02.009 http://resources.news.com.au/files/2015/06/19/1227405/202450-fiforeport2.pdf australian population studies 2021 | volume 5 | issue 1 | pages 34-39 © botha & de new 2021. published under the creative commons attribution-noncommercial licence 3.0 australia (cc by-nc 3.0 au). journal website: www.australianpopulationstudies.org commentary proposal of a short form self-reported financial wellbeing scale for inclusion in the 2026 census ferdi botha* the university of melbourne john p. de new the university of melbourne * corresponding author. email: ferdi.botha@unimelb.edu.au. address: melbourne institute: applied economic & social research, faculty of business and economics, 111 barry street, carlton, victoria 3010, australia. paper received 12 february 2021; accepted 19 april 2021; published 31 may 2021 1. introduction the australian bureau of statistics (abs) regularly invites suggestions for new content or themes for inclusion in the next national census of population and housing. this provides a unique opportunity to include current and relevant questions, as well as to keep up to date with changes in the needs of society. following extensive consultation and reviewing more than 400 submissions (abs 2018), for the 2021 census the abs decided to include a new question on long-term health conditions and a new question on service in the australian defence force, removed a question on internet usage, and made a few minor text adjustments to existing questions (abs 2020). with the next census due in 2026, there is again a unique opportunity for new content to be included, with invitations from the abs for submissions of additional material. in this paper, we argue for the inclusion of a five-item question battery that together constitutes a validated and straightforward measure of subjective, or perceived, financial wellbeing. 2. proposed census questions on financial wellbeing based on a review of previous research (cfpb 2017; muir et al. 2017; netemeyer et al. 2018), comerton-forde et al. (2018 p. 6) defined financial wellbeing as “the extent to which people both perceive and have (i) financial outcomes in which they meet their financial obligations, (ii) financial freedom to make choices that allow them to enjoy life, (iii) control of their finances, and (iv) financial security – now, in the future, and under possible adverse circumstances.” this definition therefore captures functional, situational, and temporal dimensions. consistent with this definition and in collaboration with commonwealth bank of australia (cba), comerton-forde et al. (2018) developed a 10-item financial wellbeing scale, termed the cba-mi reported financial wellbeing scale, for australian banking customers based on rigorous analyses of http://www.australianpopulationstudies.org/ mailto:ferdi.botha@unimelb.edu.au australian population studies 5 (1) 2021 botha & de new 35 potential existing items1. using these ten items as a reference point, botha et al. (2020) subsequently created an abbreviated 5-item financial wellbeing scale, termed the cba-mi reported financial wellbeing scale 5 (or r-5 in this paper). we propose this set of five questions, listed in table 1, for inclusion in the 2026 census. each item is measured on a five-point likert scale; together the five items make up the r-5 scale. the r-5 was designed to align with comerton-forde et al.’s (2018) definition and covers immediate or daily financial outcomes (items 1, 3, and 4), buffering financial wellbeing during unexpected events (item 2), and reaching long-term financial goals (item 5). the r-5 was rigorously tested and validated via a range of methods, including factor analyses and item response theory (irt)2. the correlation between the r-5 and comerton-forde et al.’s (2018) original 10-item scale is 95.2%, and the abbreviated r-5 scale also performs very similarly to the 10-item scale in terms of its overall distribution and relationship with a range of observed variables (botha et al. 2020). the correlation between the summation r-5 scale and the predicted latent r-5 scale from irt analyses was 99.2%. the r-5 scale’s reliability is excellent, moreover, with the cronbach alpha coefficient of internal consistency (reliability) being 0.86 (botha et al. 2020) and 0.91 (botha et al. 2021) reported in previous work. importantly, in addition to having been rigorously tested and validated (botha et al. 2020), the r-5 has also been applied more broadly using australian survey data that studies the impact of the covid-19 pandemic on australians’ financial wellbeing (botha et al. 2021). publication of the latter study in the peer-reviewed academic journal journal of population economics further attests to the academic community’s increasing acceptance of the r-5 measure as a valid and reliable measure of self-reported financial wellbeing. the r-5 items have also been included in the 2020 wave of the household, income and labour dynamics in australia (hilda) survey to be released in the second half of 2021. one advantage here is that the r-5 measure can be combined with the rich questions of the 20-year running household panel. unfortunately, the sample size of the hilda panel will not allow for certain important types of analyses such as detailed regional analyses, or occupation/industry-specific analyses; that can only be done with the census. the r-5 items are easy to understand, do not impose a heavy cognitive load on respondents, and can be answered quickly. the r-5 scale is also very simple for researchers to construct: summing the responses to each of the items and multiplying the sum by five produces the r-5 scale that ranges 1 comerton-forde et al. (2018) also developed the cba-mi observed financial wellbeing scale (version 1) and haisken-denew et al. (2019) developed an adapted version termed the cba-mi observed financial wellbeing scale (version 2), which consist of five items derived from bank customers’ actual bank records. the observed scales were also subsequently used in conjunction with the 10-item reported scale to understand better australians’ financial wellbeing (haisken-denew et al. 2018; comerton-forde et al. 2020). in botha et al. (2020), the r-5 scale was compared to the observed scale (version 2), with a correlation of 42.9%. this suggests that higher self-reported financial wellbeing tends to be associated with higher observed financial wellbeing, based on actual bank records. however, self-reported and observed financial wellbeing are themselves also distinct concepts and distinct components of overall financial wellbeing. 2 for more information about the validation procedures and statistical properties of the r-5 scale, see botha et al. (2020). 36 botha & de new australian population studies 5 (1) 2021 table 1: proposed census questions on perceived financial wellbeing items item question responses 1 i can enjoy life because of the way i’m managing my finances. 0 – not at all 1 – very little 2 – somewhat 3 – very well 4 – completely 2 i could handle a major unexpected expense. 0 – not at all 1 – very little 2 – somewhat 3 – very well 4 – completely 3 i feel on top of my day to day finances. 0 – disagree strongly 1 – disagree 2 – neither agree nor disagree 3 – agree 4 – strongly agree 4 i am comfortable with my current levels of spending. 0 – disagree strongly 1 – disagree 2 – neither agree nor disagree 3 – agree 4 – strongly agree 5 i am on track to have enough money to provide for my financial needs in the future. 0 – disagree strongly 1 – disagree 2 – neither agree nor disagree 3 – agree 4 – strongly agree source: botha et al. (2020) note: all five items are summed together and multiplied by the factor 5 to obtain the overall r-5 score ranging from 0 to 100. from 0 (low financial wellbeing) to 100 (high financial wellbeing)3. as an example, for each item in table 1 person a may select responses consistent with the values of 3, 1, 2, 3, and 2. person a’s financial wellbeing score would thus be (3 + 1 + 2 + 3 + 2) x 5 = 55. person b’s responses may be consistent with the values of 4, 3, 4, 2, and 3, which will produce a financial wellbeing score of (4 + 3 + 4 + 2 + 3) x 5 = 80. clearly, person b has higher perceived financial wellbeing relative to person a. as also alluded to in the next section, however, such a result would not necessarily imply that person b has a higher income than person a. it is entirely possible, and is often the case, that people with higher (lower) levels of income may have relatively lower (higher) financial wellbeing. a simple focus on income alone can therefore provide a misleading picture of people’s financial wellbeing position, as “financial wellbeing gives us a holistic view of the true pressures felt by all individuals across the income and wealth distribution…” (botha et al. 2021 p. 656). with financial 3 botha et al. (2020, 2021) found that the simple summation of the r-5 performed equally well compared to a latent factor score predicted from irt analyses. for ease of computation and interpretation, therefore, the summation r-5 score is an acceptable and simple measure of self-reported financial wellbeing. australian population studies 5 (1) 2021 botha & de new 37 wellbeing, we capture how people are able to respond to shocks, the extent to which they are financially constrained, and people’s ability to plan for future financial needs. australian households have one of the highest private household debt rates in the world with a household debt-to-income ratio of around 200 per cent (kearns et al. 2020). much of this is driven by real-estate purchases and can leave even relatively well-off households “cash-strapped”, living month-to-month. on paper, a household’s income may be substantial, yet inflexible monthly expenses of high mortgage payments leave the household with limited options in the face of unexpected expenses or economic uncertainties. given the central role of real-estate prices in the australian economy, this effect is likely only to increase over time. 3. advantages and expected returns inclusion of the r-5 scale in the 2026 census would enable stakeholders to obtain an overarching view of perceived financial wellbeing across the entire country. for government, the financial wellbeing questions would provide an indication of how australians are doing financially in a holistic sense. for researchers, availability of the financial wellbeing questions would allow for a wide range of analyses into what personal and household factors may be contributing to financial wellbeing, or how financial wellbeing may affect other spheres of australians’ lives. this financial wellbeing scale allows one to addresses the entire distribution and does not focus simply on the left tail of the distribution as a “financial stress” measure is likely to do. rather, financial wellbeing can also address equally and quite readily improvements in the right tail of the distribution (those with higher levels of financial wellbeing). the scale is specifically designed to pick up movements in all parts of the distribution, well beyond a focus on only averages. anyone aged 15 and older can answer the financial wellbeing battery of questions. for both the australian hilda panel and the australian census, it is the same target respondent population. the financial wellbeing scale battery also easily lends itself to individuals in a household context. individuals in the same household can easily have different answers, especially if there are young adults still living at home along with their parents. also given the plethora of non-standard cohabitation relationships, the finances of two cohabiting adults may be completely separate, despite living at the same address. therefore, the r-5 questions in the census would also allow for analyses of within-household differences in perceived financial wellbeing. although previous census rounds have consistently included information on factors such as total personal income and total household income, questions on individuals’ financial wellbeing have never been asked in the australian census. as far as could be determined, no other national census in the world asks citizens about their financial wellbeing using a robust and validated financial wellbeing instrument. the 2026 australian census would thus be an ideal opportunity to collect financial wellbeing information from australian citizens, and availability of financial wellbeing measures in a national census would be a world first. given the large number of observations in a census, detailed regional analyses will also be possible, to examine financial wellbeing inequalities even within fine geographical areas. moreover, though income provides important information on people’s economic circumstances, income is explicitly not synonymous with financial wellbeing (e.g., bonke and browning 2009; brown and gray 2016; 38 botha & de new australian population studies 5 (1) 2021 haisken-denew et al. 2018), and the correlation between income and financial wellbeing (using the r-5 scale) is in the range of 0.32 to 0.38 (botha et al. 2020). according to botha et al. (2021 pp. 661662): …financial wellbeing is preferable to income [alone] in that it is multi-faceted, capturing several dimensions of individual financial enjoyment such as uncertainty and future-oriented consumption that a simple income measure cannot. it also uses several items across a range of outcomes, averaging out measurement error […] [financial wellbeing] can be constructed from only a few [inobtrusive] questions that most survey respondents will gladly answer, [in comparison to income item survey nonresponse]”. addition of the r-5 items in the census would thus provide a much broader insight into australians’ financial lives, including the day-to-day enjoyment of their income, provision of adequate funds for the future, and building resilience to adverse economic shocks, such as sudden international trade disputes, bushfires, pandemics, floods, and other natural disasters. the types of analyses possible include, for example: simple means or averages, tracking of financial wellbeing values at different quantiles of the distribution, calculating measures of inequality based on these quantile values using all of the standard inequality measures found in the income inequality literature, and multivariate linear and non-linear regression analyses allowing one to control for observable factors. beyond 2026, inclusion of the r-5 scale in future census rounds would make it possible to study changes in financial wellbeing across the australian population over time and the factors that might be driving such change. in turn, such analyses would enable relevant stakeholders and policymakers to benchmark the effects of their policies and focus on factors that are likely to improve australians’ financial wellbeing in the long run, as well as how financial wellbeing may impact on other areas of australians’ lives. references abs (2018) census of population and housing: topic directions, 2021. https://www.abs.gov.au/ausstats/abs@.nsf/lookup/2007.0.55.001main+features12021. accessed on 8 february 2021. abs (2020) 2021 census topics and data release plan. https://www.abs.gov.au/statistics/research/2021census-topics-and-data-release-plan. accessed on 8 february 2021. bonke j and browning m (2009) the distribution of financial well-being and income within the household. review of economics of the household 7(1): 31-42. https://doi.org/10.1007/s11150-008-9044-3 botha f, de new j p, and nicastro n (2020) developing a short form version of the commonwealth bank– melbourne institute reported financial wellbeing scale. commonwealth bank of australia and melbourne institute financial well-being scales technical report no 5. melbourne, university of melbourne. https://fbe.unimelb.edu.au/__data/assets/pdf_file/0003/3403722/cba-mitechnical-report-no.-5.pdf. botha f, de new j p, de new s c, ribar dc, and salamanca n (2021) implications of covid-19 labour market shocks for inequality in financial wellbeing. journal of population economics 34(2): 655689. https://doi.org/10.1007/s00148-020-00821-2 brown s and gray d (2016) household finances and well-being in australia: an empirical analysis of comparison effects. journal of economic psychology 53: 17-36. https://doi.org/10.1016/j.joep.2015.12.006 https://www.abs.gov.au/ausstats/abs@.nsf/lookup/2007.0.55.001main+features12021 https://www.abs.gov.au/statistics/research/2021-census-topics-and-data-release-plan https://www.abs.gov.au/statistics/research/2021-census-topics-and-data-release-plan https://doi.org/10.1007/s11150-008-9044-3 https://fbe.unimelb.edu.au/__data/assets/pdf_file/0003/3403722/cba-mi-technical-report-no.-5.pdf https://fbe.unimelb.edu.au/__data/assets/pdf_file/0003/3403722/cba-mi-technical-report-no.-5.pdf https://doi.org/10.1007/s00148-020-00821-2 https://doi.org/10.1016/j.joep.2015.12.006 australian population studies 5 (1) 2021 botha & de new 39 comerton-forde c, ip e, ribar d c, ross j, salamanca n, and tsiaplias s (2018) using survey and banking data to measure financial wellbeing. commonwealth bank of australia and melbourne institute financial well-being scales technical report no 1. melbourne, university of melbourne. https://fbe.unimelb.edu.au/__data/assets/pdf_file/0010/2839429/cba_mi_tech_report_no_1. pdf. comerton-forde c, de new j p, salamanca n, ribar d c, nicastro a, and ross j (2020) measuring financial wellbeing with self-reported and bank-record data. melbourne institute working paper no. 26/20. melbourne institute: applied economic & social research, university of melbourne. https://melbourneinstitute.unimelb.edu.au/publications/workingpapers/search/result?paper=3547050. consumer financial protection bureau (cfpb) (2017) cfpb financial well-being scale. scale development technical report, washington dc, cfpb. haisken-denew j p, ribar d c, salamanca n, and nicastro a (2018) using survey and banking data to understand australians’ financial wellbeing. commonwealth bank of australia and melbourne institute financial wellbeing scales technical report no. 2. melbourne, university of melbourne. https://fbe.unimelb.edu.au/__data/assets/pdf_file/0010/2836324/cba_mi_tech_report_no_2. pdf. haisken-denew j p, ribar d c, salamanca n, nicastro a, and ross j (2019) improving the commonwealth bank of australia–melbourne institute observed financial wellbeing scale. commonwealth bank of australia and melbourne institute financial wellbeing scales technical report no. 3. melbourne, university of melbourne. https://fbe.unimelb.edu.au/__data/assets/pdf_file/0009/3045708/cba_mi_tech_report_no_3. pdf. kearns j, major m, norman d (2020) how risky is australian household debt? reserve bank of australia research paper rdp 2020-05, reserve bank of australia. https://www.rba.gov.au/publications/rdp/2020/pdf/rdp2020-05.pdf. muir k, hamilton m, noone j, marjolin a, salignac f, and saunders p (2017) exploring financial wellbeing in the australian context. report for financial literacy australia, centre for social impact & social policy research centre, sydney, university of new south wales. netemeyer r g, warmath d, fernandes d, and lynch j g (2018) how am i doing? perceived financial wellbeing, its potential antecedents, and its relation to overall well-being. journal of consumer research 45: 68-89. https://doi.org/10.1093/jcr/ucx109 https://fbe.unimelb.edu.au/__data/assets/pdf_file/0010/2839429/cba_mi_tech_report_no_1.pdf https://fbe.unimelb.edu.au/__data/assets/pdf_file/0010/2839429/cba_mi_tech_report_no_1.pdf https://melbourneinstitute.unimelb.edu.au/publications/working-papers/search/result?paper=3547050 https://melbourneinstitute.unimelb.edu.au/publications/working-papers/search/result?paper=3547050 https://fbe.unimelb.edu.au/__data/assets/pdf_file/0010/2836324/cba_mi_tech_report_no_2.pdf https://fbe.unimelb.edu.au/__data/assets/pdf_file/0010/2836324/cba_mi_tech_report_no_2.pdf https://fbe.unimelb.edu.au/__data/assets/pdf_file/0009/3045708/cba_mi_tech_report_no_3.pdf https://fbe.unimelb.edu.au/__data/assets/pdf_file/0009/3045708/cba_mi_tech_report_no_3.pdf https://www.rba.gov.au/publications/rdp/2020/pdf/rdp2020-05.pdf https://doi.org/10.1093/jcr/ucx109 australian population studies 2017 | volume 1 | issue 1 | pages 55–68 © carson, punshon, mcgrail and kippen. published under the creative commons attribution-noncommercial licence 3.0 australia (cc by-nc 3.0 au). journal website: www.australianpopulationstudies.org comparing rural and regional migration patterns of australian medical general practitioners with other professions: implications for rural workforce strategies dean carson* charles darwin university and umeå university katherine punshon flinders university matthew mcgrail monash university rebecca kippen monash university *corresponding author. email: dean.carson@cdu.edu.au. address: pausele 95, blåviksjön sweden 92195 paper received 25 july 2017; accepted 7 october 2017; published 20 november 2017 abstract background the shortage of professional workers in rural and regional australia continues as a major policy challenge. there has been substantially more strategy investment for the medical general practitioner (gp) profession than for other professions, particularly at the start of their careers. aims to examine differences between domestic migration patterns of gps and other professionals to rural and regional zones in australia for younger, mid-life and older workers. data and methods data from the australian bureau of statistics (abs) 2011 census were used to examine five-year migration rates for professionals in five abs occupational classifications: generalist medical practitioners (gps); engineering professionals; legal professionals; education professionals; and other health professionals. migration volumes were benchmarked for gps and compared both for other professions and career stage. results gps were less likely than other professionals to migrate from major urban to rural zones, regional to rural zones, or rural to regional zones. younger gps had the highest rural migration rates, while midlife and older gps were least likely to migrate to rural and regional zones. in contrast, increasingly age was associated positively with migration to rural zones for those in the other four professions. conclusions despite concerted policy efforts to encourage more gps to move to rural areas, overall rural migration rates for gps are lower than for other professionals, especially for older workers. further investigation of the links between gp migration patterns and workforce policies needs to be undertaken to inform the application or otherwise of workforce strategies used by other professions. keywords professional migration; medical general practitioners (gps); counter-urbanisation; rural; regional; workforce; gravity model. http://www.australianpopulationstudies.org/ mailto:dean.carson@cdu.edu.au 56 carson d et al. australian population studies 1 (1) 2017 1. introduction encouraging professionals to work in rural and regional parts of australia remains a policy challenge. different professional groups have adopted diverse strategies for addressing geographic maldistribution and the persistent shortage of workers in rural areas (corcoran, faggian and mccann 2010). the health and education professions have developed the most comprehensive strategies (jenkins, reitano and taylor 2011). other professions, such as engineering and the legal profession, lament having few if any rural workforce strategies (campbell and lindsay 2013; sharma, oczkowski and hicks 2016). among all the professions, medical general practitioners (gps) appear to have attracted the most policy attention and resources over at least the past two decades (walters et al. 2017). in recent times it has been argued that australia has a sufficient supply of doctors for non-metropolitan areas as a whole, but that significant shortages persist in various kinds of ‘rural areas’, particularly those with smaller populations and more dispersed settlements (walters et al. 2017). national analyses of need and distribution are rare for other professions. despite differences in the extent of workforce strategies applied to them, there have been few comparative studies of rural workforce issues or migration patterns among other professions (carson et al. 2010). a better understanding of the similarities and differences in migration patterns is a necessary first step in understanding the applicability and transferability of strategies used for professionals in other sectors to the gp workforce. recruitment and retention of labour in rural and regional areas involves stimulating particular migration patterns. researchers have attempted to track gp movements between major urban, rural and remote locations (mazumdar and mcrae 2015; mcgrail and humphreys 2015; ricketts and randolph 2007). however, the extent to which the observed patterns in these studies differ from what might be considered theoretically or comparatively ‘normal’ has not been considered. notably, it is not known if the observed migration patterns are unique to gps or apply across professions, and whether observed patterns are consistent with rural migration theories. migration patterns change throughout the life course and as careers progress (kley 2011). increasing occupational mobility among all professions (perales 2014) means that understanding of life course and life stage impacts on locational choice is paramount. a life-stage perspective is thus essential to understanding the spatial distribution of professional workers in australia (mcgrail and russell 2016). this paper addresses some of the gaps in knowledge about professional migration in australia by comparing migration patterns between gps and other professionals across career stages. the paper limits its discussion to migration within australia, recognising that drivers of international migration are likely to be substantially different. the findings are interrogated in relation to leading rural migration theories. by using gps as a benchmark, the research also allows for some insights into how useful workforce strategies for the general medical practitioner profession may be for other professions. 2. rural migration theory rural migration theory suggests a number of potential drivers of domestic migration from urban to rural areas. these can be divided broadly into economic and lifestyle drivers which may include, for australian population studies 1 (1) 2017 carson d et al. 57 example, the prospect of cheaper housing, higher pay and a ‘better’ or different lifestyle. a substantial focus of the economic literature has been on so-called ‘escalator migration’ in which younger workers are attracted to rural areas by the promise of rapid career development (smith and sage 2014). it is increasingly recognised, however, that economic motivations may apply also at other stages of life (martel, carson and taylor 2013). likewise, while ‘lifestyle’ dominates in discussions of push–pull factors for older worker migration to rural areas (han 2016; argent et al. 2014), lifestyle motives may apply also to younger workers. nevertheless, a life stage framework which postulates that there are substantial opportunities for rural migration for younger and older workers, in particular, is a useful foundation for examining professional migration patterns. strategies aimed at drawing gps to rural areas have considered both economic and lifestyle factors. more recently, they have focused additionally on improved career pathways and career development (for an overview of strategies in australia in the past 20 years, see walters et al. 2017). economic inducements in the form of financial incentives may include additional service payments and support for housing and continuing education costs. these tend to be governed by the degree of ‘rurality’ of the practice location and increase accordingly. lifestyle strategies may include support for partners and children to assist the family unit to embed more fully in the community. there is some evidence of the effectiveness of career development strategies for gps, such as providing opportunities for postgraduate vocational training in rural areas, mixed evidence of the effectiveness of financial incentive programs and little known about the effectiveness of lifestyle related strategies (verma at al. 2016). nevertheless, the importance of ‘life stage’ is implicit in the suite of strategies developed for gps in australia. most rural workforce strategies focus on encouraging graduate doctors to select rural clinical training early in their career with the hope that some will ‘step down’ from the larger regional training centres to smaller rural practices in subsequent years (farmer et al. 2015; mullan, chen and steinmetz 2013). in addition to rural-based education, substantial effort has been invested in encouraging people of ‘rural origin’ or ‘rural background’ (meaning they spent part or all of their childhood in rural areas) to, firstly, undertake medicine and, secondly, specialise in rural relevant medicine, predominantly as gps (sureshkumar et al. 2017). the principal outcome of rural workforce strategies for gps has been to create a ‘rural specialisation’ which is selected very early in a doctor’s career. meanwhile, rural strategies targeting older gps have focused largely on retention with little consideration given as to how lifestyle or amenity migration may also be stimulated for gps whose family responsibilities, for example, are no longer so tightly tied to larger regional or major urban centres. other health professions have also been keen to adopt some of the strategies used to encourage spatial redistribution of the gp workforce, with particular attention to early career oriented strategies (hay et al. 2017). the education profession also has a long history of rural-focused workforce strategies which include rural practicum placements and career advancement incentives for periods of rural service (kelly and fogarty 2015). in other professions, rural workforce strategies have tended to be less systematically employed. however, there are now increasing calls for lessons from the health and education sectors to be learned in professions such as the legal profession (browne 2016) and engineering (sharma, oczkowski and hicks 2016). 58 carson d et al. australian population studies 1 (1) 2017 rural workforce strategies may well be transferable between professions because of the universality of some of the rural migration theories considered above. however, structural differences in professions may mean that strategies which are effective for one profession may be less so for others (perales and vidal 2013). clearly, arguments could be made that each profession will have its own enablers of, and constraints to, rural migration. however, empirical evidence of migration patterns between professions is yet to be published. along with differing demands for particular professions, factors such as age of entry into the workforce, gender balance within occupations and options for career pathways may provide varying opportunities for, and barriers to, rural migration at different life stages. this paper provides some insights into how such opportunities may be grown and barriers lowered. 3. data and methods this research utilises the modified monash model (mmm) developed for use in gp retention policies in australia (hudson and may 2015). this model distinguishes four types of geographical zone: major urban; regional; rural; and remote. for the purposes of our research the zones are defined at the national level (figure 1) with no distinction made between moves within or between the states and territories. figure 1: major urban, regional, rural and remote zones based on sa3s. source: abs sa3 shapefile with concordance to zones by authors. the mmm as one of many systems for classifying urban, regional, rural and remote areas in australia. it closely relates to the australian statistical geography standard – remoteness areas (asgs-ra) classification, but provides a more nuanced distinction within the two ‘regional’ categories based on population size rather than remoteness. the mmm is selected and used in this research because of its immediate relevance to gp workforce distribution strategies. australian population studies 1 (1) 2017 carson d et al. 59 in this paper, the ‘major urban’ zone consists of all major cities (mmm category 1, which is nearly identical to asgs-ra 1). mmm distinguishes between ‘regional’ centres with more than 50,000 residents (mm category 2), and more sparsely populated ‘rural’ areas with under 50,000 residents (mm categories 3–5). the mm also incorporates ‘remote’ and ‘very remote’ locations (mm categories 6–7), which are nearly identical to asgs-ra 4 and 5 and include small populations that are isolated from larger urban centres. the substantial differences in human geography between rural and remote locations (taylor 2016) mean that modelling patterns of migration to the latter requires separate attention. migration to remote and very remote locations consequently is not considered in this paper. likewise, the substantial impact of recent overseas immigrants on the geographic distribution of health and other professionals in australia warrants separate attention (golebiowska et al. 2016; negin et al. 2013; terry, woodroffe and ogden 2013). hence, professionals who had arrived in australia within the past five years were also excluded from this study. data drawn from the abs 2011 census (australian bureau of statistics 2011) included: • occupation – separately identifying ‘generalist medical practitioners’ (gps), other health professionals, education professionals, engineering professionals and legal professionals • age – divided into ‘younger’ (less than 40 years), ‘middle’ (40–54 years) and ‘older’ (55 years and over) as broad proxies for career stage • place of residence on census night 2011 • place of residence five years’ prior to census night. place of residence was defined at statistical area level 3 (sa3), which is a classification intended to represent regional agglomerations identified for district-level activities like health, education or natural resource management (peters et al. 2016). sa3 units were then categorised as major urban, regional, rural and remote zones. ‘migrants’ were identified as individuals who had a residential address in one zone on census night 2011 and in a different zone five years’ earlier. observed migration matrices were constructed separately for each profession, and for younger aged, middle aged and older aged populations in each profession. a gravity model (see anderson 2011) was used to estimate the proportion of migrants from one zone who would be expected to move to another zone, given the size of the professional group and the initial geographic distribution. the gravity model approach accounts for the substantial differences in job availability in different zones for different professions. the gravity model assumed that the distance between zones was identical, and that there were no intervening factors which would influence migration between any two zones (such as differences in income). following shen’s approach (shen 2016), 𝑀𝑖𝑗 = 𝑃𝑖 𝑃𝑗 𝑀𝑖𝑗 is the expected migration between zone 𝑖 and zone 𝑗, and 𝑃𝑖 and 𝑃𝑗 are the 2011 population of professionals in the respective zones. an estimated migration matrix was constructed using this equation for each profession and each career stage within each profession. an attractiveness error (ae) was then calculated: 60 carson d et al. australian population studies 1 (1) 2017 𝐴𝐸 = 𝑂𝑖𝑗 − 𝑀𝑖𝑗 𝑂𝑖𝑗 where 𝑂𝑖𝑗 is the observed migration and, as above, 𝑀𝑖𝑗 is the expected migration. while the error values for a single professional group do not reflect the ‘normality’ of its observed migration patterns well (simini et al. 2012), comparison of error values between professional groups highlights differences in migration patterns. the error differential (ed) for each profession was calculated as 𝐸𝐷𝑝 = 𝐴𝐸𝑝 − 𝐴𝐸𝑔𝑝 where 𝑔𝑝 is gps and 𝑝 is the comparison professional group. professional groups with positive eds were more attracted (in the technical sense of more likely to migrate) than gps to the focus zone, whereas those with negative eds were less attracted than gps to the focus zone. while there are no tests of statistical significance of these differences, given that census (whole of population) data are used, a value of +/5 per cent was assumed to have some practical importance. the gravity model accounted for differences in the geographical structure of each professional group. the most significant example was the relative lack of gps in the younger age groups in the rural zones when compared with other professional groups. this was largely a result of the concentration of postgraduate training in the regional and major urban zones. census data have previously been used to analyse the geography of the gp workforce in australia and other professions (johnston and wilkinson 2001; joyce and wolfe 2005). the use of census data for this purpose can be problematic because of the requirement for professional self-identification, aggregation of data at the sa3 level and the potential disconnect between place of residence (which is used in migration analysis) and place of work. nevertheless, the census remains the only data set in australia which allows for direct comparisons between professions. 4. results gps (10%) were less likely to be living in the rural zone in 2011 than other health professionals (15%) and education professionals (16%), but similarly likely to be living in a rural zone as engineering and legal professionals (table 1). while overall migration rates were similar among the five professional groups examined, gps were more likely to migrate to a different zone at younger ages, and legal professionals were less likely to migrate to another zone at younger ages. table 1: zone of residence (2011) and inter-zonal migration rates (2006–2011) by professional group gps (n = 36,914) other health (n = 35,1731) education (n = 418,474) engineering (n = 262,544) legal (n = 149,019) zone of residence reside: major urban zone 74% 66% 65% 74% 74% reside: regional zone 15% 17% 17% 13% 13% reside: rural zone 10% 15% 16% 11% 11% reside: remote zone 1% 2% 2% 2% 2% migration rates overall migration between zones 11% 9% 8% 10% 8% migration: early career 19% 14% 14% 14% 11% migration: mid-career 8% 6% 6% 6% 7% migration: late-career 4% 5% 4% 4% 6% source: abs 2011 census. australian population studies 1 (1) 2017 carson d et al. 61 table 2: patterns of migration into the rural zone, 2006–2011 percentage of migrants who moved to the rural zone (observed value on top line; expected on second line) standardised differences in observed and expected migration (‘error differential’, compared to gps) gps other health edu eng legal other health edu eng legal all migrants 21% (21) 28% (25) 30% (26) 24% (22) 27% (23) 8% 12% 7% 14% migrants: from major urban 33% (38) 41% (44) 45% (46) 40% (41) 43% (42) 8% 14% 12% 17% migrants: from regional 17% (11) 31% (18) 37% (19) 23% (12) 26% (12) 8% 15% 13% 20% migrants: from remote 24% (10) 26% (15) 25% (16) 19% (11) 24% (11) -18% -23% -18% -4% migrants: early career 20% (16) 25% (22) 30% (24) 23% (20) 24% (19) -10% -4% -11% -4% migrants: mid-career 23% (22) 31% (26) 29% (26) 28% (24) 28% (24) 13% 6% 11% 13% migrants: late-career 25% (25) 37% (28) 33% (28) 32% (26) 36% (26) 25% 18% 20% 27% source: authors’ construction based on gravity model analysis of abs 2011 census data. notes: edu = education; eng = engineering. table 2 shows the percentage of migrants of various types who moved to the rural zone, and the error differential for each non-gp professional group compared to gps. while the gravity model showed that all professions had a higher than expected rate of migration to the rural zone, the rural zone was substantially more attractive for all other professionals than for gps (only 21 per cent of gps overall moved to the rural zone), and particularly attractive for education (30%) and other health professionals (28%). only 33 per cent of gps migrated from the major urban zone to the rural zone, compared to at least 40 per cent of all other professionals and 45 per cent of education professionals. gps were also substantially less likely than other professionals to move from the regional zone to the rural zone, once the gravity model accounted for differences in starting distributions. legal professionals were 20 per cent more likely than gps to migrate from the regional to rural zone. in contrast, gps in remote zones were more likely than other professionals to move to the rural zone. table 3 (next page) shows that the regional zone was substantially more attractive to gp migrants than to all other professional groups except ‘other health’. gps were far more likely than other professionals to migrate from the major urban zone to the regional zone. however, gps were less likely than other professionals to migrate from the rural zone to the regional zone. 62 carson d et al. australian population studies 1 (1) 2017 table 3: patterns of migration into the regional zone, 2006–2011 percentage of migrants who moved to the regional zone (observed value on top line; expected on second line) standardised differences in observed and expected migration (‘error differential’, compared to gps) gps other health edu eng legal other health edu eng legal all migrants 36% (30) 33% (28) 28% (27) 29% (27) 30% (28) -1% -11% -11% -7% migrants: from major urban 61% (58) 49% (51) 40% (48) 45% (51) 47% (52) -9% -25% -18% -17% migrants: from rural 31% (17) 41% (20) 39% (20) 35% (15) 38% (15) 3% 2% 11% 14% migrants: from remote 14% (15) 24% (18) 22% (17) 18% (14) 19% (13) 33% 27% 30% 34% migrants: early career 40% (34) 33% (30) 27% (27) 28% (28) 31% (30) -5% -15% -16% -12% migrants: mid-career 29% (30) 34% (28) 32% (27) 31% (27) 32% (27) 20% 16% 15% 17% migrants: late-career 30% (27) 32% (26) 28% (26) 33% (26) 27% (25) 7% 0% 11% -3% source: authors’ construction based on gravity model analysis of abs 2011 census data. notes: edu = education; eng = engineering. figure 2: stylised view of migration to rural and regional zones by various professional groups source: authors’ construction based on gravity model analysis of abs 2011 census data. note: the width of the arrow represents the relative likelihood of the move. figure 2 summarises the flows of the various professions between zones. figure 3 (next page) provides a visual comparison of relative migration rates to both rural and regional zones for each profession for each age group. non-gp professions are benchmarked against gps using the error differentials in tables 2 and 3. the figure shows the relatively low likelihood of rural migration at younger ages for all professions, and the relatively high likelihood of rural migration (compared to regional migration) at older ages (40 years or above) for all professions except gps. australian population studies 1 (1) 2017 carson d et al. 63 figure 3: relative levels of migration into rural and regional zones by profession and broad age group source: authors’ construction based on gravity model analysis of abs 2011 census data. younger gps were more likely to move to a rural zone than younger engineering and other health professionals, despite a lower observed rural migration rate due to fewer rural job opportunities. gps in the middle and older age groups migrated to rural zones less frequently than other professionals. education professionals in these age groups also were relatively unlikely to migrate to the rural zone compared to those in the remaining three professions. younger gps were substantially more attracted to the regional zone than other professional groups. gps aged 40–54 years were substantially less attracted than their peers in other professions to regional practice in their middle years. regional migration patterns for individual professions were relatively stable in the latter years 64 carson d et al. australian population studies 1 (1) 2017 (55 years plus) with other health and engineering professionals being more attracted and legal professionals being less attracted to regional living than gps. 5. discussion gps were substantially less likely than other professionals to migrate to rural locations in the 2006– 2011 period (table 2), despite being slightly more mobile overall (table 1). however, they were generally more likely to migrate to locations in the regional zone than professionals in the other groups (table 3). the migration patterns of professionals in engineering, other health, education and the legal profession were broadly similar with migration to the rural zone increasing with age and relatively stable levels of migration to the regional zone. this was in contrast to gp migration patterns, which demonstrated only a small increase in rural migration after age 40 and a decrease in regional migration relative to their younger years. migration rates for gps in the middle (45–54) and older (55 plus) age groups were much lower than those of their peers in other professional groups. significantly, there was substantially lower gp migration from regional to rural areas than for other professionals. the greatest diversity of migration patterns between professions was in major urban to regional zones and remote to rural zones. gps and other health professionals were substantially more likely than other professionals to engage in the former; gps and legal professionals were substantially more likely to engage in the latter. there is some evidence of younger gps being more likely to engage in migration for rapid career advancement to the regional zone, which may be driven in part by more recent policies supporting rural pathway training for gps in such areas (mcgrail, russell and campbell 2016). however, it is of great concern that the high migration to the regional zone evidenced for gps in the under 40 age group does not appear to lead to substantial ‘step down’ or migration of gps to the rural zone in the older age groups. by way of contrast, there was great similarity in the rates of migration to rural and regional areas in the older age groups for all other professions. this research makes a case for ‘gp exceptionalism’ (gps having distinctive migration patterns compared to other professions) when it comes to professional migration to regional and rural australia. while exceptionalism cannot be ascribed to specific workforce strategies with the methods used in this study, some insights may be offered. • the focus of gp workforce training and development strategies on early career location decisions does appear to make a difference to rates of migration to both rural and regional zones. • the similar focus of the education profession on rural exposure and incentives for early career rural service might also have impacted the high levels of rural migration for neophyte teachers but is not apparent in levels of regional migration. • the lack of attention to rural pathways for older gps is likely also reflected in the data, with barriers to the sorts of occupational changes required in different zones (particularly rural) possibly a factor. in this regard, the ‘rural specialisation’ strategy adopted by the gp profession may be counterproductive when trying to attract more experienced workers. other professions may be well served by considering how to balance the benefits of early specialisation on early career location decisions with strategies that facilitate rural migration in mid or later career stages. australian population studies 1 (1) 2017 carson d et al. 65 the other crucial aspect of gp exceptionalism is the lack of movement from regional to rural areas. this may be a reflection of the increasing focus of ‘rural’ medical education programs in ‘regional’ areas. other professions may pay particular attention to this issue, and ensure that rural exposure occurs in those areas where workforce growth is most needed. for gps, at least, it is apparent that ‘regional’ is not ‘rural’. 6. conclusions despite concerted policy efforts to encourage gps to move to rural areas, overall rural migration rates remain low when compared with other professions. the literature suggests that strategies which involve the exposure of medical students and gps to rural training environments are more effective in determining ultimate work location than economic or lifestyle strategies promoting financial incentives or rural lifestyle advantages. however, most of these exposure strategies are targeted at younger gps and typically involve exposure to regional rather than rural areas. the data reveal that younger gps are more likely than professionals in equivalent age groups to work in rural locations, but are particularly more likely to locate to regional zones. a substantial challenge for the gp profession is to stimulate increased rural migration for experienced practitioners in the middle and older age groups, and to stimulate migration from regional to rural areas. that these two types of migration flow are more common in other professions suggests that those professions may receive additional benefit from strategies targeting younger workers. likewise, gp workforce policy may well benefit from an improved understanding of the drivers of migration in other professions. there must be some caution taken, however, when considering the potential of ‘exposure’ type strategies used by other professions. there is a possibility that the counterpoint to strategies for early career ‘lock in’ to regional and rural practice may be a ‘lock out’ for older workers. our research explored the impacts on a simple gravity model of the workforce age distribution for five professions: generalist medical practitioners; education professionals; legal professionals; engineering professionals; and other health professionals. there is potential to expand the model considerably to examine economic (e.g. income, housing costs, financial incentives) and lifestyle factors (e.g. working hours, leisure opportunities, quality of spousal/partner employment) which might provide for more insights into the causes of gp exceptionalism. this work should be done within the theoretical frameworks outlined in this paper. while life stage influences on migration patterns have been clearly established, evidence for ‘escalator’, ‘step’ and ‘amenity’ migration drivers is yet to be investigated. more work is required also to analyse out-migration and the net effects of inand out-migration on workforce distribution. as occupational mobility increases there will be demand for knowledge about how the processes for career development and career transition impact the identification and take up of employment opportunities and choice of residential location. analysis of these processes within and across vocations can provide insights beyond those which arise from the interrogation of a single profession. while this research has investigated migration patterns across a range of professions at the national level, further investigation is required to identify causative factors and to consider the position of more detailed geographies. 66 carson d et al. australian population studies 1 (1) 2017 key messages • while encouraging major urban to rural migration is a shared ambition of many professions, diverse strategies may be required because of pre-existing differences in migration patterns. • there is evidence of the impact of gp training strategies with increased early career migration into regional areas. however, migration to rural and remote areas for those in other professions is highest in the later career stages. • despite more than two decades of concerted policy efforts to influence gp migration patterns, overall rates of rural and regional migration are lower for this profession than for other professions. • theories about the drivers of migration at different career stages, including escalator migration, step migration and amenity migration, may provide insights into effective strategies to influence migration patterns. references anderson j (2011) the gravity model. annual review of economics 3(1): 133–160. argent n, tonts m, jones r and holmes j (2014) the amenity principle, internal migration, and rural development in australia. annals of the association of american geographers 104(2): 305–318. australian bureau of statistics (2011). tablebuilder, census of population and housing, accessed february 2017, http://www.abs.gov.au/websitedbs/censushome.nsf/home/tablebuilder. browne k v (2016) rural lawyers and legal education: ruralising and indigenising australian legal curricula. international conference on education and e-learning (eel): proceedings. singapore: global science and technology forum; 50–59. campbell s and lindsay k (2013) lawyers of the future: creating aspirations, forging connections and facilitating professional links in rural and regional contexts, occasional papers series, international journal of rural law and policy 2013(2): 1–10. carson d, coe k, zander k and garnett s (2010) does the type of job matter? recruitment to australia's northern territory. employee relations 32(2): 121–137. corcoran j, faggian a and mccann p (2010) human capital in remote and rural australia: the role of graduate migration. growth and change 41(2): 192–220. farmer j, kenny a, mckinstry c and huysmans r (2015) a scoping review of the association between rural medical education and rural practice location. human resources for health 13: 27. golebiowska k, carter t, boyle a and taylor a (2016) international migration and the changing nature of settlements at the edge. in taylor a, carson d b, ensign p c, rasmussen r, huskey l and saxinger g (eds) settlements at the edge: remote human settlements in developed nations. cheltenham, uk: edward elgar publishing. han y (2016) up and down flows of migration in national-space hierarchy over time. journal of korean society of rural planning 22(1): 49–56. hay m, mercer a m, lichtwark i, tran s, hodgson wc, aretz h t, armstrong e g and gorman d (2017) selecting for a sustainable workforce to meet the future healthcare needs of rural communities in australia. advances in health sciences education 22(2): 533–551. hudson j and may j (2015) what influences doctors to work in rural locations? medical journal of australia 202(1): 5. jenkins k, reitano p and taylor n (2011) teachers in the bush: supports, challenges and professional learning. education in rural australia 21(2): 71. johnston g and wilkinson d (2001) increasingly inequitable distribution of general practitioners in australia, 1986–96. australian and new zealand journal of public health 25(1): 66–70. http://www.abs.gov.au/websitedbs/censushome.nsf/home/tablebuilder australian population studies 1 (1) 2017 carson d et al. 67 joyce c and wolfe r (2005) geographic distribution of the australian primary health workforce in 1996 and 2001. australian and new zealand journal of public health 29(2): 129–135. kelly n and fogarty r (2015) an integrated approach to attracting and retaining teachers in rural and remote parts of australia. journal of economic and social policy 17(2): 1. kley s (2011) explaining the stages of migration within a life-course framework. european sociological review 27(4): 469–486. martel c, carson d and taylor a (2013) changing patterns of migration to australia’s northern territory: evidence of new forms of escalator migration to frontier regions? migration letters 10(1): 101–113. mazumdar s and mcrae i (2015) doctors on the move: national estimates of geographical mobility among general practitioners in australia. australian family physician 44(10): 747–751. mcgrail m and humphreys j (2015) geographical mobility of general practitioners in rural australia. medical journal of australia 203(2): 92–97. mcgrail m and russell d (2016) australia’s rural medical workforce: supply from its medical schools against career stage, gender and rural-origin. australian journal of rural health first published 21 november (early view): 1–8. mcgrail m, russell d and campbell d (2016) vocational training of general practitioners in rural locations is critical for the australian rural medical workforce. medical journal of australia 205(5): 216–221. mullan f, chen c and steinmetz e (2013) the geography of graduate medical education: imbalances signal need for new distribution policies. health affairs 32(11): 1914–1921. negin j, rozea a, cloyd b and martiniuk a l (2013) foreign-born health workers in australia: an analysis of census data. human resources for health 11: 69. perales f (2014) how wrong were we? dependent interviewing, self-reports and measurement error in occupational mobility in panel surveys. longitudinal and life course studies 5(3): 299–316. perales f and vidal s (2013) occupational characteristics, occupational sex segregation, and family migration decisions. population, space and place 19(5): 487–504. peters p, taylor a, carson d and brokensha h (2016) sources of data for settlement level analysis in sparsely populated areas. in taylor a, carson d, ensign p, huskey l, rasmussen r and saxinger g (eds.) settlements at the edge: remote human settlements in developed nations. cheltenham, uk: edward elgar publishing; 153–177. rees p, bell m, kupiszewski m, kupiszewska d, ueffing p, bernard a, charles-edwards e and stillwell j (2017) the impact of internal migration on population redistribution: an international comparison. population, space and place 23(6; e2036). ricketts t and randolph r (2007) urban‐rural flows of physicians. the journal of rural health 23(4): 277–285. sharma k, oczkowski e and hicks j (2016) skill shortages in regional new south wales: the case of the riverina. economic papers: a journal of applied economics and policy 36(1): 3–16. shen j (2016) error analysis of regional migration modeling. annals of the association of american geographers 106(6): 1253–1267. simini f, gonzalez m, maritan a and barabasi a (2012) a universal model for mobility and migration patterns. nature 484(7392): 96–100. smith d and sage j (2014) the regional migration of young adults in england and wales (2002–2008): a ‘conveyor-belt’ of population redistribution? children's geographies 12(1): 102–117. sureshkumar p, roberts c, clark t, jones m, hale r and grant m (2017) factors related to doctors’ choice of rural pathway in general practice specialty training. australian journal of rural health 25(3): 148–154. 68 carson d et al. australian population studies 1 (1) 2017 taylor a (2016) introduction. in taylor a, carson d, ensign p, rasmussen r, huskey l and saxinger g (eds) settlements at the edge: remote human settlements in developed nations. cheltenham, uk: edward elgar publishing; 3–24. terry d, le q, woodroffe j and ogden k (2013) the baby, the bath water and the future of imgs. international journal of innovative interdisciplinary research 2(1): 51–62. verma p, ford j a, stuart a, howe a, everington s and steel n (2016) a systematic review of strategies to recruit and retain primary care doctors. bmc health services research 16(126). walters l, mcgrail m, carson d, russell d, o'sullivan b, strasser r, hays r and kamien m (2017) where to next for rural general practice policy and research in australia? medical journal of australia 207(2): 56–58. australian population studies 2020 | volume 4 | issue 1 | pages 20-36 © mcdonald & moyle 2020. published under the creative commons attribution-noncommercial licence 3.0 australia (cc by-nc 3.0 au). journal website: www.australianpopulationstudies.org the cessation of rising employment rates at older ages in australia, 2000-2019 peter mcdonald* the university of melbourne and the australian national university helen moyle the university of melbourne * corresponding author. email: mcdonald.p@unimelb.edu.au. address: melbourne school of population and global health, university of melbourne, 207 bouverie st, vic 3010, australia. paper received 28 january 2020; accepted 1 may 2020; published 25 may 2020. abstract background in the first decade of the 21st century, employment at older ages surged in australia, benefitting the australian economy. subsequent to 2010, however, employment rates at older ages ceased rising for older men and the increases were much more moderate for women. aim the aim of this paper is to examine these older-age employment trends in more detail, particularly the association between older-age employment trends and the business cycle. some attention is also given to alternative explanations related to changes in the characteristics of the population and industrial structure. data and methods two main data sources are used: published tables from the monthly australian bureau of statistics labour force surveys and the australian censuses for the years 2006, 2011 and 2016. the methods used are primarily descriptive. results strong labour demand in the first decade of the 21st century stimulated the entry to employment of those out of the labour force, especially at ages 45-54 and especially for men. a cooling of labour demand following the global financial crisis terminated this process in the second decade. there were strong associations between older age employment and various socio-economic characteristics, but, in general, changes in the composition of the population or in the rates of employment by these characteristics did not contribute to the cessation of rising employment after 2010. conclusions employment rates at older ages in australia in the first two decades of the 21st century were the results of shifts in labour demand before and after the global financial crisis. policy related to the taxation of superannuation also induced workers with adequate superannuation, especially public sector workers, to continue working to at least age 60. key words older age employment, australia, labour demand, superannuation, gender differences. http://www.australianpopulationstudies.org/ mailto:mcdonald.p@unimelb.edu.au australian population studies 4 (1) 2020 mcdonald & moyle 21 1. introduction the first intergenerational report (department of the treasury 2002) promoted the benefits of an increase in employment at older ages because it would mean that people would be on the tax-paying side of the ledger rather than on the tax-disbursal side. working longer would also mean that the people concerned would be able to build up their superannuation entitlements before retiring, again reducing the government’s potential future liability. subsequent official reports made the same arguments (productivity commission 2005; house of representatives standing committee on health and ageing 2005). as a result, several policy initiatives were introduced to encourage older age employment (productivity commission 2005, swoboda 2014). the 2010 intergenerational report (department of the treasury 2010) raised the additional concern that economic growth would be under significant pressure as the baby-boom generation retired from the labour force. this effect is clearly demonstrated by economic modelling (bloom et al. 2011; temple and mcdonald 2017). in the first decade of the 21st century, employment at older ages surged in australia (mcdonald 2011) and it has been demonstrated that this surge of older age employment benefitted the australian economy as had been predicted (temple et al. 2017). subsequent to 2010, however, employment rates at older ages ceased rising for men and the increases were much more moderate for women (see section 3). the aim of this paper is to examine older-age employment trends in australia over the 20002019 period, and in particular the association between older-age employment and the business cycle. 2. abs labour force definitions the analysis focuses upon ‘employment’ rather than ‘unemployment’ or ‘labour force participation’. this is because the distinction between unemployment (not employed but actively looked for work at some time in the past four weeks and available for work in the week preceding the survey) and not being in the labour force (not employed or unemployed) can be somewhat arbitrary at older ages. this is especially the case because ‘looked for work’ does not include scanning job advertisements in the newspaper or the internet. persons in employment are defined as those aged 15 years and over who, during the week preceding the data collection, were engaged for at least one hour in any activity to produce goods or provide services for pay or profit, or were temporarily absent from work so defined (abs 2018a). by these definitions a person very actively engaged in voluntary work (not for pay or profit) is not considered to be employed. criticisms are sometimes made of the low, one-hour criterion to qualify as ‘employed’ but this is a very minor issue because, in australia, less than five percent of employed persons are employed for under 10 hours per week (abs 2019a). while these definitions are consistent across australian bureau of statistics data collections, the abs advises that the definitions are not as tightly applied in the australian censuses as they are in the labour force surveys, the two sources used in this paper. however, as shown in the next section, the same old-age employment trends are evident from both sources. 22 mcdonald & moyle australian population studies 4 (1) 2020 3. older-age employment trends, 2000-2019 monthly employment rates from january 2000 to november 2019 from the abs labour force survey are shown for men and women aged 55-59, 60-64 and 65+ in figure 1. changes in employment rates for the age group 65+ may be misleading if the age structure within the age group 65+ shifts to the younger or older ends of this range. however, scrutiny of age structure changes in australia indicates that this was not an issue for the period under consideration here. for men in all three age groups, the employment rate peaked around the end of 2010 and then flattened. for those aged 60-64 and 65+, the levels seem to have risen a little in 2018 and 2019. rates for men aged 55-59, however, appear to have fallen a little since 2010. the very recent increases in employment in the two older age groups for men may have been affected by the increase in the pension eligibility age from 65 to 67 years between july 2017 and july 2023. for women aged 55-59 and 60-64, the rises in employment were even stronger than those for males in the first decade and, after levelling off for a few years immediately after the crisis in late 2008 and early 2009, women’s rates of employment continued to rise from around 2014 onwards albeit at a slower pace than before the crisis. figure 1: percentage employed by sex, age groups 55-59, 60-64 and 65+, australia, january 2000 to november 2019 source: abs 2019a figure 2 shows that the increases in older-age employment in the first decade were very likely to have been influenced by earlier rises in employment at younger ages. employment participation at ages 45-49 and 50-54 rose strongly from around 1998 through to the time of the global financial australian population studies 4 (1) 2020 mcdonald & moyle 23 crisis. subsequently, participation at these ages fell for both men and women, with rises for women only from 2014 onwards. figure 2: percentage employed by sex, age groups 45-49 and 50-54, australia january 1998 to november 2019 source: abs 2019a figure 3: percentage employed by single years of age, ages 45-79, australian censuses, 2006, 2011 and 2016 source: abs 2016 24 mcdonald & moyle australian population studies 4 (1) 2020 using census data (figure 3), it is possible to examine trends across time by single years of age. importantly for subsequent analysis, this figure indicates a strong degree of consistency between the trends in the labour force surveys and the censuses. in addition, it shows that for both men and women, the rate of employment tends to drop monotonically with age from age 50 onwards without much acceleration around the pension eligibility ages or any other age. a potential argument is that the trends in older-age employment rates may have been the result of changes in labour supply at these ages: that supply was constrained in the first decade but increased strongly in the second decade. counter to this argument, however, table 1 shows that, for age groups 55-59 and 60-64, the rates of population growth were much higher in the first decade than in the second decade. the reverse was the case for the 65 and over age group. for those aged 55-64, employment levelled off in the second decade even though the population of potential workers was growing more slowly than in the first decade. table 1: rates of population growth for older age groups, persons, 2001-10 and 2010-19 age group annual rate of population growth (%) 2001-09 2010-19 55-59 3.0 1.8 60-64 4.2 1.7 65+ 2.3 3.4 source: abs 2019a in relation to the hours of work for those employed, table 2 shows that, as expected, women were much more likely to work part-time than men in all three older age groups. however, across time, there were small increases in the percentage of men working part-time and small decreases in parttime employment among women, in all age groups. the long-term rise (from the 1990s) in the pension eligibility age for women from 60 to 66 may have had a bearing on the rises in employment, particularly for women aged 60 and over. table 2: percentage of employed persons working part-time, males and females, age groups 55-59, 60-64 and 65+, june 2000 and june 2019 sex age group june 2000 june 2019 % males 55-59 11.9 13.2 60-64 18.6 23.5 65+ 42.6 48.4 females 55-59 49.8 44.0 60-64 56.2 54.0 65-+ 70.1 64.6 source: abs 2019a the downward trend in both employment and full-time employment for men aged 55-59 is very unlikely to have been due to incentives to retire early because these ages are well below the pension eligibility age and below the age at which superannuation can be withdrawn tax-free (60 years). thus, this trend suggests involuntary causes. australian population studies 4 (1) 2020 mcdonald & moyle 25 4. retention analysis from the abs labour force surveys, it is possible to examine the net probability of retention in employment for cohorts as they age from 55-59 years in one survey to 60-64 years in the survey five years later. figure 4 shows these retention rates averaged for the months of june and july. averaging was applied to reduce the effects of sampling variation, but the rates for individual months show very similar trends. while this is a ‘net’ probability with new entries balanced off against exits, longitudinal analysis from the censuses presented below indicates that entries to employment at these ages are uncommon. figure 4: net cohort probability of employment retention, age group 55-59 to age group 60-64, by year aged 55-59, average of months of june and july source: derived from abs 2019a the interesting feature of this chart is that the probability of retention for both men and women levelled off from as early as the 2002 to 2007 period for men and 2004 to 2009 for women, earlier than the point at which the employment rates flattened (around 2010). this means that the flattening of employment rates for age group 60-64 was the result of the cessation of increases of employment rates in age group 55-59, rather than a drop in the retention rate between 55-59 and 60-64. similar data are shown by single years of age and for a wider range of ages for the two intercensal periods, 2006-11 and 2011-16 in figure 5. the more refined, single year of age data show some effect of the pension eligibility ages, particularly for men. interestingly, the charts for both men and women indicate that retention rates level off from age 65 onwards (retention from 60 to 65 in the chart) at about 50-60 per cent, although, at these ages, retention fell somewhat from the first to the second intercensal period. for men, the retention rates were lower in the 2011-16 period than in the 200611 period from age 55 upwards. thus, to a small extent, lower retention across time, especially for men, was a contributing factor to the flattening of employment rates. 26 mcdonald & moyle australian population studies 4 (1) 2020 figure 5: net cohort probability of retention in employment across five-year intercensal periods by age at the start of the intercensal interval source: derived from abs 2016 5. leaving and entering employment at older ages the longitudinal census databases enable the tracking of the employment of individuals across intercensal periods. rather than ‘net’ retention as applied in the previous section, the separate components of exit and entry can be examined from these data. the percentages leaving employment by age were quite similar across the two intercensal periods for both men and women but a little higher at older ages for the 2011-16 period compared with the 2006-11 period (figure 6a). the life table data (figure 6b) provide a perspective on the cumulative impact of the different rates of entering employment across the two intercensal periods; the impact is almost negligible. the rates of entering employment are noticeably higher at the younger ages for the 2006-11 period compared with the 2011-16 period (figure 7a) and these rates convert into higher cumulative rates of entry based on the life table data (figure 7b). indeed, most of the life table effect of entry to employment is achieved by the mid-50s ages. the conclusion to be drawn from this analysis of leaving and entering employment is that employment rates at older ages increased before 2010 not because of higher employment retention but because of high rates of entry among those who were not employed. the rates then flattened after 2010 because the rates of entry for those not employed were lower in the later period. also, the higher rates of entry to employment in the first period had their main effect at younger ages (late 40s and early 50s). these results support the hypothesis that the growth of employment at older ages was the result of strong labour demand prior to 2010 which dissipated after 2010. from a policy perspective these results imply that, if the aim is to increase older age employment, there should be a focus on increasing employment participation below age 55. higher rates would then persist into older ages because of the relative constancy of retention rates. to confirm this hypothesis, individual australian population studies 4 (1) 2020 mcdonald & moyle 27 level statistical analysis could examine whether people who entered employment during the economic boom in middle age continued to be employed into the future. figure 6: a. percentage leaving employment between censuses by single years of age, 2006-11 and 201116; b: the same data in a cumulative life table format source: abs 2018b notes: a: this is the rate (%) at which employed persons by their age at the first census are not employed five years later at the succeeding census. b: from an initial cohort of 100,000 employed persons at exact age 48, b shows the number still employed at subsequent exact ages when subjected cumulatively to the rates shown in a. 28 mcdonald & moyle australian population studies 4 (1) 2020 figure 7: a. percentage entering employment between censuses by single years of age, 2006-11 and 201116; b: the same data in a cumulative life table format source: abs 2018b notes: a: this is the rate (%) at which persons not employed by their age at the first census are employed five years later at the succeeding census. b: from an initial cohort of 100,000 not employed persons at exact age 48, b shows the number still not employed at subsequent exact ages when subjected cumulatively to the rates shown in a. 6. reasons not employed the abs monthly labour force surveys include a question on reasons for not being in the labour force. in figure 8, these responses for persons aged 55-64 are combined with those persons reported to be employed and unemployed. the result shown in figure 8 is a distribution from april 2001 to november 2019 of all persons into four categories: employed, unemployed (formal) plus others who looked for work in any way, those who did not look for work, and those institutionalised or permanently unable to work. australian population studies 4 (1) 2020 mcdonald & moyle 29 figure 8: percentage employed and reasons for not being employed, persons aged 55-64, australia, april 2001 to november 2019 source: abs 2019b 30 mcdonald & moyle australian population studies 4 (1) 2020 the figures for both men and women show that change from 2001 to 2010 was related to a shift out of the category ‘did not look for work’ into the two categories, ‘employed’ and ‘institutionalised and permanently unable to work’. from 2010 onwards, there is almost no change, especially for men. there was little change in the ‘unemployed or looking for work’ category. the strong labour demand from 2001 to 2010 appears to have led many in the ‘did not look for work’ category into employment – and this may have occurred at ages earlier than 55. others from the same category in the same period shifted into ‘institutionalised or permanently unable to work’. the percentages institutionalised were stable across time, so this was a result of people reporting themselves as ‘permanently unable to work’ – presumably persons receiving the disability support pension which increased in value substantially during this period compared with the benefit paid to those unemployed. once more, a policy of promoting employment among discouraged job seekers aged 45 and over, particularly 45-54, seems appropriate especially if labour demand is strong. 7. occupational change having concluded that older-age employment increased in the first decade of the 21st century because of higher rates of employment at younger ages (45-54) from about 1998 onwards combined with retention of those workers across time, it is useful to consider this in terms of occupations. figure 9 shows two groupings of occupation for men and women that account for around 75 per cent of all workers around age 55. the two groups can be classified as higher (managers and professionals for both men and women) and lower (technicians, trades, machine operators and drivers for men; clerical, administrative and community and personal services workers for women). the first point to make is that leaving employment around the pension age is much more prevalent for the lower occupations than for the higher occupations. for men, these are ‘manual’ jobs that may become more difficult to perform with increasing age. interestingly, however, the occupation of labourer, the most physically demanding occupation, maintained its percentage of total male and female employment with increasing age (not shown). for women, leaving employment is more frequent around the pension age for those in the lower, white collar group. given that men with higher level occupations are more likely to be married to women with higher level occupations, joint decision making may have a role in these patterns. more importantly for the theme of this paper, across time the lower level occupations for both men and women make up a larger proportion of employment at the oldest ages (63-73). this strongly suggests, as might be expected, that those who entered employment at younger ages from 1998 (during the boom years) and subsequently retained employment primarily entered the lower level occupations. 8. other employment differentials it has been demonstrated with earlier data that there are some significant differences in employment status at older ages by socio-economic characteristics (mcdonald 2011). there is a potential that employment rates flattened at older ages after 2010 because of changes in the australian population studies 4 (1) 2020 mcdonald & moyle 31 figure 9: percentage of total employment working in selected major occupation groups by age and sex, 2006, 2011 and 2016 source: derived from abs 2016 structure of the older age population by characteristics associated with employment. to consider this possibility, this section briefly examines more recent data on the socio-economic determinants of employment at older ages. earlier withdrawal from employment is heavily associated with being an employee rather than being an employer or self-employed for both men and women (figure 10). however, the changes across time are small and are unlikely to have contributed to the flattening of participation at older ages after 2010. public sector employment as a percentage of total employment falls off very sharply at older ages for both men and women (figure 11). public sector employees normally have good superannuation and the time trend data suggest that superannuation policy changes commencing on 1st july 2007 which 32 mcdonald & moyle australian population studies 4 (1) 2020 figure 10: employees as a percentage of total employment by age and sex, 2006, 2011, and 2016 source: derived from abs 2016 figure 11: public sector employees as a percentage of total employment by single years of age and sex, 2006, 2011 and 2016 source: derived from abs 2016 provided major superannuation incentives to continue working to age 60 had a large impact on the retention of public sector workers. the chart clearly shows that employment for public sector workers fell off from age 55 in 2006 but this shifted to 60 from 2011 onwards. as public sector employment is a much higher percentage of total employment for women than for men, this would have contributed to better employment participation outcomes for women overall in the post-2010 period. this analysis suggests that public sector employees stop working in the public sector because of a positive retirement income effect. australian population studies 4 (1) 2020 mcdonald & moyle 33 figure 12: percentage employed by partner status, sex and age, 2016 source: derived from abs 2016 for both older men and women, being partnered has a strong positive association with being employed (figure 12). for those not partnered, employment rates for men and women are the same at all ages. while these differences are interesting, the same patterns are observable over time and disaggregation by partner status does not contribute to an explanation of the flattening of employment rates at ages 55 and over after 2010. however, employment rates for not partnered men between ages 45 and 54 were much lower in 2016 than they had been in 2006 and 2011 (not shown). in addition, the proportion of men at older ages who are not partnered is increasing over time. this will provide a small impetus towards lower employment for men aged 55-59 in the immediate future. figure 13: percentage employed by housing tenure, males by age, 2016 source: derived from abs 2016 34 mcdonald & moyle australian population studies 4 (1) 2020 previous research based on 2006 census data has shown that, at older ages, persons holding a mortgage are more likely to be employed than full owners or renters (mcdonald 2011). these differences were still evident in 2016 (figure 13). however, within each tenure category, there was very little change in employment participation from 2006 to 2016 at older ages and, where there was change, it was consistent with the pattern for all men (not shown). however, the proportions of men holding a mortgage at older ages is increasing a little across time which could have a small positive impact on employment at older ages in the future. 9. conclusion the central conclusion of the analysis is that, for both men and women, older age employment increased in the first decade of the 21st century because expanding labour demand from 1998 drew people into employment who were not in the labour force, especially at ages 45-54. once in employment, those recruited in this way appear to have remained in employment. entry rates rose while retention rates remained stable. this pattern was more evident for those who entered lowerlevel occupations. this suggests that people out of the labour force who gain employment gather self-confidence and work experience that enables them to continue in employment. when labour demand cooled after the global financial crisis, entry rates stopped rising and, consequently, employment participation at older ages levelled off. specific to those with adequate superannuation, especially public sector workers, the analysis suggests that the 2007 policy changes to the taxation of superannuation induced more people to continue working until at least age 60. as the coverage of adequate superannuation grows, these policies should have a wider effect. there are substantial socio-economic differences in employment at older ages. married men are much more likely to be employed than single men; those with a mortgage are much more likely to be employed than full home owners or renters; those in good health are much more likely to be employed that those who are not (cai and kalb 2007; zucchelli et al. 2010); and employees and public sector workers are much more likely to retire earlier than other workers. however, the analysis here suggests that these differences along with structural changes in the characteristics of the population were insufficient to account for very much of the main employment changes at older ages. accordingly, there is little evidence of any long-term, social trend towards employment at older ages. key messages • for both men and women, older age employment increased in the first decade of the 21st century because expanding labour demand from 1998 drew people into employment who were not in the labour force, especially at ages 45-54. • subsequently, retention rates remained stable. this pattern was more evident for those who entered lower-level occupations. • the 2007 policy changes to the taxation of superannuation seem to have induced more people to continue working until at least age 60, especially public sector workers. australian population studies 4 (1) 2020 mcdonald & moyle 35 • changing socio-economic characteristics of the population and changes in employment rates associated with those characteristics had only minor impacts on aggregate employment trends. • there is little evidence of any longer-term social trend to higher employment at older ages. acknowledgements the paper was prepared as part of the research endeavour of the arc centre of excellence in population ageing research (cepar). references abs (2016) census of population and housing 2006, 2011 and 2016 (accessed via tablebuilder). canberra abs. https://www.abs.gov.au/websitedbs/d3310114.nsf/home/about+tablebuilder abs (2018a) labour statistics: concepts, sources and methods, feb 2018. catalogue no. 6102.0.55.001. canberra: abs. https://www.abs.gov.au/ausstats/abs@.nsf/mf/6102.0.55.001 abs (2018b) australian census longitudinal dataset (acld) (accessed via tablebuilder). canberra: abs. https://www.abs.gov.au/websitedbs/d3310114.nsf/home/about+tablebuilder abs (2019a) behind the numbers: how many people work one hour/week? abs media release, 14 february 2019. https://www.abs.gov.au/ausstats/abs@.nsf/lookup/6291.0.55.001media%20release1dec%2020 18 abs (2019b) labour force australia, detailed – electronic delivery. catalogue no. 6291.0.55.001. table 01 canberra: abs. https://www.abs.gov.au/ausstats/abs@.nsf/detailspage/6291.0.55.001nov%202019?opendoc ument abs (2019c) labour force australia, detailed – electronic delivery. catalogue no. 6291.0.55.001. table nm1 canberra: abs https://www.abs.gov.au/ausstats/abs@.nsf/detailspage/6291.0.55.001nov%202019?opendoc ument bloom d, canning d, and fink g (2011) implications of population ageing for economic growth. nber working paper series, working paper 16705. cambridge ma: national bureau of economic research. cai, l and kalb, g (2007) health status and labour force status of older working-age australian men. australian journal of labour economics. 10(4): 227-252. department of the treasury (2002) intergenerational report 2002-03. 2002-03 budget paper no.5. canberra: department of the treasury. department of the treasury (2010) australia to 2050: future challenges (the third intergenerational report). canberra: department of the treasury. house of representatives standing committee on health and ageing (2005) future ageing: inquiry into long-term strategies to address the ageing of the australian population over the next 40 years. canberra: parliament of the commonwealth of australia. mcdonald p (2011) employment at older ages in australia: determinants and trends. in: griffin t and beddie f (eds) older workers: research readings. adelaide: national centre for vocational education research, department of education, employment and workplace relations; pp. 25-41. productivity commission (2005) economic implications of an ageing australia. canberra: commonwealth of australia. swoboda k (2014) chronology of major superannuation and retirement income changes in australia. canberra: parliamentary library, department of parliamentary services. https://www.abs.gov.au/websitedbs/d3310114.nsf/home/about+tablebuilder https://www.abs.gov.au/ausstats/abs@.nsf/mf/6102.0.55.001 https://www.abs.gov.au/websitedbs/d3310114.nsf/home/about+tablebuilder https://www.abs.gov.au/ausstats/abs@.nsf/lookup/6291.0.55.001media%20release1dec%202018 https://www.abs.gov.au/ausstats/abs@.nsf/lookup/6291.0.55.001media%20release1dec%202018 https://www.abs.gov.au/ausstats/abs@.nsf/detailspage/6291.0.55.001nov%202019?opendocument https://www.abs.gov.au/ausstats/abs@.nsf/detailspage/6291.0.55.001nov%202019?opendocument https://www.abs.gov.au/ausstats/abs@.nsf/detailspage/6291.0.55.001nov%202019?opendocument https://www.abs.gov.au/ausstats/abs@.nsf/detailspage/6291.0.55.001nov%202019?opendocument 36 mcdonald & moyle australian population studies 4 (1) 2020 temple j and mcdonald p (2017) population ageing and the labour force: 2000-2015 and 2015-2030. australasian journal of ageing 36(4): 264-270. temple j, rice j, and mcdonald p (2017) ageing and the economic life cycle: the national transfer account approach. australian journal on ageing 36(4): 271-278. zucchelli e, jones a, rice n, and harris a (2010) the effects of health shocks and labour market exits: evidence from the hilda survey. australian journal of labour economics 13(2): 191-218. abstract background aim data and methods two main data sources are used: published tables from the monthly australian bureau of statistics labour force surveys and the australian censuses for the years 2006, 2011 and 2016. the methods used are primarily descriptive. results strong labour demand in the first decade of the 21st century stimulated the entry to employment of those out of the labour force, especially at ages 45-54 and especially for men. a cooling of labour demand following the global financial crisis termina... conclusions employment rates at older ages in australia in the first two decades of the 21st century were the results of shifts in labour demand before and after the global financial crisis. policy related to the taxation of superannuation also induced workers wi... key words older age employment, australia, labour demand, superannuation, gender differences. 2. abs labour force definitions key messages acknowledgements references australian population studies 2019 | volume 3 | issue 2 | pages 16-28 © tan, cebulla, ziersch & taylor 2019. published under the creative commons attribution-noncommercial licence 3.0 australia (cc by-nc 3.0 au). journal website: www.australianpopulationstudies.org australia’s state specific and regional migration schemes: exploring permanent and temporary skilled migration outcomes in south australia george tan * the university of adelaide; torrens university australia andreas cebulla the university of adelaide anna ziersch flinders university andrew taylor charles darwin university * email: ghim.tan@adelaide.edu.au. address: hugo centre for migration and population research, the university of adelaide, sa 5005 paper received 1 august 2019; accepted 29 september 2019; published 18 november 2019 abstract background recent concerns about population growth and its consequences in sydney and melbourne have added momentum to the debate on ways to achieve a more even geographic distribution of population. however, there is little contemporary evidence about the impact of regionally-focused immigration policies in delivering positive migrant outcomes and easing pressures in major cities. aims the aim of this paper is to compare migration, employment and settlement outcomes between permanent and temporary skilled migrants to south australia (sa) as well as the factors influencing migrants’ decisions to move into and out of the state. data and methods data in this paper draws on the south australian general skilled migrant survey of state-sponsored skilled migrants conducted by the university of adelaide in 2015. results lifestyle and employment factors were important in decisions to come to, stay or leave sa. permanent migrants were more likely to choose sa as a destination because it was perceived as a good place to raise a family, while temporary migrants were more likely to cite employment. temporary visa holders had relatively poor employment outcomes. conclusions temporary and permanent visa holders experienced different settlement and employment outcomes, demonstrating that a more detailed understanding of migrant characteristics and outcomes may be useful in designing and evaluating regionally-focused migration initiatives. key words international migration; permanent migration; temporary migration; skilled migration; regional migration; south australia; australia. http://www.australianpopulationstudies.org/ mailto:ghim.tan@adelaide.edu.au australian population studies 3 (2) 2019 tan et al. 17 1. introduction skilled migration streams are important in the context of skills shortfalls and demographic changes in many advanced economies. this is more pronounced in regional areas of high-income countries (hugo and moren-alegret 2008) and australia is not immune to these challenges. australia’s current immigration policy favours young and skilled applicants who meet specific criteria, including being qualified to work in occupations listed on the national skilled occupation lists (sol) (department of home affairs 2019); the outcome of the policy is substantial migration flows to the largest cities. a raft of immigration policies and regional migration schemes introduced since the mid-1990s, such as the state specific and regional migration (ssrm) scheme, the regional sponsored migration scheme (rsms) scheme, and the designated area migration agreements (dama), have aimed to address difficulties in filling occupations on the sol in regional areas. for some regions this has helped arrest regional and rural population decline. in the northern territory for example, alice springs was successful in attracting skilled migrants to the health sector between 2006 and 2011, changing the demographic trajectory of the town (taylor et al. 2014). population growth in australia, both numerically and in terms of growth rates, is generally higher in the larger states. for example, over the decade to 2018 victoria experienced the largest percentage population growth of any jurisdiction (an increase of 1,203,390 or 23%) followed by queensland (792,671; 19%) and new south wales (1,043,803; 15%). moreover, the capital cities of these states (melbourne, brisbane and sydney accounted for most of the total national population growth. at the other end of the scale, tasmania (increasing by 29,529 or 6%) and south australia (147,724; 9%) registered the smallest and second smallest percentage increases respectively over the decade (abs 2019a). higher growth in the rest of australia, particularly in the eastern states, has seen sa’s share of the national population decline from 8.0 per cent in 1998 to 7.5 per cent in 2008 and 6.9 per cent in 2018 (abs 2018). this is significant as sa’s demographic challenges are not only highlighted by net interstate migration losses, averaging -4,531 annually during 2008-09 to 2017-18 (abs 2019b), but also significant population ageing. twenty per cent of the population were aged 65 years and over in 2018, making sa australia’s second oldest jurisdiction behind tasmania at 21 per cent (abs 2019c). these challenges highlight the importance of international migration and regional immigration initiatives for addressing demographic challenges facing sa and other slower growing states and territories. while schemes such as the ssrm have been successful in diverting “a small but significant part of the australian immigration intake” to areas outside the major cities (hugo 2008a p.143), recent debate on population pressures in melbourne and sydney has highlighted the need for a national population policy (allen 2018; burton 2018). channelling, settling and retaining migrants in regional australia (including rural areas, remote areas and low population growth metropolitan areas) requires an integrated approach across all levels of government, an approach also promoted in the national population policy launched by the australian government in 2019, ‘planning for australia’s future population’ (department of the prime minister and cabinet 2019). because of regular modifications to australian immigration policy, evaluating migrant outcomes is a difficult but important task. research on migration in australia has tended to focus on populations from specific regions or countries such as china and india (e.g. hugo 2008b) or specific types of migrants like temporary migrants (e.g. khoo et al. 2007) and international students (e.g. tan and hugo 2017). research focusing 18 tan et al. australian population studies 3 (2) 2019 on migration, employment and settlement outcomes within a scheme such as the ssrm through dissection of characteristics, such as temporary versus permanent streams, is limited. we extend the work of khoo (2014) to compare skilled temporary migrants and permanent migrants to sa and analyse migration motivations, employment and future intentions in a regional migration scheme. such evidence has the potential to augment policy-making decisions on attracting and retaining migrants to regional areas and low population growth metropolitan regions, such as adelaide, that qualify for regional status. this evidence is timely as the australian government reviews its objectives with the establishment of a national centre for population. in this paper we present research comparing permanent and temporary migration outcomes in sa for migrants within the state-sponsored/nominated general skilled migration (gsm) program, one of the ssrm programs. our aim is to provide evidence on the effectiveness of regional immigration initiatives in states/territories wholly defined as regional for immigration purposes by focusing on the heterogeneity of migrants and factors influencing their migration, employment and settlement outcomes. this will enable the identification of opportunities for maximising the benefits of skilled migration to sa and to other similar region areas in relation to the attraction and retention of skilled migrants, migrant outcomes and population distribution. 2. data and methods this study adopted a mixed-methods approach consisting of a large survey and in-depth interviews to investigate migration outcomes for state-sponsored/nominated skilled migrants in sa under the gsm programme during 2010-2014. sa is a good case study because it provides a contemporary perspective on states and territories on the periphery of the australian migration system as all of sa is defined as ‘regional’ under the ssrm scheme. the survey was sent to all state-nominated gsm migrants who, as primary applicants: i. received sa state-nomination in the state migration plan period of july 2010 to december 2014; and ii. had been granted either a skilled regional provisional visa (subclasses 489, 475 and 487) or a skilled sponsored permanent visa (subclasses 190, 176 and 886). the survey commenced (with ethics approval from the university of adelaide) in early november 2015 and closed in mid-december 2015. it was sent by email on behalf of the research team by immigration sa to all skilled migrants on their administrative database who received statesponsorship/nomination in the study period. the response rate was 43.3 per cent (n=3,222) of the total sampling frame with valid email addresses (n=7,440). two-thirds of respondents (n=2,114) answered all survey questions relevant to them, leaving a useable, complete return response rate of 28.4 per cent. the survey data were weighted to immigration sa’s administrative database of the total sample population based on visa subclasses, age, country of birth and nominated occupation. twenty volunteers from a wide range of nationalities were also recruited via the survey for interviews that explored personal settlement experiences in greater depth. in the paper we report statistics based on analysis of the weighted sample of the 2,114 complete responses. australian population studies 3 (2) 2019 tan et al. 19 3. results 3.1. overview of survey respondents the number of survey respondents in each state-nominated visa subclass is presented in table 1 below. two permanent stream visas, the 190 skilled nominated–permanent visa (38%) and 176 skilled sponsored–permanent visa (27%), accounted for almost two-thirds of visa applicants. this was indicative of a response to the shift towards a ‘demand driven model’ for permanent skilled migrants by the australian government in 2010 which prioritised employer and government sponsored applicants for processing (spinks 2010). for the purposes of the analysis, we classified the visa categories into ‘permanent’ and ‘temporary’ visa holders. table 1: state-nominated visa subclass held by survey respondents number of respondents % total respondents permanent 190 skilled nominated-permanent 835 39.5 176 skilled sponsored-permanent 576 27.3 886 skilled sponsored-permanent 70 3.3 temporary 475 skilled regional-provisional 431 20.4 489 skilled regional-provisional 186 8.8 487 skilled regional sponsored-provisional 13 0.6 495 skilled independent regional-provisional 2 0.1 total 2,114 100.0 source: sa-gsm survey 2015 table 2: demographic characteristics of survey respondents % permanent visa holders (n=1,440) % temporary visa holders (n=674) % total respondents (n=2,114) male 68.7 70.0 69.1 female 31.3 30.0 30.9 under 25 years 2.9 3.0 2.9 25-29 years 20.8 28.7 23.3 30-34 years 33.0 28.9 31.7 35-39 years 24.4 23.1 24.0 40 and over years 18.9 16.3 18.1 postgraduate degree 36.7 37.7 37.1 graduate diploma 8.4 5.8 7.6 bachelor’s degree (incl. honours) 38.9 44.0 40.5 diploma/certificate/other post school qualification 14.9 11.9 13.9 no post school qualification 1.1 0.7 1.0 source: sa-gsm survey 2015 20 tan et al. australian population studies 3 (2) 2019 as shown in table 2, the migrants were disproportionately male (69%) with half of all visa holders aged 34 years or younger. in line with visa requirements, they were highly educated with the majority holding an undergraduate degree (41%) followed by 37 per cent with a postgraduate degree. the main source countries were india (26%), united kingdom (14%) and china (7%), mirroring a national pattern of recent arrivals (excluding new zealand) from 2012 to 2016 (abs 2017). the majority of visa holders were nominated or sponsored in ‘professional’ occupations (67%), followed by ‘technicians and trades workers’ (15%) and ‘managers’ (13%) (table 3). a higher proportion of technicians and trades workers obtained permanent skilled visas, whereas community and personal service workers, and clerical and administrative workers were more strongly represented amongst temporary visa holders. table 3: nominated occupations (anzsco major group) % permanent visa holders (n=1,429) % temporary visa holders (n=668) % total respondents (n=2,097) managers 12.7 12.0 12.5 professionals 67.3 67.3 67.3 technicians and trades workers 16.8 12.0 15.2 community and personal service workers 2.0 5.0 3.0 clerical and administrative workers 1.1 3.5 1.8 sales workers 0.1 0.3 0.2 source: sa-gsm survey 2015 3.2. settlement outcomes this section presents key finding relating to settlement and migration outcomes for survey participants, augmented with quotes from in-depth interviews. motivations for moving to sa about three quarters of migrants (74%) indicated sa was their first-choice destination when migrating to australia. they cited ‘lifestyle’ (14%) and a ‘good place to raise a family’ (18%) amongst their top five reasons for choosing sa (figure 1). the perception that sa offered ‘employment opportunities’ (8%) and ‘cheaper housing’ (6%) also featured, but less so. a notably larger proportion (15%) indicated they had ‘no alternative’ but to move to sa because they were unable to secure visa sponsorship elsewhere. this was particularly the case for temporary visa holders. positives and negatives of living in sa respondents were asked to rate aspects of their life in sa. of those who responded to this question (n=1,941), the majority (61%) indicated that their lives had improved and only 13 per cent said their lives had worsened, with the remainder saying it had stayed the same. in rating their levels of satisfaction with life in sa across a range of issues, respondents were most satisfied with the ‘lifestyle’ in sa (81%), ‘community safety’ (83%) and ‘travel time/traffic’ (83%) (figure 2). other australian population studies 3 (2) 2019 tan et al. 21 figure 1: reasons for moving to sa (up to five responses permitted) source: sa-gsm survey 2015 figure 2: levels of satisfaction with various aspects of life in sa source: sa-gsm survey 2015 22 tan et al. australian population studies 3 (2) 2019 factors such ‘climate’ (73%) and the state offering a positive environment for ‘raising a family’ (72%) also ranked highly. migrants were much less satisfied, though, with their incomes earned in sa (with 30% expressing dissatisfaction) and employment opportunities in the state (with 58% dissatisfied). overall, permanent and temporary visa holders expressed similar levels of satisfaction for all aspects. respondents were also asked about their onward migration intentions. as table 4 shows, of those still living in sa (n=1,887), 17 per cent intended to relocate (mostly to interstate destinations), with a further third (37%) undecided. overall, 37 per cent intended to migrate or had already migrated out of sa. the proportion intending to migrate from sa was lower than the estimated two-thirds of skilled migrants expected to leave sa within two years of arrival, according to the 2013 south australian parliamentary inquiry into new migrants (parliament of south australia 2013). although onward migration intentions are not necessarily accurate predictors of future movements, these statistics highlight early indicators of potential ‘migrant loss’ that may be susceptible to intervention. table 4: intention to migrate interstate or overseas in next 3 years % permanent visa holders (n=1,260) % temporary visa holders (n=617) % total respondents (n=1,877) yes interstate 16.2 14.9 15.8 yes overseas 1.0 0.8 0.9 no 46.4 47.1 46.6 undecided 36.5 37.3 36.8 source: sa-gsm survey 2015 eighty-two per cent of respondents intending to move attributed this to a lack of employment in sa, and/or better career opportunities elsewhere. interviews also indicated that the search for employment rather than living conditions shaped the thinking about whether migrants wished to remain in sa. “if i can find a job, get a contract, we will definitely stay here. but if i start to struggle with finding a job, or my husband. if he finds it difficult to find an ongoing job or ongoing contract, then we might think of moving to another city.” female, china, permanent visa for some, the quality of life in sa and other factors oriented around the household could weigh heavily in the decision to remain in sa even in the absence of desirable employment outcomes. “i have seen that other people from india i know, even if they are working in a job that is not to their skill level, they want to stay in adelaide... primarily because most of the people who are migrating here, they have family, they have kids and this seems to be one of the best places to stay with your kids...” male, india, permanent visa employment outcomes employment was a matter of concern for migrants for three reasons. first, despite their relatively high levels of qualification and nomination in occupations that were considered in short supply in sa, both temporary and particularly permanent visa holders, experienced rates of unemployment that were two to three times the state average (table 5) which ranged from 5.6 to 6.7 per cent from australian population studies 3 (2) 2019 tan et al. 23 2010-2014 (abs 2019d). second, those who had managed to obtain employment, excluding those self-employed, (n=1,440) often found themselves working in casual positions (19%) or on fixed-term contracts (19%), with the remainder in permanent positions (not shown). table 5: employment status % permanent visa holders (n=1,207) % temporary visa holders (n=621) % total respondents (n=1,828) self employed 5.5 6.4 5.8 employed in one job 69.0 72.6 70.3 employed in more than one job 8.0 10.7 9.0 unemployed 17.4 10.2 14.9 source: sa-gsm survey 2015 median income amongst migrants was about $65,000 per annum (figure 3), with more permanent skilled visa holders represented among the above-median income categories and more temporary visa holders (who had also typically been nominated in lower level occupations) reporting incomes below the median. although incomes were similar to the typical income distribution across more highly skilled occupations in sa (abs 2011), about one fifth of visa holders expressed dissatisfaction with their incomes (table 6; see also figure 2). figure 3: annual income levels source: sa-gsm survey 2015 temporary visa holders reported poorer employment outcomes, lower levels of satisfaction with their main job, and employment outside their nominated occupations, than permanent visa holders (table 6). similarly, 65 per cent of temporary visa holders reported working in jobs at lower levels than their previous occupations before migrating to sa, compared to 47 per cent of permanent visa holders. 24 tan et al. australian population studies 3 (2) 2019 table 6: satisfaction with main job, occupation (mis)match, comparison with previous occupational level % permanent visa holders (n=925) % temporary visa holders (n=516) % total respondents (n=1,441) very satisfied 10.3 9.3 9.9 satisfied 44.1 39.9 42.6 neither satisfied nor dissatisfied 25.7 30.0 27.2 dissatisfied 13.0 13.2 13.1 very dissatisfied 6.9 7.6 7.1 employed in nominated occupation 61.2 46.2 55.9 not employed in nominated occupation 38.8 53.8 44.1 current occupation at higher level than previous occupation 10.6 5.3 8.7 current occupation at same level than previous occupation 36.2 27.5 33.1 current occupation at lower level than previous occupation 47.3 64.5 53.4 not sure 5.9 2.7 4.8 source: sa-gsm survey 2015 key barriers to employment reported by migrants (figure 4) included a ‘lack of australian work experience’ (34%), followed by ‘no jobs available’ (24%), ‘no recognition of prior skills’ (10%) and ‘not [being a] permanent resident/australian citizen’ (6%). with the obvious exception of residency or citizenship status, the lack of australian work experience was prominent for both permanent and temporary visa holders. figure 4: barriers to employment source: sa-gsm survey 2015 australian population studies 3 (2) 2019 tan et al. 25 in our in-depth interviews, migrants elaborated on their experiences when seeking employment in sa, finding that employers were reluctant to recognise work experience and qualifications gained outside australia. “i've found that, for skilled positions, most employers only interview candidates with local experience. and are reluctant to spend time and effort trying to understand overseas qualifications and experience. they seem to be risk averse and prefer to hire someone with familiar credentials.” male, colombia, temporary visa faced with few alternatives, some migrants resorted to pursuing and accepting employment in less qualified positions: “we have both had to significantly tone down our experience…take up contracts in roles that we did maybe 10-12 years ago. so that was very disappointing.” female, india, permanent visa a complication that primarily, but not exclusively, affected temporary visa holders was a perception of employers’ preference for employees with permanent status: “before i had permanent residency being on a temporary visa was always an issue. people think you have an end date so that made some employers hesitant…until you get the permanent residency it's not 100% sure that you're going to be allowed to stay. so you're more likely to have to stay on a contract...” female, uk, previously on a temporary visa 4. discussion and conclusion there is little doubt that international migration and regionally-focused immigration policies have an important role in attracting and retaining migrants in australia’s regions. findings from this study underline how employment and lifestyle feature in the decision to move to and remain in places like sa. survey and interview findings highlight attributes relating to lifestyle, environment and climate, particularly when structured around the family, were important in the decision to move to sa. it was also revealing, particularly through interviews, how lifestyle and family-related factors for some migrants outweighed their careers in the decision-making process. this was more prevalent among permanent visa holders. temporary visa holders’ choices were more constrained as one of their top reasons for moving to sa was their inability to secure visa sponsorship from other state and territory governments. these differences highlight the importance of analysing skilled migrants across a range of migrant types and characteristics. here we were able to tease out other differences in the settlement and employment outcomes between permanent and temporary visa holders. our findings showed skilled migrants to be experiencing employment mismatch with their nominated occupations and/or working at levels of expertise lower than they had worked at prior to moving to sa, issues substantiated by cameron et al. (2019). however, our research also highlighted these issues were felt more keenly by temporary visa holders. 26 tan et al. australian population studies 3 (2) 2019 our findings indicated that poor employment outcomes were to linked to skilled migrants ‘leaking’ interstate and, while the extent of ‘leakage’ is not conclusive, the substantial proportion of undecided respondents in our survey indicates an opportunity for good policy to intervene and improve retention rates. sa was the first-choice destination for the majority of respondents and, coupled with the finding that most migrants had remained in sa and did not have any firm intentions to onward migrate, this challenges the perception that sa functions simply as a gateway to other parts of australia. multiple factors, some of which were identified in this study, influence the migration and settlement outcomes of permanent and temporary visa holders. these may be useful starting points for developing policies to attract and retain skilled migrants as we were able to distinguish differences in the motivations and experiences of those in different visa categories. in this context, the lack of local experience and non-recognition of overseas skills is crucial and needs to be addressed collaboratively between employers, peak industry bodies and authorities assessing the overseas work experience and qualifications of migrants (e.g. vocational education and training assessment services). the reported reluctance of some employers to hire migrants on temporary visas concurs with other research on international student graduates and other migrants on various temporary visas (coffey et al. 2018; robertson 2014) and is clearly an issue that needs to be further explored in a wider context. there was some evidence in this study to suggest that, given the right mix of conditions relating to employment and lifestyle, skilled migrants, particularly permanent visa holders, were inclined to remain in sa even when employment outcomes did not entirely meet expectations. this indicates that improving employment outcomes for skilled migrants only forms part of the answer in relation to retention, with lifestyle and family circumstances also significant in the decision-making process. this is consistent with the literature on the role of lifestyle and other non-economic factors in skilled migration in australia (khoo 2014). however, the challenge is to identify these conditions without losing sight of the need to enhance migrants’ employment prospects and outcomes (wickramarachchi and butt 2014). in summary, skilled migrants as a whole experienced a range of employment challenges in sa, with key differences in employment outcomes for temporary and permanent skilled migrants underscoring the importance of a nuanced approach to developing polices to attract and retain skilled migrants. more broadly, in the context of regionally-focused immigration policies, one should not generalise research findings to all skilled migrants simply because they enter australia under the same immigration regulatory framework and selection process. a comparative approach recognises the heterogeneity of migrants and the complex interplay between settlement location and migrant heterogeneity. it calls for the development of policies sensitive to the outcomes of migrant groups appropriate for south australia and other similarly defined ‘regional’ areas in australia in order to maximise the benefits of skilled migration. in a broader context, this approach can feed into regional migration schemes seeking to channel and divert migrants to the regions, and should be an important consideration for australia’s national population policy. australian population studies 3 (2) 2019 tan et al. 27 key messages • employer attitudes and structural factors can make it difficult for skilled migrants to obtain employment in sa, although these issues can impact temporary and permanent visa holders differently. • temporary visa holders experience relatively poorer employment outcomes. • it is important to address barriers to employment by targeting employers, industries and relevant assessing bodies. an understanding of potential differences in experiences of temporary and permanent visa holders should inform these approaches. • employment issues are important factors affecting the decision to remain rather than leave sa, but lifestyle and family-related reasons could be equally important, especially for permanent visa holders. • differences in migration and settlement outcomes between temporary and permanent visa holders highlight how measuring migration outcomes across a range of indicators and migrant types can be useful in developing policies for sa and other similar areas defined as regional. acknowledgements this research was funded by the department of state development – government of south australia and by the university of adelaide’s interdisciplinary research grant. we also thank hannah hia and dr romy wasserman for their assistance in preparing the primary data for analysis. references abs (2011) 2011 census of population and housing. canberra: abs. abs (2017) migrant data matrices 2017. catalogue no. 3415.0. canberra: abs. https://www.ausstats.abs.gov.au/ausstats/subscriber.nsf/0/62124eff57a0c8f8ca25836700813 678/$file/34150ds0090_2016_census_migrants.xls. accessed 20 june 2019. abs (2018) regional population growth, australia, 2016-17. catalogue no. 3218.0. canberra: abs. abs (2019a) australian demographic statistics. catalogue no. 3101.0. canberra: abs. abs (2019b) migration, australia, 2017-18. catalogue no. 3412.0. canberra: abs. abs (2019c) quarterly population estimates (erp) by state/territory, sex and age extracted from abs.stat. http://stat.data.abs.gov.au/. accessed on 5 june 2019. abs (2019d) labour force, australia. catalogue no. 6202.0. canberra: abs. allen l (2018) a whole-of-government approach to population policy for australia. australian population studies 2(2): 22-23. burton p (2018) spills and city deals: what turnbull’s urban policy has achieved, and where we go from here. the conversation. https://theconversation.com/spills-and-city-deals-what-turnbulls-urbanpolicy-has-achieved-and-where-we-go-from-here-102184. accessed on 19 june 2019. coffey j, farivar f and cameron r (2018) the job seeking experiences of international graduates in the host country: australia’s lost opportunity? the international journal of human resource management https://doi.org/10.1080/09585192.2018.1504106. department of home affairs (2019) skilled occupation list. https://immi.homeaffairs.gov.au/visas/working-in-australia/skill-occupation-list. accessed on 20 june 2019. https://www.ausstats.abs.gov.au/ausstats/subscriber.nsf/0/62124eff57a0c8f8ca25836700813678/$file/34150ds0090_2016_census_migrants.xls https://www.ausstats.abs.gov.au/ausstats/subscriber.nsf/0/62124eff57a0c8f8ca25836700813678/$file/34150ds0090_2016_census_migrants.xls http://stat.data.abs.gov.au/ https://theconversation.com/spills-and-city-deals-what-turnbulls-urban-policy-has-achieved-and-where-we-go-from-here-102184 https://theconversation.com/spills-and-city-deals-what-turnbulls-urban-policy-has-achieved-and-where-we-go-from-here-102184 https://doi.org/10.1080/09585192.2018.1504106 https://immi.homeaffairs.gov.au/visas/working-in-australia/skill-occupation-list 28 tan et al. australian population studies 3 (2) 2019 department of the prime minister and cabinet (2019) planning for australia’s future population. https://www.pmc.gov.au/sites/default/files/publications/planning-for-australias-futurepopulation.pdf. accessed on 18 june 2019. hugo g (2008a) australia’s state-specific and regional migration scheme: an assessment of its impacts in south australia. international migration and integration 9(2): 125-145. hugo g (2008b) in and out of australia: rethinking chinese and indian skilled migration to australia. asian population studies 4(3): 267-291. hugo g and morén-alegret r (2008) international migration to non-metropolitan areas of high income countries: editorial introduction. population, space and place 14(6): 473–477. khoo s (2014) attracting and retaining globally mobile skilled migrants: policy challenges based on australian research. international migration 52(2): 20-30. khoo s, mcdonald p, voight-graf c and hugo g (2007) a global labor market: factoring motivating the sponsorship and temporary migration of skilled workers to australia. international migration review 41(2): 480-510. parliament of south australia (2013) inquiry into new migrants, thirty fourth report of the social development committee of the south australian parliament, second session fifty-second parliament. parliament house, adelaide. robertson s (2014) time and temporary migration: the case of temporary graduate workers and working holiday makers in australia. journal of ethnic and migration studies 40(12): 1915-1933. spinks h (2010) australia’s migration program. parliamentary library. https://www.aph.gov.au/about_parliament/parliamentary_departments/parliamentary_library/ pubs/bn/1011/austmigration. accessed on 18 august 2015. tan g and hugo g (2017) the transnational migration strategies of chinese and indian students in australia. population, space and place 23(6): e2038. taylor a, bell l and gerritsen r (2014) benefits of skilled migration programs for regional australia: perspectives from the northern territory. journal of economic and social policy 16(1): 35-69. wickramaarachchi n and butt a (2014) motivations for retention and mobility: pathways of skilled migrants in regional victoria, australia. rural society 23(2): 188-197. https://www.pmc.gov.au/sites/default/files/publications/planning-for-australias-future-population.pdf https://www.pmc.gov.au/sites/default/files/publications/planning-for-australias-future-population.pdf https://www.aph.gov.au/about_parliament/parliamentary_departments/parliamentary_library/pubs/bn/1011/austmigration https://www.aph.gov.au/about_parliament/parliamentary_departments/parliamentary_library/pubs/bn/1011/austmigration abstract background aims data and methods data in this paper draws on the south australian general skilled migrant survey of state-sponsored skilled migrants conducted by the university of adelaide in 2015. results lifestyle and employment factors were important in decisions to come to, stay or leave sa. permanent migrants were more likely to choose sa as a destination because it was perceived as a good place to raise a family, while temporary migrants were more... conclusions temporary and permanent visa holders experienced different settlement and employment outcomes, demonstrating that a more detailed understanding of migrant characteristics and outcomes may be useful in designing and evaluating regionally-focused migrat... key words international migration; permanent migration; temporary migration; skilled migration; regional migration; south australia; australia. 2. data and methods key messages acknowledgements references australian population studies 2020 | volume 4 | issue 1 | pages 57-69 © wilson & temple 2020. published under the creative commons attribution-noncommercial licence 3.0 australia (cc bync 3.0 au). journal website: www.australianpopulationstudies.org to what extent is australia’s population ageing? the application of traditional and alternative ageing measures tom wilson* the university of melbourne jeromey temple the university of melbourne * corresponding author. email: wilson.t1@unimelb.edu.au. address: melbourne school of population and global health, the university of melbourne, 207 bouverie st, melbourne, vic 3010, australia. paper received 28 february 2020; accepted 11 may 2020; published 25 may 2020. abstract background most studies of population ageing apply traditional ageing measures, such as the number or percentage of the population aged 65 and above. in the context of gradually improving health and mortality at age 65, the use of a fixed age cut-off to define ‘older age’ needs to be revisited. aim the aim of this paper is to re-assess the extent of population ageing in australia and the states and territories over past decades and in the future as indicated by both traditional and alternative ageing measures. data and methods both numerical and structural ageing was measured using age cut-offs for the older population of (i) age 65, (ii) the age at which there is 15 years life expectancy remaining, and (iii) the age at which the mortality rate is above 0.01. the data consisted of life tables, population estimates and population projections. results both traditional and alternative ageing measures indicate considerable past and future numerical ageing. structural ageing has been strong since the 1970s in terms of the percentage aged 65+, but the alternative ageing measures paint quite a different picture of structural ageing both in the past and in the future. conclusions the use of a traditional measure of population ageing in combination with a mortality-based measure, such as the population with remaining life expectancy of under 15 years, is helpful for demographic analyses of ageing. key words population ageing; characteristics approach; australia; states and territories. http://www.australianpopulationstudies.org/ mailto:wilson.t1@unimelb.edu.au 58 wilson & temple australian population studies 4 (1) 2020 1. introduction in common with almost all countries of the world (un 2019), australia’s population is undergoing population ageing. traditional measures of population ageing include the numbers and proportions aged 65 and over, the elderly dependency ratio (the ratio of the population aged 65+ to those aged 2064), the ageing index (the ratio of the population aged 65+ to those aged 0-19), and median age (hobbs 2004; rowland 2003). according to these traditional measures, australia’s population is clearly undergoing ageing – both numerical ageing in which the absolute number of older people is growing, and structural ageing in which the proportion of the population in the older ages is increasing. for example, over the 20 years to 2018 the population aged 65+ increased from 2.3 million to 3.9 million, while the share of the total population in this age group rose from 12.2% to 15.7% (abs 2019a). traditional ageing measures implicitly refer to the overall ‘burden’ of an older population, including the costs associated with services such as health care, aged pension payments, home care packages, and residential aged care. but australians aged 65 today are generally in a much better position in terms of health and mortality relative to those of the same age many decades ago. for example, australian men aged 65 in 2018 had a life expectancy of 20.3 remaining years compared to 12.1 years 50 years earlier (authors’ life table calculations). the use of a chronological age cut-off for defining the older population which is invariant over time and space is not useful for many policy purposes. over the last decade or so several authors have proposed alternative measures of population ageing which reflect the shifting levels and age patterns of mortality, health, wellbeing, and economic activity (e.g. chomik et al. 2016; falkingham 2016; sanderson and scherbov 2007; sanderson and scherbov 2015; skirbekk et al. 2012; spijker 2015; spijker and macinnes 2013). sanderson and scherbov (2013) argue that the threshold of old age is best defined by a selected population characteristic, such as a certain number of years of life expectancy remaining, a particular mortality rate, or pension eligibility ages (which are slowly rising in some countries). the characteristic of 15 years remaining life expectancy has proved popular and been used in several studies as the starting age of old age (e.g. lutz et al. 2008; sanderson and scherbov 2008; un 2019). this “characteristics approach” to the measurement of population ageing therefore permits the onset of old age to vary over time and space, or between different sub-groups of a national population. a large number of studies on population ageing in australia exist, most of which employ traditional ageing measures. chomik et al. (2016) is one of the few exceptions. these authors use alternative dependency metrics for australia based on health and economic data. the aim of this paper is to reassess the extent to which the populations of australia and the states and territories are ageing as indicated by both traditional and alternative ageing measures. ageing is evaluated using actual data from 1921 to the present at the national scale, and from 1971 to the present at the state and territory scale, and then out to 2066 using our own projections. in section 2 below we describe the data we use, including projected populations and life tables, and define the chosen traditional and alternative measures of population ageing that are applied. the results are presented and discussed in section 3, followed by some concluding remarks about the alternative ageing measures and australia’s population ageing in the final section. in a rapidly ageing society, the demand for amenity, social support, and health and aged care facilities are of growing concern (piggott et al. 2016), and the regional differences in migration flows have increasingly significant socioeconomic, political, and australian population studies 4 (1) 2020 wilson & temple 59 developmental implications (jackson & felmingham 2002). as understanding the past trends of elderly migration can assist in predicting and planning for future growth, the aim of this paper is to understand how the impact of migration for older australians has changed over the past forty years in comparison to the total population. to achieve this aim, this paper first establishes the national-level intensity, effectiveness, and impact of migration for each of the eight census periods from 1976-81 to 2011-16. the paper then maps the impact of these moves at the regional level to identify which regions are experiencing gains and losses. finally, the paper speculates on possible explanations for the observed trends. 2. data and methods 2.1. data the population estimates used in this study include national and state/territory estimated resident populations (erps) from the abs from 1971-2019 (abs 2019a) as well as national population estimates from 1921-1970 obtained from the human mortality database (hmd 2019). national and state/territory death counts by sex and single years of age were sourced from the abs for the period 1971-2018 (abs 2019b) while national deaths for 1921-1970 were obtained from the hmd (2019). national scale population estimates above age 85 for 1971-2018 were those estimated by wilson and temple (2020) using extinct cohort and survivor ratio methods, which tend to be more accurate than erps at the highest ages. life tables were created for each year from 1921 to 2018 for australia and from 1971 to 2018 for each of the states and territories. these were based on the formulae set out in preston et al. (2001). age-specific death rates were calculated for abridged life table age groups for the two territories and tasmania due to the small numbers of deaths by age and sex in those jurisdictions. the grouped death rates were then interpolated to single years of age using cubic splines (above age 25) and linear splines at younger ages (where cubic splines work less well). population projections for australia and the states and territories were created using a cohortcomponent projection model. the projections started with 2016 erps and extended out to 2066, with projections in the short-run up to 2018 constrained to known demographic parameters. the projections were based on ‘business as usual’ assumptions after 2018, reflecting fertility and migration trends of recent years. immigration and emigration flows were constrained to an annual net overseas migration assumption of 225,000, while projected interstate inand out-migration flows were constrained to the average annual net interstate migration values of the 2011-16 intercensal period. national fertility was set to a long-run total fertility rate of 1.75, with state and territory tfrs varying by fixed differentials based on recent trends. mortality projections were derived from a long-run extrapolation of age-specific death rates using ediev’s (2008) method. state and territory life expectancy at birth was assumed to move in parallel with national life expectancy into the future. age-specific death rates corresponding to each life expectancy assumption were selected from the national mortality projections (wilson 2018) for every jurisdiction except the northern territory. a separate mortality projection was prepared for the nt because of its unique mortality rate age profile. 60 wilson & temple australian population studies 4 (1) 2020 2.2. ageing measures the numerical ageing measures used in this paper are: • population aged 65+ • population at ages with remaining life expectancy of under 15 years (rle<15) • population at ages where age-specific death rates are greater than 0.01 (asdr>0.01). the population aged 65+ is a traditional ageing measure and widely applied. the rle<15 indicator was selected because of its growing popularity and therefore ability to enable comparisons with other studies using it (sanderson and scherbov 2013). the asdr>0.01 measure was selected because it provides an approximate and easily calculable way of defining the older population more likely to experience health problems (falkingham 2016). although both 15 years of remaining life expectancy and a death rate of 1% remain arbitrary measures, they do possess the important feature of being defined by mortality rather than a fixed age. they are also relatively easy to calculate, and do not possess the additional data requirements (such as age-specific labour force participation or disability) of some more complex ageing measures. both our selected ‘characteristics’ ageing measures first require estimation of the exact age where life expectancy is exactly 15 years and the age where the asdr equals 0.01. this was calculated separately for males and females and interpolated linearly between single year age intervals in annual period life tables. the male and female populations above those age cut-offs were then aggregated. more refined age cut-offs could have been obtained with cohort life tables, but due to their unavailability and the small difference it would have made, we made use of period life tables. the equivalent structural ageing measures used in the paper are: • percentage of the population aged 65+ • percentage of the population in age groups with rle<15 • percentage of the population in age groups with asdr>0.01 all of which use the numerical ageing measures above as numerators and the total population as denominators. 3. the extent of population ageing in australia 3.1. national-level trends the extent of australia’s past and projected population ageing is illustrated in figure 1. the upper graph shows the growth of the older population in terms of absolute numbers (numerical ageing), while the lower graph shows the proportion of the total population in the older age group (structural ageing). the traditional ageing measure, based on an older age group cut-off of age 65, is shown in black. the red lines indicate the older population defined by less than 15 remaining years of life expectancy (rle<15), which vary from 63.7 and 60.8 years for females and males respectively in 1921, through to 74.4 and 71.8 years by 2018, and then 78.9 and 77.8 years by the end of the projection horizon. the blue lines refer to the older population defined by age-specific death rates above 0.01 (asdr>0.01). in 1921 rates of 0.01 were reached by ages 53.0 and 47.6 years for females and males respectively, by ages 70.8 and 65.5 in 2018, and are projected to rise to ages 79.4 and 78.4 years by the end of the projections. australian population studies 4 (1) 2020 wilson & temple 61 figure 1: numerical and structural ageing of australia’s population, 1921-2066 source: authors’ calculations from abs data (solid lines) and own projections (dashed lines) the number of people aged 65+ has grown substantially since 1921 and has been growing particularly fast since around 2011 when the oldest of the baby boom generation started turning age 65 (figure 1, upper graph). currently the 65+ population stands at about 4 million. the projections suggest growth will continue over the coming decades, with the population of this age group reaching just over 10 million by 2066. this is a very substantial increase. however, the two alternative ageing measures paint a less dramatic picture. the population with rle<15 has grown at a roughly steady pace over the long-run, with just a slight uptick in growth observed very recently. growth of the older population defined as rle<15 has been notably slower in the past than that of the population aged 65+. the projections suggest faster numerical ageing in coming decades, with the rle<15 population increasing from about 2 million today to 4.5 million by the end of the projections. the population with asdr>0.01 has also grown at a fairly gradual pace over the longrun, with growth interrupted during the 1990s and 2000s, followed by a return to growth in the most recent decade. the population defined in this way is currently about 3 million. by the end of the projections it is anticipated to have grown to around 4.4 million. 62 wilson & temple australian population studies 4 (1) 2020 in terms of structural population ageing (figure 1, lower graph), the proportion of the population aged 65+ has increased substantially in recent years and is currently around 16%. it is projected to pass 23% by the mid-2060s. this rapid structural ageing trajectory indicated by the traditional ageing measure stands in stark contrast to ageing as measured by the proportions of the population with rle<15 and asdr>0.01. the proportion of the population with rle<15 has been slowly declining from about 10% since the mid-20th century, and is projected to increase modestly over the next two decades to return to about the same level and then remain fairly steady. the proportion of the population with asdr>0.01 has also been mostly declining since the mid-20th century, from around 18% then to about 12% today, and is projected to fall further over the coming decades. 3.2. state and territory trends state and territory ageing trends are illustrated in figures 2 and 3. figure 2 shows numerical ageing while figure 3 focuses on structural ageing. with respect to numerical ageing, all jurisdictions have experienced substantial growth of their 65+ populations over the past half century shown in the graphs, with an acceleration of ageing occurring in the last decade. south australia’s 65+ population has grown the least over the most recent decade (increasing by 32% between 2009 and 2019) while the northern territory’s numerical ageing has been the most rapid (72% growth). as the graphs show, all states and territories are projected to undergo rapid growth of their 65+ populations over the next 50 years. however, from the perspective of the size of the population with rle<15, most states and territories have experienced more modest amounts of numerical ageing over the past half century. for some jurisdictions there are periods of little change in this population in the last decade of the 20th century and first decade of the 21st. victoria has aged the least according to this measure (with the population of rle<15 increasing by 43% between 1971 and 2019) while the two territories’ populations have aged the most, albeit from a low base. the population with rle<15 in the act has increased by 445% between 1971 and 2019 while in the northern territory the growth was 223%. numerical ageing as measured by growth in the population with asdr>0.01 is also relatively modest, and, again for some jurisdictions, includes periods of no growth during the 1990s and 2000s. as figure 2 shows, projected growth of the asdr>0.01 population is variable between states and territories. in the future south australia and tasmania are expected to experience growth, and then decline, of this population. projections for the northern territory are always highly uncertain, but it appears the asdr>0.01 population in the territory will not increase. figure 3 presents past and possible future trends in structural ageing amongst the state and territory populations. it is clear that the percentage of the population aged 65+ has increased substantially in all jurisdictions over the past half century, with an acceleration of ageing observed over the last decade or so. increases are projected to continue for all states and territories for the projection horizon shown in the graphs. by 2066 the proportion of the population aged 65+ is expected to vary from 13% in the northern territory to 31% in tasmania. australian population studies 4 (1) 2020 wilson & temple 63 figure 2: numerical ageing of state and territory populations, 1971-2066 source: authors’ calculations from abs data (solid lines) and own projections (dashed lines) 64 wilson & temple australian population studies 4 (1) 2020 figure 3: structural ageing of state and territory populations, 1971-2066 source: authors’ calculations from abs data (solid lines) and own projections (dashed lines) australian population studies 4 (1) 2020 wilson & temple 65 but while the share of the population aged 65 and over will rise considerably in all jurisdictions, the share of the older population as defined by the mortality measures rle<15 and asdr>0.01 is unlikely to rise in many cases (figure 3). the percentage of the population with rle<15 is projected to increase in all jurisdictions initially before decelerating and then remaining roughly steady. the share of the population with asdr>0.01 is projected to decline gradually, or increase slightly before declining, across all states and territories. 3.3. understanding the alternative ageing trends as shown above, population ageing can be measured in a variety of ways. if ageing is measured in the traditional way using an age cut-off of 65 it gives very different results from ageing measured with a variable cut-off age dependent on a selected mortality characteristic. why does ageing measured by the traditional and alternative ageing measures differ so much? in short, they measure different aspects of population ageing. projected growth in the numbers of people aged 65+ can be attributed to increasing sizes of birth cohorts over time, net overseas plus interstate migration gains and improved survival (preston and stokes 2012). understanding growth in older populations defined by a mortality characteristic, such as the population with rle<15 or asdr>0.01, is slightly less straightforward. the growth of the older population defined in this way depends on the interaction between the growth of age-sex population groups above age 65 and the rate of progress in mortality which determines the older age group starting age. take the example of the population with rle<15. australia’s population in the 65+ age group is projected to grow substantially over the coming decades. at the same time, declining death rates mean that the starting age of the rle<15 group rises, increasing from about 74 years for females and 71 years for males in 2016 to about 79 and 78 years, respectively, by the end of the projection horizon. figure 4 illustrates the population aged 65+ in 2016, 2041 and 2066 divided into the older population with less than 15 years of life expectancy remaining (rle<15; shown in pink) and the younger population with more than 15 years remaining (rle>15; shown in grey). with the older population defined as rle<15, there is still substantial projected numerical ageing because the growth of population at higher ages is greater than the population excluded at lower ages due to the upward shift in the starting age of this group. as a proportion of the 65+ population, however, the rle<15 group is projected to occupy a declining share. the increasing share of the 65+ population with more than 15 years life expectancy remaining reflects projected progress in reducing mortality at older ages. explaining the trends in structural ageing requires attention be paid to overall population size, which forms the denominator of the proportions of the population in the rle<15 and asdr>0.01 age groups. as shown in figures 1 and 3, structural ageing according to the rle<15 and asdr>0.01 populations is projected to be modest or non-existent in the future. in australia as a whole, and many of the states and territories, overall population growth is projected to be strong. this is due to the combination of net overseas migration gains, mid-range fertility, and population momentum. it means that, although the size of populations with rle<15 and asdr>0.01 are projected to grow, they will not always increase as a share of total population. 66 wilson & temple australian population studies 4 (1) 2020 figure 4: the age-sex structure of australia’s population aged 65+ in 2016, 2041 and 2066, divided into those with remaining life expectancy under 15 years (rle<15) and those with more than 15 years remaining (rle>15) source: authors’ projections note: the cut-off ages for rle<15 had to be rounded to the nearest integer in these graphs 4. conclusion this paper has presented past and projected future trends in population ageing in australia. from the perspective of traditional ageing measures, australia’s population is currently undergoing both numerical and structural population ageing and will continue to do so for the entire projection horizon reported here (to 2066). ageing measured from a characteristics approach using the mortality definitions of rle<15 and asdr>0.01 presents a complementary perspective. the rle<15 population has grown steadily over the past half century and is projected to grow more strongly into the future nationally, with some variation by state and territory. the population with asdr>0.01 has grown modestly over the past half century, though with some interruptions, and at the national scale is projected to grow for much of the next half century. amongst the states and territories future growth of the asdr>0.01 population growth is more varied. structural population ageing according to the two alternative ageing measures has not occurred since the mid-20th century. at the national scale some ageing is projected to occur in the future according to the rle<15 measure but not according to the asdr>0.01 measure. the projections therefore present good news: the population australian population studies 4 (1) 2020 wilson & temple 67 above age 65 will grow but populations in the higher mortality groups of rle<15 and asdr>0.01 are not expected to grow as fast. we argue that this combination of ageing measures (traditional and mortality-based) provides demographers with a more detailed understanding of population ageing. in particular, the use of mortality-based measures may help when examining levels of structural and numerical ageing in different sub-groups of national populations. one example is the ageing of non-indigenous and aboriginal and torres strait islander populations. within the gerontological and public health literature, there is considerable debate about what constitutes ‘old age’ for australia’s first nations peoples. a general consensus, however, has formed that using a younger age cut-off of 45 or 50 is more appropriate for gerontological studies due to (i) a considerable gap in life expectancy between aboriginal and non-indigenous australians (abs 2013; aihw 2017), (ii) many conditions and comorbidities as well as frailties commonly associated with ageing occurring at earlier ages in this population (aihw 2016; gubhaju et al. 2013; hyde et al. 2016), and (iii) government programs such as those governing access to specific aged care services being available to aboriginal and torres strait islanders from earlier ages compared to non-indigenous australians. despite the potential benefits of the measures we outline, projections are uncertain, and this is one of the limitations of the paper. as a general rule, projections are more uncertain the further into the future they extend, the smaller the population being projected, and the greater the amount of migration populations experience. past projections for northern territory have been shown to be especially prone to error (wilson 2012). and demographic conditions may well change; different migration policy scenarios can make substantial differences to projected structural ageing (e.g. mcdonald 2016). major external shocks can also affect demographic change. because these projections were prepared before the emergence of covid-19, the demographic impacts of the pandemic have not been incorporated in our projection assumptions. the lower net overseas migration which australia is likely to experience for a while will affect structural population ageing, but the size of this impact will depend heavily upon government policy regarding the migration program over the medium term. given the young age structure of net overseas migration, covid-19 shocks will have a negligible impact on numerical ageing. finally, a test run of our projection model over the period 2011 to 2016 revealed the 65+ numerical and structural ageing measures to be projected very accurately after 5 years, with slightly less accuracy achieved for rle<15 measures, and slightly less accuracy still for asdr>0.01 measures. in summary, this paper demonstrates that the extent of past and future population ageing in australia, and in the states and territories, depends on how it is viewed. just two alterative ageing measures were considered, though many other characteristics-based measures could have been chosen. the main conclusion for researchers and policy makers is that population ageing can be viewed from a variety of perspectives, and that the most appropriate measure depends on the purpose of the study. it is important that the extent and speed of population ageing not be misinterpreted as a result of relying solely on one ageing measure. if the focus is on the retired population, then measures based on the economically inactive population at the older ages are recommended; if the study is about residential aged care needs, then the population requiring this care should be the population of interest in any ageing measure; if the focus is on health costs, then an ageing measure focusing on age-specific health costs is most appropriate. for general 68 wilson & temple australian population studies 4 (1) 2020 demographic analysis the use of a traditional measure (using age 65 as the cut-off age) in combination with a mortality-based measure, such as one founded on rle<15, is helpful. key messages • australia has been experiencing numerical population ageing for a long time according to both traditional and alternative characteristics-based ageing measures. there will be many more people in the older population (however defined) in coming decades. • structural population ageing according to the traditional measure of the percentage of the population aged 65+ has been occurring since the 1970s and is projected to continue in the decades ahead. • the proportion of australia’s population with less than 15 years remaining life expectancy (rle<15), and the proportion of the population with age-specific death rates above 0.01 (asdr>0.01), have generally declined over recent decades. structural population ageing is projected for the next two decades according to rle<15, but not according to asdr>0.01. • population ageing can be viewed from a variety of perspectives, and the most appropriate ageing measure depends on the purpose of the study. for general demographic analysis a combination of both traditional and mortality-based ageing measures is recommended. acknowledgements this research was supported by the australian research council centre of excellence in population ageing research (project number ce170100005). we are grateful to the anonymous reviewers for their helpful comments on the original version of the paper. data availability the population projections, including age cut-offs for the rle<15 and asdr>0.01 measures, are available in an excel workbook at https://figshare.com/articles/population_ageing_projections_for_australia/12228008. references abs (2013) life tables for aboriginal and torres strait islander australians, 2010-2012. catalogue no. 3302.0.55.003. canberra: abs. abs (2019a) australian demographic statistics, jun 2019. catalogue no. 3101.0. canberra: abs. abs (2019b) deaths, australia, 2018. catalogue no. 3302.0. canberra: abs. australian institute of health and welfare (aihw) (2016) australian burden of disease study: impact and causes of illness and death in aboriginal and torres strait islander people. australian burden of disease study series no.6. catalogue no. bod7. canberra: australian institute of health and welfare. australian institute of health and welfare (aihw) (2017) trends in indigenous mortality and life expectancy, 2001-2015: evidence from the enhanced mortality database. catalogue no. ihw 174. canberra: australian institute of health and welfare. chomik r, mcdonald p and piggott j (2016) population ageing in asia and the pacific: dependency metrics for policy. the journal of the economics of ageing 8: 5-18. https://figshare.com/articles/population_ageing_projections_for_australia/12228008 australian population studies 4 (1) 2020 wilson & temple 69 ediev d m (2008) extrapolative projections of mortality: towards a more consistent method. part i: the central scenario. working paper 3/2008. vienna institute for demography, austria. falkingham j (2016) the changing meaning of old age. esrc centre for population change briefing 31. university of southampton, uk. available at: http://www.cpc.ac.uk/docs/bp31_the_changing_meaning_of_old_age.pdf gubhaju l, mcnamara j, banks e, joshy g, raphael b, williamson a and eades s (2013) the overall health and risk factor profile of australian aboriginal and torres strait islander participants from the 45 and up study. bmc public health 13: 661. hmd (2019) human mortality database. university of california, berkeley (usa), and max planck institute for demographic research (germany). available at: www.mortality.org. hobbs f (2004) age and sex composition. in siegel j s and swanson d a (eds) the methods and materials of demography. second edition. amsterdam: elsevier; pp 125-173. hyde z, flicker l, smith k, atkinson d, fenner s, skeat l and lo giudice d (2016) prevalence and incidence of frailty in aboriginal australians, and association with mortality and disability. maturitas 87: 8994. jackson n and felmingham b (2002) as the population clock winds down: indicative effects of population ageing in australia’s states and territories. journal of population research 19(2): 97-117. kendig h, mcdonald p and piggott j (eds) population ageing and australia’s future. canberra: anu press. lutz w, sanderson w and scherbov s (2008) the coming acceleration of global population ageing. nature 451: 716-719. mcdonald p (2016) ageing in australia: population changes and responses. in kendig h, mcdonald p and piggott j (eds) population ageing and australia’s future. canberra: anu press; pp 65-83. preston s h, heuveline p and guillot m (2001) demography: measuring and modeling population processes. oxford: blackwell. preston s h and stokes a (2012) sources of population aging in more and less developed countries. population and development review 38(2): 221-236. rowland d (2003) demographic methods and concepts. oxford: oxford university press. sanderson w c and scherbov s (2007) a new perspective on population aging. demographic research 16(2): 27-58. sanderson w c and scherbov s (2015) are we overly dependent on conventional dependency ratios? population and development review 41(4): 687-708. skirbekk v, loichinger e and weber d (2012) variation in cognitive functioning as a refined approach to comparing aging across countries. proceedings of the national academy of sciences 109(3): 770774. spijker j (2015) alternative indicators of population ageing: an inventory. working papers 4/2015. vienna institute of demography, austria. spijker j and macinnes j (2013) population ageing: the timebomb that isn’t? british medical journal 347: f6598. un (2019) world population ageing 2019. new york: un. wilson t (2012) forecast accuracy and uncertainty of australian bureau of statistics state and territory population projections. international journal of population research. volume 2012 wilson t (2018) evaluation of simple methods for regional mortality forecasts. genus 74:14. wilson t and temple j (2020) the rapid growth of australia’s advanced age population. unpublished paper. http://www.cpc.ac.uk/docs/bp31_the_changing_meaning_of_old_age.pdf http://www.mortality.org/ abstract background aim data and methods both numerical and structural ageing was measured using age cut-offs for the older population of (i) age 65, (ii) the age at which there is 15 years life expectancy remaining, and (iii) the age at which the mortality rate is above 0.01. the data co... results both traditional and alternative ageing measures indicate considerable past and future numerical ageing. structural ageing has been strong since the 1970s in terms of the percentage aged 65+, but the alternative ageing measures paint quite a different... conclusions the use of a traditional measure of population ageing in combination with a mortality-based measure, such as the population with remaining life expectancy of under 15 years, is helpful for demographic analyses of ageing. key words population ageing; characteristics approach; australia; states and territories. 2. data and methods key messages acknowledgements data availability references australian population studies 2020 | volume 4 | issue 2 | pages 1-13 © xiong and han 2020. published under the creative commons attribution-noncommercial licence 3.0 australia (cc by-nc 3.0 au). journal website: www.australianpopulationstudies.org will my parents come to australia when retired? later-life transnational migration intentions of chinese parents xueying xiong university of new south wales hoon han* university of new south wales * corresponding author. email: h.han@unsw.edu.au. address: school of built environment, the university of new south wales, sydney, nsw 2052, australia paper received 25 april 2020; accepted 6 august 2020; published 16 november 2020 abstract background china is the second largest source country of immigrants to australia every year. the elderly parents of these working-age immigrants in australia usually visit their children on a tourist visa, which allows short term family gatherings. these visits do not require much effort. however, when it comes to long-term transnational migration, the decision becomes hard to make and often involves complex factors, such as personal preferences and lifestyle choices. aims this paper aims to examine the later-life transnational migration intentions of elderly chinese parents and how the parents’ intentions are affected by personal preferences, personality and lifestyle differences between the generations. data and methods this paper uses data from a two-stage study including an online survey and semi-structured interviews conducted in sydney from october 2018 to may 2019 with both caregivers (adult children who are first-generation immigrants living in australia) and their elderly parents. results the study found that external contributors such as language barriers and transport dependence in australia, and existing social ties as well as another adult child living in china, have a negative influence on moving to australia. conclusions elderly parents with an optimistic and outgoing personality are more likely to consider moving to australia compared to those who are more conservative. foremost, conflicts due to different lifestyles between the generations reduces the possibility of parents’ later-life migration to australia. key words transnational migration; population ageing; china http://www.australianpopulationstudies.org/ mailto:h.han@unsw.edu.au 2 xiong and han australian population studies 4 (2) 2020 1. introduction australia absorbs thousands of skilled migrants every year (spinks 2010). since 2011, china has been the second largest immigrant source country, after india (australian department of home affairs 2019a). over a quarter of the resident population in australia were born overseas (28.5%, or 6.9 million), of which 2.2% (0.5 million) were born in china (excluding sars) (australian bureau of statistics 2017). in 2018, 109,713 permanent residence visas were granted to immigrants through the skill stream of the migration program, of which 13.6% were from china (australian department of home affairs 2019b). in contrast to australian culture, where aged parents and adult children tend to live separately and independently, china has a long tradition of ‘filial piety’ (hui and yu 2009), the virtue of respecting parents and old people. it includes material and emotional requirements including “support, memorialising, deference, compliance, respect, and love” (wang et al. 2009 p. 22). as ageing requires elderly people to adapt to declining health with more demand for support (näre et al. 2017), adult children in china generally believe they have an obligation to take care of their aged parents in their later life (bryant and lim 2013; lai 2005; wilmoth 2012). the median age of working-age chinese migrants in australia is 33.6 years old (australian department of home affairs 2019a), and their parents are retirement age. elderly parents of immigrants usually enter australia on tourist visas which allow them to visit their children regularly for a short time (wilding and baldassar 2018). however, it is a difficult decision for elderly parents to migrate to australia for long-term living or permanently. personal attributes, such as personal preferences and lifestyle choice, are the main individual factors that contribute to this later-life decision, as well as external objective factors such as visa restrictions and financial status. this study examines how the later-life transnational migration intentions of elderly chinese parents whose adult children have immigrated to australia is affected by personal preferences, personality, and lifestyle differences between the two generations. previous studies on later-life transnational migration are reviewed in section 2. section 3 presents the interview methodology and sections 4 to 6 analyse data collected through semi-structured interviews. section 7 concludes the paper. 2. later-life transnational migration older people tend to be less mobile and more satisfied with their existing homes and neighbourhood (chapman and lombard, 2006), and age in place (zhou 2012). residential mobility is usually a response to lifecycle events (the triggers). for old people, these triggers are usually divorce, becoming empty nesters, retirement, physical health degradation, income loss, and widowhood (clark 2011; walters 2002). reasons for later-life transnational migration are diverse according to previous studies (han et al. 2019), including seeking better amenities or a warmer climate, dualnationality marriages (warnes, 2009), seeking support from, and living with, children (thang et al. 2012), seeking a lower cost of living and more natural environment (sunil et al. 2007), and sometimes seeking the freedom to study and work (thang et al. 2012). in this research, familism is seen as the primary reason for elderly parents to move internationally: to reunite with their adult children (sadarangani and jun 2015), and to provide childcare to grandchildren. australian population studies 4 (2) 2020 xiong and han 3 studies have explored why later-life transnational migration occurs, but very little research has investigated the elements contributing to this transnational migration decision. for chinese elderly parents, moving to australia where their adult children live seems to be an obvious choice, especially for those from a ‘one-child’ family1 and bounded by filial piety. however, deciding to move abroad is a complex process and the decision is constrained not only by the residential location of their adult children but also by many other objective factors. at the macro level, factors include immigration and settlement policies, visa restrictions, and the international relationship between the destination and home country. micro factors, on the other hand, include both subjective and objective factors. subjective factors include lifecycle and family expectations, particularly “cultural expectations and social roles that inform kin relations which are often fractured by distance and absence” (baldassar 2007 p. 278); and seeking better air and food quality. objective factors are related to elderly parents’ health status for long-distance mobility, and their knowledge of the english language. the family’s financial ability to sponsor the migration is also essential. nonetheless, the first and foremost factor is still personal attributes, including personal preference, personality, and lifestyle choice which often vary from person to person. do elderly parents with particular personal preference or personality traits tend to make different migration choices? adult children play a vital role in the migration decision and process since they are normally the inviters and financial sponsors for parent visa applications. when adult children and their elderly parents have different lifestyles in daily life, conflicts may arise. will this lifestyle conflicts affect the migration intention of elderly parents? in the following sections, these questions will be investigated. 3. data and methods as indicated earlier, different lifestyles and personality between two generations are vital. to better investigate this issue, studies need to explore both parties’ understanding. however, the majority of previous studies focus on elderly parents themselves and neglect the adult children’s role in this decision-making process. in addition, some adult children refuse to invite their parents to australia due to conflicts in lifestyle and attitudes to daily life, but they may hide the reason from their parents due to the ‘filial piety’ moral restraint. by interviewing the adult children, we are hoping to undercover these hidden reasons that cannot be revealed by interviewing the elderly parents. in addition, many of the elderly parents still live in china which makes it more difficult to have them agree to and to participate in interviews. therefore, this paper uses data from a two-stage study including an online survey and semistructured interviews conducted from october 2018 to may 2019 (unsw human research ethics approval number hc17479). the study was conducted to understand the immigration decision of elderly chinese parents from the perspective of the adult children in australia and how intergenerational conflict between the two generations impacts the later-life migration intention of elderly chinese parents. 1 the ‘one-child’ policy was introduced in the 1980s and lasted until 2015 in china; under this fertility control policy, many families in china could only have one child. 4 xiong and han australian population studies 4 (2) 2020 stage one was a random sampling online questionnaire survey targeting caregivers (adult children) who are first-generation chinese immigrants living in australia. a multi-stage sampling method was designed using australian bureau of statistics 2016 census data. as most chinese migrants in australia live in sydney (39%) and melbourne (35%) (australian bureau of statistics 2014), sydney is used as a representative city for this target population group. a bilingual questionnaire in english and mandarin was developed and delivered to 38 statistical areas level-1 in sydney with a high proportion of china-born residents. a total of 142 valid responses were received. of all survey participants, 31 agreed to participate in semi-structured interviews. after further contact, ten were interviewed in stage two of the study. the others either did not reply or changed their mind. the semi-structured interviews were conducted from june to august 2019 in mandarin by the first author. table 1 summarises the characteristics of the ten interviewees. table 1: characteristics of the interviewees and their parents code adult children interviewees elderly parents of interviewees gendera age marital statusb visa statusc siblingd age cohort marital statusb parents’ immigration intentione visited aus 1 f 36 m pr yb (1) 56-65 m m: yes, pr visa applied f: not sure yes 2 m 29 m pr ys (1) 56-65 divorced (mother); remarried (father); not sure yes 3 f 30 m citizen ys (1) 56-65 m not sure yes 4 f 28 m pr oc 56-65 m yes, pr visa applied yes 5* m 29 m pr oc 56-65 m m: yes, pr visa received f: not sure father yes 6* m 32 nm citizen oc 56-65 m not sure yes 7 f 35 m pr oc 56-65 m not sure yes 8* f 33 m pr oc 66-75 m yes, intended to apply yes 9 f 43 df citizen ob (1) 65-75 m not sure yes 10 f 33 m pr oc 56-65 m not sure no notes: a. gender: f = female, m = male; b. marital status: m = married, df = de-facto, nm = never married, d = divorced, r = remarried; c. visa status: pr = permanent resident visa of australia; citizen = citizen of australia; d. sibling: yb = younger brother, ys = younger sister, ob = older brother, oc = only child. number in brackets is the number of siblings; e. m = mother, f = father, pr = permanent resident * interviewed with parents. of the ten interviews, seven were conducted with an adult child only and three were with both the adult child and their elderly parents, based on the participants’ willingness. all ten interviewee groups expressed to some degree that personal preferences, or lifestyle conflicts between australia and china and between the two generations, or the personality of the elderly parents impacted the migration intention to australia. some mentioned this influence as an additional factor to other essential ones such as financial affordability and care responsibilities in china for the parents’ own parents (i.e. the grandparents of the adult children in australia). three factors that emerged from the australian population studies 4 (2) 2020 xiong and han 5 interviews of personal preferences, personality, and lifestyle conflicts are discussed in the following three sections. 4. impact of elderly parents’ personal preferences personal preference is one of the most important factors in elderly parents’ intention to migrate to australia or stay in china. this subjective factor is a combination of personal feeling and some external contributors. it is hard to isolate it from other external factors such as environment, quality of life, language ability, and number of adult children. for instance, interviewee 4 indicated her parents-in-law would like to stay in china and her father-inlaw did not want to move because of his personal preference. according to her, her father-in-law used to take business trips worldwide and had been to many different places. he had experienced different lifestyles at home and abroad, and enjoyed the idea of living in china more than moving abroad. her parents-in-law used to live in cities; now, they have returned to the countryside, where their relatives live. her parents-in-law have another son who lives in china which also contributes to their decision to stay in china, although the other son does not live with them. in this case, having relatives and another son and enjoying the lifestyle in china formed the parents’ personal preference. interviewee 9 said she had asked her parents about migration as an alternative to retirement in china six years ago. at that time, it was much easier to apply for a parent visa for australia than now2; but her parents refused, although they agreed that australia is a nice place. the main reason was that her parents felt they could not adapt to life in australia. they have all their friends in china and are used to chinese culture. i know many elderly people here in australia. some of them immigrated to australia and can adapt to life here quite well, while others cannot assimilate. so, i do not want to push them. if they do not want to come, i will just respect their decision. (interviewee 9) her older brother lives in china which ensures her parents have support. her parents were seeking more spiritual pursuits in their retirement, not material ones. they feel confident and comfortable with their existing social ties. it would be challenging for them to rebuild new social ties if they moved to australia. they would rely more on their children for mobility in australia; in china, they could extend their period of independent living and mobility. both interviewee 3 and interviewee 5 claimed that their mothers are still busy with work in china and not being able to speak english worries them about moving to australia. although they share similar external contributors, the migration decision of their parents differs. interviewee 3’s mother felt uncomfortable in australia and would prefer to stay in china while interviewee 5’s mother 2 places available in the parent migration stream have gradually decreased while demand has increased. in 2013-2014, 2,250 non-contributory parent visas and 6,675 contributory parent visas were granted to applicants which decreased to 1,218 and 5,587 in 2018-2019. in the same period, applicants in the pipeline for the non-contributory parent visa increased from 36,000 to 49,994 and for the contributory parent visa from 12,800 to 52,805 (australian department of home affairs, 2011-2019). 6 xiong and han australian population studies 4 (2) 2020 enjoyed living in australia in general and will move to australia in the future. despite different personal feelings about living in australia, another important external contributor to their decisions is that interviewee 3 has a younger sister in china while interviewee 5 is the only child in his family. 5. impact of elderly parents’ personality during the interviews, it was observed that the personality of the elderly parents and their previous life experience greatly affected their intention to migrate. different from personal preference, which is a subjective willingness, personality is a personal characteristic that people are born with or have formed due to life experience. interviewee 1’s parents-in-law live in a medium-sized provincial capital city in china but grew up in a small town. the city has good air quality and a slower pace of life. they had never left china before visiting australia. she believed that this contributed to her parents-in-law’s decision to stay in china as they feel comfortable in their hometown. the air quality of australia is no longer an attractive factor to them; but she thought the most important reason was her parents-in-law’s personalities. they were not enthusiastic about new or unfamiliar things. my parents-in-law have compromising personalities. my parents, in contrast, always try to improve and achieve a better result. i believe their immigration decisions are associated with their experience and personalities. (interviewee 1) she provided an example: her parents learned to drive when they were teenagers. her parents-inlaw did not drive at all. she used to suggest her father-in-law learn to drive in australia and promised to buy him a car so that he could drive himself around, but her father-in-law was not interested. although interviewee 1 and her husband had been living in australia for three years, had new jobs, a new house and cars, and they and their children were used to their life in australia and felt comfortable living here, her parents-in-law still thought they were not as comfortable as living in china, as they believe their son and daughter-in-law have to work harder. eating out is more expensive (considering the exchange rate between currencies) and public transport is less convenient in australia. interviewee 1 added that she had told her parents-in-law they could stay in a chinese community where they could live without speaking english. my parents-in-law are not good at adapting to a new environment. they prefer to live close to their old friends. making new friends to them is not so attractive. they are afraid of having no common topic with new friends. i think it is associated with their personality and living habits. (interviewee 1) her own parents, on the contrary, are keen to learn new concepts and accept new things. her own parents, although also unfamiliar with sydney, were willing to ‘get the feel of the city’. for instance, her father installed a translator in his smartphone and used it to talk to neighbours or when grocery shopping, and was confident being out on his own. her mother was also an outgoing person. before interviewee 1 moved to her current suburb, she lived in a suburb popular with chinese people. her mother joined a social group during her first week in australia, and then joined in dancing, and even competitions, and quickly became involved in the neighbourhood. her parents-in-law would not even australian population studies 4 (2) 2020 xiong and han 7 bother to try these things, as in their mind, they had just come for temporary support and would return to china anyway. this was similar to interviewee 2’s parents and parents-in-law. interviewee 2’s parents travelled extensively worldwide. both of his parents could drive in australia. his stepmother was from hong kong and was very familiar with driving on the left-hand side of the road. on the contrary, his parents-in-law had difficulty driving, and limited english skills. compared to his parents-in-law, his parents are better at learning new things and could adapt to new environments faster. therefore, his parents tended to follow their own preferences, however his parents-in-law considered their children’s opinion more when selecting retirement locations. interviewee 6’s father, who had travelled widely, had plenty of experience of living and working in many cultures and languages. he also had friends in australia, including those he had met through work, or who also have children studying and living in australia. his personal experiences made him confident about living in australia, even without the company of his son and this ability also enabled his wife to feel comfortable living in a country with different cultures and languages. they are willing to move to australia. his father said that, despite the chinese ‘reunion with family’ tradition, whether you like australia and the lifestyle here [in australia] is also important. it will not work well if you come here only for your child. the personality of the parents and your financial status also count when making the decision. (interviewee 6’s father) interviewee 8’s father also enjoys travelling and has been to many other countries. he indicated that he is optimistic and willing to learn new things. he and his wife are determined to move to australia to be reunited with their daughter. interviewee 7 thought her parents might experience some emotional change if they moved to australia, due to the new external environment. this does not refer to the physical environment, such as living style or community facilities, but refers to changes in the neighbours her parents are familiar with and the communication style they have with these neighbours, accompanied by the feeling of being needed. her parents live in a ‘da yuan’3 in china. her parents were teachers and used to educate other people. everyone in the community showed great respect to her parents, were willing to talk with her parents and consulted her parents for advice if they had a problem. she thought her parents would lose the feeling of being respected and needed if they moved to australia. this will be the most important emotional change for them. i think my parents are not the ones with a strong will; not like the ones who never care what other people do or say. they have a relatively great need for recognition from other people. if they have no one to guide, they will feel uncomfortable. of course, they will deny such need. but they indeed are weak in inner strength. they do like worship from other people, although i think this admiration is kind of cheap. (interviewee 7). 3 da yuan refers to small communities fenced and guarded by security guards; people in a community usually work for the same company. 8 xiong and han australian population studies 4 (2) 2020 this personality of her parents, from her point of view, is the main reason that may affect her parents emotionally if they moved to australia. this can also lead to conflict between the two generations in daily life. the daughter did not want to be controlled by, or instructive at, while her parents were used to doing this. 6. impact of different lifestyles for some interviewee families, the different lifestyle is another key reason for their parents not to move to australia. this different lifestyle is explained at two levels: different lifestyles between china and australia; and different lifestyles, or attitudes towards life, between two generations. interviewee 10 indicated that her parents were not used to the lifestyle here in australia. they are so used to life at home. for example, they exercise every morning in small parks near home and have ‘guang chang wu’4 in the evening. they sometimes play mahjong with their friends in the afternoon. they already have social ties and a routine. they worry that they need to re-build a social network here. it is easy for us young generation, but it is hard for them to adapt to new lifestyles. (interviewee 10) interviewee 1 mentioned that her parents will not stay in australia long term because they are used to the lifestyle in china including food, methods of travel, and social ties. all these stop them from leaving china. her parents-in-law also refused to come to australia because they are not used to the lifestyle in australia and it is difficult for them to adapt to a new environment. interviewee 7’s parents could communicate in english, although not at a proficient level. her parents have social health insurance in china, and the welfare in australia, such as medicare and the age pension, is not attractive to them. better air quality and natural environment are the most attractive aspects of life in australia, compared to china. my parents are fond of seeing animals such as birds. i have purchased an apartment in rhodes [a popular suburb for chinese people in sydney] where there is a lake and a park around it. i think it is a good place for them, although the population density is high, the natural environment is good. after all, a better natural environment and air quality are what they come to australia for. my parents can exercise and see animals there. besides, the community in rhodes is relatively quiet which is different from other popular chinese community. (interviewee 7) despite australia’s better natural environment, her parents still prefer to live in china. interviewee 7 stated her parents have no friends in australia; all her parents’ relatives and social ties are in china. 4 “guang chang wu is square dancing in chinese style. it refers to the daily dances which take place in china’s public squares; an activity which has over 100 million people regularly up on their feet. each day, to music piped through mounted speakers, dancers gather in the country’s parks and central squares, to perform the choreographed routines of guangchang wu.” (https://www.southbankcentre.co.uk/blog/how-do-you-guangchang-wu-square-dancing-chinese-style. accessed 7 march 2020). https://www.southbankcentre.co.uk/blog/how-do-you-guangchang-wu-square-dancing-chinese-style australian population studies 4 (2) 2020 xiong and han 9 her parents sometimes strolled around, but normally ended in quarrels, which interviewee 7 was tired of dealing with. her parents feel more comfortable living in china with fewer ‘restrictions’, particularly the restriction on smoking in australia where residents cannot smoke in specific, public places. interviewee 7’s father is used to smoking wherever and whenever he wants in china, and he also thought cigarettes in australia are too expensive, at about five times the price in china, and the flavour also varies. her father felt like he was being controlled in daily life in australia. there are also other differences between interviewee 7’s lifestyle and her parents’ lifestyle. i have a very high standard in my own home in hygiene. i want everything in my home to be clean and in order. my parents, on the contrary, often make home chaotic. they like to store a lot of useless things at home (which makes the home a mess). they feel annoyed to clean the floor when it is dirty. [living with me] is like living in someone else’s home and they lose power at my home. that is why they do not want to come to australia. even when they come here, they will not stay long. our living habits are so different. (interviewee 7) interviewee 7 also said she and her parents had different attitudes towards life. for instance, her mother usually forwarded a lot of ‘spam’ news to her. she sometimes had phone calls with her mother, but they were usually her mother’s complaints about her father, or her grandparents. interviewee 7 felt the calls were quite negative which annoyed her. when shopping, interviewee 7 believes that affordability and benefits to her own family are most important. however, her parents care more about how other people see them. this is called ‘mian zi’5 in chinese, and represents a sense of dignity or self-respect. people who care too much about ‘mian zi’ normally mind how other people see them and will sometimes do things beyond their capacity just to earn respect or admiration from other people. interviewee 7 said she cannot understand when her parents even felt they had disgraced themselves, but insisted on buying many gifts for their relatives. interviewee 7 and her mother also had different opinions on raising the next generation. her mother wanted her to bear three children, while she thought it would be a bad idea. my mother will definitely want to come to australia to help me take care of my children. but i have determined not to let her do so. i am worried that the support they can provide is limited while i may use more energy to quarrel with them. i would rather care for the children myself, even though i will be quite busy and tired, rather than letting them come to help. (interviewee 7) she and her parents have now learnt to respect each other’s lifestyle when they visit each other. i used to want to change their thoughts and habits during the first few times when i returned to china. but now i have figured it out. it is their home, not mine. if i do not want them to be backseat drivers and i want them to respect my lifestyle, i need to learn to respect their 5 mian zi is the measure of one’s position, prestige and leverage in society; it “is built up through initial high position, wealth, power, ability, through cleverly establishing social ties to a number of prominent people, as well as through avoidance of acts that would cause unfavorable comment” (hu 1944 p. 61; aris 2017) 10 xiong and han australian population studies 4 (2) 2020 habits. we could not change each other. what we can do is to keep a certain distance. (interviewee 7) differences in lifestyles was an obvious factor influencing interviewee 7’s parents to consider whether to move to australia or not. interviewee 7 thought the primary cause for her parents to choose to stay in china was lifestyle and a spiritual demand. the natural environment and air quality in australia would not be enough to meet their lifestyle and spiritual needs in later life. they can still have a social life with their friends and leisure activities, such as attending university for elderly people, in china. if they come to australia for retirement life, it would be a passive compromise because they could no longer travel by themselves and are only capable of taking care of themselves, having no energy to dictate to other people. it would be when they have to go to nursing homes and be cared for by professional nursing staff (that they would come to australia). otherwise, they will stay in china when they still have the self-maintenance ability. their personal demand is the most important reason for their choice. (interviewee 7) this lifestyle difference does not mean the relationship between the two generations is negative. in fact, interviewee 7 said she had taken her parents travelling around australia and they very much enjoyed it. her parents had financially sponsored her education and living expenses when she was studying in australia. they are quite satisfied with their retirement life now in china. they have enough superannuation and savings. they have their own property. i always tell them to stop saving money (as it is a common habit in china) but spend more to have good food, to keep healthy, to travel and have fun. i tell them not to worry (about me). i do not need them to spend money on me. so, i believe they are having a very satisfied life now. (interviewee 7) 7. conclusions this study examined how personal preferences, the personality of elderly parents, and differences in lifestyles between two generations impact the migration intentions of elderly chinese parents regarding a potential move australia after retirement. the results show that these factors are important in the later-life transnational migration intentions of elderly chinese parents (see table 2). personal preference is a grounding factor in this migration decision-making process. some elderly parents are more than happy to move and reunite with their adult children. some who are initially unwilling to move to australia may be persuaded by their adult children and agree to move to australia, sometimes after visiting and experiencing life in australia. other parents who prefer living in china may stick to their personal location preference and not migrate. the adult children will normally respect their parents’ decision. they will visit china frequently and consider temporarily returning to china when their parents require more care. the personality of elderly parents, different from personal preference, also influences their intentions. elderly parents who are open to new things and concepts are more willing to try a new lifestyle in australia. the parents interviewed were optimistic and outgoing. they had travelled to australian population studies 4 (2) 2020 xiong and han 11 table 2: summary of impact of three factors on migration intention: personal preference, personality and lifestyle differences personal attributes adult children in australia elderly parents personal preferences negative for migration: • relatives (parents, other child) in china (4, 7, 9) • social ties in china (4, 7, 9) • relying on children for transport (1, 2) • english barrier (2, 3, 5) positive for migration: • only child (5, 6, 8) • have friends in australia (6) • english ability (2, 6, 7) personality negative for migration: • conservative (1) • less travel experiences (1) • emotional change as lose power in family (7) positive for migration: • open to new things, learning (1, 2, 8) • more travel experiences (1, 2, 6, 8) • take children’s opinion in consideration (2) lifestyle choices/conflicts negative for migration: • house hygiene (7) • different parenting concept for childcare (7) • different life concept (7) positive for migration: • demand for childcare (1) negative for migration: • food in china, travel in china, social activities in china (e.g. dancing, mah-jong) (1, 9, 10) • restrictions on life in australia (e.g. smoking) (7) positive for migration • air quality and natural environment in australia (5, 7) notes: number in brackets refers to interview numbers in table 1. ‘negative’ means the factor has negative impact in elderly parents’ migration intention to australia; ‘positive’ means the factor will positively contribute to their migration intention. many other countries previously, and were happy to learn english and make new friends. on the contrary, elderly parents who are conservative or have stayed in one place for their whole life and not travelled to other countries are less likely to migrate. they are less open to new things and tend to stay in their ‘comfort zone’, a place they are familiar with and where they have their social network close by. adult children play an important role in the later-life migration of their elderly chinese parents as australia’s parent visa requires adult children to be the sponsor. when adult children, especially those who assimilate culture and lifestyle in the host country, have a different lifestyle from their elderly parents, conflicts may appear in daily life and this reduces the likelihood of transnational migration. in addition, chinese elderly parents will expect increasing dependence on their adult children in daily activities if they move to australia. this increasing dependence results from both worsening health and also external obstacles such as language barriers and transport dependence as revealed in this study and previous research (ip et al. 2007; sadarangani & jun 2015; wilmoth 2001). 12 xiong and han australian population studies 4 (2) 2020 complex factors affect the decision-making process for transnational migration in later life. internal attributes of the migrant, such as personal preference and personality, and external factors including but not limited to visa restrictions, relationship with adult children, and the migration policies of the host and home countries jointly influence the later-life transnational migration decision. this qualitative study is based on ten interviews. further study is required to examine the trade-off between the costs and benefits of transnational migration for the two generations of chinese-born immigrants in australia and their elderly parents. 8. key messages • the difference in adopted lifestyles between two generations (adult children and elderly parents) impacts on the transitional migration intention of elderly chinese parents to australia. • adult children in australia play an important role in this later-life migration activity of the elderly chinese parents. • those parents unwilling to move to australia can be persuaded to move by their adult children in australia. • intergenerational conflicts in lifestyle decrease the propensity of transnational migration of chinese parents. acknowledgements this research was supported by the university international postgraduate award from the university of new south wales (unsw) and by the phd publication award from the school of built environment, unsw. we are grateful to the constructive comments from two anonymous reviewers. references aris t (2017) the concept of face in chinese culture and the difference between mianzi and lian. https://china-journal.org/2017/02/25/the-concept-of-face-in-chinese-culture-and-the-differencebetween-mianzi-and-lian/. accessed 7 march 2020. australian department of home affairs (2011-2019) report on migration 2009–2019. canberra: department of home affairs. https://www.homeaffairs.gov.au/research-andstatistics/statistics/visa-statistics/live/migration-program. australian department of home affairs (2019a) 2018-19 migration program report. canberra: department of home affairs. https://www.homeaffairs.gov.au/research-and-stats/files/reportmigration-program-2018-19.pdf. australian department of home affairs (2019b) country profile–china. https://www.homeaffairs.gov.au/research-and-statistics/statistics/countryprofiles/profiles/china. accessed 1 november 2019. australian bureau of statistics (2014) australian social trends, 2014: where do migrants live? http://www.abs.gov.au/ausstats/abs@.nsf/lookup/4102.0main+features102014. accessed 10 december 2017. australian bureau of statistics (2017) migration, australia, 2015-16. catalogue no. 3412.0. canberra: abs. http://www.abs.gov.au/ausstats/abs@.nsf/mf/3412.0. https://china-journal.org/2017/02/25/the-concept-of-face-in-chinese-culture-and-the-difference-between-mianzi-and-lian/ https://china-journal.org/2017/02/25/the-concept-of-face-in-chinese-culture-and-the-difference-between-mianzi-and-lian/ https://www.homeaffairs.gov.au/research-and-statistics/statistics/visa-statistics/live/migration-program https://www.homeaffairs.gov.au/research-and-statistics/statistics/visa-statistics/live/migration-program https://www.homeaffairs.gov.au/research-and-stats/files/report-migration-program-2018-19.pdf https://www.homeaffairs.gov.au/research-and-stats/files/report-migration-program-2018-19.pdf https://www.homeaffairs.gov.au/research-and-statistics/statistics/country-profiles/profiles/china https://www.homeaffairs.gov.au/research-and-statistics/statistics/country-profiles/profiles/china http://www.abs.gov.au/ausstats/abs@.nsf/lookup/4102.0main+features102014 http://www.abs.gov.au/ausstats/abs@.nsf/mf/3412.0 australian population studies 4 (2) 2020 xiong and han 13 baldassar l (2007) transnational families and aged care: the mobility of care and the migrancy of ageing. journal of ethnic and migration studies 33(2): 275-297. bryant l and lim s (2013) australian-chinese families caring for elderly relatives. ageing and society, 33(8): 1401-1421. chapman d w and lombard j r (2006) determinants of neighborhood satisfaction in fee-based gated and nongated communities. urban affairs review 41(6): 769-799. clark w a (2011) mobility and mobility contexts: modeling and interpreting residential change in australia. hilda survey research conference, melbourne. han j h, jung y m and xiong x (2019) the ageing of australian ethnic minorities. singapore: palgrave pivot. hu h c (1944) the chinese concepts of “face”. american anthropologist 46(1): 45-64. hui e c m and yu k h (2009) residential mobility and aging population in hong kong. habitat international 33(1): 10-14. ip d, lui c w and chui w h (2007) veiled entrapment: a study of social isolation of older chinese migrants in brisbane, queensland. ageing & society 27(5): 719-738. lai d w l (2005) cultural factors and preferred living arrangement of aging chinese canadians. journal of housing for the elderly 19(2): 71-86. näre l, walsh k and baldassar l (2017) ageing in transnational contexts: transforming everyday practices and identities in later life. identities: global studies in culture and power 24(5): 515-523. sadarangani t r and jun j (2015) newly arrived elderly immigrants: a concept analysis of “aging out of place”. journal of transcultural nursing 26(2): 110-117. spinks h (2010) australia’s migration program. background note. parliamentary library, parliament of australia, canberra. https://www.aph.gov.au/about_parliament/parliamentary_departments/parliamentary_library/ pubs/bn/1011/austmigration. sunil t s, rojas v and bradley d e (2007) united states’ international retirement migration: the reasons for retiring to the environs of lake chapala, mexico. ageing and society 27(4): 489-510. thang l l, sone s and toyota m (2012) freedom found? the later-life transnational migration of japanese women to western australia and thailand. asian and pacific migration journal 21(2): 239-262. walters w h (2002) later-life migration in the united states: a review of recent research. journal of planning literature 1: 37-66. wang d, laidlaw k, power m j and shen j (2009) older people’s belief of filial piety in china: expectation and non-expectation. clinical gerontologist 33(1): 21-38. warnes t (2009) international retirement migration. in: uhlenberg p. (ed.) international handbook of population aging. dordrecht: springer; pp. 341-363. wilding r and baldassar l (2018) ageing, migration and new media: the significance of transnational care. journal of sociology 54(2): 226-235. wilmoth j m (2001) living arrangements among immigrants in the united states. the gerontologist 41(2): 228-238. wilmoth j m (2012) a demographic profile of older immigrants in the united states. public policy & aging report 22(2): 8-11. zhou j (2012) urban elderly population: residential and mobility preference [in chinese]. construction economy 11: 71-75. https://www.aph.gov.au/about_parliament/parliamentary_departments/parliamentary_library/pubs/bn/1011/austmigration https://www.aph.gov.au/about_parliament/parliamentary_departments/parliamentary_library/pubs/bn/1011/austmigration australian population studies 2021 | volume 5 | issue 1 | pages 101-104 © grossman et al. 2021. published under the creative commons attribution-noncommercial licence 3.0 australia (cc bync 3.0 au). journal website: www.australianpopulationstudies.org visualising the covid-19 related disruptions to long-distance population mobility in australia irina grossman* the university of melbourne tom wilson the university of melbourne jonathan garber the university of melbourne jeromey temple the university of melbourne *corresponding author. email: irina.grossman@unimelb.edu.au. address: melbourne school of population and global health, the university of melbourne, melbourne, vic 3010, australia. paper received 6 march 2021; accepted 26 march 2021; published 31 may 2021 introduction the public health response to mitigate the spread of covid-19 in australia involved an entry ban limiting entrance to australia to its citizens and residents, a travel ban preventing australians from leaving the country, and a series of state border closures limiting travel within the country (grattan institute 2020). the aims of this paper are to document the extent of the downturn in domestic and international air mobility in australia that was associated with covid-19. we use aviation statistics as a proxy for long-distance human mobility. several papers have already been written on covid-19 impacted mobility using flight data (e.g., iacus et al. 2020; garcia-gasulla et al. 2020; lau et al. 2020; suzumura et al. 2020). our contribution focuses on the australian context, providing a summary visual representation of the huge changes in both domestic and international mobility trends which have occurred recently. data and methods the australian government’s bureau of infrastructure and transport research economics (bitre) provides publicly available monthly publications documenting domestic and international aviation activity (https://www.bitre.gov.au/statistics/aviation). the international data was sourced from the ‘table 5 scheduled international traffic by city pairs’ dataset found in the ‘table_5’ sheets in the january 2019 – december 2020 excel workbooks, published as part of the international airline activity—monthly publications (bitre 2021a). the domestic data was sourced from the ‘australian domestic airlines, traffic-on-board-by-stage passengers, top competitive routes’ dataset found in the ‘summary’ sheets in the january 2019 –december 2020 excel workbooks published as part of the australian domestic aviation activity monthly publications (bitre 2021b). matlab 2020b was used to aggregate monthly data into time series datasets for total domestic and international passenger numbers. additionally, time series data was assembled for travel between australian capital cities. rstudio 1.3.1056 was used together with the circlize package (gu et al. d e m o g ra p h ic http://www.australianpopulationstudies.org/ mailto:irina.grossman@unimelb.edu.au https://www.bitre.gov.au/statistics/aviation 102 grossman et al. australian population studies 5 (1) 2021 2014) to illustrate the geographical connections and numbers of passengers moving between australian capital cities. we created these for the months of december 2019, april 2020, august 2020, and december 2020. microsoft excel 2016 was used to create a line chart representing changes in domestic and international travel for the 2019 – 2020 period. figure 1: australian international and domestic air passengers, by month, 2019 and 2020 source: international data was sourced from the passenger numbers provided through the ‘table_5’ sheets in the excel workbooks (bitre 2021a). domestic data was sourced from the ‘revenue passengers’ column in the ‘summary’ sheets of the excel workbooks (bitre 2021b). note: passenger numbers include both inbound and outbound flights for international flights. key features the line chart in figure 1 shows that international and domestic travel both decreased dramatically in the first few months of 2020. in april 2020, international and domestic passenger numbers were at just 2.0% and 2.8% of their april 2019 numbers, respectively. this corresponds with the closure of the australian border, the international travel ban, and state border closures (grattan institute 2020). international travel remained flat through till the end of 2020. recovery in australian aviation activity has been led by domestic travel. domestic travel experienced a local maxima in july before decreasing again slightly, which was expected given the state border closures associated with the second wave of covid-19 in victoria (grattan institute 2020). since september there has been a gradual increase in travel, and this rate increased significantly in december of 2020, corresponding with the opening of state borders (grattan institute 2020) following the successful suppression of the second wave in victoria. by december 2020 domestic passenger numbers had recovered to 41% of the december 2019 numbers. 0 1 2 3 4 5 6 jan feb mar apr may jun jul aug sep oct nov dec millions of passengers per month 2019 domestic 2019 international 2020 domestic 2020 international australian population studies 5 (1) 2021 grossman et al. 103 figure 2: circular charts comparing inter-capital city travel in selected months of 2019 and 2020. source: inter-capital city passenger numbers were sourced from the ‘revenue passengers’ column in the ‘summary’ sheets of the excel workbooks (bitre 2021b). notes: the width of the links indicates the number of passengers travelling between two cities. these links are not directional. each unit on the axes of the circular charts represents 100,000 passengers. total inter-capital city passenger numbers, and inter-capital passenger numbers as a percentage of total domestic passenger numbers, are also presented. 104 grossman et al. australian population studies 5 (1) 2021 the circular charts in figure 2 provide more detail on changes in domestic travel in 2020 by depicting passenger flows between capital cities. the december 2019 circular chart depicts standard precovid-19 passenger flows, when inter-capital city travel comprised 57% of total domestic passenger numbers. the april and august 2020 circular charts show how australia became disconnected, with few passengers moving between the capital cities. the december 2020 circular chart displays the recovery in domestic air travel with the reforming of key inter-capital city links, particularly those between melbourne, sydney, and brisbane. ethics approval ethics approval for this project was granted by the melbourne school of population and global health (mspgh) human ethics advisory group (heag), id 2056902.1. acknowledgements the authors gratefully acknowledge funding support from the melbourne school of population and global health for covid-19 research. references bitre (2021a) international airline activity—monthly publications. https://www.bitre.gov.au/publications/ongoing/international_airline_activitymonthly_publications. accessed on 23 february 2021. bitre (2021b) australian domestic aviation activity monthly publications. https://www.bitre.gov.au/publications/ongoing/domestic_airline_activity-monthly_publications. accessed on 23 february 2021. garcia-gasulla d, napagao s a, li i, maruyama h, et al. (2020) global data science project for covid-19 summary report. https://arxiv.org/abs/2006.05573. accessed on 14 september 2020. grattan institute (2020) grattan: coronavirus announcements tracker. https://docs.google.com/spreadsheets/d/1zqncmsuevd26xrw1hmszi5q_aqwszje78y5hclwxq 64/edit#gid=0. accessed on 26 february 2021. gu z, gu l, eils r, schlesner m, & brors b (2014) circlize implements and enhances circular visualization in r. bioinformatics 30(19): 2811-2812. https://doi.org/10.1093/bioinformatics/btu393. iacus s m, natale f, santamaria c, spyratos s, & vespe m (2020) estimating and projecting air passenger traffic during the covid-19 coronavirus outbreak and its socio-economic impact. safety science 129: 104791. https://doi.org/10.1016/j.ssci.2020.104791. lau h, khosrawipour v, pior kocbach p, mikolajczyk a, ichii h, zacharski m, bania j, & khosrawipour t (2020) the association between international and domestic air traffic and the coronavirus (covid-19) outbreak. journal of microbiology, immunology and infection 53(3): 467-472. https://doi.org/10.1016/j.jmii.2020.03.026.2020. suzumura t, kanezashi h, dholakia m, ishii e, napagao s a, prez-arnal p, & garcia-gasulla d (2020) the impact of covid-19 on flight networks. https://arxiv.org/abs/2006.02950. accessed on 14 september 2020. https://www.bitre.gov.au/publications/ongoing/international_airline_activity-monthly_publications https://www.bitre.gov.au/publications/ongoing/international_airline_activity-monthly_publications https://www.bitre.gov.au/publications/ongoing/domestic_airline_activity-monthly_publications https://arxiv.org/abs/2006.05573 https://docs.google.com/spreadsheets/d/1zqncmsuevd26xrw1hmszi5q_aqwszje78y5hclwxq64/edit#gid=0 https://docs.google.com/spreadsheets/d/1zqncmsuevd26xrw1hmszi5q_aqwszje78y5hclwxq64/edit#gid=0 https://doi.org/10.1093/bioinformatics/btu393 https://doi.org/10.1016/j.ssci.2020.104791 https://doi.org/10.1016/j.jmii.2020.03.026.2020 https://arxiv.org/abs/2006.02950 australian population studies 2021 | volume 5 | issue 1 | pages 9-17 © davies & prout quicke 2021. published under the creative commons attribution-noncommercial licence 3.0 australia (cc by-nc 3.0 au). journal website: www.australianpopulationstudies.org commentary population mobility, ‘usual residence’ and the census: the case of australia’s grey nomads amanda davies the university of western australia sarah prout quicke* the university of western australia * corresponding author. email: sarah.proutquicke@uwa.edu.au. address: department of geography and planning, school of social sciences, the university of western australia, 35 hackett drive, crawley, wa 6009, australia. paper received 24 march 2021; accepted 27 april 2021; published 31 may 2021 1. introduction australia’s public heath response to covid-19 included the temporary closure of state and regional borders in an effort to significantly curb population mobility. for those who were travelling at the time, the assumption seemed to be that they could simply return home, and resume travel when it was safe to do so. while it was recognised that mobility restrictions would cause difficulties for international tourists as they did not have a permanent residence in australia, the assumption for domestic resident travellers was that they were ‘away from home’ and could therefore simply return to their place of ‘usual residence’. such normative framings of travellers as ‘away from home’ or away from a ‘usual residence’ present challenges for australians who engage in practices of bi-locale and multi-locale forms of residence, for those that do not identify as having a single place of usual residence, and for the permanently nomadic. this includes a diverse range of sub-population groups: those experiencing secondary homelessness1; seasonal labourers; backpackers; highly mobile indigenous peoples; longdistance commuters; travelling show families and workers; permanent house-sitters, and; grey nomads (the focus of this paper). in the case of grey nomads, as mobility restrictions resulting from covid-19 public health measures extended to the temporary closure of state and national parks and the camping grounds within, some travellers found they had little option but to move to costly private caravan parks, informal camping areas or to the properties of family or friends (fowler 2020). the exact number of people likely to have been impacted is unknown. this is because, like most population sub-groups that are highly mobile and despite the phenomenon of grey nomads being a feature of the australian population for more than four decades (holloway et al. 2011), information about the size, characteristics and movements of this population is limited (moore 2018; raven 2016). 1 those moving frequently between one type of accommodation and another. http://www.australianpopulationstudies.org/ mailto:sarah.proutquicke@uwa.edu.au 10 davies & prout quicke australian population studies 5 (1) 2021 the covid-19 pandemic has brought into focus the need for a greater understanding of the nation’s mobile populations who engage in regular, patterned and/or permanent forms of medium(i.e., periodic, seasonal) and long-term population circulation. many of these population sub-groups are considered ‘at risk’ for different reasons (e.g., ill-health, social exclusion/disconnection, socio-economic disadvantage), and are often problematised and/or statistically invisibilised within sedentarist governance framings (see e.g., prout quicke and green 2018). in an increasingly mobile world, governments need better tools for identifying the size, composition, patterns of movement and needs of such populations. australian scholars have been leading international efforts to develop standardised estimates of temporary populations at varying geographical and temporal scales (charles-edwards et al. 2020). this important work, expertly synthesised recently by panczak et al. (2020), foregrounds the necessity of such estimates because of the significant service-related impacts that temporary populations can have on different kinds of small area localities. as charles-edwards et al. (2020) identified from their survey of (mostly government) stakeholders, information about temporary populations is critical for: infrastructure and social service planning; land use planning; assessing and managing potential social, economic, and environmental impacts, and; securing fiscal equalisation grants. much of the work encompassed within this research field, therefore, understandably takes a location-based approach to the measurement of temporary populations and considers how de jure populations and de facto populations might be distinguished and tracked over time in particular locations. in this short commentary piece, we take mobile sub-populations themselves as our frame of reference and probe the concept of ‘usual residence’ as operationalised by the australian bureau of statistics (abs) in its five-yearly census of population and housing (hereafter ‘the census’), for its utility in relation to these groups, with particular reference to grey nomads as one case study group. we conclude by considering the possibilities for alternative survey logics and category framings in the 2026 census. 2. population mobility and conceptualisations of ‘usual residence’ in the census as the australian statistician david kalisch noted in his preface remarks to the census of population and housing: topics directions, 2021 paper, the census allows us to understand the cultural, economic and social diversity of our communities’ and provides vital information to “locate health, education and transport services where they are needed” (abs 2018 p. np). perhaps less obviously, but arguably just as important, the census presents a normative framework for categorising and statistically capturing social, cultural and economic trends. in other words, what is counted, and how categories of enumeration are conceived, reflect particular understandings of what counts (pun intended) in australian society and how society is structured. we argue in this paper that census topics regarding population and location are underpinned by a particular normative assumption about the concept of ‘usual residence’ that obscures the potential for capturing critical information about certain population dynamics. in essence, the census provides for only two possible population positions in relation to usual residence. the first is absence: those that have no usual residence and are characterised as ‘homeless’. the second (all ‘non-homeless’ australian population studies 5 (1) 2021 davies & prout quicke 11 people) are assumed to have one identifiable place of ‘usual residence’, defined by the census as the place they usually live over a minimum six-month period during the census year. such assumptions are problematic for many of the sub-populations identified in the introduction of this commentary. the concept of usual residence, is of course, critical to the most fundamental applications of the census, including the ability to enumerate locality populations. furthermore, the australian census offers one of the more sophisticated and robust approaches to capturing nuanced aspects of population and locality. for example, since 1971, it has included questions that enable analysts to track internal migration across and between intercensal periods, and to distinguish place of usual residence from place of enumeration, on a quinquennial basis. our central claim in this paper is not that the concept of ‘usual residence’ has no place in the census, but rather that accommodating (no pun intended) more than two positions in relation to usual residence might more accurately reflect the underlying realities of how population and locality intersect for highly mobile population groups. in particular, we are interested here in practices of residency that engage two or more places one might ‘usually live’ (i.e., bi-locale and multi-locale living) as well non-location-specific forms of residency that do not constitute conventional forms of primary or secondary homelessness such as sleeping rough or couch surfing. in its 2018 public census review, the australian bureau of statistics (abs) received 355 submissions regarding proposals to change or add census topics (abs 2018). twenty-four of these addressed the topic of location (within which ‘usual residence’ was a sub-topic) and included proposals to introduce a question regarding second residences, as well as the need for better information regarding people “staying temporarily in other household without a usual residence”. while the first proposal was deemed nationally significant by the abs, particularly in small towns and communities where seasonal population influxes impacted service provision, it was ultimately determined that after extensive research and testing, the question proved too complex for respondents and added considerable cost and complexity to processing census forms (abs 2018). with respect to the second proposal, while the abs acknowledged the importance of data and information regarding people with no ‘usual residence’ it expressed concern that changes to questions on addresses, which are foundational for census applications, “will only be considered with a view of the potential impact on quality of population estimates” (abs 2018). in practice, this meant introducing very minor amendments to the instructions regarding how to respond to questions about place of usual residence for those that have no place of usual residence (see table 1). table 1: census form changes to usual residence question instructions, 2016-2021 2016 census 2021 census for people who have no fixed or return address (for example, due to family conflict or eviction), write ‘none’ in the ‘suburb/locality’ box. for persons who have no usual address, write ‘none’ in the ‘suburb/locality’ box. source: abs census forms at the 2016 census, 45,179 people identified having no place of usual residence on the census form (abs 2021a). de-identified census records enable analysis of the socio-economic and demographic 12 davies & prout quicke australian population studies 5 (1) 2021 characteristics of these individuals. however, there remains within the current census instrument no real capacity to distinguish primary and secondary homelessness from those who are permanently ‘nomadic’ or live between two or more localities. in the next section, we turn to examining these quandaries in relation to one case-study sub-population: so-called ‘grey nomads’. in the broadest sense, grey nomads are typically characterised as older australians who travel for extended periods of time seasonally, or on a permanent basis, throughout australia. as australia’s population profile ages, and as international travel is likely to be limited and high-risk for older populations for some time to come, this sub-population is likely to continue to grow. 3. grey nomads in australia in australia, moving to high amenity and lower-cost areas for retirement is a long established and well recognised trend (davies and james 2011). while retirement migrants have typically been thought to prefer high amenity, coastal and climatically warm destinations, regional mobility data over the last decade also highlights the growth in popularity of inland and colder climate destinations amongst older australians who are not in the labour force (abs 2021c). these retirement migrations, involving a permanent one-way move from one residence to another, are relatively easy to infer from census population data. this stands in stark contrast to the capacity to accurately capture the movements of grey nomads (davies 2011). the published research on australian grey nomads, most of which comes from tourism studies, draws substantively on case study-based investigations and survey-based methodologies. the variability in the spatial, temporal and demographic parameters used to define the grey nomad population has resulted in variability in research sampling frames and sampling strategies across these studies and, in turn, variability in conclusions being draw about the grey nomad population in respect of its size, characteristics and movements. in one of the earliest studies of grey nomads, onyx and leonard (2005 p. 61), defined this subpopulation as “people aged over 50 years, who adopt an extended period of travel (at least 3 months) independently within their own country”. their definition made no stipulation regarding the retirement status of grey nomads. with respect to spatial dimensions, onyx and leonard (2005) drew comparisons between grey nomads and the better known north american snowbirds. they argued that while there were similarities between australian grey nomads and north american snowbirds, the phenomena were different. north american snowbirds seasonal moves are bi-locale and highly location-specific: they leave their colder northern ‘snowbelt’ states during winter months for trailer parks and retirement communities in the more temperate southern ‘sunbelt' states of the usa (happel and hogan 2002), returning to the same destination year after year (smith and house 2006; viallon 2012). however, onyx and leonard found in their study that grey nomads often avoided highly commodified experiences and were highly mobile, moving from destination to destination. while climate did impact destination selection, the more significant drivers were socio-cultural motivations, with the grey nomads seeking to explore the country, engage in experiences and have an adventure (onyx and leonard 2005). reporting on a large study of 400 grey nomads two years later, onyx and leonard (2007 p. 384) also noted that some grey nomads travelled for extended periods (up to several years) and thousands of kilometres during this time. australian population studies 5 (1) 2021 davies & prout quicke 13 in a large study of the population geography of grey nomads published shortly thereafter, cridland (2008) – who also defined grey nomads using a chronological age of 50+, but used shorter temporal dimensions of travelling (one month or more) – sampled almost 2,000 grey nomads staying in caravans at large/commercial caravan parks (with more than 50 powered sites) across northern australia. he also identified variability in destination and patterns of movement and developed a typology of grey nomads based on this variability and types of activities undertaken. however, studies that focus on non-random sampling from particular kinds of localities such as large caravan parks and rallies (e.g., wu and pearce, 2017; mahadevan, 2014) cannot be drawn upon (nor do they make such claims) to extrapolate the characteristics and patterns of movements of all grey nomads since a portion of the grey nomad population is known to deliberately avoid highly commodified locations such as major caravan parks (davies and james 2011). like onyx and leonard (2007) and cridland (2008), davies (2011) identified grey nomads using a chronological signifier (aged 55+). she extended previous analyses by showing that grey nomads could stay in a broader range of accommodation types during their travels, including hotels and private houses. however, in contrast to onyx and leonard (2005; 2007), cridland (2008) and davies (2011) some scholars eschew chronological parameters for defining grey nomads and instead draw on labour force status, and specifically retirement, as a key parameter (e.g., higgs and quirk 2007; hillman 2013). more recent studies and reports on grey nomads tend to use chronological age classifications and specify that grey nomads are free from travel limiting work and family obligations, but do not specify that they are retired (see for example, darley et al. 2017). the variability of these definitional parameters, and the findings subsequently produced from each study highlight the need for a more standardised set of definitional parameters for identifying the grey nomad population in order to facilitate analysis of the size, characteristics and migration practices of the grey nomad sub-population. the abs investigated this as part of a 2011 study on homelessness. it defined grey nomads as “people in dwellings where all people in the dwelling were aged 55 years and over, were not in the labour force, and were staying in caravans, cabins or houseboats on census night” (abs 2011). the abs found that on census night approximately 2,500 people in australia were grey nomads and of these, 80% were two person households (abs 2011). the analysis revealed that 45% of the grey nomad population was ‘elsewhere in australia’ compared to a year previous and 17% were at the same location as they were a year previous. interestingly, 40% recorded having no usual address the year pervious, suggesting that approximately 1,000 older australians were, at that time, permanently nomadic. however, as with the aforementioned case study-based investigations into grey nomads, the abs’s conclusions about grey nomads is limited by the definition used. specifically, by viewing grey nomads as ‘retired’ the abs’s analysis only included those not in the labour force. there is a need to revisit the definition used by the abs in 2011 and consider alternative parameters for capturing meaningful information about this population in the census. as most scholars agree that grey nomads include those living in caravans, cabins and houseboats and, with 55 years the most commonly used age marker in studies published in the last decade, these characteristics in the definition do not require reconsideration and can be easily extracted from existing census datasets. it is also most commonly agreed across the literature that grey nomads need to be traveling for an extended period to be considered a grey nomad. while the 14 davies & prout quicke australian population studies 5 (1) 2021 duration of their travels is not possible to determine from the census, it is possible to determine that the dwelling location categories ‘manufactured home estates’ and ‘retirement villages’ should be excluded from the analysis. it becomes more problematic when trying to accommodate variable retirement status among grey nomads, as well as the subset of grey nomads who are permanently nomadic. on the first point, as some grey nomads are not retired (indeed the website the grey nomads at https://www.thegreynomads.com.au/ includes testimonials of grey nomads who are ‘working’ and a classifieds section for those seeking work) it is necessary to include those who are still in the labour force in any analysis. however, in including this group, it is likely that those who travel for employment and who do not consider themselves to be grey nomads would also be included. those who are employed full-time or are unemployed seeking full-time employment can be excluded, as these people are unlikely to be grey nomads. on the second point, it initially appears possible to identify grey nomads by identifying from the census those who were ‘away from home’. as a portion of grey nomads are permanently nomadic, it is necessary to include those who were ‘away from home’ as well as those who were ‘at home’ on census night. however, including those who were ‘at home’ in a mobile dwelling at a caravan/residential park, camping ground or marina then includes those who permanently live in these places and do not travel and therefore would not fit the definition of a grey nomad. table 2 shows these two groups. table 2: people aged 55 and over in a caravan, cabin or houseboat on census night 2016 employed, part time employed, away from work unemployed, looking for part time work not in the labour force total elsewhere in australia caravan, at caravan/ residential park or camping ground 1,356 3,715 179 42,177 47,427 cabin, houseboat at a marina 32 20 4 324 380 at home caravan, at caravan/ residential park or camping ground 1,284 377 161 12,703 14,525 cabin, houseboat at a marina 835 147 87 8,634 9,703 total 3,507 4,259 431 63,838 72,035 source: abs (2021b) in considering those who were ‘at home’ in a mobile dwelling, general characteristics can be identified such as age profile, median income, education, gender, volunteer participation, and need for assistance. these data could be used to guide policies to ensure that appropriate social and https://www.thegreynomads.com.au/ australian population studies 5 (1) 2021 davies & prout quicke 15 health care services and infrastructure are developed to support grey nomads. however, as already noted, these data would include individuals who do not identify as grey nomads. in terms of revealing the mobility patterns of grey nomads, duration of travel, and bi-local and multi-local forms or residence, the extant census data are limited in their utility. 4. discussion and conclusion: where to for the census? official statistics, such as those derived from national censuses are powerful tools of governmentality (kukutai and taylor 2012). the way the census conceptualises categories of analysis determines the kinds of social realities that are rendered visible through it. the normative framing of the concepts of ‘usual residence’ and ‘home’ in the census obscures the variability in the way some sections of australia’s population experience residence and home. this paper focused on the grey nomad population to emphasise the challenges that exist in trying to develop an understanding of the size and characteristics of this subpopulation from census data. with grey nomads not fitting into normative framings of usual residence and home, researchers have relied on discrete case studybased data to infer characteristics about the grey nomad population. however, a review of the extant literature reveals variability between the definitions researchers have used to identify the grey nomad population. these differences in definitions have informed the development of different research sampling frames, which potentially skews understandings developed about the grey nomad population. of course, the grey nomad population is just one example of a mobile sub-population whose mobility practices are not well understood. the rationalities and service needs of other sub-populations who engage in bi-locale and multi-locale living, as well as those that are not homeless but have no permanent addresses(s) are also invisibilised by the abs’s normative framework for conceptualising usual residence. the invisibility of these rationalities has very real policy implications. as an example, australia’s public health response to the covid-19 pandemic, which has sought to significantly curb mobility during particular periods, rests heavily on the understanding that most people have a permanent home to which they can retreat, or are homeless. however, as we do not collect data on non-normative performances of residence, it is difficult to determine how many people might have been left significantly displaced or detrimentally impacted by such mobility restrictions. there are also broader non-covid related policy challenges for mobile population groups, from ensuring access to basic social services such as health and education services, to navigating through the common tropes often associated with mobile populations (e.g., as ‘parasitic’ in nature, as threats to conservation practices, as ungovernable etc.) (see gilbert 2014). the utility of the data collected through the census is, necessarily, limited by the nature, number and complexity of questions that can be included. however, we argue that in looking to the 2026 census, enabling residence data to be captured in such a way as to facilitate an improved understanding of bi-local, multi-local and nomadic populations would enable a richer understanding of residence and population mobility in australia and provide critical insights to better service these populations, wherever they may be located. 16 davies & prout quicke australian population studies 5 (1) 2021 references abs (2011) abs review of counting the homeless methodology, aug 2011. canberra: abs. https://www.abs.gov.au/ausstats/abs@.nsf/latestproducts/03750f7f8176784dca2578e200193 2bf?opendocument. accessed on 12 march 2021. abs (2018) census of population and housing: consultation on topics, 2021. canberra: abs. https://www.abs.gov.au/ausstats/abs@.nsf/lookup/by%20subject/2007.0.55.001~2021~main% 20features~preface~1. accessed 12 march 2021. abs (2021a) census – counting persons, place of usual residence (mb) (2016), tablebuilder pro. findings based on use of abs tablebuilder data. abs (2021b) census – counting persons, place of enumeration (mb) (2016), tablebuilder pro. findings based on use of abs tablebuilder data. abs (2021c) regional internal migration estimates, provisional. canberra: abs. https://www.abs.gov.au/statistics/people/population/regional-internal-migration-estimatesprovisional/latest-release. accessed on 12 march 2021. charles-edwards e, bell m, panczak r and corcoran j (2020) a framework for official temporary population statistics. journal of official statistics 36(1): 1-24. https://doi.org/10.2478/jos-20200001 cridland s (2008) an analysis of the winter movement of grey nomads to northern australia: planning for increase senior visitation, phd thesis. cairns: james cook university. darley t, lambert c and ryan m (2017) grey nomads’ caravanning use of social networking sites. information, technology and tourism 17(4): 379-398. https://doi.org/10.1007/s40558-017-00916 davies a (2011) on constructing ageing rural populations: ‘capturing’ the grey nomad. journal of rural studies 27(2): 191-199. https://doi.org/10.1016/j.jrurstud.2011.01.004 davies a and james a (2011) geographies of ageing: social processes and the spatial unevenness of population ageing. aldershot: ashgate. fowler m (2020) ‘just wild’: mass caravan park closures shatter nomads’ dreams. the sydney morning herald, 29 march 2020. https://www.smh.com.au/national/just-wild-mass-caravan-parkclosures-shatter-nomads-dreams-20200327-p54eml.html gilbert j (2014) nomadic peoples and human rights, london: taylor and francis. happel s and hogan t (2002) counting snowbirds: the importance of and the problems with estimating seasonal populations. population research and policy review 21(3): 227-240. https://doi.org/10.1023/a:1019560405310 higgs p f d and quirk f (2007) “grey nomads” in australia: are they a good model for successful aging and health? annals of the new york academy of sciences 1114(1): 251-257. https://doi.org/10.1196/annals.1396.030 hillman w (2013) grey nomads travelling in queensland, australia: social and health needs. ageing & society 33(4): 579-597. https://doi.org/10.1017/s0144686x12000116 holloway d, green l and holloway d (2011) the intratourist gaze: grey nomads and ‘other tourists’. tourist studies 11(3): 235-252. https://doi.org/10.1177/1468797611432043 kukutai t and taylor j (2012) postcolonial profiling of indigenous populations. espace, populations societes 2012(1): 13-27. https://doi.org/10.4000/eps.4812 mahadevan r (2014) understanding senior self-drive tourism in australia using a contingency behaviour model. journal of travel research 53 (1): 252-259. https://doi.org/10.1177/0047287513496476 https://www.abs.gov.au/ausstats/abs@.nsf/latestproducts/03750f7f8176784dca2578e2001932bf?opendocument https://www.abs.gov.au/ausstats/abs@.nsf/latestproducts/03750f7f8176784dca2578e2001932bf?opendocument https://www.abs.gov.au/ausstats/abs@.nsf/lookup/by%20subject/2007.0.55.001~2021~main%20features~preface~1 https://www.abs.gov.au/ausstats/abs@.nsf/lookup/by%20subject/2007.0.55.001~2021~main%20features~preface~1 https://www.abs.gov.au/statistics/people/population/regional-internal-migration-estimates-provisional/latest-release https://www.abs.gov.au/statistics/people/population/regional-internal-migration-estimates-provisional/latest-release https://doi.org/10.2478/jos-2020-0001 https://doi.org/10.2478/jos-2020-0001 https://doi.org/10.1007/s40558-017-0091-6 https://doi.org/10.1007/s40558-017-0091-6 https://doi.org/10.1016/j.jrurstud.2011.01.004 https://www.smh.com.au/national/just-wild-mass-caravan-park-closures-shatter-nomads-dreams-20200327-p54eml.html https://www.smh.com.au/national/just-wild-mass-caravan-park-closures-shatter-nomads-dreams-20200327-p54eml.html https://doi.org/10.1023/a:1019560405310 https://doi.org/10.1196/annals.1396.030 https://doi.org/10.1017/s0144686x12000116 https://doi.org/10.1177/1468797611432043 https://doi.org/10.4000/eps.4812 https://doi.org/10.1177/0047287513496476 australian population studies 5 (1) 2021 davies & prout quicke 17 moore t (2018) grey nomad lifestyle provides a model for living remotely. the conversation 10 december 2018. https://theconversation.com/grey-nomad-lifestyle-provides-a-model-for-living-remotely106074 onyx j and leonard r (2005) australian grey nomads and american snowbirds: similarities and differences. the journal of tourism studies 16(1): 61-68. onyx j and leonard r (2007) the grey nomad phenomenon: changing the script of aging. international journal of aging and human development 64(4): 381-398. https://doi.org/10.2190/g4n9-36273g7j-w454 panczak r, charles-edwards e and corcoran j (2020) estimating temporary populations: a systematic review of the empirical literature. humanities and social sciences communications volume 6: 87. https://doi.org/10.1057/s41599-020-0455-y prout quicke s and green c (2018) “mobile (nomadic) cultures” and the politics of mobility: insights from indigenous australia. transactions of the institute of british geographers 43(4): 646-660. https://doi.org/10.1111/tran.12243 raven m (2016) health of grey nomads: on the move and under the health sector radar. australian journal of rural health 24(3): 182-187. https://doi.org/10.1111/ajr.12236 smith s k and house m (2006) snowbirds, sunbirds, and stayers: seasonal migration of elderly adults in florida. the journals of gerontology: series b 61(5): s232-s239. https://doi.org/10.1093/geronb/61.5.s232 viallon p (2012) retired snowbirds. annals of tourism research 39(4): 2073-2091. https://doi.org/10.1016/j.annals.2012.06.001 wu m and pearce p (2017) the rally experience: exploring motivation patterns of australian grey nomads. journal of destination marketing and management 6: 407-415. https://doi.org/10.1016/j.jdmm.2016.06.008 https://theconversation.com/grey-nomad-lifestyle-provides-a-model-for-living-remotely-106074 https://theconversation.com/grey-nomad-lifestyle-provides-a-model-for-living-remotely-106074 https://doi.org/10.2190/g4n9-3627-3g7j-w454 https://doi.org/10.2190/g4n9-3627-3g7j-w454 https://doi.org/10.1057/s41599-020-0455-y https://doi.org/10.1111/tran.12243 https://doi.org/10.1111/ajr.12236 https://doi.org/10.1093/geronb/61.5.s232 https://doi.org/10.1016/j.annals.2012.06.001 https://doi.org/10.1016/j.jdmm.2016.06.008 australian population studies 2020 | volume 4 | issue 1 | pages 70-72 © kimpton 2020. published under the creative commons attribution-noncommercial licence 3.0 australia (cc by-nc 3.0 au). journal website: www.australianpopulationstudies.org visualising australia’s older population using grid maps anthony kimpton* the university of queensland * email: a.kimpton@uq.edu.au. address: queensland centre for population research, school of earth and environmental sciences, chamberlain building, the university of queensland, st lucia, qld 4072, australia. paper received 17 april 2020; accepted 28 april 2020; published 25 may 2020 introduction in 2016, 15.7 percent of australians were aged 65 or over (3.7 million out of a total population of 23.4 million; australian bureau of statistics 2016). this national age structure is already without precedent while projections suggest this could reach 22 percent by 2057 (8.8 million; australian institute of health and welfare 2018). as such, these projections suggest that it will become increasingly challenging for all tiers of australian government to fund the infrastructure and support services critical for the health and wellbeing of older australians as they grow at a faster rate than the working age population (parliamentary budget office 2019). there are four critical dimensions when examining population age structures, namely: (1) numbers; (2) characteristics and values; (3) proportions of the population in particular age groups; and (4) spatial distribution (hugo 2003). in this demographic i focus on the latter two dimensions to identify where australians aged 65 and over are most spatially concentrated. data and methods i draw on the australian bureau of statistics (abs) 2016 census of population and housing data in statistical area level two (sa2) neighbourhood units released as open data from their datapacks web page (abs 2016). the abs designed sa2s to approximate functional communities that are connected through regular social and economic interaction, although australian population density is low within the interior but relatively high towards the coast, with sa2s ranging in area between 49 hectares and 51 million hectares. given this variability in area, it has remained a challenge to visualise population structures across australia. i have therefore generated a gridded population that is typically featured in dasymetric population maps (silva et al. 2013; li et al. 2016). specifically, i converted the population into points located at random locations across all sa2 polygons, aggregated these points to a decimal degreespaced grid, calculated the gridded proportion of australians aged 65 or over, and then plotted these values using proportional circles. these gridded proportion circles are useful because they reveal the geographic spread of national age structure and normalise the area containing populations to a grid, thus ensuring that there are no overlapping symbols or polygons that are too small to read. in addition, i have introduced marginal histograms (lambrechts 2019) to this map to reveal the multid em og ra ph ic http://www.australianpopulationstudies.org/ mailto:a.kimpton@uq.edu.au australian population studies 4 (1) 2020 kimpton 71 modal distribution of the australian age structure along lines of latitude and longitude, thus enabling a unique spatial examination of australia’s older population. key features the demographic (figure 1) depicts clear spatial patterning. for instance, the symbols scaled according to population structure reveal that there are areas without symbols, indicating that fewer than 1 percent of the gridded population are aged 65 or older. symbols larger than the 30 percent symbol featured within the legend reveal that there are ageing enclaves throughout australia where the population aged 65 or older is more than twice the national average of 15 percent, such as the 32 percent circle representing approximately 25,000 australians residing along the fleurieu peninsula, which is located south of the city of adelaide. this spatial pattern highlights the regional dimension of ageing and echo other work that has outlined this pressing issue for regional communities (houghton & vonthethoff 2017). lastly, the marginal histograms that reveal the proportion of the population that is aged 65 or older for a given latitude or longitude reveals that these concentrations can range from 5 to 23 percent across the surface of australia. figure 1: a grid map and marginal histograms of the proportion of australia’s population aged 65 and older source: calculated by the author using data extracted from the 2016 census using data from the australian bureau of statistics census datapacks (abs 2016) 72 kimpton australian population studies 4 (1) 2020 supplementary material the bespoke script developed to read and prepare the abs data and figure 1 is available from the author’s online repository at https://rpubs.com/anthonykimpton/. acknowledgements this demographic is developed through a project funded by the australian research council linkage project grant lp160100031 with additional support from the queensland department of transport and main roads as industry partner. the analysis and interpretations are solely those of the author and do not necessarily reflect the views and opinions of the department or any of its employees. references australian bureau of statistics (2016) census datapacks. https://datapacks.censusdata.abs.gov.au/datapacks/. accessed on 15 april 2020. australian institute of health and welfare (2018) older australia at a glance. https://www.aihw.gov.au/reports/older-people/older-australia-at-a-glance. accessed on 15 april 2020. houghton k and vonthethoff b (2017) ageing and work in regional australia: pathways for accelerating economic growth. http://www.regionalaustralia.org.au/home/ageing-work-regional-australia/. accessed on 15 april 2020. hugo g (2003) australia's ageing population: some challenges for planners. australian planner 40(2): 109118. li s, juhász-horváth l, harrison p a, pintér l and rounsevell m d a (2016) population and age structure in hungary: a residential preference and age dependency approach to disaggregate census data. journal of maps 12(sup1): 560-569. lambrechts m (2019) how to make a grid map with histograms in r with ggplot. flowingdata. https://flowingdata.com/2019/12/16/grid-map-histogram-ggplot/. accessed on 15 april 2020. parliamentary budget office (2019) australia’s ageing population: understanding the fiscal impacts over the next decade. report no. 02/2019. canberra: parliamentary budget office. silva b, javier gallego f, and lavalle c (2013) a high-resolution population grid map for europe. journal of maps 9(1): 16-28. https://rpubs.com/anthonykimpton/ https://datapacks.censusdata.abs.gov.au/datapacks/ https://www.aihw.gov.au/reports/older-people/older-australia-at-a-glance http://www.regionalaustralia.org.au/home/ageing-work-regional-australia/ https://flowingdata.com/2019/12/16/grid-map-histogram-ggplot/ australian population studies 2021 | volume 5 | issue 1 | pages 49-55 © stevens & fozdar 2021. published under the creative commons attribution-noncommercial licence 3.0 australia (cc by-nc 3.0 au). journal website: www.australianpopulationstudies.org commentary ethnicity, race or nation? census classifications as barriers to the measurement of mixedness in australia catriona stevens* the university of western australia farida fozdar the university of western australia * corresponding author. email: catriona.stevens@uwa.edu.au. address: school of social sciences, the university of western australia, 35 stirling highway, perth, wa 6009, australia. paper received 1 march 2021; accepted 25 april 2021; published 31 may 2021 1. introduction it is an unfortunate truth that all systems of ethnic and racial classification are in some sense ‘wrong’, with each failing to adequately capture the complexity of a given population. this potentially perpetuates and legitimises structures of (dis)advantage based in culture or visible difference. regardless of approach taken, such classification systems will likely confuse, obscure and/or offend at least some members. but given australia’s diversity, its history as a settler colony dispossessing and numerically overwhelming its indigenous peoples, its decades as ‘white australia’, its more recent welcoming of migrants from a range of ethnic, racial, religious and national backgrounds, and its growing ‘mixed’ population, such systems are vital for determining who ‘we’ are and for mapping these changes and their wider correlates. our particular interest is in the (in)ability to identify the extent, composition and rate of mixed populations in australia. but effective classification and measurement of the ethnic and racial identities of census respondents will also support the accurate targeting and outcomes of multicultural service provision, and help scholars, policy makers, and ethnic communities understand demographic change over time. the current australian census classification system is a great improvement on earlier approaches that were informed solely by the logics of biological race in the service of a desired white nation (fozdar and stevens 2020). yet the australian standard classification of cultural and ethnic groups (ascceg), the statistical standard used by the australian bureau of statistics (abs) for the collection, aggregation and dissemination of data relating to the cultural and ethnic diversity of the australian population, is, like many other classificatory systems, still strange and inconsistent. the current approach is limited and anachronistic. this is not only an issue for social scientists seeking to understand these changes and related socioeconomic patterns. the census also sends a message, signalling the types of information deemed valid and worth collecting, and the sorts of personal characteristics seen as culturally important. in early census schedules, a race question was included in order to monitor the size of the noneuropean population. for example, in 1911 the categories employed were: asiatic, african, american, http://www.australianpopulationstudies.org/ mailto:catriona.stevens@uwa.edu.au 50 stevens & fozdar australian population studies 5 (1) 2021 polynesian, and indefinite, with each category subdivided into ‘full blood’ and ‘half-caste’. in this rendering, mixedness was only relevant insofar as it involved european and non-european ‘blood’. following the introduction of anti-discrimination legislation in 1975, the steady retreat from the earlier racist logic of the census categories led to a complete loss of race-based categories. the only exception is indigeneity, with a question asking whether the respondent is “of aboriginal or torres strait islander origin”. since 1986, ethnicity has replaced race as the key official measure of communities of descent, with the ascceg introduced as the statistical standard in 2000. however, the categories employed remain limited, and in some cases are downright bizarre, for example, including explicitly racial categories alongside generic national identifiers that mask significant ethnic and racial diversity (discussed shortly). while recognising the need for consistency over time, this commentary proposes some changes to permit expanded ways of measuring ethnic and particularly racial mixedness while minimising disruption to longitudinal studies. 2. limitations of ascceg classifications and aggregation mechanisms we argue the ascceg classification categories require reform due, in part, to a range of issues including: the current inconsistency and confusion in terms of the logic of categorisation; the increasing diversity of the population and its desire to acknowledge these backgrounds; and the growing levels of mixedness in australia. with regard to the latter, it is important to have the capacity to measure the mixed population due to its transformative potential (king-o’riain et al. 2014; parker and song 2001; meyer and fozdar 2016). this includes not only mapping the changing ‘face’ of australia, but also mixedness’s ability to erode entrenched spatial segregation (tindale and klocker 2014, 2017), both resulting from and generating a ‘contact’ effect – lower levels of prejudice and improved cross cultural relations (pettigrew and tropp 2006). however, considerable analytical contortionism is necessary to derive ‘mixedness’ from the current classification system. this is partly because of the lack of a ‘race’ based question. this is the focus of our discussion. for the 2021 census, the ascceg system is used to code responses to the ancestry question (household paper form question 22). it is applied to aggregate base-level four-digit ancestries, the most fine-grained level of analysis, into two-digit narrow groups and one-digit broad groups. these broad groups typically equate to continent of ancestry. this commentary primarily relates to question 22 and the related statistical standard, but also makes reference to the following 2021 census topics: main language other than english used at home (question 20); country of birth (question 16); and indigenous status (question 11). we are interested in these topics in relation to the ability to classify people of mixed backgrounds, and also in relation to race or visible difference. one of the main points we wish to emphasise is that there are significant inconsistencies in how the ascceg defines 'cultural and ethnic groups', with confusion about national, racial and ethnic categorization leading to often functionally meaningless classifications. the base level 4-digit categories variously describe subnational ethnic groups (for example, “5206 sundanese”, an austronesian ethnic group from western java, and the second largest ethnic group in indonesia), ethno-national categories (such as “6901 japanese”), national categories without ethnic or racial specificity (such as “8203 brazilian”, collapsing people of white, mixed, black, asian and indigenous australian population studies 5 (1) 2021 stevens & fozdar 51 backgrounds), and explicitly racial/panethnic categories (for example “8105 hispanic [n american]”, a category specifically for those from north america of latino or spanish ancestry). of this range of categorisations, national categories without ethnic or racial specificity are by far the most common, reinforcing a supposed alignment of race/culture/ethnicity with nation; and perpetuating the pervasive methodological nationalism that has been critiqued for its theoretical and practical inadequacy and anachronism (beck 2007). the logics of the ascceg are perhaps most strange when considering the americas. ancestry options for people of north america are essentially racial, while those from other countries are not. north american options include “8101 african american”, “8105 hispanic [north american]” and “8106 native north american indian” alongside the racially more ambiguous (although presumably white) “8102 american” and “8103 canadian”. these base level ancestries reflect some of the racial categories from the us census (herman 2020) and appear to have simply been imported wholesale from that. yet other classificatory systems from other countries, many of which are inflected with legacies of similarly racist and racialized histories (aspinall and rocha 2020), are not. for example, explicitly racial categories from the british census (e.g., asian british [meaning primarily people of south asian ancestries]; black caribbean; black african) are missing, with those identifying with british ancestries assumed to be white. thus, there is apparently little consistency, and an unclear logic, to the categories available for use, leading to a differential ability among analysts to clearly identify sub-national ethnic and racial backgrounds of respondents. this makes more detailed analysis of populations difficult. it also limits respondents in their ability to identify in ways they may prefer. there are further troubling inconsistencies produced through the ways that categories are aggregated in the ascceg’s hierarchical structure. base level categories are aggregated on the basis of two classification criteria: geographic proximity of ancestral origins, and ‘similar social and cultural characteristics’ (abs 2019). this is, however, a blunt tool, and in practice aggregation at both the broad and narrow levels conflates groups with different ethnic/racial backgrounds, or conversely, separates groups with similar backgrounds. as well as being logically inconsistent, both of these practices create barriers to measuring and analysing racial mixedness in the australian population. for example, “9201 afrikaner”, “9202 angolan” and “9203 eritrean” are the first three categories in the narrow group “92 southern and east african”. geographic, social and cultural similarity, not to mention racial, is not evident here. an analysis of census data at the oneor two-digit level reveals high rates of mixedness between north-west europeans and southern and east africans, yet a closer investigation would show that this rarely entails the racial mix that one might assume. similarly, although north american base-level ancestries are explicitly racial, when aggregated to the broad level, this obscures visible difference and visible mixedness. meanwhile, other ancestries are separated into different narrow and broad aggregated groups for reasons that are unclear whether on grounds of geography, ethnicity or language. two examples from europe and africa illustrate just how illogical some outcomes from the ascceg classification criteria may be. france and spain are geographically proximate, sharing a border, and the countries feature many ‘similar social and cultural characteristics’ including, for example, speaking romance languages and a history of adherence to roman catholicism. yet “3106 spanish” is placed in broad 52 stevens & fozdar australian population studies 5 (1) 2021 group “3 southern and eastern european”, while “2305 french” aggregates to broad group “2 northwest european”. this appears to reflect the legacy of historical othering of southern europeans, and implicit layered racism within the charter group. ascceg explanatory notes acknowledge the problems of including “4201 jewish” in broad group “4 north african and middle eastern”, with community consultation failing to determine a better approach (abs 2019). yet there are other confusing aspects to this broad group that merit further attention. for example, narrow group “49 other north african and middle eastern” includes both “4904 iranian” and “4913 nubian”. peoples of the sudan, meanwhile, are grouped together, notwithstanding the deep ethnic and cultural conflicts that have shaped and reshaped the national borders in this region, resulting in the strange situation where “4305 south sudanese” aggregates to broad group “4 north african and middle eastern” while people identifying with ancestries “9203 eritrean” or “9204 ethiopian” are considered part of broad group “9 sub-saharan african”. and of course “9215 south african”, which includes white, black, mixed and asian people who have migrated from a country of 59 million, are also part of this group, together with, among others, “9235 madi”, a small ethnic group of about 400,000 sudanic-speaking people living across sudan and uganda. neither geographic, nor cultural, nor racial logics seem to apply. finally, and perhaps of greatest significance to our interest in mixedness in australia, is that question 22 in the 2021 census unduly obscures the multiple ancestral identities that are increasingly prevalent. recognising australia’s growing diversity, the ascceg notes that “many people will relate to more than one cultural or ethnic group and will give a multiple response to a question on ancestry… data collection models should be developed to capture multiple responses to enhance usefulness of the output” (abs 2019). however, in practice the census currently only permits two responses to the ancestry question. this is inadequate as australians increasingly identify with many different cultural backgrounds (fozdar and stevens 2020; fozdar and volet 2016). as comparisons, respondents to the canadian census are instructed to ‘[s]pecify as many origins as applicable’ for the ancestry question and indicate all that apply for the ethnicity question (kwan-lafond and winterstein 2020 pp. 80-82), while census respondents in britain are invited to write in ‘any other mixed/multiple ethnic background’ in an open field (aspinall 2020). these are far more flexible options that allow more accurate identification, honouring respondents’ desire to acknowledge a range of backgrounds, and offering researchers the ability to track growing complexity. 3. alternative measures of ethnic and racial difference we propose that the ancestry question in the australian census be amended to accommodate multiple responses, perhaps four to six options, rather than two, providing useful data on the kinds of mixedness appearing in the population, while also supporting respondent engagement with the census through the official validation of diverse forms of identity and mixedness. while the ancp1 (first stated ancestry) and ancp2 (second stated ancestry) are important data outputs to preserve for longitudinal analyses, the inclusion of further ancestries should not impact the validity of these data. we also propose further consultations to explore the merits of a population group question to better understand visible difference and visible mixedness in australia. such understandings are highly australian population studies 5 (1) 2021 stevens & fozdar 53 limited when relying on the country of birth, ancestry and indigenous status data currently collected (fozdar and stevens 2020). there are precedents for this. in addition to an ancestry question very similar to that used in australia, statistics canada includes a further ‘visible minority’ question that closely approximates a race question, namely “is the person white / chinese / south asian / black / filipino / latin american / southeast asian / arab / west asian / korean / japanese?” (kwan-laford and winterstein 2020). respondents are invited to tick all that apply, allowing for those who identify with multiple backgrounds to include these. the results of these data are not perfect: arab, latin american and west asian, for example, are not considered ‘visible minorities’. however, this question does offer a range of self-identifications that could provide an indication of the current ‘racial’ make-up of australia, at least more clearly than can currently be inferred from ancestry responses aggregated at the broad and narrow level groupings. we acknowledge that the inclusion of racial terminology may be controversial, given australia’s history, but point out that racial categories already exist, in the form of indigenous classifications. current practice, which frequently mixes racial, ethnic, national, and even geographical categories, is inconsistent and unhelpful. while we would not suggest that the ascceg and ancestry question be abandoned, due to the need to be able to make longitudinal comparisons, we do suggest some refinement and a new ‘population group’ question is required. miri song (2018) has argued that racial classification remains vital for academics and policy makers to understand racial difference and mixing. official avoidance of the term in favour of ethnic identification, and in australia’s case, identification that confuses national, ethnic and racial categories, actually creates problems, as it makes it difficult to differentiate between different kinds of ethnic and racial experiences. it also makes it impossible to confidently draw conclusions about inter-ethnic and inter-racial unions. the current options thus invisibilise, or at best confuse, the dimensions on which differences of outcomes may rest. 4. discussion we acknowledge that ethnicity and race questions are difficult, given their potential to alienate, reference historical injustices, invisibilise, and to ignore specificities – as stated at the outset of this commentary – whatever choices are made will always be somehow ‘wrong’. however, we argue more choice is better than less, and that respondents are capable of engaging with the range of options to self-identify in ways that are meaningful to them. we reiterate that the census, as the official tool of the nation for data collection, and the categories it uses, sends a message, signalling the types of information seen as valid and valuable, with implications for the imagining of the nationstate and its peoples, and its orientation to diversity and inclusion. we have raised a number of questions above – why are the ascceg categories so arbitrary and imprecise, apparently importing us racial classifications for some people, and collapsing racial and ethnic categories into national options for others? although only adopted in 2000, the ascceg appears embedded in the logics of an earlier era. as such, it is oriented towards the past, and may not currently be able to accommodate the complexities of future populations. we further have queried: what is the point of allowing people four-digit specific options if these are then so poorly aggregated, erasing that specificity? is it realistic to allow people only two ancestry options, and what 54 stevens & fozdar australian population studies 5 (1) 2021 information is being lost as a result? how can mixedness be determined, with any degree of confidence, when the classifications are often at best loose and at worst arbitrary, and sometimes race-based, and sometimes not? in summary, we recommend small, possibly incremental, changes that will: • reduce inconsistencies in the ascceg, through the removal of classifications based on the racial logics of north america; and • permit more than two ancestral identifications and hence data collection that reflects the changing demographic realities of (what we think is) our uniquely diverse australian population. we also recommend exploration of a race/population group/visibility question. likely benefits include: • a more precise and accurate estimation of the racial diversity of australia (often proportion of migrants is pointed to as evidence of australia’s racial and cultural diversity, but this tells us little about region of origin and cultural or racial background); • the opportunity to map the changing profile of race in australia; • the opportunity to capture racial heritage of second and further generation migrants; and • making race visible as a dimension of social life in australia. this would allow local, state and federal governments to develop targeted interventions and services where needed, for ethnic communities to better estimate their numbers and geographic spread, and for businesses to target communities where appropriate. of course, any changes would require research to understand how they might be interpreted by the australian community. overall, the issues raised are related to our interest in mixedness, and its extent, patterning and growth in australia. this is important for the reasons raised above, as it potentially signals a movement towards a more diverse, less segregated and more tolerant society. but there are wider reasons for a review and revision of the ways the census approaches ethnic and racial diversity. social scientists need to be confident about the logic of the classification system in order to make predictions and policy recommendations. and australians need to feel their identities are appropriately acknowledged through official channels such as the census. this is vital for the development of an inclusive and just society. references abs (2019) australian standard classification of cultural and ethnic groups (ascceg), 2019. https://www.abs.gov.au/ausstats/abs@.nsf/mf/1249.0. accessed on 14 february 2021. aspinall p (2020) the path to official recognition of ‘mixedness’ in the united kingdom. in: rocha z l and aspinal p j (eds) the palgrave international handbook of mixed racial and ethnic classification. london: palgrave macmillan; 229-248. https://doi.org/10.1007/978-3-030-22874-3_12 https://www.abs.gov.au/ausstats/abs@.nsf/mf/1249.0 https://doi.org/10.1007/978-3-030-22874-3_12 australian population studies 5 (1) 2021 stevens & fozdar 55 aspinall p and rocha z (2020) introduction: north and south america. in: rocha z l and aspinal pj (eds) the palgrave international handbook of mixed racial and ethnic classification. london: palgrave macmillan; 49-74. https://doi.org/10.1007/978-3-030-22874-3_3 beck u (2007) the cosmopolitan condition: why methodological nationalism fails. theory, culture & society 24(7-8): 286-290. https://doi.org/10.1177/02632764070240072505 fozdar f and stevens c (2020) measuring mixedness in australia. in: rocha z l and aspinal p j (eds) the palgrave international handbook of mixed racial and ethnic classification. london: palgrave macmillan; 605-628. https://doi.org/10.1007/978-3-030-22874-3_32 fozdar f and volet s (2016) cultural self-identification and orientations to cross-cultural mixing on an australian university campus. journal of intercultural studies 37(1): 51-68. https://doi.org/10.1080/07256868.2015.1119674 herman m (2020) methods of measuring multiracial americans. in: rocha z l and aspinal p j (eds) the palgrave international handbook of mixed racial and ethnic classification. london: palgrave macmillan; 95-112. https://doi.org/10.1007/978-3-030-22874-3_5 king-o’riain r, small s, mahtani m, song m and spickard p (eds) (2014) global mixed race. new york: new york university press. kwan-lafond, d and winterstein, s (2020) the canadian census and mixed race: tracking mixed race through ancestry, visible minority status, and metis population groups in canada. in: rocha z l and aspinal p j (eds) the palgrave international handbook of mixed racial and ethnic classification. london: palgrave macmillan; 75-94. https://doi.org/10.1007/978-3-030-22874-3_4 meyer m and fozdar f (2016) “a mini united nations”: being mixed in multicultural australia. in: mcgavin k and fozdar f (eds) mixed race identities in australia, new zealand and the pacific islands. new york: routledge; 49-66. parker d and song m (eds) (2001) rethinking ‘mixed race’. pluto press. pettigrew t f and tropp l r (2006) a meta-analytic test of intergroup contact theory. journal of personality and social psychology 90(5): 751-783. https://doi.org/10.1037/0022-3514.90.5.751 song m (2018) why we still need to talk about race. ethnic and racial studies 41(6): 1131-1145. https://doi.org/10.1080/01419870.2018.1410200 tindale a and klocker n (2017) mapping the multiple geographies of mixed-ethnicity couples in australia. australian geographer 48(4): 473-495. https://doi.org/10.1080/00049182.2017.1353584 tindale a, klocker n and gibson c (2014) inter-ethnic partnerships: remaking urban ethnic diversity. urban policy and research 32(4): 393-415. https://doi.org/10.1080/08111146.2014.883967 https://doi.org/10.1007/978-3-030-22874-3_3 https://doi.org/10.1177/02632764070240072505 https://doi.org/10.1007/978-3-030-22874-3_32 https://doi.org/10.1080/07256868.2015.1119674 https://doi.org/10.1007/978-3-030-22874-3_5 https://doi.org/10.1007/978-3-030-22874-3_4 https://doi.org/10.1037/0022-3514.90.5.751 https://doi.org/10.1080/01419870.2018.1410200 https://doi.org/10.1080/00049182.2017.1353584 https://doi.org/10.1080/08111146.2014.883967 australian population studies 2022 | volume 6 | issue 1 | pages 41-44 © loginova & dwyer 2022. published under the creative commons attribution-noncommercial licence 3.0 australia (cc by-nc 3.0 au). journal website: www.australianpopulationstudies.org demographic indicators and smallscale solar energy technology uptake in brisbane, australia: an interactive visualisation julia loginova* the university of queensland declan dwyer the university of queensland * corresponding author. email: j.loginova@uq.edu.au. address: centre for social responsibility in mining, sustainable minerals institute & queensland centre for population research, school of earth and environmental sciences, the university of queensland, brisbane, qld 4072, australia paper received 10 march 2022; accepted 21 april 2022; published 25 july 2022 introduction as interest has increased in global low-carbon energy transitions, so too has interest in how population processes such as population ageing or urbanisation affect carbon emissions and renewable energy adoption at the local level (balta-ozkan et al. 2021; liddle 2014). local governments and communities around the world are particularly interested to understand the spatial disparities in renewable energy adoption and the complexity of demographic and socioeconomic factors behind that. the city of brisbane has the potential to be a small-scale solar photovoltaic (spv) powerhouse in australia. brisbane has the highest spv megawatt output potential (nationalmap 2021) and the highest solar radiation exposures of all australian state capitals (sommerfeld et al. 2017). increasing spv adoption has the potential to transform collective energy use and directly decrease queensland’s greenhouse gas emissions. though with more than 800 megawatts of spv installed (as of 2021), there is a high spatial variability of spv uptake among city neighbourhoods. researchers have begun to evaluate the impact of demographic and socio-economic factors on spv adoption variability across australia (lan et al. 2021). creating interactive and publicly available visualisations of such interrelationships is an important step to inform future equitable energy policies and planning. publicly available platforms offer a means to provide access to a wide range of users including decision-makers and urban residents. integration of multiple data sources provides an opportunity to reveal new insights that otherwise would remain elusive. finally, interactive maps provide an effective tool for exploring small area data and patterns of relations between indicators. the aim of this demographic is to present an interactive map to visually explore the interplay between selected demographic indicators and spv uptake across postcodes in the brisbane local government area (lga). d e m o g ra p h ic http://www.australianpopulationstudies.org/ mailto:j.loginova@uq.edu.au 42 loginova & dwyer australian population studies 6 (1) 2022 data and methods a set of key indicators was identified during a review of literature that addresses the impact of demographic and socioeconomic factors on the uptake of spv (lan et al. 2021; liddle 2014; sommerfeld et al. 2017). indicators were selected if they showed a positive relationship with the spv uptake and were broken into two groups. group i indicators consist of level of education (proportion of the population within a postcode which has obtained a tertiary qualification), household income (proportion of households within a postcode which meets or exceeds the national average weekly household income), median age (years old), and proportion of detached dwellings (proportion of total dwellings considered detached dwellings within a postcode) in each postcode. demographic data was sourced from the australian bureau of statistics (abs) at the level of postal areas which are abs’s approximation of postcodes. group ii represents spv adoption (small-scale solar installation totals and total aggregated kw output capacity per postcode) derived from publicly available postcode level data from the australian government’s clean energy regulator (2021). only postcodes (n=76) which fall within the brisbane lga boundary were included. the data was incorporated into the interactive map using r software (‘flexdashboard’ and ‘tmap’ packages) hosted on the github pages public repository (see loginova and wohland 2020 for instructions). the postal areas shapefile was downloaded from the abs (2021) australian statistical geography standard website. the map can display indicators individually as well as in combination. group i (demographic choropleth maps) and group ii (proportional symbols reflecting spv adoption) indicators can be overlayed to visualise the geographical interplay. the interactive map can be explored using this link: https://qcpr.github.io/brisbanespv/. key features there is a clear spatial pattern to spv uptake in brisbane. postcodes in inner city as well as outer suburbs have low spv uptake, while the highest spv uptake can be observed in inner suburbs. the highest values (above 6,000 small-scale spv installations and sgu rated output higher than 25,000 kw) are found in six inner suburbs, with three postcodes on the southside (4109, 4122, 4152) and three on the northside (4034, 4017, 4053) of the river (figure 1). areas of highest population densities (inner city) have some of the lowest spv uptake. individual indicators showed varying strengths of relationship to spv uptake. neither of age, income, education, and dwelling indicators demonstrate a clear pattern of spv uptake. currently, based on the visual representation of the interactive map, brisbane fits into a theory increasingly recognised in the literature which argue that ‘middle’ is best. areas most conducive to spv uptake consist of a demographic makeup where middle income, middle housing density, and middle age ranges are most prevalent (bondio et al. 2018; best and trück 2020; lan et al. 2021). according to the study by bondio et al. (2018) that surveyed more than 8,000 households in queensland, spv adopters are more likely to be middle-class households. in the ‘middle’ suburbs, spv offers the greatest incentives with enough flexibility to hurdle some of the practical barriers to spv such as upfront costs or roof space. practical and socioeconomic barriers to spv uptake appear to be greatest at the lowest (high capital cost barriers) and highest (low appeal beyond the benefits as an electricity cost-saving https://qcpr.github.io/brisbanespv/ australian population studies 6 (1) 2022 loginova & dwyer 43 measure) ends of most demographic indicators (bondio et al. 2018). postcodes at neither ends of the spectrum, seemingly provide a demographic context whereby the motivations and means of adopting spv result in higher installation totals. therefore, a more nuanced understanding of specific population dynamics and settlement patterns is important for more efficient and equitable policy measures and planning that target adoption of the renewable technologies. there are a range of limitations associated with this demographic that require noting. first, it reports on four demographic and socioeconomic indicators, recognising that the effect of various population and non-population processes is inherently more complex. inclusion of other indicators, such as political preferences or installation costs, would further enrich the understanding of spatial disparities in the adoption of renewable technologies. it is also recognised that demographics and a specific indicator’s level of influence changes over time (sommerfeld et al. 2017). the interactive map format has the flexibility to keep up with these changes. regular updates or additions to the data can maintain a current representation of the interrelationship between demographics and spv adoption. figure 1: geographical interplay between median age and spv output, kw. source: authors’ dashboard at https://qcpr.github.io/brisbanespv/ https://qcpr.github.io/brisbanespv/ 44 loginova & dwyer australian population studies 6 (1) 2022 acknowledgements the authors acknowledge zero emissions brisbane for their valuable guidance and comments on the early design of the interactive map. references abs (2021). australian statistical geography standard. https://www.abs.gov.au/statistics/standards/australian-statistical-geography-standard-asgsedition-3/jul2021-jun2026 australian clean energy regulator (2021). post code data for small-scale installations. http://cleanenergyregulator.gov.au/ret/forms-and-resources/postcode-data-for-small-scaleinstallations balta-ozkan, n., yildirim, j., connor, p. m., truckell, i., & hart, p. (2021). energy transition at local level: analyzing the role of peer effects and socio-economic factors on uk solar photovoltaic deployment. energy policy, 148. https://doi.org/10.1016/j.enpol.2020.112004 best, r., & trück, s. (2020). capital and policy impacts on australian small-scale solar installations. energy policy, 136, 111082. https://doi.org/10.1016/j.enpol.2019.111082 bondio, s., shahnazari, m., & mchugh, a. (2018). the technology of the middle class: understanding the fulfilment of adoption intentions in queensland's rapid uptake residential solar photovoltaics market. renewable & sustainable energy reviews, 93, 642–651. https://doi.org/10.1016/j.rser.2018.05.035 lan, h., gou, z., & liu, t. (2021). residential solar panel adoption in australia: spatial distribution and socioeconomic factors. australian geographer, 52(3), 315-332. https://doi.org/10.1080/00049182.2021.1964161 liddle, b. (2014). impact of population, age structure, and urbanization on carbon emissions/energy consumption: evidence from macro-level, cross-country analyses. population and environment, 35(3), 286-304. https://doi.org/10.1007/s11111-013-0198-4 loginova, j., & wohland, p. (2020). how to create an interactive dashboard using r: the example of the queensland covid-19 tracker. australian population studies, 4(2), 39-47. https://doi.org/10.37970/aps.v4i2.72 nationalmap (2021). the australian national map. renewable energy arena projects. https://nationalmap.gov.au/ sommerfeld, j., buys, l., mengersen, k., & vine, d. (2017). influence of demographic variables on uptake of domestic solar photovoltaic technology. renewable & sustainable energy reviews, 67, 315323. https://doi.org/10.1016/j.rser.2016.09.009 https://www.abs.gov.au/statistics/standards/australian-statistical-geography-standard-asgs-edition-3/jul2021-jun2026 https://www.abs.gov.au/statistics/standards/australian-statistical-geography-standard-asgs-edition-3/jul2021-jun2026 https://doi.org/10.1016/j.enpol.2020.112004 https://doi.org/10.1016/j.enpol.2019.111082 https://doi.org/10.1016/j.rser.2018.05.035 https://doi.org/10.1080/00049182.2021.1964161 https://doi.org/10.1007/s11111-013-0198-4 https://doi.org/10.37970/aps.v4i2.72 https://nationalmap.gov.au/ https://doi.org/10.1016/j.rser.2016.09.009 australian population studies 2021 | volume 5 | issue 1 | pages 40-48 © lyons et al. 2021. published under the creative commons attribution-noncommercial licence 3.0 australia (cc by-nc 3.0 au). journal website: www.australianpopulationstudies.org commentary counting gender and sexual identity in the australian census anthony lyons* la trobe university mary lou rasmussen australian national university joel anderson la trobe university and australian catholic university edith gray australian national university * corresponding author. email: a.lyons@latrobe.edu.au. address: australian research centre in sex, health and society, school of psychology and public health, building nr6, la trobe university, bundoora, victoria, 3086, australia. paper received 12 february 2021; accepted 27 april 2021; published 31 may 2021 1. introduction the australian census of population and housing (hereafter referred to as the ‘census’), serves as an important barometer of the nation. in this article, we outline why it is so important to include comprehensive gender and sexual identity questions in the census as well as several considerations that ought to be addressed in the course of including such questions. the census provides a count for electoral representation, informs government and non-government policies and services, and ultimately reveals who we are as a nation, how the population changes over time, and how the diversity of australia’s relatively small population is spread across a vast landmass. it also serves as an important resource for numerous other surveys by providing a complete count of the population to inform sampling and weighting of data and thus enabling population representative data gathering. for these and other reasons, it is crucial that the census captures as much of the diversity in australia’s population as possible. without this knowledge, important subpopulations are rendered invisible and initiatives to collect representative data of these populations in other surveys are severely limited, if not impossible. australia’s gender and sexually diverse population is one such population that is not fully counted. 2. including gender and sexual identity in the census 2.1. what we mean by gender and sexual identity questions surveys conducted in australia reveal a large diversity in gender and sexual identity (hill et al. 2020; leonard et al. 2012). although representative data are lacking, estimates of the size of australia’s gender and sexually diverse population broadly range from 3.5% to 11%, or 900,000 to 2.8 million (department of health 2012; wilson et al. 2020; zhang et al. 2020). this includes people who identify with one or more diverse sexual identities, such as lesbian, gay, bisexual, pansexual, or queer. it also includes people who are trans and gender diverse, including non-binary genders. despite this, population surveys, including the census, rarely involve comprehensive questions on gender and http://www.australianpopulationstudies.org/ mailto:a.lyons@latrobe.edu.au australian population studies 5 (1) 2021 lyons et al. 41 sexual identity that would enable identification of the many different gender and sexual identity populations living in australia (lyons et al. 2020). in simple terms, we refer to gender and sexual identity questions as demographic variables. in other words, these are questions where people have the opportunity to report their gender, such as woman, man, trans woman, trans man, or non-binary, and separately from this, their sexual identity, such as lesbian, gay, bisexual, pansexual, or queer. gender and sexual identities comprise significant elements of a person’s background that define and shape many aspects of their lives, such as raising families, employment, and engagement with services (hill et al. 2020). these identities are important aspects of australia’s diversity that need to be captured in population surveys. by gender, we refer to how people identify or describe themselves. this is an important distinction, as the terms ‘sex’ and ‘gender’ are typically conceptualised in different ways despite often being treated as though they were synonymous (e.g., connell 2002; hammarström and annandale 2012). as an example, the australian bureau of statistics (abs) recently published a standard for sex, gender, variations of sex characteristics and sexual orientation variables (abs 2021). this standard is aligned with guidelines from the australian government attorney-general’s department (australian government 2013) and outlines sex and gender in distinct ways. the standard utilises the term ‘sex’ to refer to biological characteristics. this is most commonly assigned to an individual at birth, usually male or female, or may be recorded differently in some instances when a variation in their sex characteristics is evident. in contrast, the term ‘gender’ is used to refer to how a person describes themselves, which may not necessarily align with the sex assigned to them at birth. for instance, a person who is trans may have been assigned female at birth but describe themselves as, and importantly are, a man. as shown by the recent private lives 3 national survey, many people in australia also describe themselves in different ways, such as neither exclusively male nor female (e.g., non-binary; hill et al. 2020). the term, ‘sexual identity’, typically refers to how a person experiences or expresses themselves sexually or romantically (durso and gates 2013), which includes diverse identity labels such as lesbian, gay, bisexual, pansexual, heterosexual, and queer. the term ‘sexual orientation’ is usually used as a broader application to refer to different components of sexuality, which can include sexual identity, sexual behaviour, and sexual attraction (katz-wise 2015). for example, people who engage in sex with someone of the same sex might describe themselves in different ways, such as lesbian, gay or bisexual. while different components of sexual orientation are important to consider, sexual identity is the more common focus. though depending on the particular focus of a survey (e.g., a survey of sexual behaviour), researchers might need to know about identity, behaviour, and attraction. it is worth noting that this account of conceptual distinctions is somewhat simplified for the purposes of this article. extensive theoretical and conceptual work have been conducted over many years and decades. for example, arguments have been made to suggest that gender and sexual identity are not always inseparable (butler 2006). however, for the purposes of this article, a main point to note is that ‘sex’, ‘gender’, and ‘sexual identity’ are terms that have particular shared meanings and distinctions, which need to inform the framing of gender and sexual identity questions. 42 lyons et al. australian population studies 5 (1) 2021 2.2. why it is essential these data are collected in australia australia’s diverse gender and sexual identity population continues to experience stigma-related harm (hill et al. 2020) and faces numerous significant social and health-related burdens such as physical and mental health challenges (perales 2019), suicidality (raifman et al. 2020), family-related challenges (perales et al. 2019), discrimination and victimisation (hill et al. 2020), homelessness (hill et al. 2020; hill et al. 2021), poverty (hill et al. 2020), and barriers to accessing services (alba et al. 2020). critical differences are also apparent between subpopulations in the extent to which particular burdens are experienced, such as differences between those who are cisgender and trans or bisexual and pansexual (hill et al. 2020). gathering comprehensive population-level data is therefore crucial. it is only with such data that policies and services can be designed and targeted effectively to areas of need. data collection also needs to be repeated regularly to enable monitoring of the progress of initiatives and to thus allow necessary adjustments to be made in policy settings and service delivery. significant social change has occurred over recent decades, with associated shifts in understandings and expressions of gender and sexuality. as mentioned earlier, gender and sexuality are integral aspects of people’s lives and it is not possible to have a truly informative picture of the nation without having data that reflects this diversity. recognising the significant implications to social, health, and economic outcomes, australia’s gender and sexually diverse population is identified as a key priority group in government strategies around the country. examples include the federal government’s aged care diversity framework, the forthcoming victorian lgbtiq+ strategy, western australia’s wa lgbti health strategy and the australian capital territory’s capital of equality strategy. however, effectively operationalising and monitoring the progress of these and other strategies and policies depend on the continued availability of comprehensive data of gender and sexual identity. in addition to ensuring the knowledge required to address significant national challenges, gathering comprehensive data on gender and sexual identity via population research is fundamental to understanding the demography of the nation. 2.3. why these data should be collected in the australian census the australian census is vital to providing essential information about australia’s population. the data collected in the census is used by a wide range of individuals and organisations in public, private and not-for-profit sectors to make informed planning and policy decisions for all australians. a major strength of the census is its coverage: it includes the whole population of australia, thereby providing the data required to cover small geographic areas and population groups that may be otherwise missed, or misrepresented, in smaller surveys. while a broad range of information is collected about people that is useful for identifying communities, information that would enable a comprehensive count of gender and sexually diverse populations is currently not collected. in 2018, it was announced that eight new topics would be tested for inclusion in the 2021 census. these eight topics included ‘non-binary sex and/or gender identity’ and ‘sexual orientation’ (abs 2018). these topics were developed following extensive public consultation but were later dropped from testing. despite this, the public consultation process identified the value of collecting comprehensive information on gender and sexual identity in the census. the process found that gender and sexually diverse communities often have higher levels of vulnerability including experiences of discrimination, australian population studies 5 (1) 2021 lyons et al. 43 abuse and mental health issues that require targeted support, and that information collected via the census would help to inform the development of policy, allocation of resources and delivery of services to people from these communities (abs 2020). despite the decision not to proceed with questions in the next census, as mentioned earlier the abs has since released a new standard for sex, gender, variations of sex characteristics and sexual orientation variables (abs 2021), which was developed through research, consultation, and testing, and could potentially serve as a foundation or starting point for the future inclusion of questions in the census. while specific questions for the census would need to be identified, not having these data collected in the census renders these communities invisible in national statistics. service providers are left in the unenviable position of having to make educated guesses about where to best put their resources. while some information can be gleaned from survey data, there are no benchmarks which can be used to create sample weights for these surveys. as mentioned earlier, census data are required for population estimates that reflect australia’s gender and sexual identities. this would enable population representative surveys to be conducted in areas not directly examined by the census, such as surveys of health, in addition to providing valuable knowledge on topics directly examined in the census, such as housing, income, employment, and family circumstances. 2.4. potential issues to consider in asking comprehensive questions of sex, gender, and sexual identity there are at least several considerations that need to be taken into account in identifying census questions on gender and sexual identity. a small but important literature has empirically analysed the consequences for not asking questions about sex, gender and sexual identity correctly. the literature on this topic has debated both strengths and limitations in the specific framing of questions, including the associated response options (lyons et al. 2020; sullivan 2020; waite and denier 2019). how these questions are framed might appear inconsequential, but subtle differences can result in different interpretations, and thus potentially different responses, by respondents. for instance, prescriptive terminology in questions on sexual identity (e.g., ‘are you…’ rather than ‘how do you identify…’) might result in respondents feeling obliged to select from identities that are more commonly used and understood, rather than allowing themselves to self-identify more freely. relatedly, it is important to highlight that how questions are asked by survey designers can frame the social categories that are valued and recognised in society. more specifically, it is argued that how questions are framed produces a ‘social reality’ (law et al. 2011), where a population is defined or generated by the question and its response options. accordingly, it is the responsibility of survey designers to think carefully about how they craft gender and sexual identity questions, the response options they choose to provide, and how they explain to survey respondents what is being asked of them. for example, asking a question where respondents indicate their sex might conceivably be answered in differing ways, such as the sex assigned to them at birth, their current sex (which may be different to that assigned to them at birth, such as those with a trans history), or their gender. in the absence of clear framing and explanation, respondents may be left to subjectively interpret these questions. reflecting current best practice, the abs has recommended in its latest standard for sex, gender, variations of sex characteristics and sexual orientation to apply specific question wording, such as “what was your sex recorded at birth?” and potentially asking respondents to indicate their gender in a separate question (abs, 2021). 44 lyons et al. australian population studies 5 (1) 2021 as noted elsewhere (e.g., lyons et al. 2020), there is debate about the advantages and disadvantages of allowing participants to respond with open ended questions, about the use of ‘other’ categories, and about singleversus multi-select answers. it is important to recognise that, contrary to oftenheld beliefs, gender and sexual identity are neither simple nor synonymous constructs, and that often individuals need to be able to select combinations of identities (for further discussion, see lyons et al. 2020; ruberg and ruelos 2020). although this may present some challenges when designing questions, if a question cannot be answered in a way that reflects a person’s background, then the respondent is left with the choice of either not answering the question (and thus not being counted) or answering in a way that does not reflect them (thus undermining the integrity of the data). therefore, it is not only important to include gender and sexual identity questions that better reflect australia’s diverse population, but also give thought to how these questions are designed and the conclusions that might feasibly be drawn from the resulting data. engaging in consultation with different communities and stakeholders as well as researching and testing approaches are therefore important initiatives to undertake prior to settling on an optimal design of questions that captures diversity as well as being understood by respondents. this could also include developing and testing approaches to ensure that respondents feel comfortable in answering the questions and thus minimising non-responses. it will also be important to review approaches over time to ensure that questions included in the census remain relevant to any future changes in sexual identity and gender categories. 2.5. are other countries doing work to gather similar information? statistics new zealand, in writing about their decision to incorporate more questions on sex, gender identity and sexual orientation in their next census, note that data that captures the diversity of this population in terms of sex, gender identity, and sexual orientation has been very limited around the world. in new zealand, the government has argued for the “‘gender by default’ principle … an approach that defaults to the collection of gender data as opposed to sex at birth…. collection of sex at birth information should be viewed as an exception where there is a specific need” (statistics new zealand 2020). the new zealand government is making this change because in “most cases a person’s gender – their social and personal identity – is most relevant for policy making and research rather than their sex at birth” (statistics new zealand 2020). the new zealand government has also specified that questions, similar to those relating to ethnicity, are suitable for all people, regardless of their age. these issues are being treated quite differently in the united kingdom. in december 2018, the office for national statistics (ons), which is responsible for the census in england and wales, set out its recommendations for the next census in the white paper 'help shape our future: the 2021 census of population and housing in england and wales' (ons 2018). this white paper gives the recommendation that the census includes, for the first time, a voluntary gender identity question, in addition to retaining a binary female/male sex question. the ons also recommended introducing a new question enquiring about an individual’s sexual orientation, where they are aged 16 years and over. in this proposal, the sex question would remain unchanged and only people aged 16 years and over would have the opportunity to voluntarily complete a question related to their gender identity and sexual orientation. in 2022, the scottish government will also ask everyone aged 16 years and over to voluntarily disclose their sexual orientation. respondents will be presented the opportunity australian population studies 5 (1) 2021 lyons et al. 45 to identify themselves as ‘straight/heterosexual’, ‘gay or lesbian’, ‘bisexual’ or ‘other sexual orientation, please write in…’. in discussing parliamentary debates about the design of the scottish census, kevin guyan argues that the decision to include a non-binary sex question “highlighted the existence of two schools of thought: one understood the census as an exercise in accurate, data collection; the other, understood the census as a method to construct and/or protect governable identity categories” (guyan 2021 p. 6). the contested politics relating to how a census constructs governable categories is apparent when one looks at how decisions are made about how old a person can be before they can declare their gender identity. the age at which a person can legally obtain gender recognition internationally varies from, for example, no age limit in the australian capital territory, british columbia, and argentina, to 18 years in new zealand. governments are clearly constructing governable identity categories when they are explicitly excluding young people from filling out the questions related to gender identity and sexual orientation. guyan further notes that “any minor errors in the [census] count will likely have a disproportionately large effect on findings” (guyan 2021, p. 8) because lgbtq populations are already a comparatively small percentage of the whole population. consequently, the decision to determine participation by age may be one of the ‘minor errors’ that has explicit effects in determining the way in which these populations are counted in scotland and elsewhere. 2.6. how australia might learn from others to address potential issues in collecting data related to sex, gender, and sexual identity statistics new zealand are endeavouring to be as capacious as possible in counting sex, gender and sexuality diversity within their population. this is evidenced by their inviting respondents of any age to answer sex and gender questions, incorporating maori translations to describe gender and sex classifications, as well as moving away from terms like gender diverse toward the term “another gender”. they are also restricting use of the sex at birth question, following in the footsteps of statistics canada, arguing that this type of specific information is only required in the limited circumstances where there is a need to disaggregate cisgender and trans populations. statistics new zealand are also flagging that the current consultations on the census will need to be ongoing because the ways in which people understand sex, gender and sexual identity are constantly changing, citing the example of the term non-binary, which was little used at the time of the last census, but is now regularly deployed by community members (veale et al. 2019). for australian purposes, there needs to be similar recognition that issues related to counting sexual and gender identity will likely not be resolved in one attempt, but will require ongoing attention, in line with changes in community norms in this space. 3. concluding comments as other jurisdictions prepare to improve how they count gender and sexual identity in their censuses, such as new zealand and scotland, a strong case exists for australia to do similarly and an opportunity exists to learn from the experiences of these countries. while continued work may be required to identify and adapt questions over time, it is an endeavour that has many important benefits. a more comprehensive account of the diversity of gender and sexual identity in australia 46 lyons et al. australian population studies 5 (1) 2021 would not only provide the foundation for population representative research involving gender and sexually diverse populations, it would enable a more thorough and contemporary assessment of the nation’s demography. these are both critical to advancing the knowledge required to address significant challenges faced by australia’s gender and sexually diverse population and thereby improve overall national social, health and economic outcomes. we have argued that it is important to take a range of considerations into account in formulating gender and sexual identity questions, such as the choice of response options and the wording of questions, as well as remaining cognisant that how these are framed will shape the data, including how specific sub-populations are understood. the recently released abs standard for sex, gender, variations of sex characteristics and sexual orientation variables may serve as a helpful basis for further investigation and community consultation. this will be integral to identifying questions that satisfy the overall requirements of the census while providing an important and long overdue count of australia’s diversity in relation to sex, gender and sexual identity. it is also important to consider whether questions related to sex, gender and sexual identity will require amendment at each census in order to keep pace with the rapid shifts occurring in australia in this domain. references abs (2018) abs tests topics for 2021 census to reflect changing nation. media release. canberra: australian bureau of statistics. https://www.abs.gov.au/ausstats/abs@.nsf/lookup/by%20subject/2007.0.55.001~2021~media %20release~abs%20tests%20topics%20for%202021%20census%20%20(media%20release)~100 00. accessed on 3 february 2021. abs (2020) 2021 census topics and data release plan. canberra: australian bureau of statistics. https://www.abs.gov.au/statistics/research/2021-census-topics-and-data-release-plan#topicsnot-included-in-the-2021-census. accessed on 3 february 2021. abs (2021) standard for sex, gender, variations of sex characteristics and sexual orientation variables. canberra: australian bureau of statistics. https://www.abs.gov.au/statistics/standards/standardsex-gender-variations-sex-characteristics-and-sexual-orientation-variables/latest-release. accessed on 2 february 2021. alba b, lyons a, waling a, minichiello v, hughes m, barrett c, fredriksen-goldsen k and edmonds s (2020) older lesbian and gay adults’ perceptions of barriers and facilitators to accessing health and aged care services in australia. health and social care in the community. https://doi.org/10.1111/hsc.13125 australian government (2013) australian government guidelines for the recognition of sex and gender. canberra: australian government attorney general’s department. https://www.ag.gov.au/sites/default/files/202003/australiangovernmentguidelinesontherecognitionofsexandgender.pdf. accessed on 1 february 2021. butler j (2006) gender trouble. new york: routledge. byron p, robards b, hanckel b, vivienne s and churchill b (2019) “hey, i’m having these experiences”: tumblr use and young people’s queer (dis) connections. international journal of communication 13: 2239-2259. https://ijoc.org/index.php/ijoc/article/view/9677 connell r w (2002) gender. cambridge: polity. department of health (2012) national lgbti ageing and aged care strategy. canberra: australian government. https://www.abs.gov.au/ausstats/abs@.nsf/lookup/by%20subject/2007.0.55.001~2021~media%20release~abs%20tests%20topics%20for%202021%20census%20%20(media%20release)~10000 https://www.abs.gov.au/ausstats/abs@.nsf/lookup/by%20subject/2007.0.55.001~2021~media%20release~abs%20tests%20topics%20for%202021%20census%20%20(media%20release)~10000 https://www.abs.gov.au/ausstats/abs@.nsf/lookup/by%20subject/2007.0.55.001~2021~media%20release~abs%20tests%20topics%20for%202021%20census%20%20(media%20release)~10000 https://www.abs.gov.au/statistics/research/2021-census-topics-and-data-release-plan#topics-not-included-in-the-2021-census https://www.abs.gov.au/statistics/research/2021-census-topics-and-data-release-plan#topics-not-included-in-the-2021-census https://www.abs.gov.au/statistics/standards/standard-sex-gender-variations-sex-characteristics-and-sexual-orientation-variables/latest-release https://www.abs.gov.au/statistics/standards/standard-sex-gender-variations-sex-characteristics-and-sexual-orientation-variables/latest-release https://doi.org/10.1111/hsc.13125 https://www.ag.gov.au/sites/default/files/2020-03/australiangovernmentguidelinesontherecognitionofsexandgender.pdf https://www.ag.gov.au/sites/default/files/2020-03/australiangovernmentguidelinesontherecognitionofsexandgender.pdf https://ijoc.org/index.php/ijoc/article/view/9677 australian population studies 5 (1) 2021 lyons et al. 47 durso l e and gates g j (2013) best practices: collecting and analysing data on sexual minorities. in a baumle (ed), international handbook on the demography of sexuality. dordrecht: springer. guyan k (2021) constructing a queer population? asking about sexual orientation in scotland’s 2022 census. journal of gender studies. https://doi.org/10.1080/09589236.2020.1866513 hammarström a and annandale e (2012) a conceptual muddle: an empirical analysis of the use of ‘sex’ and ‘gender’ in ‘gender-specific medicine’ journals. plos one 7: e34193. https://doi.org/10.1371/journal.pone.0034193 hill a o, bourne a, mcnair r, carman m and lyons a (2020) private lives 3: the health and wellbeing of lgbtiq people in australia. melbourne: la trobe university. https://www.latrobe.edu.au/arcshs/publications/private-lives/private-lives-3 hill a o, lyons a, jones j, mcgowan i, carman m, parsons m, power j and bourne a (2021) writing themselves in 4: the health and wellbeing of lgbtqa+ young people in australia. melbourne: la trobe university. https://www.latrobe.edu.au/arcshs/publications/writing-themselves-inpublications/writing-themselves-in-4 katz-wise s l (2015) sexual fluidity in young adult women and men: associations with sexual orientation and sexual identity. psychology and sexuality 6(2): 189-208. https://doi.org/10.1080/19419899.2013.876445 law j, ruppert e and savage m (2011) the double social life of methods (cresc working paper series, no. 95). milton keynes: open university. leonard w, pitts m, mitchell a, lyons a, smith a, patel s, couch m and barrett a (2012) private lives 2: the second national survey of the health and wellbeing of gay, lesbian, bisexual and transgender (glbt) australians. melbourne: la trobe university. https://www.latrobe.edu.au/arcshs/publications/private-lives lyons a, anderson j r, rasmussen m l and gray e (2020) toward making sexual and gender diverse populations count in australia. australian population studies 4(2): 14–29. https://doi.org/10.37970/aps.v4i2.69 ons (2018). help shape our future: the 2021 census of population and housing in england and wales. london, england: office of national statistics. https://assets.publishing.service.gov.uk/government/uploads/system/uploads/attachment_data/ file/765089/census2021whitepaper.pdf. accessed on 2 february 2021. perales f (2019) the health and wellbeing of australian lesbian, gay and bisexual people: a systematic assessment using a longitudinal national sample. australian and new zealand journal of public health 4: 281-287. https://doi.org/10.1111/1753-6405.12855 perales f, simpson reeves l, plage s and baxter j (2019) the family lives of australian lesbian, gay and bisexual people: a review of the literature and a research agenda. sexuality research and social policy 17: 43-60. https://doi.org/10.1007/s13178-018-0367-4 raifman j, charlton b m, arrington-sanders r, chan p a, rusley j, mayer k h, stein m d, austin b and mcconnell m (2020) sexual orientation and suicide attempt disparities among us adolescents: 2009-2017. pediatrics 145: e20191658. https://doi.org/10.1542/peds.2019-1658 ruberg b and ruelos s (2020) data for queer lives: how lgbtq gender and sexuality identities challenge norms of demographics. big data and society 7: 2053951720933286. https://doi.org/10.1177/2053951720933286 stats nz (2020) sex and gender identity statistical standards: consultation. wellington, new zealand: stats nz. https://www.stats.govt.nz/consultations/sex-and-gender-identity-statistical-standardsconsultation. accessed on 2 february 2021. sullivan a (2020) sex and the census: why surveys should not conflate sex and gender identity. international journal of social research methodology 23(5): 517-524. https://doi.org/10.1080/13645579.2020.1768346 https://doi.org/10.1080/09589236.2020.1866513 https://doi.org/10.1371/journal.pone.0034193 https://www.latrobe.edu.au/arcshs/publications/private-lives/private-lives-3 https://www.latrobe.edu.au/arcshs/publications/writing-themselves-in-publications/writing-themselves-in-4 https://www.latrobe.edu.au/arcshs/publications/writing-themselves-in-publications/writing-themselves-in-4 https://doi.org/10.1080/19419899.2013.876445 https://www.latrobe.edu.au/arcshs/publications/private-lives https://doi.org/10.37970/aps.v4i2.69 https://assets.publishing.service.gov.uk/government/uploads/system/uploads/attachment_data/file/765089/census2021whitepaper.pdf https://assets.publishing.service.gov.uk/government/uploads/system/uploads/attachment_data/file/765089/census2021whitepaper.pdf https://doi.org/10.1111/1753-6405.12855 https://doi.org/10.1007/s13178-018-0367-4 https://doi.org/10.1542/peds.2019-1658 https://doi.org/10.1177/2053951720933286 https://www.stats.govt.nz/consultations/sex-and-gender-identity-statistical-standards-consultation https://www.stats.govt.nz/consultations/sex-and-gender-identity-statistical-standards-consultation https://doi.org/10.1080/13645579.2020.1768346 48 lyons et al. australian population studies 5 (1) 2021 veale j, byrne j, tan k, guy s, yee a, nopera t and bentham, r. (2019). counting ourselves: the health and wellbeing of trans and non-binary people in aotearoa new zealand. hamilton, new zealand: transgender health research lab, university of waikato. https://www.countingourselves.nz/wpcontent/uploads/2020/01/counting-ourselves_report-dec-19-online.pdf. accessed 2 february 2021. waite s and denier n (2019) a research note on canada’s lgbt data landscape: where we are and what the future holds. canadian review of sociology 56(1): 93-117. https://doi.org/10.1111/cars.12232 wilson t, temple j, lyons a and shalley f (2020) what is the size of australia’s sexual minority population? bmc research notes 13: 535. https://doi.org/10.1186/s13104-020-05383-w zhang q, goodman m, adams n, corneil t, hashemi l, kreukels b, motmans j, snyder r and coleman e (2020) epidemiological considerations in transgender health: a systematic review with focus on higher quality data. international journal of transgender health 21(2): 125-137. https://doi.org/10.1080/26895269.2020.1753136 https://www.countingourselves.nz/wp-content/uploads/2020/01/counting-ourselves_report-dec-19-online.pdf https://www.countingourselves.nz/wp-content/uploads/2020/01/counting-ourselves_report-dec-19-online.pdf https://doi.org/10.1111/cars.12232 https://doi.org/10.1186/s13104-020-05383-w https://doi.org/10.1080/26895269.2020.1753136 australian population studies 2020 | volume 4 | issue 2 | pages 14-29 © lyons, anderson, rasmussen & gray 2020. published under the creative commons attribution-noncommercial licence 3.0 australia (cc by-nc 3.0 au). journal website: www.australianpopulationstudies.org toward making sexual and gender diverse populations count in australia anthony lyons* la trobe university joel anderson la trobe university; australian catholic university mary lou rasmussen australian national university edith gray australian national university * corresponding author. email: a.lyons@latrobe.edu.au. address: australian research centre in sex, health and society, school of psychology and public health, building nr6, la trobe university, bundoora, victoria, 3086, australia. paper received 30 july 2020; accepted 8 october 2020; published 16 november 2020 abstract background comprehensive data on gender and sexual identity is critical for the planning and delivery of health, education, and social support services. this paper examines ways in which sexual and gender diverse populations are being counted in research, with a view to informing discussions about how to represent these populations in future research. aims to examine approaches used for the collection of data from sexual and gender diverse populations in australia. data and methods we reviewed nine examples of large national surveys conducted in australia over the past ten years and compared the approaches used for collecting data on gender and sexual identity. results a diversity of approaches and a range of limitations were identified in how these diverse populations are counted. the proportions of survey respondents across sex, gender and sexual identity categories, and the types of categories, were also found to vary across studies. conclusions there is currently no consistent approach for collecting data involving sexual and gender diverse populations in australia despite the need for large-scale surveys that reflect sexual and gender diversity. this paper identifies conceptual and methodological questions for consideration when planning how to capture diversity related to gender and sexual identity. key words survey methods; sex; gender; sexual identity; australia http://www.australianpopulationstudies.org/ mailto:a.lyons@latrobe.edu.au australian population studies 4 (2) 2020 lyons et al. 15 1. introduction “there’s a saying: ‘you don’t count in policy circles until someone counts you.’” dr gary gates comprehensive data of sexual and gender diverse populations in australia is crucial for service planning and policymaking. these populations, such as people who identify as lesbian, gay, bisexual, or trans, experience many disproportionate health and social issues (lyons et al. 2019; national lgbti health alliance 2020). owing to complex histories of stigma and discrimination (anderson and holland 2015; mizock and mueser 2014), targeted and tailored provision of support is often required. effectively providing this depends on having comprehensive demographic and other data, such as age, geographic, and socioeconomic distributions. the australian census of population and housing (hereafter ‘the census’) is australia’s largest statistical collection, which is used for numerous purposes, such as public funding and decision-making for services and infrastructure (abs 2017). however, the census is currently limited in the extent to which it collects data on the diversity of gender and sexual identity. the australian bureau of statistics (abs) has reported it would not be including sexual and gender identity questions in the next census, due for 2021 (abs 2019). until questions are included in the census, researchers and policy makers must rely on what is available from other studies to ensure that sexual and gender diverse populations are counted and visible. an essential consideration is whether surveys ask about sexual and gender diversity in consistent and thorough ways, and in particular whether questions and response options provide capacity to build a detailed demographic profile of sexual and gender diverse populations in australia. it is important to first note that sex and gender are typically conceptualised in different ways (e.g. connell 2002; hammarström and annandale 2012; west and zimmerman 1987). for example, the abs released a ‘standard for sex and gender variables’ (abs 2016), which aligns with australian government guidelines from the attorney-general’s department (australian government 2013) and outlines sex and gender as distinct concepts. in this standard, ‘sex’ refers to biological characteristics. it is usually assigned at birth as male or female, or may be recorded differently where a person has a variation in sex characteristics. in the standard ‘gender’ refers to how a person identifies or describes themselves, which may not always align with their sex assigned at birth. for example, someone who is transgender may have been assigned male on their birth certificate but describe themselves as, and are, a woman. importantly, people might also describe themselves in different ways that are neither exclusively male nor female, such as non-binary (liszewski et al. 2018). the uk office for national statistics (ons) recently released a similar set of guidelines distinguishing sex and gender (ons 2019). however, the ons add further provision for ‘sex’ being self-defined by survey respondents, thus highlighting the importance of providing clarity in how terms are used in a survey and what respondents are being asked of them. ‘sexual identity’ typically refers to how a person experiences or expresses themselves sexually or romantically (durso and gates 2013). numerous identity labels are used such as lesbian, gay, bisexual, pansexual, queer, and heterosexual or straight (leonard et al. 2012). it is worth noting that the term ‘sexual orientation’ is often used broadly to refer to three main components: sexual identity; sexual behaviour; and sexual attraction (katz-wise 2015). for example, a person may be attracted to someone 16 lyons et al. australian population studies 4 (2) 2020 of the same sex and/or engage in same-sex behaviour, but might identify as heterosexual rather than lesbian or gay. depending on the focus of a survey, researchers may be interested in one or more components of sexual orientation. however, sexual identity tends to be the main focus in most national surveys in australia, and we therefore focus on sexual identity in this article. it needs to be noted that conceptual and empirical work on understanding sex, gender, and sexual identity is considerable and complex, and it is beyond the scope of this paper to cover the many nuances and perspectives. for example, there are arguments that gender and sexual identity are not necessarily inseparable (e.g. butler 2006). however, key points for the purposes of this article involve recognising that there is considerable diversity in gender and sexual identity, that terms such as sex and gender are typically used in different ways, and approaches to collecting gender and sexual identity data are likely to frame how populations are understood. for example, there are many differences in health and well-being between different sexual and gender diverse populations (leonard et al. 2012; perales 2019) that are unlikely to be detected by surveys that use binary options for sex or gender, or inadequately capture data on sexual identity. recognising these issues, an analysis was conducted in 2016 of national population-based surveys in the us to assess how effectively questions capture detailed information from sexual and gender diverse populations (federal interagency working group on improving measurement of sexual orientation and gender identity in federal surveys 2016). considerable variation was found across surveys in question wording and available response options. for example, some sexual identity questions only enabled responses of ‘straight’, ‘lesbian or gay’, or ‘bisexual’, while others allowed a greater range by including an ‘other’ category. the evidence is clear that there is a greater diversity of sexual identities than these, such as pansexual and queer, which are distinct from identities such as gay or bisexual (van anders 2015). similar work on assessing the strengths and weaknesses of approaches to asking questions in population surveys has also been conducted in other countries, such as canada (waite and denier 2019) and the uk (sullivan 2020). it is important to examine whether australia shares similar challenges and needs in designing surveys to effectively capture data that better reflects sexual and gender diverse populations. identifying potential implications of different approaches to questions would assist demographers and population researchers in developing strategies to ensure that sexual and gender diverse populations are appropriately counted in australia. how data is collected has important implications for statistical estimations that seek to identify these populations, estimate population sizes, and understand key differences between them, which informs policymaking in areas such as health, education, and social services, as well as how we understand their lives more broadly (cahill and makadon 2017; gates 2017). in this article, we focus on a selection of large and prominent surveys conducted over the past ten years. we had one main aim: to compare and discuss how gender and sexual identity questions have been asked across different surveys. we also note the percentages of respondents reported from each survey according to gender or sexual identity, where available. we provide a combined analysis and discussion to draw attention to implications related to specific surveys and to identify important considerations for future studies. australian population studies 4 (2) 2020 lyons et al. 17 2. data and methods 2.1. procedure our review focused on the census and other national surveys conducted in australia that are known to be important sources of data for policy and research. the team utilised their combined knowledge of existing national surveys in australia to identify potential case examples. selection was based on several considerations. first, to ensure relevance to the contemporary sociocultural environment, we only included those conducted in the past ten years. second, to examine whether survey questions are constructed differently depending on the target population, we included examples of those that either targeted a general population or sexual and gender diverse populations. third, to enhance diversity in different types of surveys examined, we focused on those designed for different purposes, such as health versus employment versus community participation. for surveys specifically targeting sexual and gender diverse populations, we further attended to those that focused on trans and gender diverse populations to examine how survey questions on gender are constructed when confined to these populations. finally, we limited the scope to those that also included sexual identity questions or were able to report some related information (e.g. the census). the scoping exercise identified five general population surveys, including the census, and four surveys focused on sexual and gender diverse populations that satisfied the above considerations. it is important to note the primary aim was to identify a broad selection of surveys as case examples rather than exhausting all possibilities, with the objective of documenting and discussing some but not necessarily all of the diversity in the framing of gender and sexual identity questions. for each survey, we document the year it was conducted, sample size, gender and sexual identity question wording (where included in the survey), available response options, and percentages of respondents for key response options. the analysis was primarily descriptive to enable a qualitative investigation. however, we focused on three key areas: (1) different terms used in the wording of questions (e.g. sex versus gender) and (2) variations in response options. we further note the percentages of respondents reporting different genders and sexual identities to provide additional context, but acknowledge that these may also vary based on other aspects of survey methodology. 2.2. overview of survey case examples table 1 displays the survey case examples, information about each survey including survey sampling, the wording of questions as available, and percentages of respondents for key response options. the five national general population surveys included the census (abs 2017), general social survey (gss; abs 2015), household, income and labour dynamics in australia survey (hilda; melbourne institute 2020), second australian study of health and relationships (ashr2; richters, badcock et al. 2014), and the australian gen z study (agz; singleton et al. 2019). the four surveys that specifically focus on sexual and gender diverse populations included private lives 2 (pl2; leonard et al. 2012), rainbow ageing (lyons et al. 2019), the australian trans and gender diverse sexual health survey (atgdshs; callander et al. 2019), and scrolling beyond binaries (sbb; byron et al. 2019). 18 lyons et al. australian population studies 4 (2) 2020 table 1: examples of sexual and gender identity questions from selected national surveys conducted in australia in the past ten years survey survey details sampling sex/gender question sexual identity question demographic composition‡ surveys focused on general populations census of population and housing (‘census’) first national census commenced in 1911, conducted every 5 years. most recent wave in 2016. self-completed online or paper n = 23,401,892 sex: is the person male or female: * male * female upon request, respondents could receive a special form with an ‘other’ option. 2016 was the first time this was made available. sexual identity: no sexual identity questions. however, counts of persons in samesex couples can be manually calculated (see wilson & shalley, 2018). sex (2016): 49.3% male, 50.7% female same-sex couples (2016): 46,800 same-sex couples, or 0.9% of all couples (23,700 male; 23,000 female) general social survey (gss) commenced in 2002, conducted every 4 years. most recent wave in 2014. sexual orientation item was included from 2014. face-to-face interview n = 12,932 sex: * male * female sexual identity: which of the following options best describes how you think of yourself? * straight (heterosexual) * gay or lesbian * bisexual * other * don’t know sex (2014): 49.3% male, 50.7% female sexual identity (2014): 97.0% heterosexual, 3.0% nonheterosexual (1.5% gay or lesbian; 1.5% bisexual, other, or don’t know) household, income and labour dynamics in australia (hilda) survey commenced in 2001, conducted annually (combination of interview [12%] and self-completion [88%]). sexual identity item was included in 2012 (wave 12) and 2016 (wave 16). interview (faceto-face or telephone) or self-complete n = 15,380 sex: are you… * male * female * other sexual identity: which of the following categories best describes how you think of yourself? * heterosexual or straight * gay or lesbian * bisexual * other * unsure/don’t know * prefer not to say sex (2018): 48.6% male, 51.4% female sexual identity (2012): 91.9% heterosexual or straight, 1.4% gay or lesbian, 1.4% bisexual, 0.8% other, 0.9% unsure/don’t know, 2.5% prefer not to say australian population studies 4 (2) 2020 lyons et al. 19 survey survey details sampling sex/gender question sexual identity question demographic composition‡ second australian study of health and relationships (ashr2) commenced in 2001-2002, then again in 2012-2013. telephone interview using modified randomdigit phone dialling n = 20,094 sex: telephone interviewers asked to interview a male or female person in the household and a tailored questionnaire was administered depending on whether the person was male or female. sexual identity: do you think of yourself as: * heterosexual or straight * homosexual (gay [asked of males]; lesbian [asked of females]) * bisexual * queer * not sure, undecided * something else/other sex (2012-3): 49.6% male, 50.4% female sexual identity (2012-3): 96.5% heterosexual, 1.6% gay/lesbian, 1.8% bisexual, 0.2% other australian gen z (agz) study a 2017 nationally representative telephone survey of young people aged 13-18 years (plus 30 follow-up interviews). telephone interview n = 1,200 gender: telephone interviewers asked: can i please confirm your gender? if you would prefer, please just say the number before each option i read out. do you think of yourself as? (multiple response) * one, male * two, female * three, trans * four, intersex * five, gender queer * other (specify) * don’t know * refused sexual identity: the next question is about your sexual identity. if you would prefer, please just say the number before each option i read out. do you think of yourself as? (multiple response) * one, straight, that is heterosexual * two, lesbian/homosexual/gay * three, bisexual * four, questioning * five, queer * something else (specify) * don’t know * refused gender: 49.8% male, 48.7% female, 0.4% trans, 0% intersex, 0.7% gender queer, 0.2% other, 0.3% no response sexual identity: 86% straight, 2% lesbian, homosexual or gay, 7% bisexual, 4% questioning, 1% queer or something else, 1% don’t know 20 lyons et al. australian population studies 4 (2) 2020 survey survey details sampling sex/gender question sexual identity question demographic composition‡ surveys focused on sexual and gender diverse populations private lives 2 (pl2) commenced in 2006, then again in 2011. target: lesbian, gay, bisexual, and transgender (lgbt) australians self-completed online and paper n = 3,835 gender: thinking about your gender and sexuality, are you: * male * female * trans (identifying as male) * trans (identifying as female) * i prefer to refer to myself as... (write in answer) sexual identity: do you think of yourself primarily as: * gay * lesbian * queer * bisexual * dyke * heterosexual/straight * not sure or undecided; * i prefer to refer to myself as... (write in answer) gender (2011): 44.4% male, 48.2% female, 4.4% transgender (3.2% identified as female; 1.2% identified as male), 3% preferred another term sexual identity (2011): 42.6% gay, 30.1% lesbian, 7.1% queer, 11.8% bisexual, 2.1% heterosexual, 1.7% not sure, 4.5% preferred another term rainbow ageing conducted in 2017, one wave only. target: lesbian, gay, bisexual, and transgender (lgbt) australians self-completed online n = 895 gender: how do you describe your gender? * male * female * trans female / trans woman * trans male / trans man * genderqueer * biqueer * agender * other (please describe) sexual identity: how do you describe your sexuality? * gay * lesbian * bisexual * queer * pansexual * asexual * straight or heterosexual * other (please describe) gender: 61.7% male, 32.1% female, 4.3% transgender (0.4% identified as male; 3.9% identified as female), 1.8% other sexual identity: 58.7% gay, 29.4% lesbian, 5.4% bisexual, 1.2% queer, 0.7% pansexual, 0.8% asexual, 1.8% heterosexual, 2.0% other australian population studies 4 (2) 2020 lyons et al. 21 survey survey details sampling sex/gender question sexual identity question demographic composition‡ australian trans and gender diverse sexual health survey (atgdshs) conducted in 2018, one wave only. target: trans and gender diverse (tgd) australians self-completed online n = 1,613 gender: participants self-described their gender in an open-ended question. sexual identity: participants self-described their ‘sexual orientation’ in an openended question. gender: responses stratified into 4 categories for reporting: 21.9% trans men, 24.6% trans women, 14.3% non-binary assigned male at birth, 39.2% non-binary assigned female at birth sexual identity: responses were organised into 9 categories: queer; pansexual; bisexual; homosexual; asexual; heterosexual; gynosexual; androsexual; and ‘no label’. proportions for each varied as a function of gender category scrolling beyond binaries (sbb) conducted in 2016, one wave only. target: lgbtiq+ young people aged 16-35 years self-completed online n = 1,304 gender identity: what is your current gender identity? * male * female * non-binary * different identity (please state) gender: what gender were you assigned at birth? * male * female sexual identity: do you consider yourself to be: * lesbian or gay * straight or heterosexual * bisexual * queer * different identity (please state) gender identity: 26.5% male, 45.6% female, 19.4% nonbinary, and 8.6% described their own gender identity (e.g., trans, agender, genderqueer). sexual identity: 33.9% lesbian or gay, 24.7% bisexual, 18% queer, and 19.8% chose to describe their sexual identity, which included pansexual, asexual, panromantic, and demisexual notes. ‡ of most recent wave or most recent wave with relevant data. n references the sample size from the data wave referenced in the ‘demographic composition’ column, and refers to respondents who contributed eligible data only. for the sex/gender and sexual identity questions, we have used the terms used in the surveys, such as ‘sex’, ‘gender’, or ‘gender identity’. where no terms were used for sex or gender (e.g. “are you…”), we have used ‘sex’ if options noted male or female. where no terms were used for sexuality, we have used ‘sexual identity’ if options noted identity labels such as lesbian, gay, bisexual etc. the question wordings for each survey are provided where available, but these were unavailable in some instances (e.g. the gss for sex). data were sourced from the following: census: abs (2017); gss: abs (2015); hilda: department of social services and melbourne institute of applied economic and social research (2019), wilkins et al. (2019), wooden (2014); ashr2: richters, altman et al. (2014), richters, badcock, et al. (2014); pl2: leonard et al. (2012); atgdshs: callander et al. (2019); scrolling beyond binaries: byron et al. (2019); agz and rainbow ageing: data from source authors. 22 lyons et al. australian population studies 4 (2) 2020 as noted in table 1, surveys varied by mode, such as interviews or self-completed surveys. the general population surveys varied most, with surveys administered either online, paper, or via interviews. the sexual and gender diverse surveys were all self-completed and mostly conducted online. mode can potentially influence responses for several reasons, such as enabling questions to be clarified when administered via interview or allowing privacy when self-completed (kuhne et al. 2019), which needs to be considered in the interpretation of findings. also, sample sizes varied between surveys. apart from the census, the sample sizes for the surveys ranged from 895 for rainbow ageing to 20,094 for ashr2. 3. analysis and discussion 3.1. sex and gender as displayed in table 1, of the general population surveys, the gss and ashr2 offered only binary options of male and female. the hilda survey offered an ‘other’ option and the census allowed respondents to request an additional form with an ‘other’ option. all labelled this category as sex (rather than gender) and all asked the question in prescriptive terms (e.g. ‘are you…’). conversely, the agz and all the sexual and gender diverse surveys labelled this category as gender or gender identity and offered additional response options. specifically, they offered four options, including male, female, and transgender, as well as a text-entry version of ‘other’ to allow respondents to enter their own description. an exception was the atgdshs, which provided a single open-ended text-entry field for respondents to describe their gender rather than choosing from fixed options. these surveys also asked the question in identity terms, e.g. ‘how do you identify’ / ‘how do you refer to yourself’, with the exception of pl2. these differences in question framing may appear subtle but could potentially be read differently by respondents. for example, asking a question in prescriptive terms may result in some respondents interpreting it as sex assigned on their birth certificate. when framed as identity, it enables respondents to indicate how they actually describe themselves. surveys may therefore need to pay close attention to precise question framing. there has also been a move toward a two-step process, where respondents are first asked for their sex assigned at birth and then their gender (pash.tm 2017), which enables identification of those who are not cisgender. importantly, this approach captures the experiences of people who have taken steps to affirm their gender, but do not identify as trans. in terms of the different responses, readers need to be mindful that differences between survey modes and sample sizes may have influenced responses, as noted above. however, despite some of the differences, when only binary options were offered, about half the sample each reported as male or female. however, when options outside the sex/gender binary were available, these were used to varying degrees ranging from 6.2% of respondents in rainbow ageing to 28.4% of respondents in sbb. importantly, we cannot estimate how many respondents might have used non-binary labels or would have used ‘other’ or ‘prefer not to say’ options if they had been available in the general population surveys, as these options were not available in those collections with the exception of hilda. this means that there are no reliable estimates of how many sexual and gender diverse individuals there are in australia. the exception to this is the agz survey, but it is limited to adolescents. australian population studies 4 (2) 2020 lyons et al. 23 there is some debate in the literature about how questions about being transgender should be asked. some theorists have argued that sex and gender are complex constructs that are inseparable from each other (e.g. fugard 2020), and others argue that they cannot be used interchangeably (e.g. gates 2014). pash.tm, the peer advocacy network for trans masculinities, states in their position statement on best practice data collection (pash.tm 2017) that trans and gender diverse people are not a homogeneous group, and should not be treated as one. they highlight the need for designing survey questions that are inclusive of all genders and of people who do not identify with binary gender, and to provide reliable data without erasing any groups. for example, some trans people identify strongly as male or female and trans, others do not identify as trans at all, whilst others identify as only trans for the purposes of visibility (see bauer et al. 2017). moreover, not everyone whose gender identities differ from their birth sex consider themselves trans (darwin 2020), and indeed in the atgdshs, approximately half the sample identified as transgender while the other half identified as non-binary. as such, they recommend that any gender identity question allows for multi-select answers, since single-option answer selections do not accurately reflect the gender experience of many trans people, and may result in under-reporting of gender diversity. when available, some respondents select more than one gender identity, or if describing themselves in their own words use combinations of categories (e.g. “trans masculine gender nonconforming femme”) or describing a temporally variant gender identity (e.g. “some days i am a girl, other days i have no gender”; see byron et al. 2019). this aligns with work from the us, in which more than 13% of non-heteronormative respondents preferred to select more than one gender identity option than to simply enter their own term (ruberg and ruelos 2020), and contradicts ideas that treat gender as synonymous with sex and as unchanging over the course of a lifetime (see westbrook and saperstein 2015). in addition, it suggests that for many sexual and gender diverse people, gender identity cannot be characterised simply by selecting one term from an inclusive list, but instead by being able to identify multiple, overlapping elements of identity. taken together, this suggests that there is a level of complexity and nuance in how survey respondents self-report their gender identities, which is unlikely to be fully captured in items that simply allow for a single response of male or female. 3.2. sexual identity of the general population surveys displayed in table 1, the majority asked respondents to selfdescribe their sexual identity by selecting either straight/heterosexual, gay or lesbian, bisexual, and a combination of either other, don’t know, unsure, or prefer not to say. the census does not allow a direct count of non-heteronormative individuals, but researchers have made counts of individuals in same-sex couples, based on the number of people in australia who reported living with their samesex partner on census night (which suggests less than 1% of couples are same-sex; abs 2017). this does not take into account any other non-heteronormative australians, such as single people who identify as lesbian or gay, bisexual-identifying people who are in an opposite-sex relationship, pansexual or asexual individuals, etc. while the census does not include data on sexual identity, the abs does collect information about sexual identity from respondents in the gss. this survey found approximately 3% self-reported as ‘non-heterosexual.’ it is also worth noting that a study by wilson and shalley (2018) estimated that 3.2% of the australian population were non-heterosexual based on data from several population surveys and the census. 24 lyons et al. australian population studies 4 (2) 2020 the agz and all the surveys that focused on sexual and gender diverse populations also asked for a self-description of sexual identity, and offered similar responses to the general population surveys, but also included further options, such as queer, questioning, asexual, and pansexual, as well as textentry options. this resulted in a much larger range of responses. for example, 19.8% of respondents in the sbb survey chose to describe their own sexual identity, rather than choosing lesbian, gay, straight, bisexual, or queer. this suggests that more limited options used in some surveys, such as ‘lesbian/gay’ and ‘bisexual’, do not capture the range of identities that australians use. ratifying this argument is the finding that in the atgdshs, where the only option was to provide a text response, the open-ended responses provided were coded into nine different categories. further follow-up analysis of these data using hierarchical clustering techniques yielded three non-exclusive clusters (callander et al. 2020). this work presents a potentially interesting scenario, where the widest range of responses are made possible to respondents and then employing analytical techniques in different ways to aggregate data. as with sex and gender questions, there is debate in the literature about how best to ask questions of sexual identity in surveys, although it seems more readily accepted that a range of response options are necessary. indeed, some have questioned the appropriateness of having a wide range of response options (see sabia et al. 2017 for a discussion). there are logistical considerations, such as balancing brevity in survey length and conciseness in response data against deeper and nuanced understandings of the sexual identities that people use. we recognise that there are also implications for asking people to choose from a range of choices when they might be unfamiliar with some of the options (e.g. demisexual, pansexual, panromantic, etc). indeed, a study by sell, wells and wypij (1995) revealed that some survey respondents select ‘unsure’ because they do not understand the question or the response options rather than because they are undecided. finally, having options to select from suggests that sexual identities are categorical and fixed, which runs counter to many contemporary conceptualisations of sexual identity which posit that sexuality is a dynamic spectrum (e.g. d’augelli et al. 1995). this is particularly important in light of research from europe that showed 5.9% of europeans identified as lgbt on a dichotomous yes/no question, but over 10% identified as more than 0 on the kinsey scale (that is, more than double chose an option that is not ‘exclusively heterosexual’ on a 7-point scale of sexual orientation; deveaux 2016). as per table 1, there are a suite of non-committal response options provided in some surveys, including ‘unsure’, ‘don’t know’, ‘prefer not to say.’ these are often necessary to avoid respondents dropping out of a survey if they are unsure how to answer the question or wish not to disclose. many of the surveys also included an ‘other’ option for those whose identity is not listed in the response options. however, this has been met with varying levels of criticism. for example, badgett (2009) criticises the use of ‘other’ as a response since it does not provide options for re-classification, and these respondents typically become excluded from data sets during analyses, which falsely decreases estimates of those who would not consider themselves heterosexual and attenuates statistical power. interestingly, 2.5% of hilda respondents, which would equal over half a million australians, indicated that they preferred not to report their sexual identity. this could mean that these respondents did not have the right options to choose from, or did not wish to disclose their sexual australian population studies 4 (2) 2020 lyons et al. 25 identity. in line with this contention, badgett recommends avoiding a “prefer not to say” response option. when asking questions about sexual identity, it is often unclear precisely what is meant. for the most part, survey respondents likely focus on the gender of the people they are sexually attracted to, as this is the typical focus in such discussions of sexuality (see van anders, 2015). however, the survey might equally be trying to measure other aspects of sexual orientation, including sexual behaviour or sexual attraction (or a combination of all the above). as such, estimates of populations who are not heterosexual will vary based on how questions are operationalised. for example, only asking about sexual identity might exclude individuals who engage in same-sex sexual behaviour or who experience same-sex sexual attraction without acting on it. it is worth considering that questions that ask about different aspects of sexual orientation will attract qualitatively different subpopulations, and that they will be different sizes. for instance, the evidence suggests that the proportion of people who have engaged in same-sex sexual behaviour is far higher than the proportion who selfidentify as gay, lesbian, or bisexual. for example, in the second australian study of health and relationships, 96.8% of men and 96.3% of women identified as heterosexual, yet 90.7% of men and 82.3% of women reported having sexual experiences exclusively with the other sex (richters, altman et al. 2014; see also spiegelhalter 2015). wilson and shalley (2018) highlight the need for purposefully selecting definitions so that population estimates from these surveys reflect the subject of the research. for example, same-sex sexual behaviour might be a more appropriate focus than self-identification for some forms of health surveillance research. 4. conclusions comprehensive population-level data is essential for many reasons, including government planning, understanding health needs, and informing research across disciplines. despite this, surveys have tended to adopt different approaches to asking questions about gender and sexual identity, including those specifically focusing on sexual and gender diverse populations. this adds to the challenge of comparing findings from one survey to another. in addition, questions are often limited to fixed options that are unlikely to adequately capture the identities that many people have, thus potentially resulting in respondents either misclassifying their identity or discontinuing the survey. in either case, there is the risk of erasure, where some respondents are made invisible or inadequately represented in survey results. this has the potential implication of confounding the categories that have been provided. it is especially the case for gender, where it might be erroneously assumed that all respondents who had no option other than to select male or female are cisgender. in addition, as we discussed, identities may not always be singular, fixed, or discrete, and may require respondents to select multiple options or describe their identity in a way that is meaningful to them. practical considerations, such as the length of questions and response sets and how to frame questions or providing definitions of terms so that all respondents understand what is being asked of them, add further challenges, but are unlikely to be insurmountable. ultimately, it would be beneficial to have guidelines developed for different data collection purposes with regard to framing questions and analysing responses. informing this, it will be important for researchers to examine specifically how results vary according to question framing. given that there are numerous gender and sexual identities, the implications for the reporting of results and potential 26 lyons et al. australian population studies 4 (2) 2020 policymaking of how identities are aggregated into larger categories, and the question of what comprises appropriate levels of aggregation, also requires investigation. consulting with key communities, such as sexual and gender diverse communities, to ensure relevance and appropriateness of questions will be further important. 5. key messages • there is wide variation in approaches to identifying sexual and gender diverse populations in australia. • developing greater consistency in approaches across surveys and refining questions will more fully capture diverse identities and variations in embodiment. • with advancements in approaches there is an opportunity to collect data that provides greater representation of sexual and gender diverse populations and a stronger basis for estimating the demography of these populations. • refining survey techniques in this space will ensure that diverse populations are more fully counted in australia. • counting sexual and gender diverse populations is especially important for informing service delivery, policymaking, and research. references abs (2015) general social survey household questionnaire. cat. no. 4159.0. canberra: australian bureau of statistics. http://www.abs.gov.au/ausstats/abs@.nsf/detailspage/4159.02014?opendocument. accessed on 14 july 2020. abs (2016) standard for sex and gender variables. canberra: australian bureau of statistics. https://www.abs.gov.au/statistics/standards/standard-sex-and-gender-variables/2016. accessed on 5 october 2020. abs (2017) census of population and housing: understanding the census and census data, australia, 2016. catalogue no. 2900.0. canberra: australian bureau of statistics. https://www.abs.gov.au/ausstats/abs@.nsf/lookup/by%20subject/2900.0~2016~main%20featu res~understanding%20the%20census%20and%20census%20data~1. accessed on 14 july 2020. abs (2019) abs media statement on sexual orientation and gender identity questions and the 2021 census. canberra: australian bureau of statistics media centre. https://www.abs.gov.au/websitedbs/d3310114.nsf/home/abs%20media%20statement%20on% 20sexual%20orientation%20and%20gender%20identity%20questions%20and%20the%202021%2 0census. accessed on 14 july 2020. anderson j r and holland e (2015) the legacy of medicalising ‘homosexuality’: a discussion on the historical effects of non-heterosexual diagnostic classifications. sensoria 11(1): 4-15. australian government (2013) australian government guidelines for the recognition of sex and gender. canberra: australian government attorney general’s department. https://www.ag.gov.au/sites/default/files/202003/australiangovernmentguidelinesontherecognitionofsexandgender.pdf. accessed on 5 october 2020. badgett m v l (2009) best practices for asking questions about sexual orientation on surveys. los angeles: ucla, the williams institute. http://www.abs.gov.au/ausstats/abs@.nsf/detailspage/4159.02014?opendocument https://www.abs.gov.au/statistics/standards/standard-sex-and-gender-variables/2016 https://www.abs.gov.au/ausstats/abs@.nsf/lookup/by%20subject/2900.0~2016~main%20features~understanding%20the%20census%20and%20census%20data~1 https://www.abs.gov.au/ausstats/abs@.nsf/lookup/by%20subject/2900.0~2016~main%20features~understanding%20the%20census%20and%20census%20data~1 https://www.abs.gov.au/websitedbs/d3310114.nsf/home/abs%20media%20statement%20on%20sexual%20orientation%20and%20gender%20identity%20questions%20and%20the%202021%20census https://www.abs.gov.au/websitedbs/d3310114.nsf/home/abs%20media%20statement%20on%20sexual%20orientation%20and%20gender%20identity%20questions%20and%20the%202021%20census https://www.abs.gov.au/websitedbs/d3310114.nsf/home/abs%20media%20statement%20on%20sexual%20orientation%20and%20gender%20identity%20questions%20and%20the%202021%20census https://www.ag.gov.au/sites/default/files/2020-03/australiangovernmentguidelinesontherecognitionofsexandgender.pdf https://www.ag.gov.au/sites/default/files/2020-03/australiangovernmentguidelinesontherecognitionofsexandgender.pdf australian population studies 4 (2) 2020 lyons et al. 27 bauer g r, braimoh j, scheim a i and dharma c (2017) transgender-inclusive measures of sex/gender for population surveys: mixed-methods evaluation and recommendations. plos one 12(5): e01780434. butler j (2006) gender trouble. new york: routledge. byron p, robards b, hanckel b, vivienne s and churchill b (2019) “hey, i’m having these experiences”: tumblr use and young people’s queer (dis) connections. international journal of communication 13: 2239-2259. cahill s r and makadon h j (2017) if they don't count us, we don't count: trump administration rolls back sexual orientation and gender identity data collection. lgbt health 4(3): 171-173. callander d, newman c e, holt, m, rosenberg s, duncan d t, pony m, timmins l, cornelisse v, duckchong l, wang b and cook t (2020) the complexities of categorizing gender: a hierarchical clustering analysis of data from the first australian trans and gender diverse sexual health survey. transgender health. callander d, wiggins j, rosenberg s, cornelisse vj, duck-chong e, holt m, pony m, vlahakis e, macgibbon j and cook t (2019) the 2018 australian trans and gender diverse sexual health survey: report of findings. sydney: the kirby institute, unsw. connell rw (2002) gender. cambridge, england: polity. darwin h (2020) challenging the cisgender/transgender binary: nonbinary people and the transgender label. gender and society 34(3): 357-380. d'augelli a r, patterson c and patterson c j (1995) lesbian, gay, and bisexual identities over the lifespan: psychological perspectives. oxford: oxford university press on demand. department of social services and melbourne institute of applied economic and social research (2019) the household, income and labour dynamics in australia (hilda) survey, general release 18 (waves 1-18). ada dataverse. https://dataverse.ada.edu.au/dataset.xhtml?persistentid=doi:10.26193/iybxhm. accessed on 14 july 2020. deveaux f (2016) counting the lgbt population: 6% of europeans identify as lgbt. https://daliaresearch.com/blog/counting-the-lgbt-population-6-of-europeans-identify-as-lgbt. accessed on 14 july 2020. durso l e and gates g j (2013) best practices: collecting and analysing data on sexual minorities. in a baumle (ed), international handbook on the demography of sexuality. dordrecht, netherlands: springer. federal interagency working group on improving measurement of sexual orientation and gender identity in federal surveys (2016) current measures of sexual orientation and gender identity in federal surveys. washington dc: federal committee on statistical methodology. fugard a (2020) should trans people be postmodernist in the streets but positivist in the spreadsheets? a reply to sullivan. international journal of social research methodology .23(5): 525-531. gates g j (2014) lgb families and relationships: analyses of the 2013 national health interview survey. los angeles: williams institute, ucla. http://williamsinstitute.law.ucla.edu/wpcontent/uploads/lgb-families-nhis-sep-2014.pdf. accessed on 5 october 2020. gates g j (2017) lgbt data collection amid social and demographic shifts of the us lgbt community. american journal of public health 102(8): 1220–1222. hammarström a and annandale e (2012) a conceptual muddle: an empirical analysis of the use of ‘sex’ and ‘gender’ in ‘gender-specific medicine’ journals. plos one 7(4): e34193 katz-wise s l (2015) sexual fluidity in young adult women and men: associations with sexual orientation and sexual identity. psychology and sexuality 6(2): 189-208. kuhne s, kroh m and richter d (2019) comparing self-reported and partnership-inferred sexual orientation in household surveys. journal of official statistics 35(4): 777-805. https://dataverse.ada.edu.au/dataset.xhtml?persistentid=doi:10.26193/iybxhm https://daliaresearch.com/blog/counting-the-lgbt-population-6-of-europeans-identify-as-lgbt http://williamsinstitute.law.ucla.edu/wp-content/uploads/lgb-families-nhis-sep-2014.pdf http://williamsinstitute.law.ucla.edu/wp-content/uploads/lgb-families-nhis-sep-2014.pdf 28 lyons et al. australian population studies 4 (2) 2020 leonard w, pitts m, mitchell a, lyons a, smith a, patel s, couch m and barrett a (2012) private lives 2: the second national survey of the health and wellbeing of gay, lesbian, bisexual and transgender (glbt) australians. monograph series no. 86. melbourne: la trobe university. liszewski w, peebles j k, yeung h and arron s (2018) persons of nonbinary gender: awareness, visibility, and health disparities. new england journal of medicine 379(25): 2391-2393. lyons a, alba b, waling a, minichiello v, hughes m, barrett c, fredriksen-goldsen k, edmonds s and blanchard m (2019) recent versus lifetime experiences of discrimination and the mental and physical health of older lesbian women and gay men. ageing and society doi: 10.1017/s0144686x19001533. melbourne institute (2012) hilda self-completion questionnaire. melbourne: the university of melbourne. https://melbourneinstitute.unimelb.edu.au/assets/documents/hildaquestionnaires/selfcompletionquestionnairew12m.pdf. accessed on 8 may 2020. mizock l and mueser k t (2014) employment, mental health, internalized stigma, and coping with transphobia among transgender individuals. psychology of sexual orientation and gender diversity 1(2): 146. national lgbti health alliance (2020) snapshot of mental health and suicide prevention statistics for lgbti people. sydney: national lgbti health alliance. https://d3n8a8pro7vhmx.cloudfront.net/lgbtihealth/pages/240/attachments/original/15954922 35/2020-snapshot_mental_health_%281%29.pdf?1595492235. accessed on 5 october 2020. ons (2019) what is the difference between sex and gender? exploring the difference between sex and gender, looking at concepts that are important to the sustainable development goals. london: office of national statistics. https://www.ons.gov.uk/economy/environmentalaccounts/articles/whatisthedifferencebetween sexandgender/2019-02-21. accessed on 5 october 2020. pash.tm (2017) position statement: data collection. sydney: pash.tm. http://www.grunt.org.au/wpcontent/uploads/pash.tm-data-collection_march2018.pdf. accessed on 14 july 2020. perales f (2019) the health and wellbeing of australian lesbian, gay and bisexual people: a systematic assessment using a longitudinal national sample. australian and new zealand journal of public health 43(3): 281-287. richters j, altman d, badcock p b, smith a m a, de visser r o, grulich a e, rissel c and simpson j m (2014) sexual identity, sexual attraction and sexual experience: the second australian study of health and relationships. sexual health 11(5): 451–460. richters j, badcock p b, simpson j m, shellard d, rissel c, de visser r o, grulich a e and smith a m a (2014) design and methods of the second australian study of health and relationships. sexual health 11(5): 383–396. ruberg b and ruelos s (2020) data for queer lives: how lgbtq gender and sexuality identities challenge norms of demographics. big data and society 7(1): 2053951720933286. sabia j j, wooden m and nguyen t t (2017) sexual identity, same-sex relationships, and labour market dynamics: new evidence from longitudinal data in australia. southern economic journal 83(4): 903-931. sell r l, wells j a and wypij d (1995) the prevalence of homosexual behaviour and attraction in the united states, the united kingdom and france: results of national population-based samples. archives of sexual behavior 24: 235-248. singleton a, rasmussen m l, halafoff a and bouma g d (2019) the agz study: project report. australia: australian national university, deakin university and monash university. spiegelhalter d (2015) sex by numbers: what statistics can tell us about sexual behaviour. london: profile books. https://melbourneinstitute.unimelb.edu.au/assets/documents/hilda-questionnaires/selfcompletionquestionnairew12m.pdf https://melbourneinstitute.unimelb.edu.au/assets/documents/hilda-questionnaires/selfcompletionquestionnairew12m.pdf https://d3n8a8pro7vhmx.cloudfront.net/lgbtihealth/pages/240/attachments/original/1595492235/2020-snapshot_mental_health_%281%29.pdf?1595492235 https://d3n8a8pro7vhmx.cloudfront.net/lgbtihealth/pages/240/attachments/original/1595492235/2020-snapshot_mental_health_%281%29.pdf?1595492235 https://www.ons.gov.uk/economy/environmentalaccounts/articles/whatisthedifferencebetweensexandgender/2019-02-21 https://www.ons.gov.uk/economy/environmentalaccounts/articles/whatisthedifferencebetweensexandgender/2019-02-21 http://www.grunt.org.au/wp-content/uploads/pash.tm-data-collection_march2018.pdf http://www.grunt.org.au/wp-content/uploads/pash.tm-data-collection_march2018.pdf australian population studies 4 (2) 2020 lyons et al. 29 sullivan a (2020) sex and the census: why surveys should not conflate sex and gender identity. international journal of social research methodology 23(5): 517-524. van anders s m (2015) beyond sexual orientation: integrating gender/sex and diverse sexualities via sexual configurations theory. archives of sexual behavior 44(5): 1177-1213. waite s and denier n (2019) a research note on canada’s lgbt data landscape: where we are and what the future holds. canadian review of sociology 56(1): 93-117. west c and zimmerman d h (1987) doing gender. gender and society 1(2): 125-151. wilkins r, laß i, butterworth p and vera-toscano e (2019) the 14th annual statistical report of the hilda survey. melbourne institute: applied economic and social research. melbourne: the university of melbourne. wilson t and shalley f (2018) estimates of australia's non-heterosexual population. australian population studies 2(1): 26-38. wooden m (2014) the measurement of sexual identity in wave 12 of the hilda survey (and associations with mental health and earnings). hilda project discussion paper series no. 1/14. melbourne: the university of melbourne. australian population studies 2023 | volume 7 | issue 1 | pages 30-33 © grossman 2023. published under the creative commons attribution-noncommercial licence 3.0 australia (cc by-nc 3.0 au). journal website: www.australianpopulationstudies.org visualising the impact of covid-19 on australian sa3 populations irina grossman the university of melbourne email: irina.grossman@unimelb.edu.au. address: melbourne school of population and global health, level 5, 207 bouverie st, the university of melbourne, victoria 3010 australia. received 12 october 2022; accepted 6 march 2023; published 6 december 2023 introduction the covid-19 pandemic, and the related health measures implemented to control its spread resulted in significant changes in australian international and internal migration. for example, net overseas migration was 192,700 in 2019-20, but decreased to -84,940 in 2020-21 (abs 2022). it is difficult to understand how this population decrease was distributed at the smaller area level, particularly as there were also changes in historic internal migration trends, including an increase in migration to regional areas from capital cities (stephens et al., 2022). the key aim of this data visualisation is to investigate how covid-19 impacted populations at the sa3 level by comparing 2021 population estimates with possible populations if covid-19 had not occurred. data and methods possible populations for a no-covid scenario were created for sa3 areas using estimated resident populations (erps) from 2016 and 2019, which are found in ‘table 2. estimated resident population, statistical areas level 3 (asgs 2021), australia’, of the ‘population estimates by sa2 and above (asgs 2021), 2001 to 2021’ data cube (abs 2021a). the no-covid scenario populations were created by applying the average growth rate from 2016 to 2019 to the 2019 data in order to project the sa3 populations to 2021. to investigate the potential impact of covid-19 on sa3 populations the absolute percentage differences between the 2021 sa3 populations of the no-covid scenario and the 2021 sa3 erps were calculated. these differences were then visualised on a mosaic cartogram of australian sa3 areas. mosaic cartograms are maps where each area is represented by a simple shape (cano et al., 2015). the maps presented here represent each sa3 area as an equally sized element, regardless of its physical size or the number of persons resident. the mosaic cartograms are presented layered on top of a standard map of australian states and territories; the state and territories – 2021 gda94 asgs edition 3 digital boundary files were used to produce this map (australian bureau of statistics, 2021b). d e m o g ra p h ic http://www.australianpopulationstudies.org/ mailto:irina.grossman@unimelb.edu.au australian population studies 7 (1) 2023 grossman 31 figure 1: a mosaic cartogram visualising the impact of covid-19 on 2021 australian sa3 populations source: sa3 erps were sourced from abs (2021a). digital boundaries of sa3 areas and of states and territories were obtained from abs (2021b). the maps were created using the ons hexmap automation tool (mitchell & tzelepis, 2017). note: the percentage difference between the 2021 sa3 erps and the populations in the no-covid scenario is presented in the top map. for reference, the state or territory that each element belongs to is presented in the bottom map and selected sa3 areas in capital cities are labelled. 32 grossman australian population studies 7 (1) 2023 to create the mosaic cartograms the 2021 sa3 gda94 asgs edition 3 digital boundary files were downloaded (abs 2021b). the boundary files were transformed into a mosaic cartogram shapefile using the ‘basic’ variant of the ons hexmap automation tool (mitchell & tzelepis, 2017). minor edits were made to the shapefile such that sa3 areas from different states were clustered together. the ‘ggplot’ (wickham, 2016) and ‘sf’ (pebesma et al., 2018) r packages were used to create two maps. the minor edits were completed using adobe illustrator 2023 and adobe photoshop 2023. the first map (figure 1, top map) visualises the absolute percentage change between the projected and estimated 2021 sa3 populations on a mosaic cartogram of australian sa3 areas. the second map (figure 1, bottom map) visualises which state or territory each element of the mosaic cartogram belongs to, and selected sa3 areas in capital cities are labelled. key features the top map in figure 1 visualises the 2021 changes in sa3 populations had covid-19 not occurred and previous population trends continued. for reference, the bottom map of figure 1 shows which states and territories each of the elements in the mosaic cartogram belongs to. the top map reveals that covid-19 may have resulted in decreased populations in most sa3 areas. nationally there was a 1.9% reduction when compared to 2021 erps and the populations in the no-covid scenario; this corresponded to a potential national decrease of 478,314 persons. however, the impact varied between states and territories. according to the no-covid scenario, western australia’s population was 10,074 persons, or 0.4%, greater than what could have been expected had pre-covid-19 population trends continued. this was the only state or territory with an increased population in the no-covid scenario relative to the 2021 erps. all other states saw reduced populations, particularly victoria and the australian capital territory which both had decreases of 3.7%, corresponding to 243,144 and 16,622 persons less, respectively. 81.7% of sa3s in greater capital city regions had decreased populations, amounting to a reduction of 453,904 persons in the capital cities of australia. whilst regional populations were not as affected by covid-19, they generally also saw population declines with 60.4% of sa3s outside of the capital city regions showing population decreases, resulting in a total of 24,410 fewer persons than what may have been expected in non-capital city regions in the absence of covid-19. it is important to note that whilst short term projections, such as the one used here to create the nocovid scenario, tend to be reasonably accurate, there are many factors that may impact sa3 populations other than covid-19, such as housing developments and local job opportunities. it is important to emphasise that the no-covid scenario presented here is just one of many potential scenarios. furthermore, it is important to consider that the process of creating our mosaic cartograms moved the relative positions of the sa3 areas. therefore, the sa3 areas which are in the same sa4, or the same greater capital city statistical area, did not always cluster together. this should be considered when interpreting the maps. acknowledgements i am grateful to tom wilson who kindly provided helpful comments on earlier versions of this demographic. however, all errors and omissions are the responsibility of the author. australian population studies 7 (1) 2023 grossman 33 references abs (2021a). regional population. https://www.abs.gov.au/statistics/people/population/regionalpopulation/latest-release abs (2021b). digital boundary files australian statistical geography standard (asgs) edition 3. https://www.abs.gov.au/statistics/standards/australian-statistical-geography-standard-asgsedition-3/jul2021-jun2026/access-and-downloads/digital-boundary-files abs (2022). overseas migration, 2021-22. https://www.abs.gov.au/statistics/people/population/overseas-migration/2021-22-financialyear cano, r. g., buchin, k., castermans, t., pieterse, a., sonke, w., & speckmann, b. (2015). mosaic drawings and cartograms. computer graphics forum, 34(3), 361-370. https://doi.org/10.1111/cgf.12648 mitchell, b., & tzelepis, g. (2017). automated generation of equal area cartograms (‘hexmaps’) at any scale. https://github.com/onsgeo/hexmapping pebesma, e. j. (2018). simple features for r: standardized support for spatial vector data. the r journal, 10(1), 439-446. https://doi.org/10.32614/rj-2018-009 stephens, r., cansdale, d., and forbes l (2022). increase in internal migration from capital cities to regional areas. https://www.abc.net.au/news/2022-02-18/migration-from-cities-to-regionalareas-doubles-during-covid/100839386 wickham h (2016). ggplot2: elegant graphics for data analysis. springer-verlag new york. https://ggplot2.tidyverse.org https://www.abs.gov.au/statistics/people/population/regional-population/latest-release https://www.abs.gov.au/statistics/people/population/regional-population/latest-release https://www.abs.gov.au/statistics/standards/australian-statistical-geography-standard-asgs-edition-3/jul2021-jun2026/access-and-downloads/digital-boundary-files https://www.abs.gov.au/statistics/standards/australian-statistical-geography-standard-asgs-edition-3/jul2021-jun2026/access-and-downloads/digital-boundary-files https://www.abs.gov.au/statistics/people/population/overseas-migration/2021-22-financial-year https://www.abs.gov.au/statistics/people/population/overseas-migration/2021-22-financial-year https://doi.org/10.1111/cgf.12648 https://github.com/onsgeo/hexmapping https://doi.org/10.32614/rj-2018-009 https://www.abc.net.au/news/2022-02-18/migration-from-cities-to-regional-areas-doubles-during-covid/100839386 https://www.abc.net.au/news/2022-02-18/migration-from-cities-to-regional-areas-doubles-during-covid/100839386 https://ggplot2.tidyverse.org/ australian population studies 2021 | volume 5 | issue 1 | pages 65-76 © temple et al. 2021. published under the creative commons attribution-noncommercial licence 3.0 australia (cc by-nc 3.0 au). journal website: www.australianpopulationstudies.org introductory guide understanding survey data available for researchers working in ageing: the cepar metadata database on ageing jeromey b. temple* the university of melbourne tanara vieira sousa the university of melbourne ruth williams the university of melbourne jay stiles the university of melbourne libby brooke the university of melbourne josh knight the university of melbourne * corresponding author. email: jeromey.temple@unimelb.edu.au. melbourne school of population and global health, the university of melbourne, melbourne, vic 3010, australia. paper received 8 february 2021; accepted 23 april 2021; published 31 may 2021 abstract background for researchers working in gerontology or the demography of ageing, knowledge of and access to population-based data, which includes mature age respondents, is critical. the collection of metadata (information describing data) supports researchers in their search for relevant data. aims this proof-of-concept project seeks to develop a metadata database including metadata on australian sample surveys relevant to ageing over the period 2010 – 2018. data and methods we used a five-stage approach to create the centre for excellence in population ageing research (cepar) metadata database on ageing: 1) identification of in-scope survey datasets; 2) indexing the in-scope surveys; 3) scraping metadata from publicly available sources; 4) appending metadata to a master database; and 5) creation of a webtool to enable users to search and export metadata and obtain contact details for the relevant data custodian. results the cepar metadata database webtool is available from: https://mspgh.unimelb.edu.au/centresinstitutes/centre-for-health-policy/research-group/metadata-database. conclusions metadata database collections can assist researchers to identify what data has been collected (for their specific research interest in ageing), how it was collected and how to gain access to the data. key words metadata; ageing; survey data; australian bureau of statistics; data archive http://www.australianpopulationstudies.org/ mailto:jeromey.temple@unimelb.edu.au https://mspgh.unimelb.edu.au/centres-institutes/centre-for-health-policy/research-group/metadata-database https://mspgh.unimelb.edu.au/centres-institutes/centre-for-health-policy/research-group/metadata-database 66 temple et al. australian population studies 5 (1) 2021 1. introduction evidence-driven gerontology and demography research relies on rigorous population-based empirical data. however, the australian association of gerontology (2017) states there is no single national independent organisation that can collate, translate and identify gaps in evidence related to ageing with the aim of informing australian policy. the collection of metadata (information describing data) can support important evidence-driven research. mapping metadata is a preparatory step towards understanding what information is being captured, how it is captured, as well as improving data collection. therefore, metadata can play a role in demonstrating the scope, foundations and gaps in current evidence-driven gerontology research which can inform future directions. more generally, a key task that all social scientists face when beginning a new project is understanding the variety of data that exists in order to explore the underlying research question. although excellent data archives are available in australia, including the australian data archive (ada) and the australian bureau of statistics (abs) microdata collections, there is currently no single point of metadata information specifically tailored to researchers working in ageing. our solution to this problem was to develop a metadata database as a proof-of-concept project. our metadata database includes metadata on australian sample surveys relevant to ageing (e.g., including data dictionary elements, variable names, data types and formats, and survey and sampling design) between the period 2010 – 2018. the scope further considers surveys collected by the abs or deposited with the ada, where metadata is publicly available. the metadata database webtool we developed enables users to search and export metadata and obtain contact details for the relevant data custodian. in the remainder of this paper, we outline components of metadata, the methodology employed to create the metadata database and provide a worked example of searching and extracting data from the provided webtool. we conclude with a summary of future uses and research priorities, as well as outlining the limitations to the approach employed. 2. what is metadata? metadata is data about data; it is information that describes other data. more precisely, it is the formalised structured description of content, quality, condition, or other characteristics of data. a principal purpose of metadata is to assist users in finding relevant information and discovering resources through bringing similar information together and providing the location of information. in addition, metadata provides access to information on the extent of completeness of the content; the context of the content; and structural relationships within and between information (gilliland 2016). metadata assists in identifying and organising electronic resources by relevant criteria, provides digital identification, and supports the archiving and preservation of resources (united nations statistical commission and economic commission for europe 1995). metadata can be stored internally, in the same file as the described data (also known as embedded metadata), or externally, in a separate file from the described data. a metadata repository generally stores the metadata separate from the data (niso 2004). many types of metadata exist including descriptive, structural, administrative, reference and statistical metadata. australian population studies 5 (1) 2021 temple et al. 67 descriptive metadata describes a resource, typically to search and locate an object, such as title, author, subjects, keywords, or publisher (gilliland 2016). for example, the survey name. statistical metadata describes information about statistical data and the processes of collection. statistical data repositories have their own requirements for metadata to describe not only the source and quality of the data but also what statistical processes were used to create the data. this is of particular importance to both validate and improve the process of statistical data production (gilliland 2016). reference metadata describes aspects of the contents of statistical data and includes explanatory information on the context of the data (eurostat 2019). for example, a data item list, reporting on the representativeness of the dataset. structural metadata describes information about the structure and interrelationships of data, such as types, versions, tables, columns, keys and indexes (gilliland 2016). administrative metadata describes information that assists to manage a resource, such as when and how it was created, acquisition and location information, file type, and who has access rights. two subsets of administrative metadata include rights management metadata, which concerns intellectual property rights, and preservation metadata, which contains information needed to archive and preserve a resource (gilliland 2016). the creation and archiving of metadata are important as it ensures that information and resources will survive and continue to be accessible in the future. archiving and preservation allow tracking the lineage of an object, detailing physical characteristics and documenting behaviour which can be mimicked in future technologies (gilliland 2016). in the cepar metadata database, we include components of descriptive, structural, and administrative metadata. the database also includes links to details on statistical and reference metadata. the metadata itself is external, i.e., detached from the underlying survey data. 3. methods we used a five-stage approach to collect and display the metadata (figure 1). ethics approval for this project was granted by the melbourne school of population and global health human ethics advisory group (heag) – ethics id: 1852885.2. stage one involved identifying relevant sample surveys to include. stage two centred on indexing the survey data. stage three comprised scraping the metadata for each survey in the index file. stage four included appending all the metadata for construction into a database. finally, in stage five, a webtool was developed to enable users to search key terms in the metadata database. for stage one, public repositories of australian sample survey data were searched with the following inclusion criteria: sample survey data, unit-record files, publicly available, collected in 2010 or later and data related to participants aged 45 years and over. surveys were excluded if they contained qualitative or mixed-methods data or had no available details of the data custodian. the stage one search was carried out from 17th of july 2018 until 1st of november 2018. 68 temple et al. australian population studies 5 (1) 2021 figure 1: process to develop the cepar metadata database in stage two all files that fulfilled the inclusion criteria were indexed with the following fields: file creation date, sample design, survey type, unit of analysis, sample frame, details of fieldwork, geographic sampling frame and survey description. during stage three, each survey was scraped for the following metadata: variable name, label, categories, and survey identifier. where possible, surveys were scraped using the statistical software programme r. surveys were manually scraped if they could not be scraped by r. selected .xml files were also made available from the ada. in stage four, all the scraped files were then appended using stata 15.1. this file was then used to create an exportable csv file. finally, in stage five, the csv file was uploaded to https://mspgh.unimelb.edu.au/centres-institutes/centre-for-health-policy/researchstage 1 •identification of relevant sources of surveys •inclusion criteria: sample survey data, unit-record files, publicly available, 2010 or later and data related to participants aged 45 years and over. •exclusion criteria: administrative, qualitative or mixed-methods data or had no details of the data custodian available. stage 2 •index the available surveys •all surveys were indexed with the following fields: file creation date, sample design, survey type, unit of analysis, sample frame, details of fieldwork, geographic sampling frame, and survey description. stage 3 •scrape metadata from surveys •all metadata include: variable name, label, categories and survey identifier. •statistical software programme r and manually. stage 4 •append all metadata in a unique database •all the scraped files were appended using stata 15.1. this file was then used to create the database. stage 5 •development of the website/repository •the metadata database was created and linked to a website interface with simple search functions. https://mspgh.unimelb.edu.au/centres-institutes/centre-for-health-policy/research-group/metadata-database australian population studies 5 (1) 2021 temple et al. 69 group/metadata-database. this simple webtool developed by the university of melbourne allows users to search and export the metadata. the principal repositories identified during stage one of this study were: • australian bureau of statistics (abs). the abs holds a very detailed collection of microdata, spanning microdata datasets from the 1970s through to the present day. microdata can be accessed as a basic curf (confidentialised unit record file) through abs microdata download, or more detailed microdata can be obtained through the abs datalab. selected microdata can also be analysed through the abs tablebuilder product. however, for the metadata database, the scope includes only curf and datalab products (or former radl – remote access data laboratory products). tablebuilder was excluded, because although the product is built from microdata, routine procedures used by researchers (e.g., regression models), are not possible with this product. the url for abs microdata collections, including details for access is: https://www.abs.gov.au/websitedbs/d3310114.nsf/home/microdata+entry+page. • australian data archive (ada). the ada, situated at the australian national university, holds over 6,000 datasets from 1838 onwards. over ,1500 datasets are available within the ada dataverse for users to apply for access. the data collections cover a range of topics including health, social attitudes, and electoral and opinion surveys among others. the household, income and labour dynamics in australia (hilda) and ten to men surveys are also available through the ada. the url for ada data collections, including details for access, is: https://ada.edu.au/. included surveys from the abs repository had a minimum of ‘expanded curf’ or ‘microdata’ available and totalled 46 surveys and 14,011 variables. thirty-five surveys were extracted from the ada repository, totalling 10,457 variables. hilda (71,014 variables available) and ten to men surveys were included in the ada repository. hilda metadata were sourced from the publicly accessible hilda data dictionary (https://www.online.fbe.unimelb.edu.au/hildaodd/default.aspx). the australian longitudinal study on women's health (alswh) included 13,174 variables from 8 waves. details of the repositories, including the number of sample surveys (upper panel) and variables (lower panel) can be found in table 1. from 86 indexed files, 49.4% were cross-sectional, 29.4% repeated cross-sectional and 11.8% longitudinal surveys. thirty-five (39.3%) surveys have ‘individual’ level as the unit of analysis and almost all were ‘national’ level surveys. 4. example of the database in this section we detail a simple example from the metadata database. the landing page for the metadata database is found here: https://mspgh.unimelb.edu.au/centres-institutes/centre-forhealth-policy/research-group/metadata-database. from the landing page (figure 2), users can access information on: (1) all surveys included in the metadata database (click on data sources); (2) the research team (click on about us); (3) outputs arising as part of this project (click on outputs); (4) details on funders of the project (click on supporters). by selecting ‘the database’ button, users can choose to download a user guide (this paper) or launch the database. https://mspgh.unimelb.edu.au/centres-institutes/centre-for-health-policy/research-group/metadata-database https://www.abs.gov.au/websitedbs/d3310114.nsf/home/microdata+entry+page https://ada.edu.au/ https://www.online.fbe.unimelb.edu.au/hildaodd/default.aspx https://mspgh.unimelb.edu.au/centres-institutes/centre-for-health-policy/research-group/metadata-database https://mspgh.unimelb.edu.au/centres-institutes/centre-for-health-policy/research-group/metadata-database 70 temple et al. australian population studies 5 (1) 2021 table 1: summary of ageing metadata australian bureau of statistics (abs) australian data archive (ada) household, income and labour dynamics in australia (hilda)1 australian longitudinal study on women's health (alswh) ten to men1 total number of surveys # surveys/files 46 35 1 1 2 85 # waves 17 8 2 survey type census 8 (17.4%) 0 (0.0%) 0 (0.0%) 0 (0.0%) 0 (0.0%) 8 (9.4%) cross-sectional 11 (23.9%) 31 (88.6%) 0 (0.0%) 0 (0.0%) 0 (0.0%) 42 (49.4%) longitudinal 2 (4.4%) 4 (11.5%)2 1 (100.0%) 1 (100.0%) 2 (100.0%) 10 (11.8%) repeated crosssectional 25 (54.4%) 0 (0.0%) 0 (0.0%) 0 (0.0%) 0 (0.0%) 25 (29.4%) survey unit of analysis business 6 (13.0%) 0 (0.0%) 0 (0.0%) 0 (0.0%) 0 (0.0%) 6 (7.1%) individual 1 (2.2%) 33 (94.3%) 0 (0.0%) 1 (100.0%) 0 (0.0%) 35 (41.2%) individual and other levels 4 (8.7%) 2 (5.7%) 1 (100.0%) 0 (0.0%) 0 (0.0%) 7 (8.2%) household 1 (2.2%) 0 (0.0%) 0 (0.0%) 0 (0.0%) 2 (100.0%) 3 (3.5%) private dwellings 34 (73.9%) 0 (0.0%) 0 (0.0%) 0 (0.0%) 0 (0.0%) 34 (40.0%) number of variables # variables 14,011 10,457 71,014 13,174 1,645 110,301 survey type census 1,837 (13.1%) 0 (0.0%) 0 (0.0%) 0 (0.0%) 0 (0.0%) 1,837 (1.7%) cross-sectional 3,666 (26.2%) 7,923 (75.8%) 0 (0.0%) 0 (0.0%) 0 (0.0%) 11,591 (10.5%) longitudinal 128 (0.9%) 2,532 (24.2%)2 71,014 (100.0%) 13,174 (100.0%) 1,645 (100.0%) 88,493 (80.2%) repeated crosssectional 8,380 (59.8%) 0 (0.0%) 0 (0.0%) 0 (0.0%) 0 (0.0%) 8,380 (7.6%) survey unit of analysis business 484 (3.5%) 0 (0.0%) 0 (0.0%) 0 (0.0%) 0 (0.0%) 484 (0.4%) individual 105 (0.8%) 7,704 (73.7%) 0 (0.0%) 13,174 (100.0%) 0 (0.0%) 20,983 (19.0%) individual and other levels 2,584 (18.4%) 2,756 (26.3%) 71,014 (100.0%) 0 (0.0%) 0 (0.0%) 76,351 (69.2%) household 780 (5.6%) 0 (0.0%) 0 (0.0%) 0 (0.0%) 1,645 (100.0%) 2,425 (2.2%) private dwellings 10,058 (71.8%) 0 (0.0%) 0 (0.0%) 0 (0.0%) 0 (0.0%) 10,058 (9.1%) notes: (1) ‘household, income and labour dynamics in australia survey – hilda’ and ‘ten to men survey’ were available in ada’s repository. (2) the 3 waves of the “building a new life in australia (bnla): the longitudinal study of humanitarian migrants” are available in 4 files at ada. australian population studies 5 (1) 2021 temple et al. 71 figure 2: metadata database on ageing landing page when the ‘launch the database’ option is selected, the following screen of the webtool is loaded as a new tab (figure 3). the key variables included in the database are as follows: variable: the original variable name provided in the source documentation. label: the description of the variable provided in the source documentation. categories: the items available in categorical variables. ds_label: name of the dataset. ds_format: dataset format. ds_link: url to find further information about the dataset, including contact details for the data custodian. ds_design: the type of survey sampling design, e.g., stratified, random sample. ds_type: whether the survey is cross-sectional, repeated cross-sectional, or longitudinal. ds_unit: unit of analyses of the survey instrument, e.g., individual, households. ds_samp: details of the survey sampling design. ds_desc_fw: simple fieldwork details where available. ds_source: where the survey is sourced from, e.g., abs, ada. 72 temple et al. australian population studies 5 (1) 2021 figure 3: metadata database webtool a number of placeholder variables (e.g., ds_name, _merge) are included for administrative purposes or for potential updates in the future. to simplify the display, click ‘select columns’. ensure variable, label, categories, ds_label and ds_link are selected (highlighted in blue). the remaining variables should be deselected or highlighted in grey (figure 4). click ‘apply’. figure 4: selection of variables in the metadata database webtool. for the purposes of this example, let us imagine a researcher is interested in investigating measures related to disability. the user can either select ‘global search’ or ‘field search’. a global search will search for the term across all variables in the database. a field search enables the user to specify the search term for single or multiple variables in the database. to conduct a global search, simply enter australian population studies 5 (1) 2021 temple et al. 73 ‘disability’ in the text box and click the magnifying glass icon. this produces 3,496 records containing disability (figure 5). as an example, the first record is from the disability, ageing and carers survey, with variable name disbstat, variable label ‘disability status’ with the link to dataset information provided in the ds_link column. as this variable is a categorical variable, the column ‘categories’ shows each category of disability considered. if the user wishes to download the results of the search, simply click ‘download all matched results in a csv file’. figure 5: example global search for the term ‘disability’ to clear the search, click ‘clear search’, delete the ‘disability’ text in the global search text box and click the magnifying glass icon. this will reload the full database. the user may only be interested in measures which include disability in the ‘label’ and ‘categories’ column. to complete this search, type ‘disability’ in the field search text box. select ‘label’ from the adjacent combo-box. next, click ‘add field’. again, enter ‘disability’ in the new field search text box. select ‘categories’ from the adjacent combo-box. clicking ‘search’ narrows the number of records from 3,496 to 146 (figure 6). figure 6: example field search for the term ‘disability’ in ‘label’ and ‘categories’ columns clicking ‘reset field’ followed by ‘search’ will once again reload the full database. with the full database reloaded, the user can click ‘download all matched results in a csv file’ to download the full database. the ‘select columns’ option may need to be reset if the user wishes to download all variables in the database, by clicking ‘select all’ followed by the ‘apply’ button. to exit the metadata database, simply close the tab. 74 temple et al. australian population studies 5 (1) 2021 5. conclusions in this paper, we have reported on our proof-of-concept project to develop a metadata database for researchers working in the field of ageing. our tool, alongside other metadata collections as they become more comprehensive, will assist with: • making unknown datasets ‘discoverable’ by researchers; • linking researchers to data custodians to facilitate data access; • increasing researcher productivity by providing a one-stop-shop to understand measures collected of interest to researchers in ageing; • assisting with the preparation of grants by pointing to all existing data on a specific topic in a very efficient manner; • assisting with understanding the potential for data linkage (when legal and/or feasible); and • facilitating new research conversations across disciplines. in short, metadata transparency can help empower researchers to identify what data exists around a given topic (e.g., a specific research interest in ageing), how the data was collected, and how to gain access to the data. understanding the structure, limitations, definition, and description of data also serves to protect against misinterpretation or misuse, thereby ensuring the integrity of the research output. metadata itself may also be of interest to academics, as a means to understand the historical evolution of these measures and point to gaps in our knowledge. understanding data collections requires analysis of taxonomies of variables and their inter-relationships, theoretical frameworks, and research methodologies employed in ageing research. thus, collecting metadata can help inform an epistemological map of the ‘state of play’ in australian gerontology. in doing so, the extraction and analysis of metadata can play a critical role in the evolution of measures relevant to ageing research. with an increasing ageing population and reliance on data, metadata on ageing has growing relevance now and into the future. 5.1. limitations our project considered a very specific scope of data for inclusion: sample survey data available at two major data repositories between the period 2010 to 2018. many other datasets are likely to be of interest to researchers in the field of gerontology and the demography of ageing. for example, large administrative datasets, surveys held by research institutions or researchers, qualitative or mixed-methods data, spatial databases as well as vital statistics. these important data sources are omitted from our project. moreover, our search was carried out from 17th july 2018 until 1st november 2018. other datasets housed at both the abs and ada are likely to be in-scope but published online after november 2018 and therefore not included in the metadata database for this study. a second major limitation of this project relates to data quality. collating even this range of metadata was a considerable task, with data dictionaries available in different formats (excel, word processing documents and pdf files). although our procedures automated through r and stata helped to streamline the process, many errors had to be manually corrected. given the size of the database, errors are likely to remain and users of the metadata database should note this limitation. a further limitation is that the webtool is a relatively simple search tool. nonetheless, users are given the option to download the full database as a *.csv file to use with other software. australian population studies 5 (1) 2021 temple et al. 75 finally, a limited number of variables have been added to the database. curating the metadata to include conceptual, theoretical or analytical methodologies would greatly improve the coverage of the database. for example, tagging variables from each survey which are of use to researchers working on particular ageing theories, or tagging variables with key areas of thematic interest from domestic and international conferences. 5.2. summary and future research in this paper we have described a proof-of-concept project which seeks to develop a metadata database including metadata on australian sample surveys relevant to ageing. we have outlined our five-stage approach to create the cepar metadata database on ageing and provided a webtool for users to interrogate the data, or to export the data for further use. nonetheless, improvements can be made with regard to coverage of data sources (to include for example administrative data), temporal coverage (increasing the time period and spatial heterogeneity of data collections), and procedures to improve the underlying data quality of the database – when for example, only pdf or word processor documents are the only public source of microdata data dictionaries. beyond these improvements, research using the metadata database itself has strong academic merit. using metadata, such queries may include: • the availability of measures of interest to gerontologists in australian datasets. • the complementarity of evolving theories and conceptual frameworks with data availability. • the historical evolution of measurement. • the strengths of current data collections. • identification of weaknesses and how they can be improved in future data collections. improving our knowledge through metadata can limit misinformation and misuse of data, avoid duplication of research effort while simultaneously enhancing reliability and integrity of research outputs. such important and wide-reaching outputs include the creation of informed social policy to improve the lives of older australians, both current and future. key messages • this introductory guide explains the concept of metadata and its utility to researchers specifically in the field of ageing. • the methodology for creating the metadata database is described and explained. • an example of searching and extracting data using the cepar metadata database webtool is described. • researchers are provided with full access to the underlying database and a webtool to interrogate the database. acknowledgements funding for this project was provided by the arc centre for excellence in population ageing research (cepar) and the melbourne school of population and global health, university of melbourne. the authors acknowledge the assistance of sebastian kocar, data archivist from the 76 temple et al. australian population studies 5 (1) 2021 australian data archive. we also acknowledge the development work of prof. richard sinnott, glenn jayaputera, zelong cong, and alwyn pan on the webtool. references australian association of gerontology (2017) submission to aged care workforce strategy taskforce. https://www.aag.asn.au/news-publications/aag-submissions/2017-1-aag-submission-to-agedcare-workforce-strategy-taskforce. eurostat (2019) glossary: metadata. https://ec.europa.eu/eurostat/statisticsexplained/index.php/glossary:metadata#:~:text=reference%20metadata%20(sometimes%20c alled%20explanatory,a%20semantic%20point%20of%20view. accessed on 4 february 2021. gilliland a j (2016) setting the stage. in: baca m (ed.) introduction to metadata. third edition. los angeles: getty publications. https://www.getty.edu/publications/intrometadata/setting-thestage/. accessed on 15 november 2018. niso [national information standards organisation] (2004) understanding metadata. https://www.niso.org/publications/understanding-metadata. accessed on 14 november 2018. united nations statistical commission and economic commission for europe (2000) guidelines for the modelling of statistical data and metadata. geneva: united nations. https://unece.org/fileadmin/dam/stats/publications/metadatamodeling.pdf. https://www.aag.asn.au/news-publications/aag-submissions/2017-1-aag-submission-to-aged-care-workforce-strategy-taskforce https://www.aag.asn.au/news-publications/aag-submissions/2017-1-aag-submission-to-aged-care-workforce-strategy-taskforce https://ec.europa.eu/eurostat/statistics-explained/index.php/glossary:metadata#:~:text=reference%20metadata%20(sometimes%20called%20explanatory,a%20semantic%20point%20of%20view https://ec.europa.eu/eurostat/statistics-explained/index.php/glossary:metadata#:~:text=reference%20metadata%20(sometimes%20called%20explanatory,a%20semantic%20point%20of%20view https://ec.europa.eu/eurostat/statistics-explained/index.php/glossary:metadata#:~:text=reference%20metadata%20(sometimes%20called%20explanatory,a%20semantic%20point%20of%20view https://www.getty.edu/publications/intrometadata/setting-the-stage/ https://www.getty.edu/publications/intrometadata/setting-the-stage/ https://www.niso.org/publications/understanding-metadata https://unece.org/fileadmin/dam/stats/publications/metadatamodeling.pdf australian population studies 2022 | volume 6 | issue 2 | pages 27-32 © johnstone 2022. published under the creative commons attribution-noncommercial licence 3.0 australia (cc by-nc 3.0 au). journal website: www.australianpopulationstudies.org apa conference demographers in a time of covid kim johnstone apa president email: dr.kim.johnstone@gmail.com. address: c/o australian population association, school of demography, australian national university, 9 fellows road, acton act 2601 received 29 november 2022; accepted 30 november 2022; published 19 december 2022 this is an edited version of the apa presidential address given by dr kim johnstone, apa president, at the 20th australian population association in canberra on 23 november 2022 1. introduction i would like to begin by recognising the ngunnawal and ngambri peoples who are the traditional owners of the land that we meet on today, land that was never ceded. i pay my respects to elders past and present, and pay my respects to aboriginal peoples from other country who are here today. as demographers, we recognise the intricate links between people and place over time – into the future and back to the dreaming. and i think it’s important we recognise these links, especially as australia talks about formalising this recognition through a voice to parliament. i would like to start by saying how fantastic it is to finally be able to hold this conference in person. in 2020 when ann evans was president, it felt like a gift to have on-line sessions about demography, but i have been ridiculously excited at the thought of seeing you all, my friends and colleagues in person again. and excited to be able to talk about population with people who get it, and who know what a cohort effect is, and who know what nom [net overseas migration] means! i can’t go any further without acknowledging that this conference wouldn’t have happened without a fantastic organising committee headed by edith gray with denise carlton, brian houle, james o’donnell and ray harris. but a particular thank you must go to edith whose work behind the scenes has been so important. i must admit it’s quite a daunting task writing a president’s address, particularly after the past couple of years we’ve all had! when i sat down to prepare, i looked to past presidents for inspiration (apa presidents, not trump). i’ve sat through several president’s addresses based on the research they’re doing, but i don’t have any research to present. when alison taylor was president, her presidential address in hobart in 2018 started with photos of her travels through europe and how demography was everywhere. http://www.australianpopulationstudies.org/ mailto:dr.kim.johnstone@gmail.com 28 johnstone australian population studies 6 (2) 2022 2. the effect of covid my first year as president had a huge lockdown in the middle of it – 107 days, not that we were counting, and i know – not as long as melbourne! and outside of lockdown our state borders were closed so there was no going anywhere, not even wa. so in my family, every saturday we picked a country, and every family member cooked a course for dinner, and we dressed up to represent that country (figure 1). it really was the highlight of our week. what strikes me when i look at the countries we picked is that we didn’t have to travel at all to see demography everywhere. every country we picked was based on our friends and where they are from. and i’d say, like many households, this range of cuisines is actually a diversity we see most weeks – either through the food we cook or buy from our local restaurants. while all covid restrictions have been lifted, we are still in the midst of a global pandemic and it exposes us all to risks of morbidity and death. we can’t escape the fact that the last couple of years have been frankly horrendous – family members stuck overseas, zoom funerals, mental illness, redundancy. and that’s just me. figure 1: johnstone family dinner outfits source: kim johnstone but, of course, everyone was affected. we know that everyone’s lives were disrupted and disrupted in ways that will see continued impacts for evermore. we are already talking about the covid generation – the group of young people who were on the cusp of adulthood – about to leave school, get their first job, leave home for the first time, travel overseas when the pandemic hit. what was i saying about cohort effects? we have good data on what happened over the past couple of years – with one in five australians experienced high or very high levels of psychological distress in 2020 and 2021, higher levels than pre-pandemic. we saw levels of household work and childcare increase. we saw incomes go up for those reliant on government pensions, and go down for people whose hours were cut but still working, and stay the same but with more hours worked for people able to work from home. australian population studies 6 (2) 2022 johnstone 29 we know that young people were disproportionately affected by higher levels of psychological distress, and by higher unemployment. women suffered greater stress and anxiety than men, were more likely to lose hours of work, and shouldered much of the responsibility for home schooling. there were notable spatial differences in how people experienced pandemic restrictions. i live in one of the harsh lockdown local government areas of sydney – what we called the plague zone in our house, an lga with lower socio-economic status, and a large population who had a very different experience to the northern beaches [of sydney] who were locked down at the end of 2020. our work as demographers and population specialists is going to be critical in ensuring that not only that we understand the impacts of covid and the inequitable impacts of covid-19 at the time, but also as they play out over the next 5, 10, and 20 years. i believe our strength as interdisciplinary researchers means we will be key contributors to understanding the longer-term effects of this pandemic across social demography, life course research and the impact on the key drivers of population change – mortality, fertility and migration. not only will covid affect our population for years to come, i’m picking it will be a key conference theme at our apa conferences for some time yet. 3. covid and demography one of the things covid has done for our profession is put population front and centre of people’s minds. it started when australia went into its first lockdown, when the international border was closed, and when unemployment lines wound around street blocks. everyone wanted to know what it meant! i know several of us in this room were on radio, tv and in print talking about what was happening and the likely impacts. it really emphasised how much we, as demography professionals, had something useful and important to say about what was happening. demographers were certainly flavour of the month. we benefited from the focus on epidemiology, being able to interpret data and what it meant for populations. it was a real joy to see younger demographers filling our screens and airways to talk about population issues. and having just seen the world population hit 8 billion, i was very happy seeing liz allen and elin charles-edwards on my tv screen to talk about this. for me, i found (and continue to find) that the challenge in responding to this media interest in population is a focus on short-term impacts, when a lot of what we do in relation to demographic dynamics is slow moving and less responsive. it’s particularly difficult when you get 30 seconds to convey a message. i kept coming back to two key messages, that still hold true. the first is – age is important. you have to understand the age of the people – and this applies to everything – the age of the people in this place, the age of the people who are moving, the age of the people who are not moving, etc. we know this as demographers. but it is the hardest thing i have to convey to the nondemographers i work with. the second message was – even though we’re seeing changes in some parts of our communities – most people are not moving, and most people’s circumstances are not changing. all the people we need to be thinking about are here and will still be here next year. and here in 10 years. 30 johnstone australian population studies 6 (2) 2022 i have noticed that now that our borders are open and restrictions lifted, there’s less interest in talking to demographers. and indeed, because of inflation rates and rba interest rate rises, economists are once again the experts most likely to be called upon. while i don’t begrudge skilled professionals being recognised for their expertise, i do get cranky when economists do our work. this is mainly because they often get it wrong – ignoring age structure and basic demographic processes. i am mighty sick of explaining that a bigger household is not a measure of lack of housing supply, or explaining that you cannot replace overseas one year ago in the census with nom data and then expect it to add up to erp [estimated resident population]. while i joke about economists, i do think that demographers need to find ways to improve how we communicate about the impacts of the pandemic and changing age profiles on our communities. this is a key issue that policymakers need to understand, and our politicians, who are not renowned for a long-term view, also need to understand. the population frameworks that we work within, without even thinking about it, are i think the most reliable tools we have to consider the wide range of factors at play. we demographers also need to be talking about the distinction between developments that are caused by covid-19 and those that are based on our population structure. this migration exodus out of sydney, for example, had an age profile of people who moved which was exactly the same as those people who left before covid. fertility rates were already going down. 4. covid and new data one of the things that has been a real boon to us as demographers since the onset of covid has been new data. and i know that we have several papers on the program at this conference looking at this very topic. a huge thank you to the abs who stepped in with more regular release of key data products to help understand the covid impacts. when i was working as a consultant they were the main source of data we used to inform clients in local government about what might happen in their communities. we saw several nation-wide surveys administered – the abs household impacts survey, the australian institute of family studies, families in australia survey, the melbourne institute taking the pulse of the nation survey, monash university did a survey and the national children’s commissioner. we’ve seen surveys about covid impacts from anu, asthma australia, architects australia … i don’t have an exhaustive list. but it’s clear we have a wealth of insights that will help our work well into the future. one of the exiting developments was the exploration of new data to fill data gaps. there is a poster [at the 2022 apa conference] from my colleagues that i encourage you all to see about how we used drivers license change of address data in nsw to understand population mobility. what was striking was that the pandemic exposed data gaps that we have long been aware of – in this case, local moves. the need to understand these moves so government could prepare appropriate pandemic responses has set up processes that have enabled data sharing, and strengthened an evidence base that we need not just for pandemics, but for all service and infrastructure planning. australian population studies 6 (2) 2022 johnstone 31 a note of caution though – i am excited by the new data, but none of it is perfect. population data gaps were exposed that remain problematic. data on temporary populations is one such gap. how do we measure people when they don’t live in a place permanently but are there for a long time, or who are living in a second residence for part of the year? this is particularly important across regional australia where the usual resident population doesn’t change but there is not enough housing or water infrastructure to meet workers who are in town. i know that addressing these and other data challenges is going to be a big part of the work of our statistical agencies here and around the world, as well as dealing with the impact of covid on our data sources that we have relied on and which may need to be looked at with more caution for at least the next few years. 5. key challenges so thinking about our theme of population and policy, i have been thinking about what are the challenges that i see for us as demographers, those things that are challenges for me in my day job. i hope they are relevant to all our work. first, there is the challenge in needing exact numbers when planning infrastructure, sometimes 60 years into the future, and managing uncertainty. we’re currently working with people putting together a budget bid for a new hospital in an area that is not projected to grow based on the changes to nom. but if we are wrong, and a new hospital is not built, we will have service gaps in another 20 or 30 years’ time. why this matters is because for those people asking for the precise numbers – when we can’t do it – the value of the demographic process, or our own professional capability, is questioned. often the policy response is to throw more data at the challenge, or more complex models – which doesn’t necessarily deliver better outcomes. the challenge to us as demographers is how to communicate those things we can be certain of – age structure for example – and explore ways to narrow some of the uncertainty gap. we also need to engage in the policy responses so they make sense. my latest talk was to a school planning conference talking about projected populations of children, and a key takeaway was the need for multi-functional space as we prepare for waves of children alongside population ageing. but this response will need legislative changes for how public schools can be used, designed, and funded. the second challenge is translating research into an outcome policy makers can use. for my current work, and for the work i did in the private sector, this often means understanding what outcomes mean at a local level. if we want to see evidence informing policy, part of the work needs to include identifying which policy lever the evidence applies to and by which level of government – state, federal or local. evidence also needs a translator – i’ve sat down with many planners and policy makers who are excited by a research idea but don’t know what to do with it. and, third, we can’t talk about challenges without the issue of how research is funded. i know many of you faced job uncertainty at the height of lockdowns when university funding was severely constrained, and this continues with short term contracts and no ongoing funding streams. i know from government it is very hard to be an industry partner for arc grants when you only get an annual budget allocated and experience ministry of government changes, and a new secretary. it’s certainly hard as a public servant to fund research we don’t know the outcome of because the first 32 johnstone australian population studies 6 (2) 2022 question is: what do we get out of it? i’m certainly hopeful that the new australian universities accord is useful in bettering what we see in this space. the fourth challenge is about aboriginal demography and population data. it’s been 10 years since i did my phd on indigenous fertility in the nt and i feel that the progress in the areas of indigenous data sovereignty, increasing the talent pool of aboriginal data experts, and ensuring communities use data to make sure they get the services they need hasn’t made much progress. i am alarmed by the glacial pace of change, and i find it disappointing that when i talk about aboriginal data and population data, i’m mostly talking with other whitefellas. i know there are a lot of people in this room trying to do more in this space, but the data challenges are significant and i think ensuring we as a professional organisation support both closing the gap data sovereignty initiatives and making sure we train more aboriginal people to be in the room when the work is done has to be a priority. i don’t have solutions to these challenges, and i know many of them are part of the amazing conference program we have the privilege to be immersed in this week. but what i do know is that this professional group we are part of through the apa, and our network of demographers and population analysts that goes beyond the association is collegial and as a collective we are well placed to address these challenges, spanning as we do public, private and academic sectors. when i was thinking about how we address these challenges as a group, i remembered all the people who have given me ideas, helped me understand different methods (very patiently), edited my work, over the years. this year the man who is the reason i am a demographer, ian pool, died. and i remember him with such fondness because of his generosity of spirit in promoting me and supporting me, and everyone he worked with. i was thinking of ian because we’ve lost other demographers over the past two years, last year iwu utomo, and this year gavin jones, who we are honouring at tomorrow morning’s plenary. what strikes me in remembering these demographers is how for each of them, it is their generosity with their intellect, support and kindness to newcomers that is talked about, as well as the work they left behind. and i think this is the thing that makes our family of demographers, and the australian population association, such an invaluable professional network to be part of and well placed to address the challenges we face in understanding australia’s population, and those of our neighbours and around the globe. we are very good at what we do, and we support each other. it’s why i think we will continue to do amazing work and address the challenges we face from surviving a pandemic, to understanding what it means for australia’s population and those across the world. i would like to close by saying thank you to everyone who is here today – this conference wouldn’t happen without you. thank you to all apa members, the long-standing members and new ones. and a thank you to the apa council – all our work is voluntary and frankly, that’s been hard yakka the past couple of years. thank you. australian population studies 2021 | volume 5 | issue 1 | pages 56-64 © allen 2021. published under the creative commons attribution-noncommercial licence 3.0 australia (cc by-nc 3.0 au). journal website: www.australianpopulationstudies.org commentary understanding ethnicity in contemporary australia using the census liz allen the australian national university email: liz.allen@anu.edu.au. address: centre for social research and methods, the australian national university, acton, act 2601, australia. paper received 22 february 2021; accepted 6 may 2021; published 31 may 2021 1. introduction australia is amongst the most ethnically and culturally diverse countries in the world. despite this diversity, little is understood about how australians identify with their ethnic heritage. data collected on a national level via the five-yearly census of population and housing do not adequately capture the complexity and richness of australia’s ethnic diversity. this inadequacy could be contributing to the lack of diversity and parity representation of the population in politics, business, and decision-making. in addition, there are benefits for individuals, business, and planning (among others) for a more comprehensive understanding of australia’s contemporary ethnic profile. how the national enumeration could better capture contemporary ethnic identity in australia is considered here, with consideration of previous attempts at collection of such information in australia and elsewhere alongside potential improvements for the 2026 census. 2. ethnicity in the census 2.1. what is ethnic identity? “ethnic identity is defined as a sense of belonging, based on ones’ ancestry, cultural heritage, values, traditions, rituals, and often language and religion” (green et al. 2015 p. 676). in practice, ethnic identity is a self-defined concept and can be a complex articulation of one’s biological background and social experiences and practices, especially as it relates to migrant and mixed-race populations. ethnicity is further complicated by the dynamic nature of individual articulation; self-reported identification can change over time, reflecting expanding experiences (united nations 2017). in this way ethnicity, a socially constructed and individually interpreted embodiment of identity, differs from the outdated and problematic notions of race based on birthplace, heritage and/or even skin colour (horn 1987). ethnicity and culture are often politically charged and highly sensitive concepts, especially for individuals and communities identified as visible minorities from, and by, the majority population. the sensitivity and loaded nature of these terms stem from their connection to race, which in australia is http://www.australianpopulationstudies.org/ mailto:liz.allen@anu.edu.au australian population studies 5 (1) 2021 allen 57 related historically to eugenic practices of distinguishing people of white origins to others (stevens et al. 2015) and their so-called racial purity. despite the history of poor practices of collecting data concerning ethnicity in australia, there has been growing expectations and attempts to better reflect contemporary ethnic identity in data collections to best capture the diversity of the nation (allen 2020; federation of ethnic communities’ council of australia 2020). 2.2. collection of ethnicity indicators in the australian census the census is vital in informing population estimates and trends, especially among hard to reach and minority populations (united nations 2017). the value of capturing ethnic identity in the census (over collection in sample surveys) is wide ranging, but reflects the need for a whole-of-population enumeration that is also accessible to all. a whole-of-population enumeration enables data across all levels of geography to help identify populations of interest and in determining needs and services relating to such things as business and government. sample survey data has the potential to overlook important aspects of the population on the topic of ethnicity, simply because the sampling frame lacks necessary insights from the census. this circular predicament of census and sampling frames is best resolved by bringing the census into line with contemporary australian ethnic identity. horn (1987, 1993) and rocha et al. (2018) identify shortcomings of the current approach to collecting ethnicity data in the census, but instead of recommending the census be updated (like khoo 1991 does) they call for ethnicity information to be collected via sample surveys. the justification for avoiding updating the national census to include self-reported ethnic identity embodies a long-enduring fear australia has had reckoning with its history. horn (1993 p. 126) highlights the limitations of the current approach to understanding cultural diversity in census, but rather than making recommendations for improvements argues australia may not be ready for a discussion about identity, saying such a move is likely to “evok[e] ancient ghosts”. khoo (1991) however argues that australia needs to confront its history. the census and its related education campaign could be a way forward for the nation in reckoning with its past and looking to the future, through the lens of ethnicity. the measurement of ethnic diversity can be methodologically and socially fraught at the best of times. but the environment in which the national statistical agency operates includes the legacies of brutal british colonisation, the discriminatory white australia policy, and enduring xenophobia. the endeavour for meaningful data to best reflect the diversity of the australian population must be understood in this complex and complicated context. the willingness of people (especially those from minority groups) to want to identify their ethnic identity should also be carefully weighed with the perceived national social benefits. a range of questions designed to capture information relating to cultural background and diversity are asked as part of the australian national population and housing census, these include: indigenous status, citizenship status, country of birth, year of arrival for people born overseas, country of parent’s birth, language other than english spoken at home, english proficiency, religion, and ancestry. these data items together are used to construct a picture of the characteristics of australians based on cultural backgrounds (khoo 1991). most commonly however, country of birth and language are used as surrogate measures for, and ancestry as a direct measure of, ethnicity or ethnic identity (horn 1987, 1993). 58 allen australian population studies 5 (1) 2021 the surrogate measures of country of birth and language, which serve as indicators of ethnicity, have been included in the australian census since 1911 and 1986 (respectively) (abs 2020; horn 1987). these data, coupled with country of birth of parents are popularly used, although they are limited to two generations of reach in detail (khoo 1991). for example, the ethnic identity of someone who identifies as asian australian would be missed from the full picture of australian’s diversity where the individual speaks english only and they and their parents were born in australia. this would be an increasingly more common phenomenon, particularly as time passes from the early waves of migration, and as mixed partnering and multiracial births increase (rocha et al. 2018). while country of birth, language, and religion all provide vital details for provision of needs and services for example, these surrogate measures are inadequate for reflecting ethnic identity (horn 1993). the inappropriateness of these surrogate measures increases with generations. moreover, it is not just the understanding of diversity that benefits from more comprehensive and contemporary measures of ethnicity; business, local governments, individuals, and communities all stand to benefit. the australian census first included a direct question on ancestry in 1986 (abs 2020), at a time when immigration from asian countries overtook that from european nations for the first time in australia’s more recent history (jupp 2001). prior to this ancestry question, the australian census included a question about race origins (horn 1987, 1993). the inclusion of the ancestry question was the result of research into the best way to gauge ethnicity in australia. in 1982, the australian statistician appointed a committee to evaluate the collection of ethnicity data in the 1986 census. headed by w.d. borrie, the committee argued the need to move away from a racebased question to the inclusion of so-called surrogate measures of ethnicity (for example, the suite of questions presently in the census) and or a self-identified direct question on ancestry (abs 1984). the committee oversaw a series of pilot tests covering a number of potential options, including both direct ancestry and ethnicity questions. the committee recommended “maximising the level and quality of responses to any ethnicity question, and of minimising any adverse reaction to such a question” (abs 1984 p. 68). in doing so, the committee opted for simplicity over accuracy. the final 1986 question asked about each person’s ancestry, and presented six example options: greek, english, indian, armenian, aboriginal, chinese. a dissenting committee member warned this question was too crude an approach over a self-identified ethnicity question, and would not deliver the data required by the nation’s “multicultural society” (abs 1984 p. 71). despite the best intentions, the self-reported ancestry variable in 1986 was unsuccessful in achieving its intended purpose, failing to adequately capture ethnicity among australians (stevens et al. 2015). following the initial inclusion, the ancestry question was removed from the census questionnaire in 1991 and 1996 (abs 1994). the reason for removal as reported by the australian bureau of statistics (abs) was due to the costs involved in processing the data and the data underutilisation (1994). the data simply were not used enough to justify their collection. moreover, the data were inconsistently reported, even by individuals within families and had the potential to change across time (khoo 1991). during the preparations for the 1996 census, the abs tested a more comprehensive approach to the collection of ancestry based on submissions received as part of the consultation process. data users and experts (including government departments, community groups and academics) gave evidence for the need for the inclusion of a question that adequately captured ethnicity information (abs 1994). the australian population studies 5 (1) 2021 allen 59 abs tested two questions prior to the 1996 census, with the different forms delivered to two alternate households in the pilot fieldwork. the trialled questions sought to consider whether ancestry differed to that of country of birth for respondents based on objective versus respondent self-identified information. ‘is the person's ancestry different from their country of birth?’ was asked in a more yes/no objective framing versus ‘does the person identify with an ancestry different from their country of birth?’. data from this trial (including follow up interviews) showed that ancestry was not reported where australians did indeed have non-australian heritage (abs 1994). in other words, the ancestry question failed to elicit complete and accurate data. in the 1994 report, the abs (1994) stated that “the ancestry question could not replace any of the surrogate measures of ancestry which were used for service delivery, and one key issue under debate was whether the additional information from the ancestry question warranted the additional respondent load and resources needed to collect and process the question”. citing concerns that the concept of ancestry was not appropriately and consistently understood by all, the abs recommended no ancestry question be included the 1996 census. yet in 2001 the ancestry question (deemed to be inadequate and of little use previously) was again included, allowing only two responses of ancestry presumably for the maternal and paternal lines (abs 2002). it appears australia has merely sought to avoid the issue of ethnicity altogether, but this does not mean the issues concerning ethnic identity cease to exist (perkins 2004). since its initial inclusion 35 years ago, little has changed in the way ancestry data is collected in the australian census, with only minor reordering of the top potential responses. such an approach does not enable the identification of ethnicity beyond one or two generations. the issues concerning adequacy and appropriateness of the ancestry question have not been addressed and remain today (rocha et al. 2018). there have been notable attempts to better reflect australia’s diversity in data, including work by hopkins (2017) in the context of law enforcement. that said the ethnic classification system proposed for victoria by hopkins focuses on the physical attributes of race and ethnicity, and as such is not an appropriate means to collect self-identified ethnic identity via the census in australia. the justification and identified benefits considered for a more comprehensive and nuanced question or suite of questions concerning ancestry failed to account for the empowering of individuals and representative nature of collecting such data. furthermore, australia’s increasing diversity through mixed partnering demands more contemporary data collection. 2.3. international context australia shares a similar demographic profile, and history of british colonisation and subsequent migration with a number of countries, including new zealand (nz) and canada. both the nz and canada censuses capture ethnic identity. additionally, ethnicity is collected in the census conducted in the united kingdom (uk). the censuses of nz, canada and the uk offer examples to help inform the formulation of contemporary collection of ethnic identity in australia. ethnicity has been collected in new zealand (nz) censuses since 1926, albeit crudely and prior to the 1976 census based on outdated biological concepts of race and blood purity. a significant shift, 60 allen australian population studies 5 (1) 2021 however, occurred in the 1986 census questionnaire when self-identified ethnicity was first asked (cormack 2010). the initial 1986 ethnicity question resembles that asked on the 2018 nz household form (see table 1), where respondents are asked to indicate all the ethnic groups they belong to. a major flaw of the nz ethnicity census question is that it does not reflect more recent, and diverse, migration flows (statistics new zealand 2001). a report by the nz statistics agency found the narrow ethnic groupings failed to adequately reflect the contemporary population and indeed led to data inconsistencies over time. table 1: ethnicity census question, new zealand 2018 english form question: which ethnic groups do you belong to? select all that apply to you. new zealand european maori samoan cook islands maori tongan niuean chinese indian other, eg. dutch, japanese, tokelauan source: statistics new zealand (2018) the office for national statistics (ons) has collected ethnicity in censuses undertaken in the united kingdom (uk) since 1991 (barton 2020). the collection of ethnicity in england comprises five main groups, each with further specified groupings (table 2). respondents choose one broad group and then select one of the more detailed groupings within. the english census form includes more detail than that of nz, yet restricts the number of options individuals can identify with. furthermore, the uk approach collapses or precodes options, which is not recommended by the united nations on the collection of ethnicity data (united nations 2017). nevertheless, a strength of the uk collection of ethnicity in the census is the inclusion of the mixed or multiple ethnic group. ethnicity is collected in the canadian census based on origins of respondents’ ancestors, rather than self-identified affiliation. since 1871, a question on ethnic and cultural origins has formed part of the census of population in canada in some form or another (statistics canada 2020). in the 2021 canadian census individuals are asked about the ethnic or cultural origins of this person’s ancestors. a list of over 500 ethnic or cultural origins are included as a guide, based on commonly indicated groups. any number of ethnic or cultural origins can be reported. the canadian census question largely conforms with international recommendations for the collection of ethnicity data. australian population studies 5 (1) 2021 allen 61 table 2: ethnicity census question, england & wales 2021 household form question: what is your ethnic group? choose one section from a to e, then tick one box to best describe your ethnic group or background a white • english, welsh, scottish, northern irish or british • irish • gypsy or irish traveller • any other white background, write in b mixed or multiple ethnic groups • white and black caribbean • white and black african • white and asian • any other mixed or multiple ethnic background, write in c asian or asian british • indian • pakistani • bangladeshi • chinese • any other asian background, write in d black, black british, caribbean or african • caribbean • african background, write in below • any other black, black british or caribbean background, write in e other ethnic group • arab • any other ethnic group, write in source: office for national statistics (2020) table 3: census question on ethnic or cultural origins, canada 2021 question: what were the ethnic or cultural origins of this person’s ancestors? ancestors may have indigenous origins, or origins that refer to different countries, or other origins that may not refer to different countries. for examples, refer to this list of ethnic or cultural origins. specify as many origins as applicable. source: statistics canada (2020) 3. proposed census question development of a census question seeking data on ethnicity must recognise the potential for, and concerns relating to, racial profiling and use of the data to cause harm. there is a persistent fear, particularly among marginalised groups, that census data might be used to identify and discriminate against minority and vulnerable groups (price 2016). the identification and maltreatment of jewish people in nazi germany (luebke & milton 1994) or internment of japanese people in the united states (anderson & seltzer 2007) are often examples used to illustrate harms. while no guarantees can be made to ensure ethnicity data (or census data more broadly) is not misused in the interpretation and use by the public, the proposed census additions do not make people more vulnerable to discrimination than the current census questions. the legislative safeguards of the census and statistics act 1905 and the privacy act 1988 are two measures which go some way to mitigate potential misuse of census data by setting out the limited uses of unit record data (abs 2021). https://www12.statcan.gc.ca/census-recensement/2021/ref/98-20-0002/982000022020001-eng.cfm#a7 62 allen australian population studies 5 (1) 2021 the risk of oversimplification of census items relating to ethnicity, on the other hand, should also be avoided to prevent any potential for promoting social fractures or othering of people. for example, collapsing data potential by precoding and lumping options together can lead to minimising identity (united nations 2017). it is thus vital that any ethnicity question for census be formulated with expert (including advice from advocacy groups) and user input to ensure appropriateness. testing of such a question is fundamental to ensure robust data while balancing the ethical implications of collecting ethnic identity information. ethnicity in contemporary australia could be understood by employing a question in census similar to the framing of that used in the nz census with the response options like that used by statistics canada. such an approach would draw on a familiarity among respondents with the way in which ancestry has been previously collected in australia, minimising reactivity and misunderstanding. table 4 presents a potential solution to overcome the problems identified in the collection of ancestry and surrogate ethnicity data in the census. the proposed question seeks to collect ethnic identity by asking which ethnic groups people belong to or identify with. rather than constraining the selection of groups, it is advised that all options should be selected by respondents as they apply (united nations 2017). in the initial application of this proposed question, the options could be listed as is the case in the canadian census. the list of options could be informed by the australian standard classification for cultural and ethnic groups (abs 2019). analysis of the data could then be used to refine the list of ethnic origins. clear instructions on how to complete the proposed census question on ethnic identity, especially for those of multiracial backgrounds and identity, should be developed in collaboration with a range of stakeholders (including experts, interest groups, and data users). table 4: proposed census question on ethnic identity, australia 2026 census question: which ethnic groups do you belong to/identify with? select all that apply to you. for examples, refer to this list of ethnic or cultural origins. specify as many origins as applicable. this proposed question does not seek to replace any questions relating to first nations status; any ethnicity question complements such information. there are likely interconnections between an ethnicity question and that of determining indigeneity, but as the current census questionnaire attests these questions can coexist on the national enumeration form. changes to ethnic identity information collected in the census is best informed by stakeholders comprised of first nations elders. 4. discussion collection of ethnicity or ethnic identity data has posed great challenges for considerable time in australia, despite the immense importance of such information. the collection of surrogate and direct measures of ethnicity in the national census has proven inadequate for completeness and contemporary applications. whether the collection of ethnic identity data is stuck in outdated notions of race and identity, being held captive to political arguments or government administration, https://www12.statcan.gc.ca/census-recensement/2021/ref/98-20-0002/982000022020001-eng.cfm#a7 australian population studies 5 (1) 2021 allen 63 it is beyond time the national census keep step with the needs and wants to truly reflect australian diversity in the national snap shot. lack of experience responding to a new question format to capture ethnicity should not be used to avoid innovation. nor should the status quo be maintained in the ancestry question because it is simply too difficult to try something else – it appears this has been the case since 2001, despite the evidence to the contrary and poor-quality data. australians are now more than ever aware of the importance of better understanding the diversity of the nation. and there is sufficient opportunity to test and prepare people via a public campaign (from the abs in collaboration with community groups and organisations) to ensure accurate and complete data. ultimately, a more comprehensive and appropriate question for the contemporary collection of ethnicity or ethnic identity is a priority for the australian census in 2026, especially if it the nation’s last traditional census undertaking. references allen l (2020) the future of us: demography gets a makeover. sydney: newsouth publishing. anderson m and w seltzer (2007) challenges to the confidentiality of us federal statistics, 1910-1965. journal of official statistics 23(1): 1-34. abs (1984) the measurement of ethnicity in the australian census of population and housing. report to the australian statistician by the 1986 population census ethnicity committee. australian bureau of statistics, canberra. abs (1994) census working paper 94/4 ancestry, 1996. https://www.abs.gov.au/ausstats/abs@.nsf/7d12b0f6763c78caca257061001cc588/a6845d677bc 8c50fca2570610032c06d!opendocument. accessed on 10 february 2021. abs (2002) 2001 census of population and housing fact sheet: ancestry, 2001. https://www.abs.gov.au/ausstats/abs@.nsf/lookup/2970.0.55.006main+features12001?open document. accessed on 11 february 2021. abs (2019) australian standard classification of cultural and ethnic groups (ascceg), 2019. https://www.abs.gov.au/ausstats/abs@.nsf/7d12b0f6763c78caca257061001cc588/8e81298ff9bf d160ca257ff1001e661e!opendocument accessed on 30 april 2021. abs (2020) 2021 census topics and data release plan. https://www.abs.gov.au/statistics/research/2021census-topics-and-data-release-plan#data-download. accessed on 10 february 2021. abs (2021) 2021 census privacy statement. https://www.abs.gov.au/census/planning-2021census/privacy. accessed on 30 april 2021. barton c (2020) preparing for the 2021 census (england and wales). house of commons library briefing paper, number 8531. 22 october. https://commonslibrary.parliament.uk/research-briefings/cbp8531/ accessed on 19 may 2021. cormack d (2010) the practice and politics of counting: ethnicity data in official statistics in aotearoa/new zealand. wellington: te rōpū rangahau hauora a eru pōmare. https://www.otago.ac.nz/wellington/otago600095.pdf. accessed on 10 february 2021. federation of ethnic communities’ council of australia (2020) if we don’t count it…it doesn’t count: towards a consistent national data collection and reporting on cultural, ethnic and linguistic diversity. https://fecca.org.au/wp-content/uploads/2020/10/cald-data-issues-paperfinal2.pdf. accessed on 11 february 2021. https://www.abs.gov.au/ausstats/abs@.nsf/7d12b0f6763c78caca257061001cc588/a6845d677bc8c50fca2570610032c06d!opendocument https://www.abs.gov.au/ausstats/abs@.nsf/7d12b0f6763c78caca257061001cc588/a6845d677bc8c50fca2570610032c06d!opendocument https://www.abs.gov.au/ausstats/abs@.nsf/lookup/2970.0.55.006main+features12001?opendocument https://www.abs.gov.au/ausstats/abs@.nsf/lookup/2970.0.55.006main+features12001?opendocument https://www.abs.gov.au/ausstats/abs@.nsf/7d12b0f6763c78caca257061001cc588/8e81298ff9bfd160ca257ff1001e661e!opendocument https://www.abs.gov.au/ausstats/abs@.nsf/7d12b0f6763c78caca257061001cc588/8e81298ff9bfd160ca257ff1001e661e!opendocument https://www.abs.gov.au/statistics/research/2021-census-topics-and-data-release-plan#data-download https://www.abs.gov.au/statistics/research/2021-census-topics-and-data-release-plan#data-download https://www.abs.gov.au/census/planning-2021-census/privacy https://www.abs.gov.au/census/planning-2021-census/privacy https://commonslibrary.parliament.uk/research-briefings/cbp-8531/ https://commonslibrary.parliament.uk/research-briefings/cbp-8531/ https://www.otago.ac.nz/wellington/otago600095.pdf https://fecca.org.au/wp-content/uploads/2020/10/cald-data-issues-paper-final2.pdf https://fecca.org.au/wp-content/uploads/2020/10/cald-data-issues-paper-final2.pdf 64 allen australian population studies 5 (1) 2021 green e, sarrasin o and n fasel (2015) immigration: social psychology aspects. in: wright j d (ed) international encyclopedia of the social & behavioural sciences, 2nd edition, volume 11. london: elsevier. hopkins t (2017) monitoring racial profiling: introducing a scheme to prevent unlawful stops and searches by victoria police. https://www.policeaccountability.org.au/wpcontent/uploads/2017/08/monitoringrp_report_softcopy_final_22082017.pdf. accessed on 11 february 2021. horn r (1987) ethnic origin in the australian census. journal of the australian population association 4(1): 1-12. https://doi.org/10.1007/bf03029575 horn r (1993) the validity of australian ancestry statistics. journal of the australian population association 10(2): 119-126. https://doi.org/10.1007/bf03029418 jupp j (2001) the australian people: an encyclopedia of the nation, its people and their origins. cambridge: cambridge university press. khoo s-e (1991) consistency of ancestry reporting between parents and children in the 1986 census. journal of the australian population association 8(2): 129-139. https://doi.org/10.1007/bf03029441 luebke d and s milton (1994) locating the victim: an overview of census-taking, tabulation technology, and persecution in nazi germany. annals of the history of computing 16(3): 25-39. https://doi.org/10.1109/mahc.1994.298418 office of national statistics (2020) census 2021 paper questionnaires. https://www.ons.gov.uk/census/censustransformationprogramme/questiondevelopment/census 2021paperquestionnaires. accessed on 11 february 2021. perkins m (2004) australian mixed race. european journal of cultural studies 7(2): 177-199. https://doi.org/10.1177/1367549404042493 price j (2016) trying to sense of the census, but i’m still not convinced on privacy. the sydney morning herald, 9 august. https://www.smh.com.au/opinion/trying-to-make-sense-of-the-census-but-imstill-not-convinced-on-privacy-20160808-gqnj9q.html. accessed on 30 april 2021. rocha l, fozdar f, acedera k and b yeoh (2018) mixing race, nation, and ethnicity in asia and australasia. social identities 25(3): 289-293. https://doi.org/10.1080/13504630.2018.1500162 statistics canada (2020) ethnic or cultural origins: technical report on changes for the 2021 census https://www12.statcan.gc.ca/census-recensement/2021/ref/98-20-0002/982000022020001eng.cfm#a7. accessed on 11 february 2021. statistics new zealand (2001) review of the measurement of ethnicity: classifications and issues. https://www.stats.govt.nz/methods/review-of-the-measurement-of-ethnicity-classifications-andissues. accessed on 30 april 2021. statistics new zealand (2018) 2018 census: design of forms https://www.stats.govt.nz/assets/reports/2018-census-design-of-forms/2018-census-design-offorms.pdf. accessed on 11 february 2021. stevens g, ishizawa h and d grbic (2015) measuring race and ethnicity in the censuses of australia, canada, and the united states: parallels and paradoxes. canadian studies in population 42(1-2): 13-34. https://doi.org/10.25336/p6pw39 united nations (2017) principles and recommendations for population and housing censuses revision 3. new york: department of economic and social affairs statistics division. https://www.policeaccountability.org.au/wp-content/uploads/2017/08/monitoringrp_report_softcopy_final_22082017.pdf https://www.policeaccountability.org.au/wp-content/uploads/2017/08/monitoringrp_report_softcopy_final_22082017.pdf https://doi.org/10.1007/bf03029575 https://doi.org/10.1007/bf03029418 https://doi.org/10.1007/bf03029441 https://doi.org/10.1109/mahc.1994.298418 https://www.ons.gov.uk/census/censustransformationprogramme/questiondevelopment/census2021paperquestionnaires https://www.ons.gov.uk/census/censustransformationprogramme/questiondevelopment/census2021paperquestionnaires https://doi.org/10.1177/1367549404042493 https://www.smh.com.au/opinion/trying-to-make-sense-of-the-census-but-im-still-not-convinced-on-privacy-20160808-gqnj9q.html https://www.smh.com.au/opinion/trying-to-make-sense-of-the-census-but-im-still-not-convinced-on-privacy-20160808-gqnj9q.html https://doi.org/10.1080/13504630.2018.1500162 https://www12.statcan.gc.ca/census-recensement/2021/ref/98-20-0002/982000022020001-eng.cfm#a7 https://www12.statcan.gc.ca/census-recensement/2021/ref/98-20-0002/982000022020001-eng.cfm#a7 https://www.stats.govt.nz/methods/review-of-the-measurement-of-ethnicity-classifications-and-issues https://www.stats.govt.nz/methods/review-of-the-measurement-of-ethnicity-classifications-and-issues https://www.stats.govt.nz/assets/reports/2018-census-design-of-forms/2018-census-design-of-forms.pdf https://www.stats.govt.nz/assets/reports/2018-census-design-of-forms/2018-census-design-of-forms.pdf https://doi.org/10.25336/p6pw39 australian population studies 2021 | volume 5 | issue 2 | pages 73–85 © irving et al. 2021. published under the creative commons attribution-noncommercial licence 3.0 australia (cc by-nc 3.0 au). journal website: www.australianpopulationstudies.org long term population decline and the impacts of covid-19 in south australia’s regions jacob irving* university of south australia sandy horne university of south australia andrew beer university of south australia *corresponding author. email: jacob.irving@unisa.edu.au. address: university of south australia, way lee level 5, room 62, 37-44 north terrace, city west campus, adelaide, south australia 5000 paper received 16 august 2021; accepted 9 november 2021; published 29 november 2021 abstract background south australian regions have been given little attention in discussions on population decline. aims this paper aims to examine the nature and incidence of population decline in south australia as well as evaluate the potential impacts of covid-19. data and methods estimated resident population data from 2001 to 2020, and census data from 2006 and 2016, were used to investigate demographic and economic change. measures of population change, age structure, employment and components of population change were used. results population decline has been a feature of south australia’s regions for decades and continues to be so as more of the population concentrates in its capital and regional centres where greater opportunities of employment and greater provisions of amenities are available. covid-19 has the potential to accelerate this change if south australia’s vulnerable regions are not able to absorb the economic impacts the pandemic poses. conclusions a strong driver of population decline in the regions is employment loss in core industries. strategies that support these industries or otherwise aim to stimulate economic activity in these communities are required to moderate further decline in south australia’s regions especially as the economy recovers from the impacts of covid-19. key words population decline; structural ageing; employment loss; loss of services; south australia; covid-19. http://www.australianpopulationstudies.org/ mailto:jacob.irving@unisa.edu.au 74 irving et al. australian population studies 5 (2) 2021 1. introduction australia’s rural population peaked in 1911 soon after the nation’s agricultural frontier was established in the wake of colonisation by europeans. by 1954, australia had developed considerable metropolitan primacy, with 50% or more of state and territory populations resident in their capitals (bitre 2014). over the past 110 years rural populations have fallen in many parts of australia, with country towns experiencing decline and, in some instances, disappearing entirely (bitre 2014). this process has been especially evident in south australia where farm aggregation, the shift of retail functions from local towns to larger urban centres, the decline of some long-established industries, and the withdrawal of financial services from many centres (beer et al. 2003) has seen populations fall. nationally, there have been a number of studies examining the decline of rural and regional populations (mckenzie 1994), but much less attention has been paid to this issue in south australia where rural populations are small even by the standards of other parts of australia. this paper sets out to shed light on the nature, incidence and distribution of population decline in regional south australia, paying attention to the processes of both demographic and economic change. the paper examines key drivers of change including structural ageing, the loss of employment, geographic distribution, overseas migration and the potential impacts of covid-19. it examines abs data as a lens into the processes of change in south australia before commenting on the capacity of both shortand long-term effects of the covid-19 pandemic to usher in a change in trajectories. its findings are used to draw conclusions on how best to address population loss in south australia. 2. data and methods this paper draws on data from the australian bureau of statistics (abs), especially the most recent estimated resident population (erp) data. the research examined erp data over the past 20 years to assess long-term patterns of population change and age structure. additionally, 2006 and 2016 census data was used to assess the change of employment over time in regions experiencing population decline. the primary geographical level chosen for analysis was local government areas (lgas). lgas were analysed as they provide an effective geography for comparison, and one which is widely accepted in policy debates. it is also readily understood by a broad audience, with many residents identifying strongly with their local government and its territory. 3. population decline in south australia 3.1. patterns of change population change in greater adelaide and non-metropolitan south australia between 2001 and 2020 is shown in figure 1. proportional population increase in the first intercensal period to 2006 was similar in adelaide and non-metropolitan regions, and again between 2011 and 2016, but it is clear that there has been a distinct divergence in growth rates between greater adelaide and regional south australia in the intercensal period between 2006 to 2011, and again between 2016 and 2020. where this divergence between metropolitan and non-metropolitan south australia has australian population studies 5 (2) 2021 irving et al. 75 occurred geographically can be seen in figure 2, which shows regions that have experienced population decline. figure 1: south australia population growth index (2001=1.00), greater adelaide and non-metropolitan south australia source: abs.stat estimated resident population database. since 2001, south australia’s population decline has been most marked in the eyre peninsula, the far north, the mid north and the murraylands and riverland. very minor population loss has occurred in the yorke peninsula and along the limestone coast. the top 10 lgas ranked by percentage population decline are listed in table 1, with the estimated numerical population loss also presented. the table shows that the south australian regions that declined in population are distinctive. the greatest losses occurred in maralinga tjarutja and coober pedy located to the north of the eyre peninsula and in the far north (both are notable for their remoteness and reduced employment prospects). peterborough and orroroo/carrieton lgas, ranked third and fourth on this metric, are located in the outer mid north. they have experienced long-term decline in key industries including, manufacturing, agriculture and mining. ranks five through to 10 were shared between the regions of the upper east eyre peninsula (kimba; elliston; wudinna and cleve); and the murraylands and riverland (karoonda east murray; southern mallee) and these localities have experienced declines in employment in agricultural and mining industries. 76 irving et al. australian population studies 5 (2) 2021 figure 2: decline in lga populations in south australia, 2001-20 source: abs.stat estimated resident population database. table 1: top ten lgas ranked by percentage population decline, 2001-20 population 2001 population 2020 population change lga region number per cent 1 maralinga tjarutja eyre peninsula 136 64 -72 -52.9 2 coober pedy far north 2,414 1,820 -594 -41.6 3 peterborough mid north 2,014 1,668 -346 -24.6 4 orroroo/ carrieton mid north 1,013 844 -169 -17.2 5 kimba eyre peninsula 1,229 1,056 -173 -16.7 6 karoonda east murray murraylands and riverland 1,276 1,101 -175 -14.1 7 elliston eyre peninsula 1,150 1,011 -139 -13.7 8 wudinna eyre peninsula 1,449 1,307 -142 -12.1 9 southern mallee murraylands and riverland 2,300 2,089 -211 -9.8 10 cleve eyre peninsula 1,939 1,780 -159 -9.2 total 14,920 12,740 -2,180 -14.6 source: abs.stat estimated resident population database. australian population studies 5 (2) 2021 irving et al. 77 of the ten lgas to experience the greatest decline in population in south australia between 2001 and 2020, nine experienced a loss of population of 10% or more, with cleve placing tenth at 9.2%. this makes for an estimated total loss of 2,180 people from the fastest shrinking lgas. the cumulative population loss of all lgas to experience a decline since 2001 was 5,348 people. in addition, unincorporated sa, while not an lga, had a population decline of 2,458 people, or 58.4%, from 5,911 to 3,453, during the same period. it is important to acknowledge, however, the highly mobile nature of many aboriginal australians and they may have been temporarily absent from the place they would identify as ‘home’ on census night affecting the data on which estimated resident population is based. figure 3: growth in lga populations in south australia, 2001-20 source: abs.stat estimated resident population database. while there has been significant depopulation within the lgas in table 1, figure 3 shows evidence of population growth in some parts of regional south australia. these include the regions that make up the adelaide metropolitan commuting region along the fleurieu peninsula, up to the lower mid north and east to the murraylands; port pirie, a mid north regional centre; a number of coastal centres along 78 irving et al. australian population studies 5 (2) 2021 the eyre peninsula; the copper coast at the top of the yorke peninsula; kangaroo island off the coast of the fleurieu peninsula; several lgas along the limestone coast; renmark paringa along the murray river in the north-eastern murraylands and riverland; and roxby downs, a mining town in the far north of south australia. 3.2. age, employment and service provision to help understand the population trends shown above, we examined population age structure, contraction of employment opportunities, and loss of ancillary services. figure 4 compares the 2020 age structure of all lgas to experience population decline between 2001 and 2020 with the age structure of the whole of south australia. in regions showing population decline, the average proportion of the population aged between 20 and 44 was significantly lower than the state average while the proportion aged 55+ was above the state average. this distribution indicates that regions experiencing population decline have lower rates of retention of young adults than the rest of south australia. furthermore, the majority of adults in declining regions were aged 45 years or older, an age group which is “dominantly post-reproductive” (smailes et al. 2019 p 107). a lower retention of young adults and a mostly post-reproductive adult population indicates these regions are vulnerable to, if not already experiencing, natural decrease where the number of deaths exceeds births. figure 4: age structure of regions with population decline compared to south australia, 2020 source: abs.stat estimated resident population database. employment loss as a driver of population decline was also examined. table 2 shows the top ten lgas ranked by loss of employed persons as a percentage. each lga, with the exception of franklin harbour, has been identified in figure 2 to have experienced a decline in resident population. in australian population studies 5 (2) 2021 irving et al. 79 aggregate 1,584 jobs were lost from these regions. there appears to be a close relationship in these areas between job loss and population loss and this represents a critical juncture in the economies of these places (beer et al. 2021). table 2: top ten lgas ranked by loss of employed persons, 2006 to 2016 employment change lga region number per cent 1 maralinga tjarutja eyre peninsula -16 -53.3 2 anangu pitjantjatjara far north -297 -36.9 3 kimba eyre peninsula -89 -14.5 4 karoonda east murray murraylands and riverland -81 -14.4 5 southern mallee murraylands and riverland -128 -11.5 6 elliston eyre peninsula -62 -11.0 7 cleve eyre peninsula -101 -10.6 8 the coorong murraylands and riverland -265 -10.4 9 franklin harbour eyre peninsula -61 -10.2 10 berri and barmera murraylands and riverland -484 -10.1 total -1,584 -12.6 source: 2006 and 2016 censuses, place of usual residence databases. to assess the relationship between the loss of ancillary services in the regions and population decline, change in employment by sector was examined (table 3). critically, some 185 jobs in the service industries of health and education were lost from high population loss lgas in rural south australia, which suggests service industry contraction had a correlative, but not determinant, role in their shrinking populations. table 3: top ten lgas ranked by loss of employed persons in health and education, 2006 to 2016 employment change lga region number per cent 1 maralinga tjarutja eyre peninsula -4 -57.1 2 elliston eyre peninsula -24 -21.2 3 karoonda east murray murraylands and riverland -17 -16.8 4 anangu pitjantjatjara far north -36 -12.9 5 franklin harbour eyre peninsula -9 -8.6 6 kimba eyre peninsula -6 -7.5 7 cleve eyre peninsula -9 -6.6 8 port augusta far north -63 -4.6 9 goyder mid north -10 -3.8 10 ceduna eyre peninsula -7 -1.8 total -185 -6.5 source: 2006 and 2016 census, place of usual residence databases – indp 1 digit level ‘health care and social assistance’ and ‘education and training’. 80 irving et al. australian population studies 5 (2) 2021 4. drivers of population change 4.1. drivers of population decline and growth in non-metropolitan south australia. in 1994, mckenzie described how rural depopulation had been occurring over many decades in australia (mckenzie 1994). beer & keane (2000) described how the spencer gulf area, which includes port pirie, port augusta and whyalla, saw a population loss of nearly 5% in the decade from 1986. this earlier work reinforced how decline has been a trend in parts of south australia for a considerable period. it is evident from the analysis that population decline in south australia is pronounced and widespread in areas of dryland, broadscale farming and in parts of the riverland not immediately adjacent to the murray river. non-metropolitan growth is concentrated around adelaide, in the southern eyre peninsula and in coastal areas. at the level of the entire state, the south australian government population projections anticipate adelaide and its outer area will account for 79% of the south australian population by the year 2041 (sa government 2019), an increase from 77.8% in the abs (2021) march erp release. this is consistent with figure 1, which highlighted a divergence in population growth rates of lgas outside greater adelaide from 2001 to 2020. major drivers of population decline in regional south australia include structural population ageing, the loss of employment opportunities, and the loss of ancillary services. these drivers are interrelated and changes in one often have implications for the other. the failure to retain young adults in many of the state’s regions is a powerful driver of population decline. it reflects the decisions of these young people to seek greater employment and educational opportunities in regional centres or in a major metropolitan area, and these are largely opportunities which their hometowns simply cannot provide (smailes et al. 2019). the loss of ancillary services in declining regions has been well documented in mckenzie (1994) and it is a process that feeds back to accelerate population decline (beer & keane 2000). this vicious cycle of interconnection between population decline, structural ageing, the loss of employment opportunities and the removal of ancillary services is represented below in figure 5, which has been adapted from sorensen (1990a in mckenzie 1994). reviewing the geographical distribution of south australian population changes from 2001 to 2020 reveals decline is most pronounced in those localities dependent on traditional regional industries, including the production of wool and grains. contrary, perhaps to expectations, mining has provided limited employment security. this is most evident in the south australian wheatbelt – the outer mid north and the eastern murraylands and riverland – and in the upper east eyre peninsula. furthermore, these vulnerable regions demonstrate common drivers of population decline such as structural ageing – where deaths outnumber births – and a correspondence with the loss of employment and services as seen in tables 2 and 3 for the regions of the eyre peninsula, the far north, and the murraylands and riverland. south australia is often said to be the ‘driest state in the driest continent’ and is consequently particularly vulnerable to climate change, severe weather events, and natural disasters such as droughts and bushfires. a 2013 study that developed a vulnerability index looking at australia’s country towns and the potential impacts of climate change noted that south australia faced unique challenges due to its highly concentrated population, alongside other factors (beer et al. 2013). australian population studies 5 (2) 2021 irving et al. 81 figure 5: the vicious cycle of population decline source: adapted from mckenzie (1994). the riverland towns of waikerie, renmark, barmera, berri and loxton, with economies tied to the health of the river murray, were scored as high risk on the vulnerability index (beer et al. 2013). agricultural production is at the centre of many country towns and climate change has been identified as a major threat to this industry with the projected higher temperatures and reduced rainfall having negative impacts. this risk will be particularly acute in south australia’s more arid zones (beer et al. 2013). changing climate could prove to be a significant driver of out-migration and the eventual abandonment of some settlements. some regions that have experienced a decline in traditional regional industries have reversed a trend of population decline due to their presence within the expanded commuting belt of adelaide. adelaide’s commuting belt has expanded over the last decades with advances in communications and technology. consequently, the parts of the mid north and murraylands and riverland closest to adelaide have grown in population while their northern and eastern neighbours have declined. alternatively, growth regions have come about due to post-productivist shifts in coastal and riverside regions which have become destinations for retirement, tourism (argent et al. 2016; smailes et al. 2019) and unique consumptionist experiences (morrison et al. 2015). the regional australia institute (2015) has argued that other drivers include the growth of remote mining communities and towns on key transport routes. regions in decline are caught in a vicious cycle in which employment and service loss are influenced by structural ageing and population decline, which then further exacerbates the loss of jobs and service provision. this phenomenon can help explain the distributions of population loss in regional south australia to date and puts question marks around how migration patterns may change as the ongoing economic and health impacts of covid-19 emerge over the next decade. 82 irving et al. australian population studies 5 (2) 2021 4.2. short and long-term effects of covid-19 in non-metropolitan south australia due to covid-19, australia’s international borders have been closed, and state borders have also been subject to period closures. south australians have consequently been restricted to holiday destinations closer to home. karoonda, a small town in the murraylands, has sought to capitalise on the popularity of this increase in domestic tourism. with its silos as a drawcard – newly painted with artwork highlighting the history of the area – the karoonda east murray council developed plans to expand and improve the town’s caravan park to accommodate the large influx of visitors. in addition, the council invested in a range of incentives for local businesses to upgrade facilities (abc news riverland 2021). the experiences of karoonda reflect those of other small towns throughout south australia with hopes expressed by many that events could boost community, industry and the local population. however, there remains a disconnect between the perception that the increased tourism in the short-term will result in any long-term population growth and reality, as this notion is inconsistent with population projections (centre for population 2020). over the coming decade, australia’s population growth is expected to slow to its lowest level in more than a century as net overseas migration has come to a standstill with the onset of the pandemic (abs 2021). furthermore, projections based on the 2021-22 federal budget forecast that overseas migration is not expected to return to pre-covid-19 levels until 2023-2024 (commonwealth of australia 2021). while overseas migration in south australia is mostly concentrated in adelaide, the concern for the regions lies in that the economic shocks of covid-19 are likely to affect south australia’s most fragile economies as the state’s economy contracts. communities caught within the vicious cycle of population decline, or at risk due to structural ageing, are more vulnerable to external events such as droughts and, economic and health crises such as the covid-19 pandemic. they face the potential of further decline and even disappearance as residents migrate to more urbanised centres to secure services in these periods (beer et al. 2011). regional agricultural areas, for example, have suffered in the past due to droughts and yield volatility (productivity commission 2005) producing work that is, when available, insecure. more broadly, political support for an ‘australia first’ migration policy may result in lower targets for the number of immigrants entering australia (betts & birrell 2020), thereby stripping away one option for growth in areas of decline as south australia recovers from the economic effects of covid-19. wilson et al. (2021) notes that “population forecasting is particularly hazardous at this time” due to the uncertainty arising from the pandemic – and it is clear some population forecasts need to be approached with caution. nevertheless, there are grounds for concern given south australia’s reliance over the last two decades on overseas migration for population growth, especially given its role in reversing projections of net state loss in the early 2000s (hugo 2008). 5. conclusions this paper has shown there has been a steady and ongoing decline in rural and remote south australia for several decades as industries have contracted, farms aggregated, and as agricultural regions have turned away from their long-standing focus on production, becoming more significant for their consumption offerings. in contrast, regions closest to the adelaide metropolitan and other large regional centres which offer high employment and amenities have experienced an increase in population. three drivers of population decline – structural ageing, loss of employment opportunities australian population studies 5 (2) 2021 irving et al. 83 and loss of ancillary services – were examined in this analysis. this paper finds that regions in decline have a positive correlation with structural ageing and the loss of employment and services. furthermore, these drivers have feedback on each other such that initial population decline reinforces the fragility of core industries, which in turn contributes to the loss of support services. this paper lastly sought to review the potential shortand long-term impacts of the covid-19 pandemic as it applies to south australia. short-term impacts include low levels of migration from capital to rest of state areas, and a reduction in overseas migration. while a short-term reduction of overseas migration is negligible, two decades worth of marked reliance on net overseas migration for the maintenance of south australia’s population growth, in addition to the historically high levels of interstate outmigration, means that the state is likely positioned for half a decade of slow growth until overseas migration resumes to pre-covid-19 levels. the resultant contraction of the state’s economy as it recovers from the pandemic may result in a shock on south australia’s most vulnerable economies and there is a substantial risk that some of these places will not recover if their core industries are adversely affected. the loss of employment in core industries is a key trigger for many regions to tip over from being at risk of population decline, to experiencing population decline and this is a central finding from this research. as we look to the future, we can only conclude that there is a need to find better ways to support these industries, or introduce new forms of economic activity into these places, in order to ensure the enduring legacy of covid-19 is not the demise of south australia’s regions. key messages • population decline has been occurring in remote south australia for a long time and this trend looks set to continue. • structural ageing and the loss of employment are significant causes of population decline in regional south australia. • loss of overseas migration due to covid-19 may result in an economic shock to some south australian regional economies which could lead to population decline. • strategies that support regional industries or introduce new forms of economic activity in these communities are needed to moderate further population decline in south australia. acknowledgments the authors thank justin mcculloch, the australian alliance for social enterprise at the university of south australia, for producing the maps. references abc news riverland (2021) karoonda looking to ensure pandemic travel boom leads to lasting growth. 22 june, https://www.abc.net.au/news/2021-06-22/karoonda-travel-boom/100223080 accessed 22 june 2021. abs (2021) national, state and territory population (march 2021). abs: canberra. https://www.abs.gov.au/statistics/people/population/national-state-and-territorypopulation/mar-2021 https://www.abc.net.au/news/2021-06-22/karoonda-travel-boom/100223080%20accessed%2022%20june%202021 https://www.abc.net.au/news/2021-06-22/karoonda-travel-boom/100223080%20accessed%2022%20june%202021 https://www.abs.gov.au/statistics/people/population/national-state-and-territory-population/mar-2021 https://www.abs.gov.au/statistics/people/population/national-state-and-territory-population/mar-2021 84 irving et al. australian population studies 5 (2) 2021 argent n, griffin t and smailes p (2016) migration and ageing processes in non-metropolitan australia: an analysis of thirty years of dramatic change. in wilson t, charles-edwards e, & bell m (eds.) demography for planning and policy: australian case studies. springer; pp. 133-154. beer a and keane r (2000) population decline and service provision in regional australia: a south australian case study. people and place 8(2): 69-76. beer a, barnes t and horne s (2021) place-based industrial strategy and economic trajectory: advancing agency-based approaches. regional studies. https://doi.org/10.1080/00343404.2021.1947485 beer a, maude a and pritchard b (2003) developing australia’s regions: theory and practice. university of nsw press. beer a, tually s, kroehn m, martin j, gerritsen r, taylor m, graymore m and law j (2013) australia’s country towns 2050: what will a climate adapted settlement pattern look like? report for the national climate change adaptation research facility. https://www.nccarf.edu.au/publications/country-towns-2050-climate-adapted-settlement beer a, tually s, rowley s, haslam mckenzie f, shlapp j, bridsall-jones c and corunna v (2011) the drivers of supply and demand in australia’s rural and regional centres. ahuri final report no.165. melbourne: australian housing and urban research institute. https://www.ahuri.edu.au/research/final-reports/165 betts k and birrell b (2020) a big australia: why is may all be over. the australian population research institute. https://tapri.org.au/wp-content/uploads/2020/10/big-australia-finalv6.pdf bureau of infrastructure, transport and regional economics (bitre) (2014) the evolution of australian towns. research report 136. department of infrastructure and regional development, canberra. https://www.bitre.gov.au/publications/2014/report_136 centre for population (2020) population statement. australian government centre for population, canberra. https://population.gov.au/docs/population_statement_2020.pdf commonwealth of australia (2021) 2021 intergenerational report: australia over the next 40 years. australian government, canberra. https://treasury.gov.au/sites/default/files/202106/p2021_182464.pdf hugo g (2008) australia’s state-specific and regional migration scheme: an assessment of its impacts in south australia. journal of international migration and integration 9(2): 125-145. https://doi.org/10.1007/s12134-008-0055-y mckenzie f (1994) population decline in non-metropolitan australia: impacts and policy implications. urban policy and research 12(4): 253-263. https://doi.org/10.1080/08111149408551645 morrison t h, lane m b and hibbard m (2015) planning, governance and rural futures in australia and the usa: revisiting the case for rural regional planning. journal of environmental planning and management 58(9): 1601-1616. https://doi.org/10.1080/09640568.2014.940514 productivity commission (2005) trends in australian agriculture. productivity commission research paper, canberra. https://www.pc.gov.au/research/completed/agriculture regional australia institute (rai) (2015) population dynamics in regional australia. regional australia institute, canberra. http://www.regionalaustralia.org.au/wp-content/uploads/2015/01/finalpopulation-dynamics-in-regional-australia.pdf smailes p, griffin t and argent n (2019) regional cities and city regions in rural australia: a long-term demographic perspective. springer. https://doi.org/10.1007/978-981-13-1111-6 south australian centre for economic studies (saces) (2019) regional profiles. https://www.adelaide.edu.au/saces/regional-profiles. accessed 29 october 2021. state government of south australia (2019) population projections for south australia and regions, 201641. department of planning, transport and infrastructure. https://dpti.sa.gov.au/__data/assets/pdf_file/0008/563813/population_projections_for_south_a ustralia_and_regions_2016-41_-_may_2019.pdf https://doi.org/10.1080/00343404.2021.1947485 https://www.nccarf.edu.au/publications/country-towns-2050-climate-adapted-settlement https://www.ahuri.edu.au/research/final-reports/165 https://tapri.org.au/wp-content/uploads/2020/10/big-australia-finalv6.pdf https://www.bitre.gov.au/publications/2014/report_136 https://population.gov.au/docs/population_statement_2020.pdf https://treasury.gov.au/sites/default/files/2021-06/p2021_182464.pdf https://treasury.gov.au/sites/default/files/2021-06/p2021_182464.pdf https://doi.org/10.1007/s12134-008-0055-y https://doi.org/10.1080/08111149408551645 https://doi.org/10.1080/09640568.2014.940514 https://www.pc.gov.au/research/completed/agriculture http://www.regionalaustralia.org.au/wp-content/uploads/2015/01/final-population-dynamics-in-regional-australia.pdf http://www.regionalaustralia.org.au/wp-content/uploads/2015/01/final-population-dynamics-in-regional-australia.pdf https://doi.org/10.1007/978-981-13-1111-6 https://www.adelaide.edu.au/saces/regional-profiles https://dpti.sa.gov.au/__data/assets/pdf_file/0008/563813/population_projections_for_south_australia_and_regions_2016-41_-_may_2019.pdf https://dpti.sa.gov.au/__data/assets/pdf_file/0008/563813/population_projections_for_south_australia_and_regions_2016-41_-_may_2019.pdf australian population studies 5 (2) 2021 irving et al. 85 wilson t, temple j and charles-edwards e (2021) will the covid‑19 pandemic affect population ageing in australia? journal of population research https://doi.org/10.1007/s12546-021-09255-3. https://doi.org/10.1007/s12546-021-09255-3 australian population studies 2020 | volume 4 | issue 1 | pages 4-19 © borsellino 2020. published under the creative commons attribution-noncommercial licence 3.0 australia (cc by-nc 3.0 au). journal website: www.australianpopulationstudies.org the changing migration patterns of the 65+ population in australia, 1976-2016 rosabella borsellino* the university of queensland * email: rosabella.borsellino@uqconnect.edu.au. address: queensland centre for population research, school of earth and environmental sciences, chamberlain building, the university of queensland, st lucia, qld 4072, australia. paper received 11 february 2020; accepted 20 march 2020; published 25 may 2020. abstract background the demand for amenity, health and aged care, and social support are of concern to rapidly ageing communities. understanding how the migration trends of the 65+ population compare to movements made by the rest of the population can assist in planning for future growth. aim the aim of the paper is to understand how the migration patterns of the 65+ population in australia have changed over the past forty years in comparison to the total population. data and methods five-year interval census migration data were used to examine the migration flows of the 65+ population between 1976 and 2016. a range of metrics was used to measure the level or intensity of internal migration and its impact on the redistribution of the population at the national level. regional net migration rates were then used to compare net gains and losses between regions and track changes over time in the direction of flows across the australian settlement system. results the propensity to migrate declined over the forty years for all ages but proportionally less for the over 65+ group. gains in the 65+ population were spread across a range of middle-density regions with losses from the most urban and remote areas. the migration patterns observed in the total population were less spatially consistent, with periods of gain in some remote regions. conclusions the system-wide metrics showed an overall decline in migration propensity between regions and changes in the redistributive impact of migration by age. the net migration maps and scatterplots revealed large changes in the regional growth and decline of the 65+ population as new destinations emerged in regional victoria and new south wales. key words migration, ageing, 65+, counter-urbanisation, australia. http://www.australianpopulationstudies.org/ mailto:rosabella.borsellino@uqconnect.edu.au australian population studies 4 (1) 2020 borsellino 5 1. introduction there has been little overall change in the spatial distribution of australia’s population over the past forty years with the majority of people living in urban centres along the coast. growth is concentrated in the urban periphery, high-amenity regions on the east and southwest coasts, resource-rich areas, and regional centres supported by tourism, while corresponding decline occurs in the wheat-sheep belt, pastoral areas, and former mining towns (hugo 2013). as regional variations in fertility and mortality are too small to contribute significantly to demographic change in these regions (jackson & felmingham 2002), shifts in the volume and composition of inflows and outflows are behind the changing age structure of regions in australia (hugo 2013; mcmanus & pritchard 2000). divergence in the age structure of urban and rural australia has been attributed to increased outmigration of young people from rural areas to pursue education and employment opportunities (davies 2008), while in-migrants to non-metropolitan australia tend to be older and attracted to selected wellserviced and amenity-rich communities (davies & james 2011). regularities in age-specific migration rates are well-established (rogers & castro 1981) and while migration research has tended to focus on the movements of the working age population, there are two distinct peaks in the migration propensity at older ages (rogers 1988), associated with different spatial patterns (conway & houtenville 2003). newly retired migrants are typically motivated by a desire for the social opportunities and recreational activities offered by high-amenity coastal and hinterland regions adjacent to metropolitan regions (argent, smailes & griffin 2007; davies & james 2011). conversely, the oldest migrants are often drawn by necessity to the healthcare and assisted living facilities offered by regional centres and metropolitan areas. in a rapidly ageing society, the demand for amenity, social support, and health and aged care facilities are of growing concern (kim et al. 2003), and the regional differences in migration flows have increasingly significant socioeconomic, political, and developmental implications (jackson & felmingham 2002; tonts & greive 2002). as understanding the past trends of elderly migration can assist in predicting and planning for future growth, the aim of this paper is to understand how the impact of migration for older australians has changed over the past forty years in comparison to the total population. to achieve this aim, this paper first establishes the national-level intensity, effectiveness, and impact of migration for each of the eight census periods from 1976-81 to 2011-16. the paper then maps the impact of these moves at the regional level to identify which regions are experiencing gains and losses. finally, the paper speculates on possible explanations for the observed trends. 2. data this paper draws on data from the quinquennial australian census of population and housing administered by the australian bureau of statistics (abs), covering eight census periods from 1976-81 to 2011-16. the analysis relies on the retrospective “place of usual residence five years ago” question to determine patterns of internal mobility and therefore omits return and repeat moves made during this five-year period. migrants who die or emigrate during the inter-censal period, 6 borsellino australian population studies 4 (1) 2020 immigrants who moved prior to the five-year period, and those aged between 0-4 at the time of the census are not counted. to accommodate changes to administrative spatial boundaries over the study period, the migration flows are organised into 69 temporally-consistent regions (figure 1). these regions are based on the statistical divisions defined by the abs: statistical local areas (sla) for 1976 to 2006 (bell et al. 1999; rees et al. 2000), and statistical areas level 2 (sa2) for 2011 and 2016. these data are held by the queensland centre for population research and can be made available to researchers on request. it should be noted that only moves made between regions are counted for the purposes of this analysis. a high degree of metropolitan primacy in the australian settlement system means that regions outside of the state capitals can cover very large areas, resulting in unequal sizes of the spatial units of analysis. this unique geography means that while intra-region moves are often local or short-distance in smaller regions, in larger regions these moves may involve a significant distance as the spatial units may encompass a large range of settlement types. figure 1: the 69 temporally consistent australian regions source: abs spatial data, created by author 3. methods as migration is a multidimensional process, three metrics are used to measure the intensity, effectiveness and impact of the migration system (bell et al. 2002). each metric is calculated for the total population, those aged 0-64, and those aged 65+. australian population studies 4 (1) 2020 borsellino 7 the first metric captures the overall volume of movement that occurs within the population. intensity is expressed using the crude migration index (cmi) which measures the number of migrants (𝑀𝑀) as a percentage of the population at risk (𝑃𝑃) of moving. 𝐶𝐶𝑀𝑀𝐶𝐶 = (𝑀𝑀 𝑃𝑃⁄ ) × 100 (1) the effect of migration on population distributions is not just a function of volume but is related to asymmetry of the migration system or the extent to which migration flows are offset by counter flows. the effectiveness of migration in redistributing the population is measured using the migration effectiveness index (mei) (shryock & siegel 1975). the mei is calculated as the absolute value of net migration (di – oi) summed across all regions divided by the sum of inflows and outflows for all regions (di + oi) (morrison, bryan & swanson 2004; plane & rogerson 1991). the mei is reported as a value between 0 and 100 with higher values indicating greater effectiveness. 𝑀𝑀𝑀𝑀𝐶𝐶 = (∑ |𝐷𝐷𝑖𝑖 − 𝑂𝑂𝑖𝑖|𝑖𝑖 ∑ 𝐷𝐷𝑖𝑖 + 𝑂𝑂𝑖𝑖𝑖𝑖⁄ ) × 100 (2) intensity and effectiveness together drive the overall impact of migration on the settlement system and the complexity of this relationship can be computed using the aggregate net migration rate (anmr). the anmr is used to summarise the system-wide impact resulting from the net balance of inter-regional flows (bell et al. 2002). the anmr is expressed as a percentage of half the absolute value of net migration summed across all regions over the population at risk of each group. 𝐴𝐴𝐴𝐴𝑀𝑀𝐴𝐴 = (0.5 × ∑ |𝐷𝐷𝑖𝑖 − 𝑂𝑂𝑖𝑖|𝑖𝑖 ∑ 𝑃𝑃𝑖𝑖𝑖𝑖⁄ ) × 100 (3) as national-level metrics obscure changes in the composition of regions, the net migration rates (nmr) for each of the 69 regions are mapped to identify spatial patterns in the attractiveness of destinations. the nmr is calculated as the balance of inflows and outflows (dij – oji) for each region 𝑖𝑖 divided by the region’s population at risk (𝑃𝑃𝑖𝑖). 𝐴𝐴𝑀𝑀𝐴𝐴 = �∑ ∑ �𝐷𝐷𝑖𝑖𝑖𝑖 − 𝑂𝑂𝑖𝑖𝑖𝑖� 2 𝑖𝑖≠𝑖𝑖𝑖𝑖 𝑃𝑃𝑖𝑖� � × 100 (4) finally, to identify the dominant direction of migration flows across the urban hierarchy, this paper modifies the approach used by rees et al. (2017). net migration rates for regions are plotted against population density which is used as a proxy for the degree of urbanisation of a region. this measure accommodates changes in the classification of urban and rural areas within australia over time. the relationship between net migration rates and population density is estimated using a second-degree polynomial. 4. results between 1981 and 2016 the number of people aged 65+ in australia increased from 1.35 million (9% of the total population) to 3.26 million (14%) (figure 2a). despite the unprecedented absolute and relative growth of this age group, the number of moves did not increase at the same rate, resulting in a decline in migration intensity for the 65+ group from 9% in 1976-81 to 7% in 2011-16 (figure 2b). the largest drop occurred in 2006-11 when the migration intensity fell 1.2%, compared to an average decline of 0.3% for each census period. overall, migration intensity declined slowly for the 65+ 8 borsellino australian population studies 4 (1) 2020 population, dropping only two per cent over forty years. in comparison, the migration intensity of the 0-64 population declined five percentage points from 17% to 12% over the same time period, while the migration intensity of the total population declined four percentage points from 16% to 12%. (a) (b) (c) (d) figure 2: comparison of (a) population growth, (b) crude migration intensity; (c) migration effectiveness index; and (d) aggregate net migration rate for the total population, those aged 0-64, and those aged over 65 in australia, 1976-81 to 2011-16 source: abs census data, author’s calculations the mei for the 65+ group was higher than that recorded by both the 0-64 group and the total population, meaning that migration was more effective at redistributing those aged 65+ than the rest of the population. however, the trend was not linear for either age group as the decline was interrupted by periods where migration became more effective at redistributing the population. an initial fluctuation in migration effectiveness for the 65+ group between 1976-81 and 1986-91 was followed by a sustained decline from a maximum of 23% in 1986-91 to a minimum of 15% in 2006-11 (figure 2c), before a final increase to 17% in 2011-16. migration became less effective at 0 5 10 15 20 25 19 76 -8 1 19 81 -8 6 19 86 -9 1 19 91 -9 6 19 96 -0 1 20 01 -0 6 20 06 -1 1 20 11 -1 6 c ru de m ig ra tio n in te ns ity (% ) census period total 0-64 65+ 0 5 10 15 20 25 19 76 -8 1 19 81 -8 6 19 86 -9 1 19 91 -9 6 19 96 -0 1 20 01 -0 6 20 06 -1 1 20 11 -1 6 m ig ra tio n ef fe ct iv en es s in de x (% ) census period total 0-64 65+ 0 5 10 15 20 25 19 81 19 86 19 91 19 96 19 01 20 06 20 11 20 16 po pu la tio n (m illi on s) census year total 0-64 65+ 0.0 0.5 1.0 1.5 2.0 2.5 3.0 19 76 -8 1 19 81 -8 6 19 86 -9 1 19 91 -9 6 19 96 -0 1 20 01 -0 6 20 06 -1 1 20 11 -1 6 ag gr eg at e ne t m ig ra tio n ra te (% ) census period total 0-64 65+ australian population studies 4 (1) 2020 borsellino 9 redistributing the total population and the 0-64 age group over time, with the mei of both declining from a maximum of 14% in 1976-81 to 8% in 2011-16. while the overall pattern of decline was similar across the three groups, the direction of the meis diverged in the final census period, with the 65+ group experiencing an increase while the total population and 0-64 group continued to decline. the anmr for all three groups exhibited a similar overall decline (figure 2d); however, while the impact the 0-64 and total groups was virtually identical, the 65+ group had a slightly different trajectory. the highest impact for the 65+ group occurred in 1976-81, where 2% of the population was redistributed through migration. while the 0-64 age group and total population initially had a greater impact on the redistribution of the population, the impact of the 65+ group was greater between 1986 and 2006. all three groups declined to a low of 1% in 2006-11, though while the impact of the 0-64 and total populations continued to decline, the 65+ group rose to 1.2% in 2011-16. these results illuminate the relationship between the number of migrants and the effectiveness of migration in redistributing the population. despite the lower volume of migrants in the 65+ group, the higher effectiveness of this group meant that the overall impact on regional settlement patterns was relatively similar to that of the total population. conversely, the impact of the 0-64 population was driven by the greater number of migrants. the impact of internal migration has not been uniform across the settlement system. the pattern of net gains and losses of the 65+ group were spatially bifurcated with net gains along the east, south, and southwest coasts, and pronounced net loss in the remote regions of each state in all census periods (figure 3). movements were initially highly directed towards the amenity-rich regions on the northern coast of new south wales and southeast queensland coast receiving high net gains between 1976-81 and 1991-96; however, the intensity of these gains declined over time. the suburban peripheries of brisbane, adelaide, and perth declined further, changing from net gains to net losses between 2001-06 and 2006-11. concurrently, inland new south wales and victoria flipped progressively over time from net loss to net gain, suggesting a pattern of counter-urbanisation away from the traditional retirement destinations and towards regional centres adjacent to the capital centres. notably, inner sydney, brisbane, melbourne, and perth also experienced net losses. more remote regions in each state experienced consistent loss with the highest net losses generally from resource-rich regions of western australia and western queensland. the broad pattern of net migration for the 0-64 age group (figure 4) and total population (figure 5) was similar to that of the 65+ group, with gains along the east, south, and southwest coasts, and widespread losses in remote inland regions. however, there were also additional gains in key outback resource regions in 1976-81, 1981-86, 2006-11, and 2011-16, and gain in southeast inland regions was not as temporally consistent or spatially extensive as seen in the 65+ group. plotting the net migration rate against the population density for each region reveals a broad trend of net gain in moderately densely populated regions and corresponding losses in highand low 10 borsellino australian population studies 4 (1) 2020 1976-81 1981-86 1986-91 1991-96 1996-2001 2001-06 2006-11 2011-16 net migration rate < -10 -10 to -5 -5 to 0 0 to 5 5 to 10 > 10 figure 3: net migration rate of the 65+ population for regions in australia, 1976-81 to 2011-16 source: abs census data, author’s calculations australian population studies 4 (1) 2020 borsellino 11 1976-81 1981-86 1986-91 1991-96 1996-2001 2001-06 2006-11 2011-16 net migration rate < -10 -10 to -5 -5 to 0 0 to 5 5 to 10 > 10 figure 4: net migration rate of the 0-64 population for regions in australia, 1976-81 to 2011-16 source: abs census data, author’s calculations 12 borsellino australian population studies 4 (1) 2020 1976-81 1981-86 1986-91 1991-96 1996-2001 2001-06 2006-11 2011-16 net migration rate < -10 -10 to -5 -5 to 0 0 to 5 5 to 10 > 10 figure 5: net migration rate of the total population for regions in australia, 1976-81 to 2011-16 source: abs census data, author’s calculations australian population studies 4 (1) 2020 borsellino 13 density regions for the 65+ population (figure 6). the strength of the relationship declined slightly between 1976-81 (r2=0.49) and 2011-16 (r2=0.44), likely as a result of reduced net gains in middledensity regions and increased net losses in selected low-density regions. although the presence of this trend is a persistent feature over time, there are notable changes in the magnitude of the gains for regions in southeast queensland and northern new south wales, while several suburban capital regions flipped from net gain to net loss. the relationship between net migration and population density was weaker in 1976-81 (r2=0.23) for the 0-64 age group, likely owing to the higher gains in middle-density areas (figure 7). this relationship weakened further (r2=0.04) as a number of middledensity regions experienced net loss in 2011-16. (a) (b) figure 6: net migration and population density for the 65+ population in (a) 1976-81 and (b) 2011-16 source: abs census data, author’s calculations (a) (b) figure 7: net migration and population density for the 0-64 population in (a) 1976-81 and (b) 2011-16 source: abs census data, author’s calculations inner sydney richmond-tweed nsw mid-north coast nsw outer brisbane gold coastcaboolture north west … outer perth pilbara wa kimberley wa act y = -1.8326x2 + 5.8466x 0.8913 r² = 0.4881 -40 -30 -20 -10 0 10 20 -2 -1 0 1 2 3 4 n et m ig ra tio n ra te log population density inner sydney richmond-tweed nsw mid-north coast nsw outer brisbane gold coast caboolture north west qld pilbara wa kimberley wa darwin act y = -1.7189x2 + 4.8447x 0.7671 r² = 0.4356 -40 -30 -20 -10 0 10 20 -2 -1 0 1 2 3 4 n et m ig ra tio n ra te log population density inner sydney outer sydney richmond-tweed nsw mid-north coast nsw outer brisbane gold coast north west qld outer perth pilbara wa kimberley wa darwin nt balance act y = -2.0944x2 + 5.2834x + 1.1525 r² = 0.2299 -20 -10 0 10 20 30 40 -2 -1 0 1 2 3 4 n et m ig ra tio n ra te log population density inner sydney inner brisbane outer brisbane gold coast caboolture wide bay-burnett qld outer perth pilbara wa kimberley wa y = -0.4296x2 + 0.7241x 1.1268 r² = 0.0367 -20 -10 0 10 20 30 40 -2 -1 0 1 2 3 4 n et m ig ra tio n ra te log population density 14 borsellino australian population studies 4 (1) 2020 the relationship between net migration of the total population and regional population density (figure 8) strengthened between 1976-81 (r2=0.24) and 2011-16 (r2=0.58), likely as a result of the considerable decline in the magnitude of net gains in middle-density regions. the weak to moderate r2 values shown here for both age groups as well as the population overall, indicates that while a relationship exists, there is a substantial amount of variation in the net migration rates that is not explained by regional population density. (a) (b) figure 8: net migration and population density for the total population in (a) 1976-81 and (b) 2011-16 source: abs census data, author’s calculations 5. discussion the aim of this paper was to understand how the migration patterns of australia’s 65+ population changed over the past forty years in comparison to the 0-64 and total populations. this was achieved by first establishing the system-wide dynamics before investigating the underlying spatial distribution of flows between regions. the migration intensity of the 65+ group followed a similar downward trend to that observed in the 0-64 and total populations. though the share of older people increased, the likelihood of the 65+ age group moving was approximately half that observed in the 0-64 and total populations and the decline in mobility was less rapid for the 65+ population than the population overall. conversely, the higher mei indicates greater asymmetry in the 65+ migration system with a number of regions gaining population at the expense of losses in others (shryock & siegel 1975). the temporal pattern of increase and decrease was similar to that seen in the 0-64 and total populations until 1996-2001, after which the trends diverged. notably, the effectiveness of the 65+ group increased in the latest census period after a prolonged period of decline. this could potentially signal a reversal of a longterm trend or simply be an anomaly as seen in 1986-81. for the 65+ group, the 0-64 group, and the total population, the overall decline in effectiveness indicates that the migration exerts increasingly less influence in the redistribution of the population over time. finally, the impact of migration on population redistribution summarised by the anmr was similar between all three groups despite inner sydney richmond-tweed nsw mid-north coast nsw outer brisbane gold coast caboolture north west qld outer perth pilbara wa kimberley wa act y = -1.3753x2 + 4.3206x 1.3181 r² = 0.5765 -20 -10 0 10 20 30 40 -2 -1 0 1 2 3 4 n et m ig ra tio n ra te log population density inner sydney richmond-tweed nsw mid-north coast nsw outer brisbane gold coast caboolture north west qld outer perth pilbara wa kimberley wa darwin y = -2.0074x2 + 5.2251x + 0.9443 r² = 0.2433 -20 -10 0 10 20 30 40 -2 -1 0 1 2 3 4 n et m ig ra tio n ra te log population density australian population studies 4 (1) 2020 borsellino 15 differences in the intensity and effectiveness. this indicates that the asymmetry indicated by the mei contributes more to the redistribution of the 65+ age group, while the higher cmi of the 0-64 and total populations indicate that it is the volume of moves that contributes most to the migration impact of these groups. while periods of increase in an otherwise ongoing decline may be attributed to short-term structural economic change (stillwell et al. 2000), the overall decline can likely be attributed to population ageing, changes to the demographic structure of regions, and the long-term stability of australia’s settlement system (champion et al. 2018). further examining the regions that are gaining and losing reveals a directionality to this shift in the spatial orientation of migration over the forty year period. the 1970s and 1990s were characterised as periods of pronounced counter-urban movement with flows generally directed towards highamenity regions in northern new south wales (richmond-tweed and mid-north coast), southeast queensland (caboolture and the gold coast) and southwest western australia (argent, smailes & griffin 2007). however, emerging trends indicate that regions along the urban periphery and select inland rural regions in victoria and new south wales are becoming key destinations for the 65+ group. the scatterplots reveal that moderately dense regions experienced the highest net gain with accompanying losses in very high and low density regions. however, shifts in the regional patterns of gains and losses suggest that the attractiveness of places changes over time so gains may be linked to migrant rather than regional characteristics (haas & serow 1993). for example, moves to the suburban peripheries might be to be near family while increasing gains in less-serviced regional towns – such as those in inland victoria – may be driven by post-war cohorts returning to their place of birth upon retirement (jauhiainen 2009). the spatial pattern of this rural encroachment was not matched by flows from the 0-64 age group, which exhibited a distinct resurgence towards low-density regions in 2001-06 and 2006-11. the accompanying substantial drops in the magnitude of net migration gains for this age group were likely linked to growth of the working-age population in resource-rich outback towns – namely the pilbara and central western australia – and larger regional bases along the mid-queensland coast during the mining boom (robertson, blackwell & mcfarlane 2017). the lack of an inflection point suggests that the 65+ population was largely unaffected by these economic-driven flows, though the devaluing of retirement nest eggs during the global financial crisis may have marginally accelerated the decline in migration as people were financially unable to move (wahlquist 2017). while increasing affordability of coastal living in the 1970s and 1980s initially attracted retirees to non-metropolitan regions along the east and southwest coasts (sander 2011), the declining popularity of urban and peri-urban coastal regions with the 65+ population suggests that they may be being priced out of these regional housing markets. lifestyle-driven population growth coupled with the slow release of land has reshaped housing and labour markets in these traditional retirement locations (particularly those in southeast queensland and northern new south wales) (beer 2017; beer et al. 2011). understanding housing choices will be more important in the coming years, particularly as key retirement destinations change (painter & lee 2009). moves by older people are often associated with a change in housing tenure or dwelling size (clark & deurloo 2006; painter & lee 2009), and triggered by changes to personal circumstances or household composition (abramsson & andersson 2015). the desire to continue living independently despite deteriorating 16 borsellino australian population studies 4 (1) 2020 health will necessitate purpose-built accommodation that is often in short supply in regional areas (beer et al. 2011). the age-specific dynamics of gain and loss will have significant consequences for local communities as older people have different service needs and asset levels from the general population. while most regions experienced consistent net gain or loss across all ages and census periods, a number of mostly rural and remote regions along the new south wales and victoria border experienced net gain of the 65+ population and the net loss of the 0-64 population during one or more census periods. however, these inflows of the 65+ age group were rarely sufficient to combat the long-term trends of net loss driven by younger age groups. migration can improve socioeconomic growth in rural areas (davies 2008), though large flows of older people can also increase the pressure on already limited health care resources (costello 2009). the concurrent loss of the working-age population may undermine social and economic support systems, further accelerated by the withdrawal of key government services from regional communities (alston 2004). while this paper is primarily concerned with understanding the spatial and temporal dynamics of movement, regions must be equally concerned with the increasing proportion of older people who choose to age-in-place. the decline in migration intensity despite increasing numbers of older people suggests a reluctance to relocate, a lack of suitable housing choices, or an inability to meet the costs of moving (hansen & gottschalk 2007). alternatively, changes in the timing of key life course events such as the trend of working longer may reduce the migration freedom of older people, with a portion of moves simply reflecting late-life job relocations (day 2019). 6. conclusion the system-wide metrics revealed an overall decline in the likelihood and impact of between-region moves for the 65+ population over the past forty years, while comparing change in net migration rates showed shift in the direction and magnitude of moves that differentiate this age group from the 0-64 and total populations. although the trend of counter-urbanisation established in the 1970s has persisted for the 65+ group, the destinations have become less focused over time, reflecting the shifting demographic and socioeconomic landscape. these regional communities will need to invest in appropriate amenities, services and infrastructure to meet the needs of the growing 65+ population (denny 2018). further stratifying the analysis to distinguish between moves made by those aged 65-74, 75-84 and 85+ and/or including within-region moves would help to explain how the search for amenity or healthcare and social support changes as people age. extending the analysis to encompass a suite of younger age groups could further illuminate the causal relationships that underscore how migration behaviour changes across the life course (sander & bell 2008) and help us understand how age-differentiated patterns of migration accommodate changes to subnational population age structures. key messages • older australians are consistently less mobile than the overall population. • the decline in migration propensity between regions is less pronounced for the 65+ population. australian population studies 4 (1) 2020 borsellino 17 • migration towards traditional retirement destinations along the eastern coast has declined in favour of increasing growth in inland regions. • emerging destinations need to be prepared to provide appropriate housing, services and infrastructure to accommodate growth of older population. acknowledgements the author thanks dr elin charles-edwards, dr aude bernard, and professor jonathan corcoran for providing comments on a draft version of this paper. this research was supported by an australian government research training scholarship. references abramsson m and andersson e (2015) changing preferences with ageing – housing choices and housing plans of older people. housing, theory and society 33(2): 217-241. alston m (2004) ‘you don’t want to be a check-out chick all your life’: the out-migration of young people from australia’s small rural towns. the australian journal of social issues 39(3): 299-313. argent n, smailes pj and griffin, t (2007) the amenity complex: towards a framework for analysing and predicting the emergence of a multifunctional countryside in australia. geographical research 45(3): 217-232. beer a (2017) a housing affordability crisis in regional australia? yes, and here’s why. the conversation. https://theconversation.com/a-housing-affordability-crisis-in-regional-australia-yes-and-hereswhy-71808. accessed on 27 january 2020. beer a, tually s, rowley s, haslam mckenzie f, schlapp j, birdsall-jones c and corunna v (2011) the drivers of supply and demand in australia’s rural and regional centres. australian housing and urban research institute, melbourne. champion t, cooke t and shuttleworth i (eds.) (2018) internal migration in the developed world: are we becoming less mobile? london: routledge. bell m, blake m, boyle p, duke-williams o, rees p, stillwell j and hugo g (2002) cross-national comparison of internal migration: issues and measures. journal of the royal statistical society series a (statistics in society) 165(3): 435-464. bell m, rees p, blake m and duke-williams o (1999) an age-period-cohort database of inter-regional migration in australia and britain, 1976-96. university of adelaide, adelaide. https://core.ac.uk/download/pdf/51003.pdf. accessed 17 january 2020. clark wa and deurloo m c (2006) aging in place and housing over-consumption. journal of housing and the built environment 21(3): 257-270. conway k s and houtenville a j (2003) out with the old, in with the old: a closer look at younger versus older elderly migration. social science quarterly 84(2): 309-328. costello l (2009) urban–rural migration: housing availability and affordability. australian geographer 40(2): 219-233. davies a (2008) declining youth in-migration in rural western australia: the role of perceptions of rural employment and lifestyle opportunities. geographical research 46(2): 162-171. davies a and james a (2011) geographies of ageing: social processes and the spatial unevenness of population ageing. london: routledge. day c (2019) house prices post-gfc: more household debt for longer. economic analysis and policy 64: 91-102. https://theconversation.com/a-housing-affordability-crisis-in-regional-australia-yes-and-heres-why-71808 https://theconversation.com/a-housing-affordability-crisis-in-regional-australia-yes-and-heres-why-71808 https://core.ac.uk/download/pdf/51003.pdf 18 borsellino australian population studies 4 (1) 2020 denny l (2018) migration is slowing australia’s rate of ageing, but not necessarily in the regions. the conversation. https://theconversation.com/migration-is-slowing-australias-rate-of-ageing-butnot-necessarily-in-the-regions-94970. accessed on 27 january 2020. haas w h and serow w j (1993) amenity retirement migration process: a model and preliminary evidence. the gerontologist 33(2): 212-220. hansen e b and gottschalk g (2007) what makes older people consider moving house and what makes them move? housing, theory and society 23(1): 34-54. hugo g (2013) the changing demographics of australia over the last 30 years. australasian journal on ageing 32(s2): 18-27. jackson n and felmingham b (2002) as the population clock winds down: indicative effects of population ageing in australia’s states and territories. journal of population research 19(2): 97-117. jauhiainen j s (2009) will the retiring baby boomers return to rural periphery? journal of rural studies 25(1): 25-34. kim s, kim h and kim w g (2003) impacts of senior citizens’ lifestyle on their choices of elderly housing. journal of consumer marketing 20(3): 210-226. mcmanus p and pritchard b (2000) geography and the emergence of rural and regional australia. australian geographer 31(3): 383-391. morrison p a, bryan t m and swanson d a (2004) internal migration and short-distance mobility. in: siegel j s and swanson d a (eds.) the methods and materials of demography. san diego: elsevier academic press; 493-521. painter g and lee k (2009) housing tenure transitions of older households: life cycle, demographic, and familial factors. regional science and urban economics 39(6): 749-760. plane d a and rogerson p a (1991) tracking the baby boom, the baby bust, and the echo generations: how age composition regulates us migration. professional geographer 43(4): 416-430. rees p, bell m, duke-williams o & blake m (2000) problems and solutions in the measurement of migration intensities: australia and britain compared. population studies 54(2): 207-222. rees p, bell m, kupiszewski m, kupiszewska d, ueffing p, bernard a, charles-edwards e and stillwell j (2016) the impact of internal migration on population redistribution: an international comparison. population, space and place 23(6): e2036. robertson s a, blackwell b d and mcfarlane j a (2017) the viability of remote mining communities: insights from community perceptions and employment impact assessments. impact assessment and project appraisal 35(4): 310-324. rogers a (1988) age patterns of elderly migration: an international comparison. demography 25(3): 355370. rogers a and castro l j (1981) model migration schedules. international institute for applied systems analysis, laxenburg, austria. sander n (2011) retirement migration of the baby boomers in australia: beach, bush or busted? phd thesis, the university of queensland, australia. sander n and bell m (2008) migration and the life course: does retirement trigger interregional moves? paper presented to european population conference, barcelona, spain, 9-12 july. shryock h s and siegel j s (1975) the methods and materials of demography. us bureau of the census, washington dc. stillwell j, bell m, blake m, duke-williams o and rees p (2000) net migration and migration effectiveness: a comparison between australia and the united kingdom, 1976–96 part 1: total migration patterns. journal of the australian population association 17(1): 17-38. tonts m and greive s (2002) commodification and creative destruction in the australian rural landscape: the case of bridgetown, western australia. australian geographical studies 20(1): 58-70. https://theconversation.com/migration-is-slowing-australias-rate-of-ageing-but-not-necessarily-in-the-regions-94970 https://theconversation.com/migration-is-slowing-australias-rate-of-ageing-but-not-necessarily-in-the-regions-94970 australian population studies 4 (1) 2020 borsellino 19 wahlquist c (2017) ‘they’ve lost the lot': how the australian mining boom blew up in property owners' faces. https://www.theguardian.com/australia-news/2017/may/12/theyve-lost-the-lot-how-theaustralian-mining-boom-blew-up-in-property-owners-faces. accessed on 27 january 2020. https://www.theguardian.com/australia-news/2017/may/12/theyve-lost-the-lot-how-the-australian-mining-boom-blew-up-in-property-owners-faces https://www.theguardian.com/australia-news/2017/may/12/theyve-lost-the-lot-how-the-australian-mining-boom-blew-up-in-property-owners-faces abstract background aim data and methods five-year interval census migration data were used to examine the migration flows of the 65+ population between 1976 and 2016. a range of metrics was used to measure the level or intensity of internal migration and its impact on the redistribution of ... results the propensity to migrate declined over the forty years for all ages but proportionally less for the over 65+ group. gains in the 65+ population were spread across a range of middle-density regions with losses from the most urban and remote areas. the... conclusions the system-wide metrics showed an overall decline in migration propensity between regions and changes in the redistributive impact of migration by age. the net migration maps and scatterplots revealed large changes in the regional growth and decline o... key words migration, ageing, 65+, counter-urbanisation, australia. 2. data key messages acknowledgements references australian population studies 2021 | volume 5 | issue 2 | pages 15–28 © kerr et al. 2021. published under the creative commons attribution-noncommercial licence 3.0 australia (cc by-nc 3.0 au). journal website: www.australianpopulationstudies.org measuring area-level disadvantage in australia: development of a locally sensitive indicator jessica a. kerr* murdoch children’s research institute; university of melbourne; university of otago suzanne mavoa murdoch children’s research institute; university of melbourne ralf-dieter schroers deakin university; university of canberra serryn eagleson frontiersi daniel j. exeter university of auckland alison watkins university of canterbury; auckland university of technology lukar thornton deakin university; university of antwerp *corresponding author. email: jessica.kerr@mcri.edu.au. address: murdoch children’s research institute, the royal children’s hospital, 50 flemington road, parkville, victoria 3052, australia paper received 27 july 2021; accepted 1 november 2021; published 29 november 2021 abstract background in australia, the socio-economic indexes for areas (seifa), which includes the index of relative socioeconomic disadvantage (irsd), captures the socioeconomic characteristics of areas. because seifa rankings are relative to the country or state, the decile categorisations may not reflect an area’s socioeconomic standing relative to areas nearby. aims the aim of the research was to explore whether irsd rankings could be re-ranked to become locally sensitive. data and methods using existing seifa data to redistribute the membership of current decile irsd groups, we tested three methods to re-rank all sa1 areas relative to the nearest areas capped at: (1) the nearest 99 neighbours, (2) a population threshold of 50,000 (3) a distance threshold of 10 km. results the reclassification of seifa irsd deciles was largest (up to 8 decile points of change) when comparing the nearest neighbour and population threshold local methods to current state-based rankings. moreover, compared to using current national and state seifa irsd rankings, the use of local rankings resulted in more evenly distributed deciles between cities, regional, and remote areas. conclusions because seifa irsd rankings are used to allocate resources and health services, we encourage the combined use of a state and local ranking to refine areas considered the most disadvantaged. key words disadvantage; deprivation, neighbourhood; local; area-level; australia. http://www.australianpopulationstudies.org/ mailto:jessica.kerr@mcri.edu.au 16 kerr et al. australian population studies 5 (2) 2021 1. introduction using data from the 1981 uk census, townsend established that area-level circumstances are an important measure of deprivation independent of poverty or individual socioeconomic position (sep) (townsend 1987). deprivation was defined as “…observable and demonstrable disadvantage relative to the local community or the wider society or nation to which an individual, family or group belongs” (townsend 1987 p. 125). therefore, in comparison to individual level measures of socioeconomic position such as income, education and occupation, area-based measures more commonly capture neighbourhood deprivation or disadvantage by aggregating social and material conditions of residential areas (e.g., % residents unemployed, % residents that own a car) into a single figure estimate. such area-level socioeconomic conditions are often strongly associated with health outcomes both at the individualand population-level, including long-term illness and mortality (galobardes et al. 2004; meijer 2013; king et al. 2006). these effects can accumulate over the life course, influencing an increased allostatic load (i.e., cumulative biological risk) and subsequently health status in adulthood (cornaz et al. 2009; gustafsson et al. 2014). following townsend’s pioneering work, contemporary area-level socioeconomic indicators are measured in a variety of ways including composite indexes derived from census (atkinson et al. 2014; department of the environment, transport and the regions 2000; norman et al. 2019; townsend et al. 1988) or routinely collected administrative data (exeter et al. 2017; noble et al. 2006; smith et al. 2015). in australia, the socio-economic indexes for areas (seifa) are used to determine the relative socioeconomic characteristics of different areas in australia (abs 2011; abs 2016b). this set of indexes is a product developed by the australian bureau of statistics using data gathered in the fiveyearly census. broadly, to determine the collective socioeconomic characteristics of people living in an area, advantage and disadvantage are defined by people’s access to social and material resources, including their ability to participate in society (abs 2011). constrained by census questions, each of the four seifa indexes is a weighted combination of selected variables (see data and methods) measuring the aggregated attributes of the area’s residents. much of the methodological discussion in relation to such indexes in australia and internationally relates to the choice of variables that make up each index or the size of the area-level unit being explored (jordan et al. 2004). this latter issue is the well-recognised modifiable areal unit problem (maup) where the scale, or aggregation, one uses to analyse the information can produce different or distorted findings (brunsdon & comber 2020). indexes can also be problematic for areas with very diverse neighbourhoods, such as the australian capital territory (act). indeed, the act government has highlighted that seifa critically underestimates disadvantage in the act (act government 2012) because in the act many disadvantaged households are adjacent to advantaged households (due to mixed tenure housing policy), which seifa does not capture because it averages across larger geographic areas (tanton et al. 2015). many issues related to small-area measures of disadvantage have been discussed by allik and colleagues (allik et al. 2020). however, a largely unrecognised problem for australia’s seifa indexes and other international examples (atkinson et al. 2014; exeter et al. 2017; townsend et al. 1988) is that the current rankings, and subsequent decile groups, are based on relative rankings across the whole country or within each individual state or territory. therefore, indexes based on relative ranks australian population studies 5 (2) 2021 kerr et al. 17 within a country or state may not reflect an area’s standing relative to its neighbours and nearby areas. consequently, regional areas are ranked against both regional and urban areas (and vice versa) and regional areas with low-levels of disadvantage relative to other nearby areas cannot be distinguished because their level of deprivation is ranked relative to all other areas within the state or within the country. the use of the decile groups (as recommended by the abs (2011, 2016b)) makes it difficult to isolate the most disadvantaged areas relative to neighbouring areas and may misrepresent the levels of disadvantage present, especially in regional areas of australia. the abs recognise some of these limitations and note that users should be cautious when using the current seifa indexes for very remote areas of australia (abs 2011). the use of state-based relative rankings and resulting decile groups can also be problematic for localities bordering states or territories. these limitations are important because place and health research relies on, and often assumes, locally sensitive measures of socioeconomic disadvantage. this is important to resolve as it is critical to employ suitable area-disadvantage indicators that are meaningful (hugo 2007). researchers and policy makers rely heavily on the seifa indexes to inform local and national policy development, to determine areas in need of funding and essential services, and to inform research examining relationships between area-level sep and various socialand health-related outcomes (abs 2011). therefore, there are potentially major consequences to misinterpreting local socioeconomic disadvantage and misdirecting resources. we direct interested readers to previous research that considers small area data zones (allik et al. 2020; noble et al. 2006; zhao and exeter 2016), and to researchers who are beginning to consider relative or local ‘neighbouring disadvantage’ within definitions of deprivation (cox et al. 2007; graif et al. 2016; pearson et al. 2013; zhang et al. 2011; zhang et al. 2013). however, most relevant to our own concept is rae’s alternative approach to measuring neighbourhood deprivation within the united kingdom (rae 2009). advocating an approach to better understand the relationships areas have with their wider neighbours, rae developed a method to assess deprived areas as integrated neighbours rather than isolated entities. despite its use in evidence-based policy development across england, the uk index of multiple deprivation (imd) lacked metrics specific to locally sensitive spatial dimensions. by calculating the mean value of imd for three different levels of adjacent neighbourhoods, rae developed a nearest neighbour imd as an alternative method to interpret local level area deprivation. similarly, we contend that the current use of seifa rankings and the subsequent decile groups is limited for locally sensitive research questions and policy decisions. the shortcomings cannot be overcome through limiting and re-ranking areas within local government areas due to issues that would be introduced through edge-effects. similar discussions have been had in australia where the late professor graeme hugo (hugo 2007) previously advocated a move away from the traditionally employed administrative units in order to define more relevant spatial units when considering social, economic and environmental planning purposes. this work led to the restructuring of the statistical geography of australia with the intention of using topography and built environments to homogenously group people with like characteristics. for example, in rural areas it is common for communities to form around the intersection of major roads and such a community should be kept 18 kerr et al. australian population studies 5 (2) 2021 together for analysis. in urban areas, a similar approach has been undertaken where topographic features such as roads and rivers can unite or divide an area – depending on the number of crossing points (bridges/traffic lights). to define the statistical geography, criteria were assigned by an expert panel with a weighting (high, medium or low) to facilitate the drawing of the spatial boundaries (abs 2003; eagleson et al. 2003). the intention of the restructured statistical geography based on mesh blocks (the smallest area defined by the abs) was to allow flexibility (e.g., customised units) of data aggregation for the output of demographics and calculation of indexes by both researchers and policy makers. however, whilst broad land use characteristics, counts of dwellings and population have been published at the mesh block scale, seifa output remains confined to larger administration areas (e.g., sa1 or sa2). due to limited data availability on the mesh block scale, we explored whether seifa at the sa1 level could be re-ranked to become locally sensitive. overall, research directly related to seifa indicators is remarkably scarce. by developing methods to re-rank the index of relative socioeconomic disadvantage (irsd), we aimed to redistribute the membership of current decile groups to establish a locally sensitive indicator of relative disadvantage for australia. in this proof-of-concept approach, to investigate whether existing seifa irsd rankings could be redistributed to become locally sensitive, we explore three methods by re-ranking areas relative to: (1) the nearest administrative (sa1) areas capped at the nearest 99 neighbours (nearest neighbour), (2) the nearest areas capped at a population threshold of 50,000 (population threshold), and (3) the nearest areas capped at a distance threshold of 10km (distance threshold). unlike the approach used by rae (2009), our measure does not alter or recalculate the seifa scores allocated to areas, but simply re-ranks them. ultimately our vision is for a local irsd ranking to be tested, adapted, and adopted for wide use to inform relevant policy and research within australia. 2. data and methods 2.1 description of current seifa irsd rankings the construction of the irsd rankings have been outlined in detail elsewhere (abs 2011, 2016b). briefly, seifa scores are derived for almost all statistical area level 1 (sa1) units across australia (n=55,028), with a few excluded mainly due to issues such as low population size (abs 2016a). sa1s are built from whole mesh blocks and are intended to remain relatively stable over time, although the boundaries are updated when there has been substantial population growth in given areas. sa1s are the smallest geographical unit for the release of seifa with an average population size of 400 residents (range 200 to 800 residents) (abs 2016a). by nature of design, sa1s outside of cities are often smaller in population and larger in geographical coverage. the 2016 version of the irsd score for each sa1 is a weighted combination of 16 census-collected variables that indicate levels of social and material disadvantage at the sa1 level. scores have been standardised to have a mean of 1,000 and a standard deviation of 100. for an area to receive a score of 1,000, all 16 indicators must be equal to the national average. with this score, sa1 areas are ranked on a continuum and then allocated to decile groups from 1 (most disadvantaged, i.e., a high proportion of relatively disadvantaged people in this area) to 10 (least disadvantaged). abs australian population studies 5 (2) 2021 kerr et al. 19 recommend using these index rankings and deciles, rather than the numerical score. the rankings and decile groups are made available at a national level and at a state/territory level. 2.2 methods to re-rank seifa irsd to achieve a more locally sensitive irsd, membership of decile groupings was redistributed by reranking each sa1 across australia according to the three methods described below (code currently available from authors upon request). to test our concept, we employed arbitrary cut off points for each method described below. it is important to note that each of these methods rely on first generating a geometric centroid for each sa1 polygon (arcgis – points forced to be inside polygon) and that the measures are not restricted to state borders (i.e., a neighbouring sa1 could be on the other side of the state boundary). method 1: re-rank sa1s according to neighbours (nearest neighbour) to explore whether existing seifa irsd rankings (that are derived from nearly all sa1 units across australia) could be re-ranked to become locally sensitive, this first method scored each sa1 against its 99 nearest neighbours (total of 100 inclusive of the sa1 that the nearest 99 was measured from). once generated, the 99 nearest neighbours, i.e., 99 points in closest proximity, were identified using r and the ‘knearneigh’ function (r package ‘spdep’) and an origin-destination matrix was created. the scores for each sa1 were given a fixed rank based on how each sa1 scored in relation to its 99 nearest neighbours. this process resulted in the creation of new membership of decile groups relative to the sa1s nearest neighbours. method 2: re-rank sa1s according to population (population threshold) in method 2, we constructed a nearest-neighbour matrix similar to method 1, with a threshold of minimum 50,000 residents, rather than 99 neighbours. this was achieved using r and the ‘knn.index’ function of the ‘fnn’ package, by iteratively identifying nearest neighbouring areas and summing up corresponding population values, until the value of at least 50,000 was reached. the final sa1 neighbour added may cause the total to slightly exceed this threshold. method 3: re-rank sa1s according to distance (distance threshold) nearest neighbours were identified within a distance of 10km measured from the centroid of the sa1 polygon. this was implemented through applying the ‘dnearneigh’ function (r package ‘spdep’), identifying neighbour points by euclidean distance (10km). for this method, the minimum number of neighbouring areas was set at 30. this means that areas that have <30 sa1s within 10km are not included in this distance-based ranking. thus, updated decile groups are available at the national and state level (abs generated) and the local level (generated using these three methods above). comparisons of these decile groups were undertaken and graphs produced using stata v16.1. 20 kerr et al. australian population studies 5 (2) 2021 3. results we provide detail on the variations by state in the distribution of sa1s to decile groups according to the national, state and three local methods in figure 1. further details of the irsd score breakdown within decile categories for each of the ranking methods is available from the authors on request. data presented in figure 1 demonstrates subtle and sometimes large differences in the distribution of sa1s to decile groups depending on the ranking method used. whilst at times the overall distribution seems to not have shifted a large amount, a more detailed look at the changes in figure 2 illustrates the degree to which the decile membership of sa1s can shift when using the various local methods compared to the state-based decile groups using the state of victoria as an example (possible range -9 to +9). for the nearest neighbour and population threshold methods some sa1s differed by up to 8 decile categories, indicating that their decile group would vary greatly depending on the use of state or local decile groups. for the distance threshold method, the differences at the tail ends of the distribution were not as great. national and state decile groups were also compared in the same way for victoria (not presented) and it is clear that the distribution does not shift as meaningfully between state and national decile groupings, as it does between the state and the three local decile groups. figure 1: state-based differences in the percentage of sa1 in each decile of disadvantage using the national, state, and local ranking methods source: authors’ calculations. australian population studies 5 (2) 2021 kerr et al. 21 (a) nearest neighbour decile grouping (b) population-based decile grouping (c) distance-based decile grouping figure 2: distribution of shift in decile group membership between state decile groups and local decile groups within victoria source: authors’ calculations. note: values represent how far a sa1 shifted in terms of decile group table 1 presents the percentage of each state’s sa1s that were in both the state and local nearest neighbour most disadvantaged decile. the combined use of a state and local ranking is important for resource allocation purposes as it allows further refinement of areas considered the most disadvantaged. for sa1s in major city regions, excluding the act (82%), less than half of the sa1s categorised as most disadvantaged according to the state rankings were also in the local nearest neighbour most disadvantaged decile. these percentages were generally higher in inner regional and outer regional areas, but still suggested that between half and three quarters of the sa1s were in the most disadvantaged decile when combining both the state and nearest neighbour rankings compared to just the state-based ranking. 22 kerr et al. australian population studies 5 (2) 2021 table 1: number and percentage of sa1s in state’s most disadvantaged decile (decile 1) that were also in the most disadvantaged decile (decile 1) for the local nearest neighbour measure major city inner regional outer regional state no. of sa1s in state decile 1 no. of sa1s in nearest neighbour decile 1 no. (%) of sa1s in state decile 1 that are also in nearest neighbour decile 1 no. of sa1s in state decile 1 no. of sa1s in nearest neighbour decile 1 no. (%) of sa1s in state decile 1 that are also in nearest neighbour decile 1 no. of sa1s in state decile 1 no. of sa1s in nearest neighbour decile 1 no. (%) of sa1s in state decile 1 that are also in nearest neighbour decile 1 act 101 105 83 (82.2%) 0 0 0 nsw 1071 1090 501 (46.8%) 438 366 267 (61.0%) 219 127 118 (53.9%) nt 2 32 2 (100%) qld 418 566 207 (49.5%) 385 213 183 (47.5%) 229 170 124 (54.1%) sa 256 216 96 (37.5%) 33 64 24 (72.7%) 95 51 47 (49.5%) tas 94 89 64 (68.1%) 39 48 33 (84.6%) vic 889 811 319 (35.9%) 353 276 204 (57.8%) 119 72 68 (57.1%) wa 307 339 141 (45.9%) 68 62 51 (75.0%) 79 44 44 (55.7%) source: authors’ calculations figure 3: remoteness area differences in the percentage of sa1s in each decile of disadvantage using the national, state, and local ranking methods source: authors’ calculations. australian population studies 5 (2) 2021 kerr et al. 23 using the national and state rankings, over one third of major city sa1s are within deciles 8 to 10 (the least disadvantaged) whereas for inner regional, outer regional and remote areas over one third of sa1s are within deciles 1 to 3 (the most disadvantaged) (figure 3). for each of these four areas, the distribution of sa1s within decile categories is more even when using the local ranking methods. we are cautious to make conclusions on the very remote sa1s given the unique characteristics of these areas. a visual example of the change in decile rankings for sa1s within and around the inner regional city of bendigo in victoria by comparing the national, state and local ranking methods is presented in figure 4. this highlights a substantial reclassification of sa1 deciles using the local ranking methods in a single regional township and results presented earlier indicate this is likely to be replicated elsewhere. figure 4: comparison of the national, state, and local decile groupings around bendigo (victoria) source: authors’ calculations. 4. discussion area-based measures of disadvantage are critical for resource allocation, advocacy, and equitable planning (salmond & crampton 2012), and therefore it is imperative to accurately reflect relative levels of disadvantage. current indicators of area-level disadvantage in australia rank areas based on two separate geographic levels, national and state/territory. in this study we created a third, local, level for ranking the seifa disadvantage index and proposed three approaches to achieve this. the methods we proposed in this research utilised existing seifa data and expanded upon current approaches to ranking areas to better capture relative disadvantage. we focused on australian data since the large geographic area of australia and distribution of the population across this large area makes this issue even more pertinent, particularly outside of major cities as demonstrated by our findings against the categories of remoteness. however, we also note that international measures of 24 kerr et al. australian population studies 5 (2) 2021 deprivation and disadvantage potentially suffer similar limitations (allik et al. 2020; braveman et al. 2005; greco et al. 2019; quick & luan 2021) and whilst many ideas have been put forward to address these (mostly in relation to the creation of the composite indexes themselves), we believe our local ranking approach provides a unique contribution. thus, as a proof of concept, we created three local ranking methods and reallocated the decile categorisation used in the seifa irsd based on these new rankings. after this initial conceptualisation and validation work, we anticipate future research will further develop and refine the most appropriate local measure. our results highlight noticeable differences in the local rankings decile categories allocated to sa1s compared to that provided by national and state level rankings. current seifa indicators may guide resource allocation, and with a high number of areas considered to have high levels of disadvantage under the state and national ranking system, it likely means that not all areas will be provided with the level of support required. by combining the state and local nearest neighbour ranking (as an example), a lower number of sa1s were considered most disadvantaged in comparison to using the state measure alone. consequently, using this combination ranking may help to refine and identify priority areas with the greatest need of resource allocation. this may be particularly important in smaller regional local government areas where social support mechanisms may be underfunded, meaning that the prioritisation of resource allocation is critical (parliament of australia 2006). we provide proof of concepts for three unique methods to rank area-level disadvantage at a local level in australia. unlike the currently available irsd state decile groupings, our options for local groupings are less subject to edge-effects as they can cross state boundaries. furthermore, we demonstrate that our methods improve the ability to distinguish areas of disadvantage relative to other nearby areas and populations, particularly outside of major cities. we acknowledge that the methods and thresholds selected could be altered in a number of ways. conclusions reached can depend on the input data and, similar to maup (brunsdon & comber 2020; openshaw 1979; openshaw 1984), the results of this study are specific to the methods and thresholds we used. to promote progress toward the development of more standardised methods for the measurement of local area-level disadvantage in australia, we propose that research be undertaken to test alternative approaches. these could include further exploring the use of k-means clustering techniques to generate typology of deprivation classes that may be applied locally within urban and rural settings across australia. based on our best judgement, we employed arbitrary cut off points for what constitutes a local neighbourhood (e.g., 99 neighbouring areas; areas within a 10km radius; nearest 50,000 population). all three methods were based on the distance to the centroid created for the nearest neighbour decile grouping, when in fact the road layout and presence of physical boundaries may create barriers between the sa1s. using mesh blocks, it may have been possible to create an estimate of the population-weighted centroid which may alter results slightly for some of the more sparsely populated and therefore geographically larger sa1s in regional and remote areas. additionally, because we set the minimum number of neighbouring areas at 30 for the distance-based decile groupings, we acknowledge that this method of re-ranking does not work as well in more sparsely populated areas and those near natural barriers such as coastlines. australian population studies 5 (2) 2021 kerr et al. 25 in this study we use decile categorisation but do not address related problems. categorisation of exposures measures can cause a wide range of issues (lamb & white 2015) and, as recently highlighted by allik et al. (2020), it is important to recognise that values near to the cut-off points between categories could easily belong to another category. this limitation means that caution is needed when using of decile categorisation as the deciding factor motivating a policy intervention. 5. conclusions the relationship between space and place is complex. while data collection systems and technology for processing data have evolved, we still need better ways to understand disadvantage and the process for equitable distribution of resources that help reduce the impact of socioeconomic disadvantage. the current methods demonstrate that relative levels of socioeconomic disadvantage at the national or state level differ to that at a local level. because current rankings make it challenging to isolate disadvantaged areas relative to neighbouring areas, policy decisions relating to the distribution of health and other services may be misguided. as a first step to improve this limitation, we offer three methods to better measure local disadvantage throughout australia and we encourage researchers to trial these in future research. rather than replacing current seifa rankings, we envision these local ranking will complement existing national or state rankings. key messages • australia’s area-based socioeconomic indexes (seifa) rank areas relative to country or state. • to better reflect disadvantage relative to nearby areas, we developed local rankings. • results are presented for three newly developed local rankings for australia. • we encourage the combined use of a state and local rankings for resource allocation, to ensure that the most disadvantaged areas are not overlooked. acknowledgments jak is supported by a melbourne children’s lifecourse postdoctoral fellowship, funded by royal children’s hospital foundation grant #2018-984. lifecourse acknowledges all collaborators, cohort representatives and participants; https://lifecourse.melbournechildrens.com/contact/ for further details. sm was supported by an australian national health and medical research council early career fellowship #1121035. this research did not receive any specific grant from funding agencies in the public, commercial, or not-for-profit sectors. research at the murdoch children’s research institute is supported by the victorian government’s operational infrastructure support program. the views expressed in this paper are those of the authors and may not reflect those of their organisational affiliations, nor of other collaborating individuals or organisations. references abs (2003) information paper: mesh blocks australia. catalogue no. 1209.0. canberra: abs. abs (2011) technical paper: socio-economic indexes for areas (seifa). catalogue. no. 2033.0.55.001. canberra: abs. https://lifecourse.melbournechildrens.com/contact/ 26 kerr et al. australian population studies 5 (2) 2021 abs (2016a) australian statistical geography standard (asgs): volume 1 main structure and greater capital city statistical areas. catalogue no. 1270.0.55.001. canberra: abs. abs (2016b) technical paper: socio-economic indexes for areas (seifa). catalogue. no. 2033.0.55.001. canberra: abs. allik m, leyland a, ichihara m y t, & dundas r (2020) creating small-area deprivation indices: a guide for stages and options. journal of epidemiology & community health 74(1): 20-25. http://dx.doi.org/10.1136/jech-2019-213255 atkinson j, salmond c, & crampton p (2014) nzdep2013 index of deprivation. wellington: department of public health, university of otago. https://www.otago.ac.nz/wellington/otago069936.pdf australian capital territory government (2012) detecting disadvantage in the act: report on the comparative analysis of the seifi and seifa indexes or relative socio-economic disadvantage in the australian capital territory. canberra: act government. https://www.cmtedd.act.gov.au/__data/assets/pdf_file/0006/464775/dectectdisadvantage.pdf braveman p a, cubbin c, egerter s, chideya s, marchi k s, & metzler m (2005) socioeconomic status in health research: one size does not fit all. journal of the american medical association 294(22): 2879-2888. https://doi.org/10.1001/jama.294.22.2879 brunsdon c & comber a (2020). opening practice: supporting reproducibility and critical spatial data science. journal of geographical systems 23: 477-496. https://doi.org/10.1007/s10109-02000334-2 cornaz s, taffé p, & santos-eggimann b (2009) life-course socioeconomic environment and health risk behaviours. a multilevel small-area analysis of young-old persons in an urban neighbourhood in lausanne, switzerland. health & place 15(1): 273-283. https://doi.org/10.1016/j.healthplace.2008.05.007 cox m, boyle p j, davey p g, feng z, & morris a d (2007) locality deprivation and type 2 diabetes incidence: a local test of relative inequalities. social science & medicine 65(9): 1953-1964. https://doi.org/10.1016/j.socscimed.2007.05.043 department of the environment transport and the regions (2000) measuring multiple deprivation at the small area level: the indices of deprivation 2000. london: department of the environment, transport and the regions. https://webarchive.nationalarchives.gov.uk/ukgwa/20100407204456/http://www.communities. gov.uk/archived/general-content/communities/indicesofdeprivation/indicesofdeprivation/ eagleson s, escobar f, & williamson i (2003) automating the administration boundary design process using hierarchical spatial reasoning theory and geographical information systems. international journal of geographical information science 17(2): 99-118. https://doi.org/10.1080/713811748 exeter d j, zhao j, crengle s, lee a, & browne m (2017) the new zealand indices of multiple deprivation (imd): a new suite of indicators for social and health research in aotearoa, new zealand. plos one 12(8) e018260. https://doi.org/10.1371/journal.pone.0181260 galobardes b, lynch j w, & davey smith g (2004) childhood socioeconomic circumstances and causespecific mortality in adulthood: systematic review and interpretation. epidemiologic reviews 26(1): 7-21. https://doi.org/10.1093/epirev/mxh008 graif c, arcaya m c, & roux a v d (2016) moving to opportunity and mental health: exploring the spatial context of neighborhood effects. social science & medicine 162: 50-58. https://doi.org/10.1016/j.socscimed.2016.05.036 greco s, ishizaka a, tasiou m, & torrisi g (2019) on the methodological framework of composite indices: a review of the issues of weighting, aggregation, and robustness. social indicators research 141(1): 61-94. https://doi.org/10.1007/s11205-017-1832-9 http://dx.doi.org/10.1136/jech-2019-213255 https://www.otago.ac.nz/wellington/otago069936.pdf https://www.cmtedd.act.gov.au/__data/assets/pdf_file/0006/464775/dectectdisadvantage.pdf https://doi.org/10.1001/jama.294.22.2879 https://doi.org/10.1007/s10109-020-00334-2 https://doi.org/10.1007/s10109-020-00334-2 https://doi.org/10.1016/j.healthplace.2008.05.007 https://doi.org/10.1016/j.socscimed.2007.05.043 https://webarchive.nationalarchives.gov.uk/ukgwa/20100407204456/http:/www.communities.gov.uk/archived/general-content/communities/indicesofdeprivation/indicesofdeprivation/ https://webarchive.nationalarchives.gov.uk/ukgwa/20100407204456/http:/www.communities.gov.uk/archived/general-content/communities/indicesofdeprivation/indicesofdeprivation/ https://doi.org/10.1080/713811748 https://doi.org/10.1371/journal.pone.0181260 https://doi.org/10.1093/epirev/mxh008 https://doi.org/10.1016/j.socscimed.2016.05.036 https://doi.org/10.1007/s11205-017-1832-9 australian population studies 5 (2) 2021 kerr et al. 27 gustafsson p e, san sebastian m, janlert u, theorell t, westerlund h, & hammarström a (2014) lifecourse accumulation of neighborhood disadvantage and allostatic load: empirical integration of three social determinants of health frameworks. american journal of public health 104(5): 904910. https://doi.org/10.2105/ajph.2013.301707 hugo g (2007) space, place, population and census analysis in australia. australian geographer 38(3):335357. https://doi.org/10.1080/00049180701639760 jordan h, roderick p, & martin d (2004) the index of multiple deprivation 2000 and accessibility effects on health. journal of epidemiology & community health 58(3): 250-257. http://dx.doi.org/10.1136/jech.2003.013011 king t, kavanagh a m, jolley d, turrell g, & crawford d (2006) weight and place: a multilevel crosssectional survey of area-level social disadvantage and overweight/obesity in australia. international journal of obesity 30(2): 281-287. https://doi.org/10.1038/sj.ijo.0803176 lamb k e & white s r (2015) categorisation of built environment characteristics: the trouble with tertiles. international journal of behavioral nutrition and physical activity 12(1): 19. https://doi.org/10.1186/s12966-015-0181-9 meijer m (2013) neighbourhood context and mortality: an overview. in stock c & ellaway a (eds) neighbourhood structure and health promotion. boston, ma: springer; pp. 11-37. https://doi.org/10.1007/978-1-4614-6672-7_2. noble m, wright g, smith g, & dibben c (2006) measuring multiple deprivation at the small-area level. environment and planning a 38(1): 169-185. https://doi.org/10.1068/a37168 norman p, berrie l, & exeter d j (2019) calculating a deprivation index using census data. australian population studies 3(1): 30-39. https://doi.org/10.37970/aps.v3i1.45 openshaw s (1979) a million or so correlation coefficients: three experiments on the modifiable areal unit problem. in wrigley n (ed.) statistical applications in the spatial science. london: pion; pp. 127144. openshaw s (1984a) the modifiable areal unit problem. norwich: geobooks. openshaw s (1984b) ecological fallacies and the analysis of areal census data. environment and planning a 16(1): 17-31. https://doi.org/10.1068/a160017 parliament of australia (2006) the blame game: report on the inquiry into health funding. canberra: house of representatives standing committee on health and ageing. https://www.aph.gov.au/parliamentary_business/committees/house_of_representatives_com mittees?url=haa/./healthfunding/report.htm. accessed on 26 january 2021. pearson a l, griffin e, davies a, & kingham s (2013) an ecological study of the relationship between socioeconomic isolation and mental health in the most deprived areas in auckland, new zealand. health & place 19: 159-166. https://doi.org/10.1016/j.healthplace.2012.10.012 quick m & luan h (2021) the spatial structure of socioeconomic disadvantage: a bayesian multivariate spatial factor analysis. international journal of geographical information science 35(1): 63-83. https://doi.org/10.1080/13658816.2020.1759807 rae a (2009) isolated entities or integrated neighbourhoods? an alternative view of the measurement of deprivation. urban studies 46(9): 1859-1878. https://doi.org/10.1177/0042098009106019 salmond c e & crampton p (2012) development of new zealand's deprivation index (nzdep) and its uptake as a national policy tool. canadian journal of public health 103: s7-s11. https://www.jstor.org/stable/41995682 smith t, noble m, noble s, wright g, mclennan d, & plunkett e (2015) the english indices of deprivation 2015. london: ministry of housing, communities and local government. https://www.gov.uk/government/statistics/english-indices-of-deprivation-2015 https://doi.org/10.2105/ajph.2013.301707 https://doi.org/10.1080/00049180701639760 http://dx.doi.org/10.1136/jech.2003.013011 https://doi.org/10.1038/sj.ijo.0803176 https://doi.org/10.1186/s12966-015-0181-9 https://doi.org/10.1068/a37168 https://doi.org/10.37970/aps.v3i1.45 https://doi.org/10.1068/a160017 https://www.aph.gov.au/parliamentary_business/committees/house_of_representatives_committees?url=haa/./healthfunding/report.htm https://www.aph.gov.au/parliamentary_business/committees/house_of_representatives_committees?url=haa/./healthfunding/report.htm https://doi.org/10.1016/j.healthplace.2012.10.012 https://doi.org/10.1080/13658816.2020.1759807 https://doi.org/10.1177/0042098009106019 https://www.jstor.org/stable/41995682 https://www.gov.uk/government/statistics/english-indices-of-deprivation-2015 28 kerr et al. australian population studies 5 (2) 2021 tanton r, vidyattama y, & mohanty i (2015) disadvantage in the australian capital territory. policy studies 36(1): 92-113. https://doi.org/10.1080/01442872.2014.981062 townsend p (1987) deprivation. journal of social policy 16(2): 125-146. https://doi.org/10.1017/s0047279400020341 townsend p, phillimore p, & beattie a (1988) health and deprivation: inequality and the north. routledge. zhang x, cook p a, jarman i, & lisboa p (2011) area effects on health inequalities: the impact of neighbouring deprivation on mortality. health & place 17(6): 1266-1273. https://doi.org/10.1016/j.healthplace.2011.05.009 zhang x, cook p a, lisboa p j, jarman i h, & bellis m a (2013) the effects of deprivation and relative deprivation on self-reported morbidity in england: an area-level ecological study. international journal of health geographics 12(1): 5. https://doi.org/10.1186/1476-072x-12-5 zhao j & exeter d j (2016) developing intermediate zones for analysing the social geography of auckland, new zealand. new zealand geographer 72(1): 14-27. https://doi.org/10.1111/nzg.12110 https://doi.org/10.1080/01442872.2014.981062 https://doi.org/10.1017/s0047279400020341 https://doi.org/10.1016/j.healthplace.2011.05.009 https://doi.org/10.1186/1476-072x-12-5 https://doi.org/10.1111/nzg.12110 australian population studies 2022 | volume 6 | issue 2 | pages 1-13 © martinus 2022. published under the creative commons attribution-noncommercial licence 3.0 australia (cc by-nc 3.0 au). journal website: www.australianpopulationstudies.org australian labour force changes under the reduced mobility of covid-19 kirsten martinus university of western australia email: kirsten.martinus@uwa.edu.au. address: school of social sciences, the university of western australia, 35 stirling hwy, perth, western australia, 6009 paper received 11 march 2022; accepted 19 august 2022; published 19 december 2022 abstract background the geographic mobility of labour has long facilitated a well-functioning labour market for australia, being of importance in skill-matching and jobs in regional economies. disrupting the long-distance labour commute, covid-19 border closures and community lockdowns had an immediate and significant impact on the australian labour market. aims the aim is to understand australian labour force demography and provide an empirical understanding of how regions, and their respective states and territories, faired through the pandemic. data and methods using australian bureau of statistics sa4 level labour force participation and unemployment data, the paper highlights regional changes between 2018 and 2021 covering periods immediately before and after the emergence of covid-19. its analysis is contextualised by the respective state and territory and employment conditions underpinning labour demand via proxies of gross national product and state and territory gross product, gross real income and job vacancies. results the paper finds variations in labour force change are dependent on regional industry economic profiles between and within states and territories. this was in part due to state and territory lockdown and border closure policies as well as respective industry economic profiles. conclusions a more comprehensive mapping and understanding of labour force shifts over time will better capture the trajectories of regional labour markets. this will enable better targeting of specific policy outcomes at various levels of government, including to encourage industry diversity, support labour reskilling and the uptake of technologies. such policies will be better placed to assist australian labour force transitions post-covid and efficient labour market functioning. key words labour mobility, labour participation, unemployment, labour force, covid-19. http://www.australianpopulationstudies.org/ mailto:kirsten.martinus@uwa.edu.au 2 martinus australian population studies 6 (2) 2022 1. introduction the geographic mobility of labour has been a long-term means of facilitating a well-functioning labour market across australia. long-distance commuting is particularly important for skill-matching and jobs in regional economies, where access to a diverse range of skilled labour is difficult (productivity commission, 2013). indeed, modelling by the productivity commission (2013) found labour mobility during a mining boom enabled higher than average real wages and economic growth rates. the prevalence of long-distance commuting contractual arrangements in australia has seen a range of acronyms emerge describing different mobility types – dido (drive-in/drive-out), fifo (flyin/fly-out), bibo (bus-in/bus-out) and siso (ship-in/ship-out). most often associated with mining, long-distance commuting is used in a range of industries such as hospitality, transport, medical and health care and agriculture (hutton, 2021; martinus, 2016). despite the frequency of long-distance commuting in australia, it is difficult to know how many workers engage in such work practices. during the 2012 mining boom, there were about 276,000 fifo workers nationally (the west australian, 2019). in 2018, around 60,000 of the nation’s fifo jobs were in the western australian mining sector (the west australian, 2019), with pre-covid 2019 estimates showing between 6,000 and 7,000 flew in from other states (tyson, 2020). largely halting long-distance labour commuting, covid-19 border closures and community lockdowns had an immediate and significant impact on the demography of the australian labour market. aside from the sharp decline in long-distance commuting, several other factors also influenced the labour market of australian states and territories under covid restrictions. first, there was a decline in workers changing jobs (i.e., ‘churn’) at the beginning of the pandemic (abs, 2021a) due to sudden decreases in labour demand as businesses closed. second, a sharp decrease in long-distance commuting occurred as workers obeyed stay-at-home mandates. some workplaces adopted hybrid working models, such as office workers whose meetings and conferences went fully online; while others changed their mode of customer engagement, such as restaurants which shifted to online sales and take-way orders. third, demand for australian-made products increased due to global supply chain disruptions, and onshoring of processes added to the need for domestic labour in specific areas (productivity commission, 2021). as such, the decline in labour mobility due to border closures and community lockdowns measures did not necessarily lead to uniform declines in labour participation and unemployment, but rather variegated national experiences. this paper aims to shed light on the different labour force trends that emerged in regions, states and territories across australia due to pandemic restrictions. indeed, the only workers who commuted throughout the pandemic, irrespective of stay-at-home mandates, were those considered “essential” such as front-line health care and construction workers. but the definition of ‘who’ was essential also varied by state and territory. from the onset of the pandemic in australia, the individual australian states and territories had clear autonomy in setting rules regarding border closures and community lockdowns. each implemented covid-19 restrictions according to respective community sentiments and industry needs, reducing the overall impact on gross national production. some also closed internal regional borders to protect vulnerable indigenous communities and enable the on-going operation of essential sectors. for example, western australia had notably the most stringent mobility restrictions both intraand interaustralian population studies 6 (2) 2022 martinus 3 state, but it classified fly in/fly out workers of the mining sector as ‘essential’ ensuring continued production of one of the nations’ largest export sectors. combined with hard international border controls and quarantine measures, australia managed to emerge with one of the strongest economies of all oecd nations. it was also one of few nations to record a gross domestic product above pre-pandemic levels, and it eased labour mobility restrictions faster than most other oecd nations – a factor linked to higher gdp and economic recovery (abs, 2021b). nonetheless, the changes to labour mobility had an immediate impact on the labour demographics of specific regions, particularly in relation to labour participation and unemployment profiles. using australian bureau of statistics (abs) labour participation and unemployment rates at the statistical areas level 4 (sa4) spatial area across the years 2018 to 2021, this paper aims to unpack australian labour force trends and provide an empirical understanding of how regions – and respective states and territories faired through the pandemic. the research highlights the different labour force participation and unemployment trends between and within states and territories against a backdrop of individual state economic profiles to understand the economic and employment conditions underpinning labour demand in the respective states and territories. these economic profiles were compiled via proxies of gross national product and state and territory indicators of gross product, gross real income and job vacancy. the research finds regional labour market variations reflected the industry economic profiles of each state and territory. 2. data and methods as noted by kpmg’s terry rawnsley (2021), the 2021 australian bureau of statistics (abs) journey-towork data will be of limited use in exploring labour force trends under covid restrictions due to the increase in people working from home. for this reason, this paper used alternative abs data of unemployment and labour participation between 2018 to 2021 to understand these changes. following the abs definition, “labour force” refers to persons aged 15 to 65 years old. the timeframe of the study covered the two years before the first case of covid was confirmed in australia and the two years after. abs unemployment and labour participation data were collected at the sa4 level (abs, 2022a), as this spatial unit represents ‘labour markets or groups of labour markets within each state and territory’ (abs, 2016) and were therefore appropriate to analyse labour force trends. unemployment and labour force participation are key economic and demographic indicators. in this study, changes in labour force participation and unemployment levels of the different sa4s of each state and territory across the time period was assumed to provide insights into the impact of state and territory border closures and community lockdowns on regional labour force trends. the participation rate is of interest because higher levels often signal more women choosing to work instead of performing unpaid domestic duties. in terms of age groups, labour force participation tends to be higher in the 35 to 55 age group as those younger may choose to study rather than work and those older may chose early retirement (gustafsson, 2021). the unemployment rate is the proportion of people who are not in paid work but in the labour force actively looking for work (cf. reserve bank of australia, 2021). the participation rate is the proportion of employed workers out of 4 martinus australian population studies 6 (2) 2022 the total working population age, including retirees, those unable to work and not looking for work for other reasons, such as study or engaged in carer activities (cf. reserve bank of australia, 2021). australian gross national product (gdp), individual state and territory gross product (gsp), job vacancy and real gross income data from the abs were used as proxies to assess the economic and employment conditions underpinning labour demand in respective states and territories. these were appropriate as covid border closures and community lockdowns constricted labour mobilities within state and territory borders. 3. results the australian labour force participation rate has been steadily increasing over the decades, moving from around 60% in 1983 to 65% in 2011 and 66% in january 2020 (abs, 2022d; gustafsson, 2021). once covid began to impact the economy, participation rates dropped dramatically from the 66% in march 2020 to 62.6% in may 2020, slowly recovering to a level of 66% in november 2020. the surge of the virus in july 2021 across the nation saw participation rates again fall, being 64.5% in september 2021 and rising again to 66% in november. figure 1 demonstrates the changes in the australian labour force participation rate, unemployment rate and job vacancies. figure 1: national unemployment rate (%), labour force participation rates (%) and job vacancies (‘000s) source: adapted from abs (2021c, 2022b). the impact of covid restrictions is evident with the sharp dip in workforce participation corresponding to a rise in national unemployment and a sudden drop in job vacancies by may 2020. however, by august 2020, the job market had recovered to almost the same pre-pandemic levels. it surpassed this by november 2020, almost doubling by november 2021 to around 400,000 vacant australian population studies 6 (2) 2022 martinus 5 positions with unemployment rates dropping to around 4.2% in january 2022 compared to a march 2020 rate of 5.3% (abs, 2022a). january 2022 had the lowest unemployment rate in 13 years. coupled with extraordinary increases in job vacancies, the rate drop pointed to overall national labour shortages experienced differently by the states and territories due to their respective border restrictions and lockdowns. to fully understand state and territory labour force participation, variations in gross domestic product (gdp) and gross state product (gsp) were considered. figure 2 demonstrates the consistent rise in australia’s gdp over the 2018 to 2021 period, with the sharpest jumps occurring in 2019 and 2021. 2020 saw a relatively lower rise in gdp than experienced by 2019 volumes. this same pattern in gdp increases was not found across all states and territories. most showed slight increases between 2018 to 2019, with victoria and queensland generally steady until 2021. new south wales and south australia experienced rises in 2021 after plateauing between 2019 and 2020. the largest and most consistent rise across the entire period was found in western australia. these patterns were more prominent in the percentage changes in gdp and gsp of table 3, which shows that western australia far outstripped the increases of other states and territories. figure 4 shows the average annual unemployment rate by statistical area 4 (sa4), where a darker colour demonstrates lower unemployment. figure 5 illustrates the average annual participation rate, with darker colours meaning higher labour participation, and figure 6 shows the overall percentage change in annual participation rate across the 2018 to 2021 period. comparing these unemployment and participation rates across sa4s, the areas experiencing the highest levels of disadvantage from covid were noticeably the northern territory and south australia, which both have relatively small figure 2: australian gross domestic product (gdp) and gross state products (gsp), by chain volume measure and current prices source: adapted from abs (2022c). 6 martinus australian population studies 6 (2) 2022 figure 3: australian gdp and gsp, percentage change source: adapted from abs (2022c). figure 4: average annual unemployment rate (%) sa4, 2018-2021 source: adapted from abs (2021c). note: darker green is improvement in unemployment rates (decrease), darker blue is worsening (increase). australian population studies 6 (2) 2022 martinus 7 figure 5: average annual labour participation rate (%) by sa4, 2018-2021 source: adapted from abs (2021c). note: darker green is improvement in labour participation rates (increase), darker blue is worsening (decrease). figure 6: percentage change in average labour participation rate by sa4, 2018-2021 source: adapted from abs (2021c). note: darker green shows more labour participating in employment over the period; darker red is negative change. 8 martinus australian population studies 6 (2) 2022 economies by gsp. the northern territory was the only state or territory with negative growth in gsp, experiencing greater unemployment and significant decline in the labour force participation rate of change across the periods for all sa4s. south australia and tasmania had comparatively higher percentage gsp growth rates over the period, though the major metropolitan areas of these states and territories experienced overall 2018-2021 unemployment decreases and positive labour participation changes. other states and territories also exhibited uneven gsp increases. regional areas in queensland (e.g., outback) experienced higher unemployment rates than the rest of the nation, despite having the highest national labour participation growth rate across 2018-2021. queensland coastal areas of cairns, townsville, mackay and fitzroy showed marked improvements in unemployment numbers, and little overall percentage change in their comparatively high labour participation over the 20182021 period. this perhaps pointed to the residential attractiveness of these regions for queensland’s workforce. the experiences of regional victoria and new south wales were more varied. some locations saw unemployment declines and labour force participation rate rises (riverina, central west nsw; hume vic); others saw rises in unemployment and falls in participation rates (far west and orana, murray – nsw; gippsland – vic) through 2018-2021. again, populations around sydney and melbourne were varied in their experiences in unemployment and participation movements. western australia appeared to operate differently to the rest of the country, with all sa4s seeing extremely low unemployment rates – even in highly remote locations. perth also had relatively low unemployment across all sa4s compared to that of other metropolitan areas, with most of wa experiencing growth in participation rates – except for bunbury which had one of the largest drops in the nation. the negative change in participation rate across wa regions is due to their high levels before the pandemic, and that there was a slight decline due to those who lost jobs as parts of the economy shut down. this is evident as most negative change is in the state’s southwest where there was an increase in labour as many took refuge to escape perth restrictions. overall, there were no drops in jobs available once covid arrived. figure 7 shows the percentage change in job vacancies of all states and territories, with the northern territory, south australia and western australia all surpassing 100%. the rise in jobs may partly explain observed declines in unemployment and rises in labour force participation across the sa4s given that job vacancies were not evenly distributed across the states and territories or industries. table 1 provides a measure of state and territory income using real gross income. the substantially higher percentage change as a whole and per capita points to the extraordinary wealth of western australia as underpinning why workforce trends function differently to those of other states and territories. australian population studies 6 (2) 2022 martinus 9 figure 7: percentage change in state and territory job vacancies, feb 2020 to nov 2021 source: adapted from abs (2022b). table 1: change in real gross state and territory income (rgsi), 2020-2021 rgsi % rgsi per capita % new south wales 1.6 1.2 victoria -0.3 -0.3 queensland 0.2 -0.9 south australia 5.2 4.6 western australia 18.1 17.0 tasmania 5.5 4.8 northern territory -3.9 -4.4 australian capital territory 2.9 2.2 australia 3.7 3.2 source: adapted from abs (2022c) 4. discussion and conclusions covid has profoundly changed how australians live and work. while the technologies and capabilities existed before covid, 2020 saw much of the australian workforce became digitally hyper-mobile with the mass and sudden adoption of virtual meetings and working from home. the 2020 families in australia survey found 67% were working from home compared to 42% before the pandemic (baxter and warren, 2021). nonetheless, the degree of change depended on the industry sub-sector and an individuals’ occupation type. not all work was easily transferred to the online space, for example, those in the university sector adapted relatively quickly to online teaching whereas those in primary and secondary education found it more difficult. while some workplaces were forced to shut down as they were unable to operate virtually, other workplaces were deemed to deliver essential goods or services and continued to operate. for example, gyms and restaurants paused but those in health care and the resources industry commuted throughout the pandemic. 10 martinus australian population studies 6 (2) 2022 the changes experienced had different labour demographic implications across australia, particularly given the extent that national industries relied on long-distance commuting before the pandemic. some of the influencing factors were the rules around shutdowns due to community lockdowns and border restrictions. these varied by state and territory, demonstrating the respective autonomies of states and territories over their citizens. the experiences of regions across the australia states and territories were influenced by different industry profiles and border restrictions (gilfillan, 2020; national skills commission, 2020). given the relative limitations of abs journey-to-work data to capture working-from-home arrangements, this paper examined regional labour force trends using abs sla4 labour participation and unemployment rates. the period of analysis, from 2018 to 2021, captures the timeframe immediately before and after the emergence of covid-19 in australia. this data is contextualised by the economic conditions of the respective states and territories using state and territory gross product (gsp), job vacancies and gross income data. the results show that labour demand across the entire western australia, even remote areas, was much higher than other states and territories and resulted in the most significant improvements in unemployment and labour participation. this was reflected in a high increase in gsp, job vacancies and real gross state income due to a strong export demand for wa resources. further, the wa labour market demonstrated a distinct pattern which was likely linked to its strict interand intra-state border closures inhibiting long-distance commuting – particularly from other states and territories. interestingly, despite wa unemployment rate declines, bunbury sa4 labour participation had the sharpest decline. this may be due to the increase in people moving away from greater perth metropolitan areas to wa’s southwest region during covid restrictions. the large positive changes in unemployment and labour participation of the wa outback sa4 may also reflect wa’s fly-in / flyout workers moving to the region rather than living in perth, given intra-wa regional travel restrictions. the resource state of queensland did not follow this pattern, with the most remote sa4 (queensland outback) retaining the highest national unemployment rate despite gains in the labour participation rate. interestingly, this area contains resource towns (mt isa and cloncurry) which did not shut down like other queensland towns dominated by tourism, leisure and entertainment. the uneven labour force changes of queensland sa4’s was different to wa as intra-state regional movement enabled fifo workers to live and work in separate locations. nonetheless, fifo workers who had previously commuted from homes in new south wales were forced to locate to queensland. despite the vulnerability of industries such as tourism and entertainment, the unemployment declines and relatively stable and high labour participation of coastal lifestyle areas likely reflects more people living in these regions and working elsewhere (e.g., mining). the states of victoria and new south wales were the hardest hit by border closures and prolonged lockdowns (jamaldeen, 2020; mizen, 2021), experiencing the lowest nationally recorded gross state income changes, low job vacancies and percentage gsp changes. the labour demographic patterns in the regional and metropolitan areas of these states were varied, though the coastal lifestyle sa4’s generally had the worst unemployment and labour participation rates. this was perhaps due to the household spending contractions of extended state lockdowns impacting the dominant industries (retail, tourism, leisure) in these towns (cf. brinsden, 2021). australian population studies 6 (2) 2022 martinus 11 the higher unemployment and lower labour participation profile of the northern territory reflects a significant decline in job opportunities due to negative economic growth and that it was a ‘covid refugee’ location. indeed, despite hard northern territory border closures, many people entered as it was relatively covid-free and provided a migration pathway into western australia (cf. thompson, 2020). south australia and tasmania performed relatively well economically given their small economic sizes and that neither experienced extended lock downs (cf. wright, 2021). though improvements in labour market trends were concentrated in the metropolitan sa4s of hobart, launceston and northeast, and adelaide. regional areas of these states did not demonstrate the same positive changes in unemployment and labour participation across the study period. the findings of this paper show the varied experiences between and within states and territories. this was in part due to state and territory lockdown and border closure policies as well as respective industry economic profiles. nonetheless, this research has limitations as it does not specifically breakdown findings by industry sector. this offers several avenues for future research exploration: first, around the relationship between industry sector and indicators of labour demographic change, such as labour participation and unemployment; and second, around the appropriate government policy responses to these changes at the different federal, state, territory, regional or local levels. more comprehensive mapping and understanding labour force shifts over time will better capture the trajectories of regional labour markets, providing evidence to target specific policy outcomes at various levels of government. indeed, the oecd (2021) suggests that policies must account for the diversity of industry sector, support labour reskilling and the uptake of technologies to assist labour demographic transitions and labour market functioning across australia for different economy types. key messages • geographic mobility of labour facilitates a well-functioning labour market for australia, particularly in terms of skill-matching and jobs in regional economies. • labour force change varied between and within states and territories due to differences in state and territory industry economic profiles and lockdown and border closure policies. • more comprehensive mapping and understanding labour demographical shifts over time will better capture the trajectories of regional labour markets, enabling more specifically targeted policy outcomes. acknowledgments i would like to acknowledge the contributions of bo guo, phd candidate at the university of western australia, for his work on the figures and maps in this paper. in addition, i would like to acknowledge the support of the australian research council (grant number de170100727). references abs (2016). australian statistical geography standard (asgs): volume 1 main structure and greater capital city statistical areas, july 2016. abs cat. no. 1270.0.55.001. abs: canberra. abs (2021a). job mobility falls during the first year of the pandemic. media release. abs: canberra https://www.abs.gov.au/media-centre/media-releases/job-mobility-falls-during-first-yearpandemic https://www.abs.gov.au/media-centre/media-releases/job-mobility-falls-during-first-year-pandemic https://www.abs.gov.au/media-centre/media-releases/job-mobility-falls-during-first-year-pandemic 12 martinus australian population studies 6 (2) 2022 abs (2021b). international economic comparisons after a year of the pandemic. media release. abs: canberra https://www.abs.gov.au/articles/international-economic-comparisons-after-yearpandemic abs (2021c). impact of lockdowns on household consumption insights from alternative data sources. abs: canberra. abs (2022a). labour force, australia, detailed. abs cat. no. 6291.0.55.001. abs: canberra. abs (2022b). job vacancies, nov 2021. abs: canberra. https://www.abs.gov.au/statistics/labour/employment-and-unemployment/job-vacanciesaustralia/latest-release abs (2022c). australian national accounts: state accounts 2020-21 financial year. abs: canberra. abs (2022d). labour force, australia, december 2021. abs: canberra. https://www.abs.gov.au/statistics/labour/employment-and-unemployment/labour-forceaustralia/dec-2021 baxter, j., & warren, d. (2021). towards covid normal: employment & work-family balance in 2020. australian institute of family studies: canberra, australia. https://aifs.gov.au/research/researchreports/towards-covid-normal-employment-work-family-balance brinsden, c. (2021). covid lockdowns in nsw and victoria cause economy to shrink but manufacturing bouncing back. 7news. https://7news.com.au/business/covid-lockdowns-in-nsw-and-victoriacause-economy-to-shrink-but-manufacturing-bouncing-back--c-4769084 gilfillan, g. (2020). covid-19: labour market impacts on key demographic groups, industries and regions. research paper series, 2020-21, 23 october, 2020. parliament of australia: canberra. https://www.aph.gov.au/about_parliament/parliamentary_departments/parliamentary_library/ pubs/rp/rp2021/covid-19-stat_snapshot gustafsson, l. (2021). australian labour force participation: historical trends and future prospects. australian government, the treasury: canberra. https://treasury.gov.au/publication/p2021164860 hutton, r. (2021). everything you need to know about fifo work. labour solutions australia. https://www.laboursolutions.com.au/blog/2021/06/fifo? jamaldeen, m. (2020). nsw economy benefits from opening up. deloitte access economics. https://www2.deloitte.com/au/en/pages/economics/articles/nsw-economy-benefits-openingup.html martinus, k. (2016). regional development in a resource production system: long distance commuting, population growth and wealth redistribution in the wa goldfields. geographical research, 54(4), 420-432. https://doi.org/10.1111/1745-5871.12173 mizen, r. (2021). pandemic’s economic toll weights most heavily on victoria. the financial review. https://www.afr.com/policy/economy/pandemic-s-economic-toll-weights-heaviest-on-victoria20211119-p59ae9 national skills commission (2020). the shape of australia’s post covid-19 workforce. australian government: canberra. https://www.nationalskillscommission.gov.au/publications/shapeaustralias-post-covid-19-workforce oecd (2021). oecd policy responses to coronavirus (covid-19): designing active labour market policies for the recovery. oecd publishing: paris. https://www.oecd.org/coronavirus/en/policy-responses productivity commission (2013). geographic labour mobilities issues paper. productivity commission: canberra. https://www.pc.gov.au/inquiries/completed/labour-mobility/report productivity commission (2021). vulnerable supply chains, interim report. productivity commission: canberra. https://www.pc.gov.au/inquiries/completed/supply-chains#report https://www.abs.gov.au/articles/international-economic-comparisons-after-year-pandemic https://www.abs.gov.au/articles/international-economic-comparisons-after-year-pandemic https://www.abs.gov.au/statistics/labour/employment-and-unemployment/job-vacancies-australia/latest-release https://www.abs.gov.au/statistics/labour/employment-and-unemployment/job-vacancies-australia/latest-release https://www.abs.gov.au/statistics/labour/employment-and-unemployment/labour-force-australia/dec-2021 https://www.abs.gov.au/statistics/labour/employment-and-unemployment/labour-force-australia/dec-2021 https://aifs.gov.au/research/research-reports/towards-covid-normal-employment-work-family-balance https://aifs.gov.au/research/research-reports/towards-covid-normal-employment-work-family-balance https://7news.com.au/business/covid-lockdowns-in-nsw-and-victoria-cause-economy-to-shrink-but-manufacturing-bouncing-back--c-4769084 https://7news.com.au/business/covid-lockdowns-in-nsw-and-victoria-cause-economy-to-shrink-but-manufacturing-bouncing-back--c-4769084 https://www.aph.gov.au/about_parliament/parliamentary_departments/parliamentary_library/pubs/rp/rp2021/covid-19-stat_snapshot https://www.aph.gov.au/about_parliament/parliamentary_departments/parliamentary_library/pubs/rp/rp2021/covid-19-stat_snapshot https://treasury.gov.au/publication/p2021-164860 https://treasury.gov.au/publication/p2021-164860 https://www.laboursolutions.com.au/blog/2021/06/fifo https://www2.deloitte.com/au/en/pages/economics/articles/nsw-economy-benefits-opening-up.html https://www2.deloitte.com/au/en/pages/economics/articles/nsw-economy-benefits-opening-up.html https://doi.org/10.1111/1745-5871.12173 https://www.afr.com/policy/economy/pandemic-s-economic-toll-weights-heaviest-on-victoria-20211119-p59ae9 https://www.afr.com/policy/economy/pandemic-s-economic-toll-weights-heaviest-on-victoria-20211119-p59ae9 https://www.nationalskillscommission.gov.au/publications/shape-australias-post-covid-19-workforce https://www.nationalskillscommission.gov.au/publications/shape-australias-post-covid-19-workforce https://www.oecd.org/coronavirus/en/policy-responses https://www.pc.gov.au/inquiries/completed/labour-mobility/report https://www.pc.gov.au/inquiries/completed/supply-chains#report australian population studies 6 (2) 2022 martinus 13 rawnsley, t. (2021). census 2021: what will journey to work data mean when we aren’t leaving home? kpmg newsroom. https://newsroom.kpmg.com.au/will-journey-work-data-mean-arent-leavinghome/ reserve bank of australia (2021). unemployment: its measurement and types. https://www.rba.gov.au/education/resources/explainers/unemployment-its-measurement-andtypes.html storen, r., & corrigan, n. (2020). covid-19 a chronology of state and territory government announcements (up until 30 june 2020). parliament of australia, canberra. https://www.aph.gov.au/about_parliament/parliamentary_departments/parliamentary_library/ pubs/rp/rp2021/chronologies/covid-19stateterritorygovernmentannouncements the west australian (2019). striking the balance for distance workers. https://towards2030.thewest.com.au/striking-the-balance-for-distance-workers/ thompson, j. (2020). the nt government says ‘coronavirus refugees’ are heading to the territory. will they fix its shrinking population? abc news. https://www.abc.net.au/news/2020-09-17/nt-covid19-refugees-solve-northern-territory-population-woes/12651268 tyson, r. (2020). looming changes for fifo employees in western australia. mining international. https://www.mining.com/looming-changes-for-fifo-employees-in-western-australia/ wright, s. (2021). the little economies that could: sa and tasmania lead the nation. the sydney morning herald. https://www.smh.com.au/politics/federal/the-little-economies-that-could-sa-andtasmania-lead-the-nation-20211119-p59afd.html https://newsroom.kpmg.com.au/will-journey-work-data-mean-arent-leaving-home/ https://newsroom.kpmg.com.au/will-journey-work-data-mean-arent-leaving-home/ https://www.rba.gov.au/education/resources/explainers/unemployment-its-measurement-and-types.html https://www.rba.gov.au/education/resources/explainers/unemployment-its-measurement-and-types.html https://www.aph.gov.au/about_parliament/parliamentary_departments/parliamentary_library/pubs/rp/rp2021/chronologies/covid-19stateterritorygovernmentannouncements https://www.aph.gov.au/about_parliament/parliamentary_departments/parliamentary_library/pubs/rp/rp2021/chronologies/covid-19stateterritorygovernmentannouncements https://towards2030.thewest.com.au/striking-the-balance-for-distance-workers/ https://www.abc.net.au/news/2020-09-17/nt-covid-19-refugees-solve-northern-territory-population-woes/12651268 https://www.abc.net.au/news/2020-09-17/nt-covid-19-refugees-solve-northern-territory-population-woes/12651268 https://www.mining.com/looming-changes-for-fifo-employees-in-western-australia/ https://www.smh.com.au/politics/federal/the-little-economies-that-could-sa-and-tasmania-lead-the-nation-20211119-p59afd.html https://www.smh.com.au/politics/federal/the-little-economies-that-could-sa-and-tasmania-lead-the-nation-20211119-p59afd.html australian population studies 2020 | volume 5 | issue 2 | pages 43–57 © mckenzie et al. 2021. published under the creative commons attribution-noncommercial licence 3.0 australia (cc by-nc 3.0 au). journal website: www.australianpopulationstudies.org beyond population growth: bringing connection, renewal and agency into regional development fiona mckenzie* victorian government fiona mckenzie orange compass fiona mckenzie university of western australia *corresponding author. email: fiona.mckenzie@delwp.vic.gov.au. address: department of environment, land, water and planning, po box 500, east melbourne, vic 8002, australia paper received 9 august 2021; accepted 30 october 2021; published 29 november 2021 abstract background boosting the population growth of towns and cities across regional australia has long been a popular aspiration. however, the reality of many regional locations is long-term population decline. aims regional development has been underpinned by a growth paradigm which conflates population growth with success. this article explores how some regional communities have experienced renewal, fostered innovation, and enhanced community wellbeing without population growth. data and methods a mixed methods approach informs this paper. five case studies are used to illustrate ways in which regional communities can experience successful economic and social outcomes without population growth. literature is provided for context and empirical data are used to illustrate key features of the selected towns. results the case studies reveal themes of connection, place, agency and wellbeing which contribute to ‘successful’ communities. for example, developing connections with a wider network within the region or across sectors enables economic growth through innovation, irrespective of population growth. conclusions being able to respond to, or take advantage of, change is an important element in regional development and one which creates community renewal and wellbeing even if not population growth. key words population growth; population decline; regional australia; community wellbeing. http://www.australianpopulationstudies.org/ mailto:fiona.mckenzie@delwp.vic.gov.au 44 mckenzie et al. australian population studies 5 (2) 2021 1. introduction boosting the growth of towns and cities across regional australia has long been a popular, if unsuccessful, strategy (guaralda et al. 2020). this strategy has been fuelled by the growth-oriented paradigm underpinning the economic conceptualisation of regional development (fudge et al. 2021). however, this paradigm does not address the reality of many regional locations which have an older age structure and little in migration – two demographic factors which make population growth unrealistic. the growth paradigm also conflates economic growth and population growth with an underlying assumption that both forms of growth will lead to “success”, a somewhat unclear objective which does not always consider aspects of community or individual wellbeing. this article examines ways in which regional communities can experience renewal, foster innovation, and enhance community wellbeing without population growth. case studies have been selected to illustrate this. while literature is provided for context and empirical data are used to illustrate key features of the selected towns being discussed, the paper also draws upon the combined experience of the authors, each of whom has had more than 20 years’ experience in regional development and research. a mixed methods approach therefore informs this paper. the selected communities have had to adapt to economic and social change over the past four decades and all have had stable or declining rates of population growth. their experience can inform debates about the elements of regional development that matter most to communities: connection, agency, wellbeing, renewal, and sustainability. the case studies presented in this article focus on towns in which there has been limited or no population growth over recent decades (figure 1). yet in terms of economic development, social diversification, and wellbeing, each may be regarded as successful. there is no single approach that can fully explain such success. for birchip, networks, information flows and innovation are key ingredients. katanning and nhill have both experienced major changes in social profile which have created new economic opportunities, while in lake figure 1: population of selected case study towns, 1981 to 2016 source: abs census data, urban centre and localities. australian population studies 5 (2) 2021 mckenzie et al. 45 cargelligo and condoblin community-led place-based development has increased local agency to identify and achieve their objectives for local wellbeing. 2. networks and connections 2.1. literature the australian settlement system is characterised by large state capital cities which contain high proportions of their states’ population. nevertheless, australian settlement patterns are not just about capital cities. there are hundreds of small and medium sized towns across the country. some have experienced population growth, others decline, and others have shown little change – at least in terms of overall population numbers. regional economic development narratives that highlight population growth as a key element of success would seem to imply that lack of population growth means economic or social demise. the recognised role of large cities as economic drivers in modern economies further entrenches this view. because large cities are seen to have high levels of human capital, dense networks and information spillovers, they are seen as being a driving force of innovation and development (florida 2002; scott & storper 2003). very few authors have challenged the view that large cities are the only places where innovation can occur. shearmur (2012, 2015) is an exception in this regard. he challenges the notion that innovation always requires physical proximity. furthermore, proximity itself may mean different things apart from physical closeness. it may include social, organisational, or cultural connections which form strong networks across a variety of distances (shearmur 2015). this fluidity of proximity is further explored in a range of academic works (rallet & torre 1999; blanc & sierra 1999; urry 2003a; 2004; boschma 2005a, 2005b; kesselring 2014) which examine the increasing levels of actual and virtual mobility in the modern world which change the nature and perception of distance and space. this is important for regional economic development because increasing connectivity has the potential to widen networks and overcome many of the disadvantages of distance or remoteness. an outcome is that “[t]hings are made close through these networked relations” (hannam et al. 2006 p. 14, original emphasis). networks such as these highlighted above are clearly advantageous in a dynamic modern economy, both for businesses and individuals. however, such dynamism is not traditionally associated with small rural communities. many rural communities are characterised by strong ties – bonding social capital that can create resilience, but also parochialism. from a sociological point of view, communities with strong bonds of social capital may be seen to have key strengths in terms of mutual support (granovetter 1983). these ‘strong ties’ connect individuals through frequent contact and similar interests or values. in contrast are ‘weak ties’ which can be thought of as acquaintances – people seen or contacted infrequently and perhaps with fewer interests in common. granovetter, in his seminal 1973 paper, showed that, while strong ties provide cohesion for a particular group or community, their ultimate weakness is in the insularity and isolation which can occur as a result of that group having few contacts with other worlds or world views (granovetter 1973). this type of insularity presents particular problems in a rapidly changing world where it can rapidly lead to isolation and detachment from mainstream events (economic, social or political). by contrast, 46 mckenzie et al. australian population studies 5 (2) 2021 individuals who have many weak ties are better placed to defuse new ideas since their ties can bridge different communities across many locations (granovetter 1973). the implication for non-metropolitan communities is that links with the nonlocal are likely to lead to new ideas and perspectives (rallet & torre 1999; fitjar & rodríguez-pose 2011). these in turn can fuel innovation and future development within the region. digital technology can maintain weak-tie relationships across broad networks (mckenzie 2017). these ‘weak ties’ are important for development and innovation. it is important to highlight here the growth of distributed knowledge networks, as distinct from social networks, which reach far beyond the local due to virtual connectivity. research conducted in australia found that those undertaking transformational change had more far-reaching information and knowledge network connections yet less extensive social links to family, friends and colleagues compared with incremental-scale adaptors. strong access to knowledge and weak social ties actually increased the ability to facilitate action that differs from established social norms (dowd et al. 2014). bringing these ideas together – connectivity, fluid concepts of proximity, and the importance of weak social ties but strong knowledge networks – has important implications for regional development. work by fitjar & rodríguez-pose (2011) highlighted the factors that made regional businesses in the peripheral region of southwest norway wealthy and innovative. local factors such as collaboration, agglomeration and clustering were not found to be significant. instead, it was international networks and long-distance cooperation which accounted for much of the product innovation occurring in the region (fitjar & rodríguez-pose 2011). the education levels and attitudes of managers were also found to be important in the creation and retention of international networks, suggesting that human capital and individual initiative can play a significant role in bringing innovative practices to peripheral regions. many of these managers had been forced to go elsewhere in norway for their higher education due to the absence of a local university. this experience had expanded their horizons, experience and networks which could continue to be used upon their return to a more peripheral location. the general open-mindedness associated with the life experience of moving in order to improve their level of education can be considered as a fundamental factor in overcoming place-boundedness in their work relationships (fitjar & rodríguez-pose 2011). these alternative views to the mainstream narrative of city-based innovation open interesting avenues for considering the prospects for human capital and economic potential in regional australia. the pattern of moving from regional areas to capital cities for education is relatively common. one estimate suggests that around half of school leavers from non-metropolitan areas move to capital cities and 29% of this group of leavers have returned to non-metropolitan areas by the age of 23 (delwp 2016). others may well return to non-metropolitan areas at later ages. this pattern of spending time in a capital city and then returning to non-metropolitan areas of australia, has the potential for links to be made in the way that fitjar & rodríguez-pose (2011) envisage. not only have such individuals gained human capital in a formal sense through education and work experience, but the networks gained while in the city can continue to be of benefit for future regional development. the city of orange in nsw shows how mobility and networks have been used to increase economic development opportunities for local growth. research undertaken by mckenzie (2015) found that many of the town’s leaders expressed the value of having left the city to gain australian population studies 5 (2) 2021 mckenzie et al. 47 education, skills or experience, then bringing those benefits back to the community (mckenzie 2015). this type of mobility has been found to play an important role in regional development in australia as well. work by plowman et al. (2003) found that more innovative communities had higher proportions of residents who had lived elsewhere. social and professional networks are important for regional development in that highly skilled people within a network can be viewed as “potential resources to be mobilised and capitalised at a distance” (rerat 2016 p. 280). in other words, they need not be geographically close to provide a potential economic benefit. such networks can deliver “exponential increases in output (and rewards or wealth) that are spread throughout a network of relationships within which a variety of enterprises are located and operate” (urry 2003b p. 53, original emphasis). 2.2. case study: birchip even small towns with little or no population growth can succeed through connectivity and local initiative. the town of birchip in victoria’s mallee region is small, with an enumerated population of 676 recorded at the 2016 census. the nearest centre with more than 2,500 people is located more than 100 kilometres away. yet the connectivity of birchip to the outside world is evident through innovative farmers and businesses in the field of dryland agriculture. the birchip cropping group was established in the early 1990s as a way to improve farming research, knowledge and practice across the region. its stated mission is “to improve the prosperity of australia’s broadacre farmers through applied science-based research and extension” (bcg 2020 p. 1). the group delivered over 80 projects in 2019-20 with the aim of delivering benefits to its members through attracting research and extension funding which benefits farmers, communities and the environment in the local region (bcg 2020). the birchip cropping group has been at the forefront of applying new technologies and farming practices and has developed a network of more than 50 investors and collaborators across government, universities, and private enterprise. the bcg’s investors and collaborators include a wide range of organisations and sectors, both local and global as shown in table 1. the local primary to year 12 (p-12) college has developed close links between its curriculum and the local economy. its ‘structured workplace learning’ program provides opportunities for students to combine job experience in agriculture with secondary school study (batters 2021). the development of digital skills for modern precision agriculture involves gaining experience with geographical information systems (gis), variable rate technology (for application of water fertilisers and pesticides), soil sensors, yield monitors and remote sensing using drones. understanding these technologies is of particular benefit for students’ education and the future of the local economy (bcg 2018). 48 mckenzie et al. australian population studies 5 (2) 2021 table 1: number of investors and collaborators with birchip cropping group by sectors and locations, 2019-20 sector no. location no. research / academia 17 interstate 28 business 8 victoria 24 state government 8 local (within birchip) 2 industry association 6 overseas 1 philanthropic 4 secondary education 3 federal government 3 peak body 2 other 4 total 55 total 55 source: compiled from data in bcg 2020, p. 9. 3. renewal 3.1. literature for much of the twentieth century, many rural-based communities enjoyed population growth. public investment in transport infrastructure, utilities and social services, along with tariffs and import restrictions supported and protected local industry. trade liberalisation policies introduced in the mid-1980s significantly affected their local economies (tonts & haslam mckenzie 2005). the restructuring of the australian economy caused many services in rural communities to be rationalised and centralised in larger, regional centres, with many people in small towns following the services. depressed commodity prices and increased costs of farm inputs in the 1980s and 1990s triggered many to leave agriculture and small communities struggled. structural changes were starting to create major changes in rural australia. farms were, on average, larger and more productive, but fewer workers were needed as capital intensive methods superseded labour intensive farming practices (stayner & reeve 1990; garnett & lewis 1999). declining terms of trade in the agricultural sector also fuelled the trend towards consolidation, as productivity could be increased through expansion into larger farms (barr 2009; smailes et al. 2019). the social composition of the countryside was also changing, and economic activities broadened from productive to consumptive activities such as second homes and tourism uses (smailes 2002; mckenzie et al. 2008). this growing complexity in rural land use was variously termed the ‘postproductivist’ countryside (wilson 2001; argent 2010) or ‘multifunctional countryside’ (smailes 2002). increasing diversity of land use was both fuelled by, and generated, increasing temporary mobilities. importantly, these economic changes were happening at a time of increasing personal mobility. car ownership and improvements in road quality played a major factor in changing population distribution in regional areas. small towns which had previously offered a range of basic services could now be bypassed as people were able to access a wider geographical area. this favoured larger centres where businesses could increasingly gain access to a wider market. conversely, businesses in small towns saw a contraction in demand for their goods and services. australian population studies 5 (2) 2021 mckenzie et al. 49 as regional australia became increasingly integrated into global economic systems, a parallel process of economic policy reform was occurring (tonts & haslam mckenzie 2005; beer 2012). in the 1980s, the hawke and keating governments led a period of major economic reform which effectively ended the high levels of protectionism in australian trade and industry. fiscal reforms, such as floating the australian dollar and allowing entry of foreign banks into the country, made australia much more open to the global economy. the subsequent coalition government undertook further economic reform from the mid-1990s. this neoliberal agenda continued through the 2000s and was characterised by fiscal conservatism, labour-market deregulation, mutual-obligation pacts, introduction of a national competition policy, and a move away from the welfare state (o’neill & fagan 2006; hogan & young 2015). trade liberalisation policies significantly affected their local economies (tonts & haslam mckenzie 2005). the restructuring of the australian economy caused many services in rural communities to be rationalised and centralised in larger, regional centres, with many people in small towns following the services. depressed commodity prices and increased costs of farm inputs triggered many to leave agriculture and small communities struggled throughout the 1980s and 1990s (smailes et al. 2019). despite these challenges, however, some communities responded with innovative ideas for renewal. 3.2. case study: nhill the town of nhill in western victoria has experienced population decline over several decades. in 2016 the town’s population was 1,752 according to the census, down from 1,879 in 2006 (abs 2016). despite the lack of growth in total population, the town has seen a major change in its composition over the period. the proportion of overseas-born has doubled from 7% to 14% during the decade (abs 2016). since the late 2000s, the resettlement of refugees from myanmar has created significant change in the town. the resettlement was driven largely by the owner of a local food processing plant who was seeking to expand operations. more than 160 refugees have settled in the district with around 30 children enrolled in local schools. the local economic benefit of the karen refugees has been estimated at $40 million by deloitte access economics (ames & deloitte 2015). clearly, this influx of new people has added to the local labour force, school population and ethnic diversity of the town. while it may have slowed the rate of decline of the town’s population, it has not created population growth. structural population decline has played out over half a century. any turnaround may take as long to achieve. nevertheless, nhill has gained a great deal through refugee settlement. regeneration can occur through social and economic change rather than in an increase in overall numbers. 3.3. case study: katanning the shire of katanning is a municipality in western australia, 286 km south-east from perth. historically, agricultural production, particularly sheep and wool industries, were central to the economy. katanning was not immune to the challenges, already cited, presented by australia’s engagement with global markets and the commitment to a neoliberal policy agenda. katanning experienced hardship and struggled to make headway for almost two decades as farmers took advantage of their neighbours’ departure from the industry and aggregated their properties to optimise economies of scale. many young people did not see a future for themselves and they too left, causing a general ageing of the population, especially in rural communities (tonts et al. 2008). 50 mckenzie et al. australian population studies 5 (2) 2021 nonetheless, the industry profile of the katanning local government area continues to be dominated by agriculture however two key investments have diversified local employment in the industry and created a sense of renewal. the katanning abattoir was established in the 1970s and, after restructuring in 1999, it became a producer-owned and controlled co-operative which has thrived, unlike many others across regional western australia. the abattoir pursued a niche in the meat market by focusing on premium quality lamb slaughtered in compliance with halal practices, opening up a lucrative international market which has continued to expand. this strategy transformed the local community, with several malay families recruited from the cocos and christmas islands to perform the halal slaughtering. subsequent family groups followed and the business grew. as early as 1975, the muslim community built a mosque in the town, the first in regional western australia. this and the promise of semiand unskilled work were pull factors for arrivals and there has been a steady flow of migrants and refugees who have subsequently settled in the shire. in 2020, the local shire boasted that more than 45 different nationalities live in the local government area (walga 2019). while the social adjustment has not always been easy because of cultural or language differences, the social change has enlivened katanning and supported the local economy. another innovation in katanning was the 2014 replacement of the former sheep saleyards with a state-of-the-art covered facility capable of trading 1.5 million sheep annually. the roof covers more than a hectare of stockyards. located alongside this are offices and amenities for farmers, stockmen and truck drivers who need to stay overnight. the enhanced conditions have been the catalyst for a new, unexpected tourism attraction with visitors coming to watch the rapid-fire sales routines and the dogs working in the yards and trucks each tuesday and wednesday (shire of katanning 2021). the new canteen has also become a regular destination for visitors to rub shoulders with the stockmen and many local farmers, with up to three sittings on tuesday evenings and throughout the sale day each wednesday. it has become busy enough that local women have now started a catering business running the canteen. hence what was part of the traditional primary sector has spawned a number of service industries related to accommodation, hospitality and tourism – all of which have been expanding economic sectors in recent decades and which offer a range of employment opportunities. when the saleyards were constructed, the facility adopted a number of environmental and laboursaving innovations which generated other new businesses. manure from the saleyards flows into holding tanks and subsequently collected for drying and processing by a local migrant family who now employ additional labour. the product is sold to horticultural wholesalers and nurseries across southern western australia. the saleyard roof covers 44,000m2 and rainfall is collected, stored, treated, then gravity pushed to a new lake, arboretum and botanic garden in the centre of the town, transforming a low lying, unsightly area to prized real estate (shire of katanning 2021). the revival of the sheep industry in katanning inspired a local entrepreneur to buy the dilapidated historical flour mill in the centre of the town. the local shire sold the property for one dollar in 2014, on condition that the new owner restore the building (acott 2018). considerable funds were invested to renovate it as a high-end hotel with the dome café opening at street level in 2019. the business plan took account of katanning’s location in the centre of the great southern region with no other accommodation like it within at least two hour’s drive of the town. in addition to the nascent sheep australian population studies 5 (2) 2021 mckenzie et al. 51 saleyard attraction, katanning is located midway between perth and albany with two national parks and a well-established wine region less than an hour away. despite the obvious financial set back associated with covid 19 in 2020, the domestic tourism market boom in 2021 has certainly compensated with occupancy rates being steady throughout 2020-2021 (the premier mill 2021). katanning, like nhill, struggles to maintain its population base because of structural population ageing, yet the social profile and economic prospects of both towns have experienced renewal through migration, social diversity and innovative industry development. regeneration can occur through social and economic change rather than in an increase in overall population numbers. 4. place-based approaches 4.1. literature in the literature, there has been a move from an economic development focus to a more holistic sustainable development focus when considering regional development (fudge et al. 2021), however formal regional policies at state, federal or even local government level still have an assumption (implicit or explicit) that population growth is an essential part of regional development. it is increasingly acknowledged that, in addition to job creation, wage rates and other quantitative measures of economic development, other factors such as sense of place, social capital and social and environmental wellbeing are also important (plummer et al. 2018). place-based policies, as the term implies, have a focus on specific locations or regions. but as noted by beer et al. (2020) they represent far more than just a label for already established programmes of government activity, or the concentration of public sector resources in specific locations. place-based policies embody an ethos about, and an approach to, the development of economies and society that acknowledges that the context of each and every city, region and rural district offers opportunities for advancing well-being (beer et al. 2020 p. 12). while recognising that there are many different types of place-based approaches, here we are referring to collective, community-led initiatives. they are characterised by collaboration, community-driven agency and ownership at their heart. connection and agency are central themes which affect the success of place-based initiatives – they give power to the community to guide their own change agenda. it is critical to note that “place” is not homogenous. it is something that is encountered and experienced differently by different people in the community (mackay et al. 2021; rainnie 2021). community is also not homogenous, yet it remains central to place-based approaches. such approaches require support from skilled local teams that help to create meaningful connections and partnerships across the community, businesses, local service providers, agencies and funders. these teams recognise that there is no ‘one size fits all’ recipe and ensure that any action taken is the right action for that community’s unique context. key features of good community-focussed placebased development initiatives include: • fostering cross-sector collaborations that harness local resources and place community priorities at the centre; 52 mckenzie et al. australian population studies 5 (2) 2021 • implementing timely and targeted responses that meet the specific needs of community, especially those most vulnerable; • creating trusted relationships as an intermediary between the wider system and local communities; • empowering the community through local participation and representation; and • drawing upon local knowledge, community networks, and relationships to help people deal with uncertainty. all of this is built upon a foundation of connection and trust amongst different community networks (formal and informal) that has taken time and effort to build, underpinned by appropriate social infrastructure to support collaboration. 4.2. case study: growing lachlan (condobolin and lake cargelligo) growing lachlan has been a pioneer of community-led place-based initiatives in nsw. over six years, growing lachlan has developed into an important community asset, with trusted relationships that allow it to play a range of crucial roles. the initiative began in 2014 with a request from lachlan shire interagency groups for a dedicated resource to improve communication and collaboration between services, increase service uptake and influence government funding to address local needs. over the first year, it developed into a community initiative that sought to identify and action opportunities that locals see as important to build on the shire’s strengths (dusseldorp forum 2020). across 2015 and 2016, the team gathered data on key indicators of community wellbeing, conducted surveys and held a series of community conversations to understand how the community is faring, and identify key challenges and opportunities across the shire. the information came together in the growing lachlan report which outlines the community’s shared priorities towards: 1. investing in community renewal 2. supporting economic sustainability and development 3. improving community safety and crime prevention 4. taking care of older and younger generations and 5. strengthening cultural recognition, respect and genuine participation. (mckenzie 2019). since 2017, growing lachlan has worked with all shire communities, supporting a wide variety of people, groups and services to activate their ideas towards advancing the community priorities. it is not a service provider itself, but rather plays a unique linking and supporting role, and encourages new ways of working and collaboration, challenging groups to think and act on how things can be done better, and better together. in addition to their growing lachlan work over the last two years, the team was instrumental in the development of the regional innovators’ network, an award-winning model of regional capacity building, with tacsi and capp dubbo (rin 2021). growing lachlan now holds sophisticated skills in authentic community engagement, community co-design and facilitation methods. it is growing a culture of innovation in the lachlan shire by supporting local groups to ‘test and try’ new approaches to addressing community priorities. growing lachlan has supported, partnered with and in some cases catalysed a range of locally specific innovations that have improved outcomes for individuals, families and the wider community. australian population studies 5 (2) 2021 mckenzie et al. 53 more and more communities are looking towards models of place-based collaboration to address their development needs (mckenzie 2020). adopting place-based collaborative development approaches also improves alignment of government and community aspirations, leading to more efficient and productive allocation of resources (taylor et al. 2017). such initiatives do need to be adequately resourced. empowerment of community is a necessary but not sufficient move towards a different destiny (rainnie 2021). furthermore, the challenges of community development in small regional towns including resource scarcity, volunteer burnout and serious socioeconomic disadvantage for population segments may still need to be addressed. 5. conclusions for communities facing population stability or decline, there are alternative ways to approach regional development and measure “success”. the case studies presented in this article show ways in which communities are more than simply the total number of people living there or whether that number is increasing or decreasing. instead, the selected case studies reveal themes of connection, place, agency and wellbeing. these factors contribute to ‘successful’ communities even in the face of population decline. for example, developing connections with a wider network within the region (condobolin and lake cargelligo) or across sectors (birchip) enables economic growth through innovation, irrespective of population growth. place-based approaches are emerging as a new way of unlocking local agency and capability. there has been growing interest in the question of agency, particularly at the local level. this has often materialised in the form of what has become known as placebased leadership (rainnie 2021). it is important to rethink existing assumptions about regional development through these lenses of place and local experience (fudge et al. 2021). when talking about agency, it is important to acknowledge the challenge of agency in the context of regional economies exposed to global fluctuations, and how local businesses and firms usually sit in a very low position on most value chains. regional policy narratives have been somewhat contradictory in this regard. on the one hand, rural communities are encouraged to be part of the global economy for their successful development, while on the other hand they are told that the future is in their own hands with local leaders expected to steer the fortunes of their communities (plummer et al. 2018). it is true that communities may not have full agency over their futures. global forces – economic and social – play an important role in shaping the fortunes of regional communities. yet, as shown in the case studies in this paper, being able to respond to, or take advantage of, change is an important element in regional development and one which creates community renewal and wellbeing even if not population growth. key messages • regional development has been underpinned by a growth paradigm which conflates population growth with success. • seeking population growth has been a popular regional policy aspiration, however, the reality of many locations is long-term population decline. 54 mckenzie et al. australian population studies 5 (2) 2021 • some towns with limited or no population growth can be successful by utilising networks, information flows and community-led place-based development. • being able to respond to, or take advantage of, change is an important element in regional development and one which creates community renewal and wellbeing even if not population growth. disclaimer the views expressed in this paper are those of the authors and should not be regarded as representing the views of the victorian government or the department of environment, land, water and planning. references abs (2016) census of population and housing 2016, australian government, canberra. https://www.abs.gov.au/websitedbs/censushome.nsf/home/2016 abs (various years) census of population and housing, australian government, canberra. https://www.abs.gov.au/census/find-census-data/historical. accessed on 9 july 2021. acott k (2018) history of 127-year-old katanning flour mill preserved as luxury premier mill hotel, the west australia. https://thewest.com.au/news/tourism/history-of-127-year-old-katanning-flourmill-preserved-as-luxury-premier-mill-hotel-ng-b88844136z. accessed on 17 november 2021. ames (adult multicultural education services) and deloitte access economics (2015) small towns big returns. economic and social impact of the karen resettlement in nhill. https://www2.deloitte.com/content/dam/deloitte/au/documents/economics/deloitte-aueconomics-small-towns-big-returns-nhill-resettlement-270415.pdf argent n (2010) from pillar to post? in search of the post-productivist countryside in australia. australian geographer 33(1): 97-114. https://doi.org/10.1080/00049180220125033 barr n (2009) the house on the hill: the transformation of australia’s farming communities. land and water australia in association with halstead press, canberra. batters j (2021) structured workplace learning at bcg. birchip cropping group news 28 june 2021. https://www.bcg.org.au/structured-workplace-learning-at-bcg/. accessed on 17 october 2021. bcg (birchip cropping group) (2018) agricultural stem in classrooms. birchip cropping group news 27 february 2018. https://www.bcg.org.au/agricultural-stem-in-classrooms/. accessed on 17 october 2021. bcg (birchip cropping group) (2020). annual report 2019-20. https://issuu.com/jemma35/docs/bcg_annual_report_201920-bcg-39jctn2. accessed on 22 june 2021. beer a (2012) the economic geography of australia and its analysis: from industrial to post-industrial regions. geographical research 50(3): 269-281. https://doi.org/10.1111/j.1745-5871.2012.00771.x beer a, mckenzie f, blažek j, sotarauta m, & ayres s (2020) every place matters: towards effective placebased policy. taylor and francis. https://doi.org/10.4324/9781003110118 blanc h & sierra c (1999) the internalisation of r&d by multinationals: a trade-off between external and internal proximity. cambridge journal of economics 23(2): 187-206. https://doi.org/10.1093/cje/23.2.187 boschma r (2005a) role of proximity in interaction and performance: conceptual and empirical challenges. regional studies 39(1): 41-45. https://doi.org/10.1080/0034340052000320878 https://www.abs.gov.au/websitedbs/censushome.nsf/home/2016 https://www.abs.gov.au/census/find-census-data/historical https://thewest.com.au/news/tourism/history-of-127-year-old-katanning-flour-mill-preserved-as-luxury-premier-mill-hotel-ng-b88844136z https://thewest.com.au/news/tourism/history-of-127-year-old-katanning-flour-mill-preserved-as-luxury-premier-mill-hotel-ng-b88844136z https://www2.deloitte.com/content/dam/deloitte/au/documents/economics/deloitte-au-economics-small-towns-big-returns-nhill-resettlement-270415.pdf https://www2.deloitte.com/content/dam/deloitte/au/documents/economics/deloitte-au-economics-small-towns-big-returns-nhill-resettlement-270415.pdf https://doi.org/10.1080/00049180220125033 https://www.bcg.org.au/structured-workplace-learning-at-bcg/ https://www.bcg.org.au/agricultural-stem-in-classrooms/ https://issuu.com/jemma35/docs/bcg_annual_report_201920-bcg-39jctn2 https://doi.org/10.1111/j.1745-5871.2012.00771.x https://doi.org/10.4324/9781003110118 https://doi.org/10.1093/cje/23.2.187 https://doi.org/10.1080/0034340052000320878 australian population studies 5 (2) 2021 mckenzie et al. 55 boschma r (2005b) proximity and innovation: a critical assessment. regional studies 39(1): 61-74. https://doi.org/10.1080/0034340052000320887 delwp (department of environment, land, water and planning) (2016) attraction and retention. education and labour force pathways of metropolitan and regional school leavers in victoria. summary of findings from the arc (australian research council) linkage project lp120100212, undertaken by the university of queensland; victorian government; graduate careers australia, and latrobe university bendigo, report prepared by forward policy and research branch delwp, victorian government, melbourne. https://www.planning.vic.gov.au/__data/assets/pdf_file/0033/97683/attraction-and-retention2016.pdf dowd a, marshall n, fleming a, jakku e, gaillard e, & howden m (2014) the role of networks in transforming australian agriculture. nature climate change 4: 558-563. https://doi.org/10.1038/nclimate2275 dusseldorp forum (2020) place based resilience: community driven response and recovery in a time of covid-19. dusseldorp forum, sydney. https://www.dusseldorp.org.au/wpcontent/uploads/2020/11/place-based-resilience_community-driven-response-and-recovery-ina-time-of-covid_19.pdf fitjar r & rodríguez-pose a (2011) innovating in the periphery: firms, values, and innovation in southwest norway. european planning studies 19(4): 555-574. https://doi.org/10.1080/09654313.2011.548467 florida r (2002) the economic geography of talent. annals of the association of american geographers 92(4): 743-755. https://doi.org/10.1111/1467-8306.00314 fudge m, ogier e, & alexander k a (2021) emerging functions of the wellbeing concept in regional development scholarship: a review. environmental science & policy 115: 143-150. https://doi.org/10.1016/j.envsci.2020.10.005 garnett a & lewis p (1999) trends in rural labour markets. the centre for labour market research, discussion paper 99/9, murdoch university, perth. granovetter m (1973) the strength of weak ties. american journal of sociology 78(6): 1360-1380. https://doi.org/10.1086/225469 granovetter m (1983) the strength of weak ties: a network theory revisited. sociological theory 1: 203233. https://doi.org/10.2307/202051 guaralda m, hearn g, foth m, yigitcanlar t, mayere s, & law l (2020) towards australian regional turnaround: insights into sustainably accommodating post-pandemic urban growth in regional towns and cities. sustainability 12. hannam k, sheller m, & urry j (2006) mobilities, immobilities and moorings. mobilities 1(1): 1-22. https://doi.org/10.1080/17450100500489189 hogan a & young m (2015) the making of rural and regional australia: an introduction. in hogan a and young m (eds.) rural and regional futures. routledge; pp. 1-15. https://doi.org/10.4324/9781315775333 kesselring s (2014) mobility, power and the emerging new mobilities regime. sociologica 1(1): 1-30. mackay s, klaebe h, hancox d, & gattenhof s (2021) understanding the value of the creative arts: placebased perspectives from regional australia. cultural trends. https://doi.org/10.1080/09548963.2021.1889343 mckenzie f (2015) understanding regional cities: combining quantitative and qualitative methods in case studies of orange and goulburn, nsw. australasian journal of regional studies 21(3): 303-322. https://www.anzrsai.org/assets/uploads/publicationchapter/mckenzie-final.pdf https://doi.org/10.1080/0034340052000320887 https://www.planning.vic.gov.au/__data/assets/pdf_file/0033/97683/attraction-and-retention-2016.pdf https://www.planning.vic.gov.au/__data/assets/pdf_file/0033/97683/attraction-and-retention-2016.pdf https://doi.org/10.1038/nclimate2275 https://www.dusseldorp.org.au/wp-content/uploads/2020/11/place-based-resilience_community-driven-response-and-recovery-in-a-time-of-covid_19.pdf https://www.dusseldorp.org.au/wp-content/uploads/2020/11/place-based-resilience_community-driven-response-and-recovery-in-a-time-of-covid_19.pdf https://www.dusseldorp.org.au/wp-content/uploads/2020/11/place-based-resilience_community-driven-response-and-recovery-in-a-time-of-covid_19.pdf https://doi.org/10.1080/09654313.2011.548467 https://doi.org/10.1111/1467-8306.00314 https://doi.org/10.1016/j.envsci.2020.10.005 https://doi.org/10.1086/225469 https://doi.org/10.2307/202051 https://doi.org/10.1080/17450100500489189 https://doi.org/10.4324/9781315775333 https://doi.org/10.1080/09548963.2021.1889343 https://www.anzrsai.org/assets/uploads/publicationchapter/mckenzie-final.pdf 56 mckenzie et al. australian population studies 5 (2) 2021 mckenzie f (2017) keeping up with the capital. how regional australia has responded to economic and demographic change since the 1990s. phd by portfolio of publications applied economics, university of south australia business school, school of commerce. mckenzie f (2019) growing lachlan: community driven place-based change. orange compass, sydney. https://www.orangecompass.com.au/images/growing_lachlan_outcomes_report.pdf o'neill p & fagan b (2006) geographical takes on three decades of economic reform in australia. geographical research 44(2): 204-219. https://doi.org/10.1111/j.1745-5871.2006.00377.x plowman i, ashkanasy n, gardner j, & letts m (2003) innovation in rural queensland: why some towns thrive while others languish. main report. university of queensland business school and queensland department of primary industries, brisbane. https://cep.org.au/wpcontent/uploads/2021/03/why-small-towns-prosper-1.pdf plummer p, tonts m, & argent n (2018) sustainable rural economies, evolutionary dynamics and regional policy. applied geography 90: 308-320. https://doi.org/10.1016/j.apgeog.2017.01.005 rainnie a (2021) regional development and agency: unfinished business. local economy. https://doi.org/10.1177/02690942211023877 rallet a & torre a (1999) is geographical proximity necessary in the innovation networks in the era of the global economy? geojournal 49(4): 373-380. https://doi.org/10.1023/a:1007140329027 rérat p (2016) migration and post-university transition. why do university graduates not return to their rural home region? geographica helvetica 71(4): 271-282. https://doi.org/10.5194/gh-71-271-2016 rin (regional innovators’ network) (2021) the australian centre for social innovation. https://regionalinnovation.com.au/about. accessed on 20 july 2021. scott a & storper m (2003) regions, globalization, development. regional studies 37(6/7): 549-578. https://doi.org/10.1080/0034340032000108697a shearmur r (2012) are cities the font of innovation? a critical review of the literature on cities and innovation. cities 29(s2): s9-s18. https://doi.org/10.1016/j.cities.2012.06.008 shearmur r (2015) far from the madding crowd: slow innovators, information value, and the geography of innovation. growth and change 46(3): 424-442. https://doi.org/10.1111/grow.12097 shire of katanning (2021) katanning regional sheep saleyards. https://www.katanning.wa.gov.au/explore/things-to-see-and-do/regional-saleyards.aspx. accessed on 17 november 2021. smailes p (2002) from rural dilution to multifunctional countryside: some pointers to the future from south australia. australian geographer 33(1): 79-95. https://doi.org/10.1080/00049180220125024 smailes p, griffin t, & argent n (2019) regional cities and city regions in australia: a long-term demographic perspective. springer. https://doi.org/10.1007/978-981-13-1111-6 stayner r & reeve i (1990) 'uncoupling': relationships between agriculture and the local economies of rural areas in new south wales. rural towns project, the rural development centre, university of new england, armidale, nsw. taylor b, walton a, loechel b, measham t, & fleming d (2017) strategic foresight for regional australia: megatrends, scenarios and implications. canberra: csiro and the australian government department of infrastructure, regional development and cities. https://doi.org/10.25919/5be328254aaa7 the premier mill (2021) premier mill hotel: our story, an icon restored. https://premiermillhotel.com/the-restoration-1. accessed on 17 november 2021. tonts m & haslam mckenzie f (2005) neoliberalism and changing regional policy in australia. international planning studies 10(3-4): 183-200. https://doi.org/10.1080/13563470500378861 https://www.orangecompass.com.au/images/growing_lachlan_outcomes_report.pdf https://doi.org/10.1111/j.1745-5871.2006.00377.x https://cep.org.au/wp-content/uploads/2021/03/why-small-towns-prosper-1.pdf https://cep.org.au/wp-content/uploads/2021/03/why-small-towns-prosper-1.pdf https://doi.org/10.1016/j.apgeog.2017.01.005 https://doi.org/10.1177/02690942211023877 https://doi.org/10.1023/a:1007140329027 https://doi.org/10.5194/gh-71-271-2016 https://regionalinnovation.com.au/about https://doi.org/10.1080/0034340032000108697a https://doi.org/10.1016/j.cities.2012.06.008 https://doi.org/10.1111/grow.12097 https://www.katanning.wa.gov.au/explore/things-to-see-and-do/regional-saleyards.aspx https://doi.org/10.1080/00049180220125024 https://doi.org/10.1007/978-981-13-1111-6 https://doi.org/10.25919/5be328254aaa7 https://premiermillhotel.com/the-restoration-1 https://doi.org/10.1080/13563470500378861 australian population studies 5 (2) 2021 mckenzie et al. 57 tonts m, argent n, & plummer p (2012) evolutionary perspectives on rural australia. geographical research 50(3): 291-303. https://doi.org/10.1111/j.1745-5871.2011.00745.x urry j (2003a) social networks, travel and talk. british journal of sociology 54(2): 155-176. https://doi.org/10.1080/0007131032000080186 urry j (2003b) global complexity. polity. urry j (2004) connections. environment and planning d: society and space 22(1): 27-37. https://doi.org/10.1068/d322t walga (western australian local government association) (2019) snapshot of multi-cultural katanning. https://walga.asn.au/getattachment/policy-advice-and-advocacy/people-and-place/accessibleand-inclusive-communities/multicultural-katanning-presentation-walga-oct-2019final.pdf?lang=en-au. accessed on 2 june 2021. https://doi.org/10.1111/j.1745-5871.2011.00745.x https://doi.org/10.1080/0007131032000080186 https://doi.org/10.1068/d322t https://walga.asn.au/getattachment/policy-advice-and-advocacy/people-and-place/accessible-and-inclusive-communities/multicultural-katanning-presentation-walga-oct-2019-final.pdf?lang=en-au https://walga.asn.au/getattachment/policy-advice-and-advocacy/people-and-place/accessible-and-inclusive-communities/multicultural-katanning-presentation-walga-oct-2019-final.pdf?lang=en-au https://walga.asn.au/getattachment/policy-advice-and-advocacy/people-and-place/accessible-and-inclusive-communities/multicultural-katanning-presentation-walga-oct-2019-final.pdf?lang=en-au australian population studies 2022 | volume 6 | issue 1 | pages 37-40 © guan, raymer & gray 2022. published under the creative commons attribution-noncommercial licence 3.0 australia (cc by-nc 3.0 au). journal website: www.australianpopulationstudies.org identifying different sources of school enrolment change in the australian capital territory qing guan* the australian national university james raymer the australian national university edith gray the australian national university *corresponding author. email: qing.guan@anu.edu.au. address: school of demography, research school of social sciences, the australian national university, act 2601 paper received 10 march 2022; accepted 7 june 2022; published 25 july 2022 introduction to provide more accurate enrolment projections for local schools and to facilitate long-term education infrastructure planning, practitioners and researchers require detailed enrolment data to analyse school enrolment changes (gla demography 2015; national center for education statistics 2016; pajankar & srivastava 2019; raymer, biddle & guan 2017). we adopt the multiregional cohort component project model developed for regional population projections (rogers 1995) to analyse and project school enrolments in the australian capital territory (act). this approach allows users to track the sources of school enrolment change over time. in this data visualisation, we present the application of a multiregional cohort component model to identify sources of school enrolment change in the act. data and methods the act public school system consists of preschools, primary schools containing kindergarten to year 6 students, high schools containing year 7 to year 10 students, and colleges containing year 11 and year 12 students. we have developed a school enrolment projection model, known as the school transition estimation and projection (step) model, based on a multiregional demographic accounting framework. the step model uses linked administrative school enrolment data to track student movements across academic levels and between all schools in the school system, including those in the catholic and independent sectors. the specification of the step model is as follows: 𝐄𝐭+𝟏 = 𝐆𝐄𝐭 + 𝐏 + 𝐈 (1) where 𝐄𝐭 and 𝐄𝐭+𝟏 are two vectors of student enrolment at times t and t+1, respectively. 𝐏 is a vector of preschool enrolments at t+1. 𝐈 is a vector of new students who have migrated to the school system at time t+1. 𝐆 contains transition probabilities between all schools and academic levels from time t to time t+1, taking into account graduates and students who left the school system prior to year 12 (i.e., students who have out-migrated). d e m o g ra p h ic http://www.australianpopulationstudies.org/ mailto:qing.guan@anu.edu.au 38 guan, raymer & gray australian population studies 6 (1) 2022 as a basis for annual projections for the next ten years, the most recent five years of observed enrolment data are used. observed enrolment data and transition information were provided by the act education directorate, containing administrative data on all individual students in the act from 2011 to 2021. future preschool enrolments are based on assumptions regarding the relationship with observed or projected suburb-level fertility four years prior to the projection year. the fertility projections are provided by the act chief minister, treasury and economic development directorate. both the student enrolment data and fertility projection data are not publicly available and were provided directly for the purposes of this analysis. the step model framework allows one to decompose year-to-year school enrolment growth by different sources of enrolment change. year-to-year school enrolment change is the difference between 𝐄𝒕+𝟏 and 𝐄𝒕. for the whole act school system, there are four components contributing to school enrolment changes: (i) new preschool entries (p); (ii) graduates from year 12 (embedded in g); (iii) student in-migration to kindergarten to year 12 (i); and (iv) student out-migration from preschool to year 11 (embedded in g). when examining school enrolment changes amongst individual schools or between districts, there are two additional sources of enrolment change (both embedded in g): (v) transitions from the school / district of interest to other schools / districts; and (vi) transitions from other schools / districts to the school / district of interest. the decomposition allows users to analyse and project each enrolment component separately. key features in figure 1, we present the 2011-12 to 2030-31 sources of school enrolment change for eight statistical area level 3 (sa3) districts in the act. the time periods are separated into three segments depending on whether the data are considered background information, inputs or outputs of the step model. the distribution of sa3 boxes represents their relative geographic locations in the act, created using the geofacet r package. north canberra, south canberra, woden valley, weston creek, and tuggeranong are established districts with no new suburbs planned. they experienced relatively slow or minimum total enrolment changes (the diamonds) over the last ten years and are projected to experience very conservative total changes in the next ten years. with the multiregional cohort component model, we differentiate the dynamics of enrolment changes in these five districts. though they have similar levels of total enrolment change, these are driven by different components. the two central districts north canberra and south canberra both gain from net migration. north canberra has more preschool entries contributing to its enrolment change (around 600 students each year), whereas south canberra has around 400. instead, south canberra gains over 300 students each year from inter-district transfers. woden valley also gains from inter-district transfers projected for 2022-2031, as well as from net migration. weston creek does not have a public college, australian population studies 6 (2) 2022 guan, raymer & gray 39 figure 1: year-to-year sources of school enrolment growth across act by sa3 district: 2011-12 to 2015-16 background data, 2016-17 to 2020-21 model input data, and 2021-22 to 2030-31 projected data source: step model (15 december 2021 version) 40 guan, raymer & gray australian population studies 6 (1) 2022 thus there are very few year 12 graduates and very high transfers to other districts (around 500 students each year). preschool entry is the largest booster to enrolment growth in weston creek, contributing to around 400 new students each year. tuggeranong has high levels of preschool entries (over 1,000 each year), low levels of inter-district transfers, and high levels of student out-migration or graduation. gungahlin and belconnen are two established districts with new suburbs built and planned. both districts gain from preschool entries and migration, but lose students from inter-district transfers. gungahlin exhibited the highest observed enrolment growth across all districts, including substantial growth in preschool entries. school enrolment growth in gungahlin is expected to slow in the next ten years, driven by drops in preschool entries, increases in year 12 graduates and losses from interdistrict transfers. preschool entry is the main contributor to belconnen’s school enrolment growth (over 1,200 students in observed data). with several new residential suburbs in the process of being developed, this component is expected to further increase to over 1,500 students after 2024 and to 1,600 students after 2029. the substantial growth in preschool entries is expected to drive enrolment change in belconnen, counterbalancing increases in year 12 graduates and net losses from interdistrict transfers. finally, molonglo is the newest district in the act with several more residential suburbs and new schools planned. school enrolment in molonglo is projected to substantially increase in the next ten years, driven by preschool entries. multiregional demographic accounting frameworks applied to analyse and project school enrolments, like the step model, produce more accurate projections and make better use of available enrolment data (raymer, biddle & guan 2017). traditional school enrolment projection models on the other hand are more inefficient and prone to error as they focus on particular schools, independent of other schools. acknowledgements the step model and sources of growth outputs are developed for the anu school of demography / act education directorate joint research project on school enrolment analysis and modelling. references gla demography. (2015). projected demand for school places: briefing 2015-02. london: greater london authority. retrieved 21/02/2022 from http://dera.ioe.ac.uk/24855/2/projected%20demand%20for%20school%20places_redacted.pdf national center for education statistics. (2016). projections of education statistics to 2024. washington dc us department of education. retrieved 21/02/2022 from https://nces.ed.gov/pubs2016/2016013.pdf pajankar, v., & srivastava, s. (2019). an approach of estimating school enrolment with reconstructive cohort approach. journal of physics: conference series, 1366(012116). https://doi.org/10.1088/1742-6596/1366/1/012116 raymer, j., biddle, n., & guan, q. (2017). a multiregional sources of growth model for school enrolment projection. australian population studies, 1(1), 26-40. https://doi.org/10.37970/aps.v1i1.10 rogers, a. (1995) multiregional demography: principles, methods and extensions. wiley, chichester. http://dera.ioe.ac.uk/24855/2/projected%20demand%20for%20school%20places_redacted.pdf https://nces.ed.gov/pubs2016/2016013.pdf https://doi.org/10.1088/1742-6596/1366/1/012116 https://doi.org/10.37970/aps.v1i1.10 australian population studies 2021 | volume 5 | issue 2 | pages 58–72 © argent 2021. published under the creative commons attribution-noncommercial licence 3.0 australia (cc by-nc 3.0 au). journal website: www.australianpopulationstudies.org dissecting the drivers of rural demographic decline: the case of the new england & north west region of new south wales neil argent university of new england email: nargent@une.edu.au. address: department of geography and planning, university of new england, armidale, nsw, 2351, australia paper received 21 september 2021; accepted 14 november 2021; published 29 november 2021 abstract background in the contemporary academic literature, rural population decline has generally been regarded as a long-running and almost natural phenomenon. aims this paper examines the complex temporal, spatial and cultural dynamics of the population of an inland, largely agriculturally-dependent rural region, the new south wales new england & north west statistical division (sd), from the late 1990s to the 2016 census. it investigates the key demographic processes that have driven the region’s spatially and temporally-uneven experiences of population change – including decline – over this tumultuous period, using these as portents of the regional population’s likely future trajectories. data and methods the analysis draws on australian bureau of statistics census data for the sd’s population as a whole, and for the non-indigenous and indigenous segments of the population. results the analysis identifies that a profound ageing process is underway across the entire region, is becoming more severe with the passing years, and leading to natural decrease for some shires. however, the sd’s indigenous population presents a striking contrast to the non-indigenous one, growing rapidly, increasing its share of the population and is a force for demographic rejuvenation. conclusions the analysis reveals that a profound ageing process is underway across the entire sd and is becoming more severe with the passing years. the sd’s indigenous population presents a striking contrast to the non-indigenous, growing rapidly, increasing its share of the population – particularly in the western-most lgas – and is a force for demographic rejuvenation. key words new england & north west region; indigenous populations; rural population ageing; natural decrease. http://www.australianpopulationstudies.org/ mailto:nargent@une.edu.au australian population studies 5 (2) 2021 argent 59 1. introduction not long ago, the idea of ‘rural decline’ was something of a tautology; during the 1980s, 1990s and into the 2000s the two terms almost pre-supposed each other. the era of the ‘rural crisis’ triggered enormous media interest in and coverage of, inter alia, the dire struggle of family farmers against international commodity markets and national banks, dying country towns, disappearing public and private services and sporting and social clubs, along with the fate of those residents left behind to cope. academic interest in ‘the plight of the bush’ was indicated by a number of key publications seeking to disentangle the myriad causes of the crisis from its effects, and to offer policy-relevant remedies (see, for example, pritchard & mcmanus 2000; gray & lawrence 2001). australian rural regions were not alone in experiencing the chill winds of economic and political restructuring, with similar processes of restructuring noted in the us, canada, the uk and numerous european nations where neoliberalised political reforms to macro-economic management in general, and to agricultural industry in particular, combined to pressure rural regions and towns (see, for example, ramsey 2010; nelson 2001; bryant & joseph 2001; ilbery & bowler 1998). from a demographic perspective, the phenomenon of rural population decline in the contemporary era is neither surprising nor especially newsworthy. ravenstein’s laws of migration and the hypothesis of the mobility transition (zelinsky 1971) the latter providing the missing migration and spatial dimension of the theory of demographic transition foresaw the tendency for societies become more urbanised as they undergo modernisation under the conditions of industrial capitalism. the relative growth of the world’s urban population at the expense of rural areas has continued since the onset of the industrial revolution and, by 2005, the global population had become, for the first time in recorded history, predominantly urban (united nations 2012). however, such a finding tells us little, if anything about the complex ways in which rural regions and places are changing around the world. due to their own particular initial resource endowments, politicaleconomic and institutional histories, different rural areas will have their own unique experience of economic, technological, political and demographic change. in this paper, i examine the complex temporal, spatial and cultural dynamics of the population of an inland, largely agriculturally-dependent rural region, the new england & north west region of new south wales, from the late 1990s to the 2016 census. the paper seeks to tease out the key demographic processes that have driven the region’s spatially and temporally-uneven experiences of population change – including decline – over this tumultuous period, using these as portents of the regional population’s likely future trajectories. notably, the paper explores these processes and trajectories for the region’s population as a whole, then for the non-indigenous and indigenous segments, though the comparative analysis is constrained somewhat by a lack of local demographic data for the indigenous population. 2. rural population decline in the literature in the contemporary academic literature, rural population decline has generally been regarded as a long-running and almost natural phenomenon that commenced from the end of the famed postwwii “long boom”. often, decline is associated with the increasing mechanisation of agriculture, associated expanded economies of scale in farming, and concomitant increasing pressures on the 60 argent australian population studies 5 (2) 2021 family farming system that prevailed across much of the developed world. such accounts give primacy to economic forces in explaining the push factors for demographic change, i.e., for people to leave rural employment and hence rural areas altogether. levels of urbanisation across the globe increased throughout the post-wwii era, although they were already high in so-called metrocentric settler societies such as australia, new zealand and canada by the 1950s (smailes et al. 2019; argent 2016). the fortunes of agriculture and of rural society and settlement have generally been intimately intertwined with one another but also bound up with the institutions of government and subject, to a greater or lesser degree, to their political objectives. in australia, colonial and then federal and state governments from the 1850s to the 1980s explicitly and actively encouraged rural versus urban economic, demographic and social development via “closer settlement” policies and a swathe of support mechanisms for agriculture and rural living (williams 1975; argent 2002), a stance consistent with the so-called productivist regime that pertained across western europe, the uk and north america (ilbery & wilson 1998; wilson 2001). nonetheless, while productivist supports for farming were thought to be a tide that lifts all boats (read individual farmers and rural regions) population trends painted a slightly more complex picture. rural population trajectories across much of the socalled ‘global north’ have been anything but homogenous over the past few generations with, for example, net population loss through the 1950s to 1970s across western europe and the united states coinciding with slow growth in australia but relatively robust net increase in the uk (argent 2016). subsequent international trends are similarly heterogenous in spite of the generally widespread occurrence of so-called counterurbanisation (a.k.a. the population turnaround) during the 1970s and 1980s (beale 1975; hugo 1994). in some nations such as australia, counterurbanisation did indeed have a measurable impact for a decade or so on destination regions and towns almost across the rural spectrum, but then became more spatially and socially selective, and tied much more to higher amenity regions (argent et al. 2007). as noted in the literature, peri-metropolitan fringe zones, coastal, lacustrine and riverine settings, and well-watered and well-connected inland regions have been favoured by a new generation of counterurbanisation migrants (burnley & murphy 2004). elsewhere, a range of population pathways have been documented, depending somewhat on local economic bases, land uses and relative accessibility/remoteness. counterurbanisation migration flows bolstered the population stocks of select rural regions but growing concern was expressed for the more remote, natural resource dependent and ‘deep rural’ (lowe & ward 2009) regions of europe, north america and australasia. in tacit recognition of the interrelated nature of rural populations, land use and economy (see brown & argent 2016), a ‘desertification’ phenomenon was observed, where net population loss and demographic decline (e.g., loss of young cohorts, ‘left behind’ aged people) combined with public and private service withdrawals to effectively eviscerate the demographic, social and economic life of more remote regions and villages (cawley 1994; sorensen 1990). consistent with a myrdalian perspective on capitalist regional development, in which cycles of circular and cumulative causation drive regional divergence, “(i)n many cases, demographic change is taken as a proxy indicator for regional performance, and hence population decline is equated with weak economic development and/or less attractive living conditions” (dax & fischer 2018 p. 301). australian population studies 5 (2) 2021 argent 61 for those regions and localities not favoured by amenity-led in-migration and counterubanisation flows, then, local economic dependence on natural resources – particularly agriculture but also mineral and energy extraction – can generate demographic responses (e.g., age-specific outmigration) that further drive down factor conditions for the local economy. a corollary of these processes is the marginalisation of the ‘left-behind’ – those ageing-in place, the relatively immobile due to physical incapacity and/or lack of access to reliable transport – constrained in their ability to access even basic public services as these retreat to localities further up the local urban hierarchy, and unable to liquidate their housing asset and move to a better location due to the collapse of the local housing market (cawley 1994). 3. diverging demographic dynamics in an agriculturally-dependent region 3.1. overview of the new england & north west region the empirical focus of this paper is the broad northern statistical division (sd) of new south wales, an official spatial unit that comprises the northern tablelands, north central plain and north west slopes subregions (see figure 1). the region was re-named the new england and north west region in 2011, and for simplicity’s sake is referred to as new england & north west hereafter. covering an expanse of approximately 99,100 square kilometres (australian bureau of agricultural and resource economics and sciences 2015) in the northern sector of the state, the sd’s centroid is an approximate six hour drive north of the state’s capital, sydney, and five hours drive south from the queensland capital, brisbane. as discussed elsewhere (argent et al. 2010), a good understanding of the region’s historical geography cannot be had without recognising: (1) the process of colonisation (or invasion, as the indigenous inhabitants and others would see it), and (2) the particular role that the region fulfilled in the colony’s (later the state’s) and the nation’s development as a supplier of vital food and fibre and export income. in relation to the first point, the indigenous peoples of the area – principally the anaiwan and the kamilaroi – knew the tablelands well and it is they who felt the full force of the extension of pastoralism into the region from the 1830s. a squatting ‘mode of occupance’ (holmes 2006) developed in the region, facilitated by environmental conditions, locational and regulatory factors and the technological limits to settlement. population densities and overall population levels remained comparatively low through the 1800s and early 1900s (premier’s department 1951). ‘closer settlement’ policies of the late 1800s and early 1900s facilitated both the amalgamation of small properties and the subdivision of large runs. amidst this flux, the regional settlement system that expanded from the railway and major roads that ran along the spine of the new england tablelands was consolidated (premier’s department 1951; smailes & molyneux 1965; atchison 1977). figure 2 shows the changing population levels of the new england & north west sd – incorporating the statistical sub-divisions of the new england tablelands, the north central plain, and the north west slopes – from 1961 to 2016. as is clear from this graph, the new england & north west sd experienced sustained population growth (20%, or 1% per annum) over the two decades from 1961 to 1981. the new england tablelands’ population growth exceeded that of the aggregate sd (21.1%) while the drier, broadacre farming-dependent north central plain, incorporating moree and 62 argent australian population studies 5 (2) 2021 figure 1: map of new england & north west statistical division, showing major sub-divisions source: argent (2016). narrabri, grew more slowly (14.5%). over the same period, each major region’s population share shifted little: an approximate two percentage point decline for the north central plain was absorbed by its neighbour, the north west slopes. surprisingly, levels of urbanisation remained largely unchanged. in 1961, just over 65% of the sd’s population resided in towns or other urban centres; by 2011 this level had increased only slightly to 66.5%. throughout the 1980s and 1990s, though, the new england & north west sd experienced a period of slow, persistent population loss (see table 1). for the remainder of this paper, i explore: (1) the spatial unevenness of demographic dynamics within the region since the 1990s; (2) the key demographic drivers of regional and sub-regional population change, including net decline; (3) the extent to which key drivers provide an indication of future demographic change; and, finally, (4) the extent to which these demographic indicators have manifested themselves differently for indigenous and non-indigenous populations within the region, to the extent that data permit. australian population studies 5 (2) 2021 argent 63 figure 2: population levels in new england & north west sd and ssds, 1986-2016 source: australian bureau of statistics, various censuses. note: ssd = statistical subdivision. table 1: indigenous and non-indigenous population change, new england & north west sd and ssds, 1986-2016 (% rate of change over period) regions/populations 1986-1996 1996-2006 2006-2016 1986-2016 new england & north west -1.2% -1.6% 5.3% 2.4% non-indigenous -3.0% -3.8% 2.7% -4.1% indigenous 39.3% 33.1% 35.4% 151.0% northern tablelands ssd 0.9% 1.0% 1.7% 3.6% north central plan ssd -9.2% -8.5% -3.1% -19.4% northern slopes ssd -0.3% 2.0% 8.8% 10.7% source: australian bureau of statistics, various censuses. note: ssd = statistical subdivision. 3.2. disentangling different scales and trends in population change figure 2 and table 1 show that different sectors of the new england & north west sd have experienced contrasting population trends since the mid-1980s, at least. the entire sd reached its hitherto population peak of about 181 000 people at the 1991 census, a level that it would not see again until a generation later. from the mid-1980s to the mid-1990s population loss was felt across almost the entire region but loss was concentrated particularly in the drier and more broadacre farming dependent north central plain. while individual towns and local government areas on the tablelands and the northern slopes did experience decline during this period – smaller shires such as barraba and bingara losing over ten per cent of their residents between 1991 and 1996 – the overall population levels of these sub-regions exhibited slow growth over the entire three decades. shires contiguous with the major regional centres of armidale and tamworth generally grew the fastest. however, the two westernmost local government areas of moree and narrabri underwent consistent depopulation over the thirty years from 1986, though the rate of loss did taper off in the decade prior to 2016. between them, these two last-mentioned shires lost close to one-fifth of their populations over the study period. 64 argent australian population studies 5 (2) 2021 in many respects, then, the new england & north west sd’s last quarter century or so of fluctuating trends – first substantial decrease then slow but steady recovery – disguise some complex intra-regional processes. while net loss was consistently felt across all local government areas through the late 1980s and 1990s, the turn of the new century saw the eastern and southern portions of the sd – the northern tableland and the northern slopes – on the one hand, and the north-westernmost lgas of narrabri and moree, on the other, diverge in their demographic trajectories. this emerging divide somewhat reflects the growing importance of the regional centres of armidale, inverell, tamworth and gunnedah relative to the smaller, more agriculturallydependent and lower amenity towns to the north and west. further investigation of the dimensions of demographic divergence between the major regional centres and the smaller towns and rural remainder is not possible within the confines of this paper but would be a worthwhile exercise. as discussed above, the sd is home to the indigenous people of the anaiwan and kamillaroi nations (amongst others) as well as a non-indigenous population of more recent provenance. for some decades it has been evident that australia’s indigenous and non-indigenous population are substantially different to each other in key demographic terms. while there has been some convergence noted between them over recent years in the areas of fertility and life expectancy (carmichael 2019), australia’s indigenous population exhibits a generally much younger age structure, is much more likely to live in rural and remote regions, and is growing substantially faster than the non-indigenous (taylor 1997; markham & biddle 2018; taylor et al. 2020). as table 1 and figure 2 reveal, such stark contrasts between the two populations can also be discerned within the new england & north west sd. a key aspect of indigenous population growth, but one that is difficult to precisely quantify, relates to the relative propensity of people to identify as indigenous; a tendency that varies spatially and temporally. as shown by biddle & markham (2018), the national indigenous population grew by around 80 000 across the 2011-16 intercensal period due to this factor alone. gains due to changes in identification have been strongest in more accessible parts of the country and negligible – at best – in remote regions (taylor et al. 2020). between 1986 and 2016, while the sd population grew by close to 2.4 per cent in aggregate – though noting the losses experienced through the 1980s and 1990s – the indigenous population increased by just over 150 per cent (i.e., 5 per cent per annum). the non-indigenous population thus dropped by over four per cent overall for the same period. although the indigenous population still constitutes a relatively small minority within the sd its share of the population has more than doubled between 1986 and 2016, from just over four per cent to over ten per cent; though it comprises nearly 22 per cent of the moree plains lga population. the reasons for this substantial proportional shift are difficult to pinpoint due to the unavailability of indigenous fertility and mortality data at detailed spatial scales but, as is discussed in more detail below, a younger age structure, higher fertility and higher rates of natural increase should feature prominently in any explanation. 3.3. seven steps of rural population decline a diagnostic framework of impending population decline for new zealand regions based on detailed demographic analysis was developed by jackson (2014). consisting of seven broad indicators that encompass the changing balance between mortality and fertility, and between australian population studies 5 (2) 2021 argent 65 young and old, jackson’s (2014) model demonstrates how shifts in age-sex structure over time can lead to natural decrease and almost irreversible population decline (see table 2). for this paper, the full set of measures was applied to the statistical division for most of the study period (19962016), but only the first six indicators were used in the separate analyses of the indigenous and non-indigenous populations of the broad region and for a slightly shorter period due to unavailability of data (e.g., no indigenous births and deaths data at the regional scale prior to 2001). table 2: jackson’s (2014) de-population diagnostic framework indicator demographic measure youth deficit 15-24 year old population < 15% of population labour market entry/exit ratio 15-24 / 55-64 age group ratio < national average 15-24 / 55-64 age group ratio < 1 prime reproductive ages* per cent 25-39 age group < national mean per cent 25-39 age group < 15% of local population older population per cent aged over 65 > national mean per cent aged over 65 > 20% of local population prime working age population aged 20-69 declining over time: numerically population aged 20-69 declining over time: proportionally older/youth ratio ratio of over 65s / 0-14 year olds > national mean ratio of over 65s / 0-14 year olds > 1 natural increase/decrease natural increase < 0.25% local population size source: jackson (2014). * note: for the indigenous population, the prime reproductive age measure encompassed the 15-34 year age group given its different fertility behaviour (carmichael 2019). table 3: jackson’s (2014) diagnostic framework applied to new england & north-west sd and australia 1996 2001 2006 2011 2016 youth deficit australia 14.4% 13.7% 13.6% 13.3% 12.8% new england & n w 13.7% 13.1% 12.8% 12.7% 12.2% entry/exit ratio australia 1.73 1.46 1.23 1.15 1.09 new england & n w 1.44 1.23 1.04 0.98 0.90 prime reproductive australia 23.2% 22.1% 20.9% 20.9% 21.1% new england & n w 20.8% 18.9% 17.0% 16.2% 16.4% % aged 65+ australia 12.0% 12.6% 13.3% 14.0% 15.1% new england & n w 12.5% 13.8% 15.5% 17.2% 19.6% prime working age australia 62.5% 63.0% 63.8% 64.5% 64.6% new england & n w 60.6% 60.3% 60.5% 60.5% 60.5% older/youth ratio australia 0.56 0.61 0.67 0.73 0.84 new england & n w 0.53 0.61 0.72 0.83 1.00 source: australian bureau of statistics, various censuses. note: shaded cells indicate results that exceed the indicator’s threshold. 66 argent australian population studies 5 (2) 2021 table 3 displays the results for six of the jackson indicators applied to the nation as a whole and the new england & north west sd. of course, it needs to be recognised that regional age-sex structures are influenced, to a greater or lesser degree, by those of the broader nation, along with political, economic and social processes operating at national and other geographical scales. while the table does clearly demonstrate the gradual decline of the youthfulness of the sd and the concomitant impact of that change on other segments of the population it is important to recognise that australia as a whole is also ageing and has been for some time. for instance, in relation to the youth deficit indicator, both australia and the new england & north west sd have fallen below the 15 per cent threshold since 1996 at least, though the size of the sd’s deficit has expanded more rapidly than that of the nation. the workforce entry/exit ratio was well above 1.0 for both at the period’s onset but has progressively declined over time and fallen below 1.0 (i.e., more people are notionally moving out of the workforce than entering) for the new england & north west sd from around the time of the 2011 census. not surprisingly, then, the proportion of the population in the post-retirement ages has increased at national and new england & north west sd scales. equally, the older/youth ratio has steadily grown from around one older person to every two youths in 1996 to 1.0 for the new england & north west sd in 2016, and is approaching that level for australia overall. the prime reproductive ages’ proportional share of the population has shrunk slowly for the nation and the sd, though did jump slightly for both in 2016. nonetheless, the new england & north west’s level is nearing the 15 per cent threshold for this indicator. finally, the prime working age indicator has actually increased for the nation – possibly due to substantial international migrant intakes during the 2000s – while remaining largely unchanged for the sd. overall, table 3 shows that the sd’s population has been more affected by ageing and other processes likely to lead to depopulation than the nation – another key dimension of jackson’s (2014) measures. needless to say, perhaps, migration has played a vital role in shaping population change and agesex structures across the new england & north west sd (see argent & walmsley 2008). table 4 shows that natural increase has been maintained across the broad region since the turn of the current century, though natural decrease emerged in two lgas (glen innes/severn and tenterfield), and nearly occurred in another (gwydir) during the 2011-16 intercensal period. such an outcome reflects declining fertility – itself an outcome of age-selective net migration loss – which also helps drive the structural ageing dynamic within the region. as is obvious in the table, net out-migration was especially severe in the more remote, broadacre agriculture-dependent areas of moree and narrabri during the early 2000s. however, in terms of its implications for the youthfulness of local populations, and their capacity to replace themselves, net migration loss may have had the most substantial impacts on some of the more remote tablelands and slopes shires, such as glen innes/severn, tenterfield and gwydir. australian population studies 5 (2) 2021 argent 67 table 4: natural change and net migration, new england & north west sd, 2001-2016 natural increase (per 1000) net migration rate (per 1000) natural increase (per 1000) net migration rate (per 1000) natural increase (per 1000) net migration rate (per 1000) 2001-06 2006-11 2011-16 northern tablelands ssd +17.83 -20.82 +18.72 +6.93 +14.04 +6.99 north central plain ssd +40.01 -119.13 +42.90 -69.39 +41.64 -46.20 northern slopes ssd +18.15 +1.44 +22.19 +13.98 +22.44 +26.34 new england and n w +21.73 -27.66 +24.08 -1.86 +22.02 +8.35 source: australian bureau of statistics, 2006, 2007, 2020a, 2020b. note: ssd = statistical sub-division. to convey some sense of how ageing processes have affected different parts of the sd over time and space, table 5 shows the results of applying the jackson framework to the local government areas (lgas) that comprise the new england & north west sd, showing the proportion of lgas whose population fell below each indicator. working down the table, one can clearly see the relative decline of the youth and adolescent cohorts in the sd over the two decades. almost all lgas in the region have experienced a youth deficit for the entire period, and the workforce entry/exit ratio has rapidly fallen below 1.0 for all but one lga by 2016, and was below the national mean for almost the full two decades. a further consequence of this shift for the region’s age structure can be seen in the growing proportion of lgas with over 20 per cent of their populations aged over 65 years and, on the same measure, which exceeded the national mean. table 5: jackson’s framework applied to lgas of new england & north west sd, 1996-2016 (% of lgas that fell below the critical threshold for each indicator) 1996 2001 2006 2011 2016 youth deficit 95% 95% 92% 92% 92% entry/exit ratio < 1 37% 58% 75% 75% 92% entry/exit ratio < national level 95% 84% 92% 92% 92% prime reproductive ages < 15% 0% 0% 8% 50% 42% prime reproductive ages < national mean 95% 95% 92% 100% 100% % aged 65+ > 20% 0% 11% 0% 42% 58% % aged 65+ > national mean 63% 79% 83% 100% 100% older/youth ratio > 1 0% 11% 0% 42% 58% older/youth ratio > national mean 53% 68% 67% 83% 92% source: australian bureau of statistics, 1998, 2007a, 2021. the gradual temporal and spatial ageing of the regional population can be seen in the increasing proportion of lgas whose older/youth ratio fell below 1 (from zero in 1996 to over half by 2016). similarly, there has been a slow but sure expansion of the number of shires whose ‘reproductive ages’ accounted for less than 15 per cent of their populations; from zero in 1996 and 2001 but up to around half by 2011. such trends point to an almost inexorable decline in fertility in the region, itself obviously linked to the out-migration and net migration loss of young adults, as well as retirement-focused in-migration, which further accelerates local ageing and predisposed the regions to future natural decrease. 68 argent australian population studies 5 (2) 2021 table 6 shows the same set of indicators applied to the new england & north west sd’s indigenous population for the four census years from 2001, while table 7 does the same in relation to the region’s non-indigenous population (i.e., last mentioned is derived by subtracting the indigenous population from the total population for the sd for each indicator). the indigenous population presents a striking contrast to the non-indigenous population across virtually all measures; the proportion of its population that is deemed young is growing rather than declining, and the same is true for the prime working age population. while the workforce entry/exit ratio has declined it has done so from a relatively high level, and is still over double that of the nation. on all other indicators the sd’s indigenous population easily exceeds the jackson indicator thresholds, although some slow convergence is occurring between the indigenous and nonindigenous populations in this regard. notwithstanding its minority status in the region the indigenous population is a source of rejuvenation. this point is borne out by table 7 which shows that the youth deficit would have been more severe, and the workforce entry/exit ratios and youth/aged ratios would have been lower, and would have fallen below the critical thresholds earlier, if not for the indigenous population. table 6: jackson’s framework applied to the new england & north west sd indigenous population 2001 2006 2011 2016 youth deficit 17.9% 17.9% 18.5% 18.1% entry/exit ratio < 1 3.73 3.26 2.95 2.31 entry/exit ratio < national level yes yes yes yes prime reproductive ages < 15% 31.5% 29.7% 29.8% 29.8% prime reproductive ages < national mean no no no no % aged 65+ > 20% 2.9% 3.5% 4.5% 6.0% % aged 65+ > national mean no no no no prime working age 49.0% 49.5% 51.3% 53.9% older/youth ratio > 1 0.07 0.09 0.12 0.17 older/youth ratio > national mean no no no no source: australian bureau of statistics (1998, 2007a, 2021) table 7: jackson’s framework applied to the new england & north west sd non-indigenous population 2001 2006 2011 2016 youth deficit 12.8% 12.4% 12.2% 11.5% entry/exit ratio < 1 1.15 0.96 0.89 0.82 prime reproductive ages < 15% 18.8% 16.9% 16.2% 16.4% % aged 65+ > 20% 14.6% 16.5% 18.4% 20.9% prime working age 61.1% 61.5% 61.4% 61.3% older/youth ratio > 1 0.69 0.83 0.97 1.18 source: australian bureau of statistics (1998, 2007a, 2021). australian population studies 5 (2) 2021 argent 69 4. conclusions in the context of the major ideological and regulatory shifts that took place in rural australia from the 1970s, along with occasional if severe climatic conditions (e.g., protracted droughts), this paper sought to explore some of the key factors that underlay its spatially and temporally uneven experiences of population decline through the 1990s and 2000s. in addition, but perhaps even more importantly, it has investigated some of the diverse trajectories of demographic change across the new england & north west sd over the past thirty years, taking into account indigeneity and relative location. unfortunately, and this is a weakness of the paper’s overall argument, data on indigenous fertility and mortality are unavailable at the local scale and this has precluded the measurement of natural increase/decrease and net migration in the same manner as for the non-indigenous population. in applying jackson’s (2014) diagnostic framework to the region as a whole, the analysis has revealed how ageing processes in combination with age-selective migration has affected the age structure, and eventually the capacity of individual lgas to, for example, replace retiring workers and experience natural increase. a profound ageing process is underway across the entire region and is becoming more severe with the passing years. however, the sd’s indigenous population presents a striking contrast to the non-indigenous one, growing rapidly, increasing its share of the population – particularly in the western-most lgas – and is a force for demographic rejuvenation, and boosting local school enrolments, the demand for post-school training and education and workforce entry rates. however, the continued heavy migration losses in the same regions where nature increase is highest suggest that a lack of employment and social opportunities, together with changing environmental factors, perhaps (e.g., droughts, heatwaves), are forcing residents to seek their fortunes elsewhere. nonetheless, it is vital for future demographic planning for rural regions that greater attention is paid to this trend by academics and policy-makers. key messages • structural population ageing is affecting the demographic trajectories of the broad rural region of the new england & north-west sd of new south wales, leading to eventual natural decrease in some local government areas. • net migration loss has been particularly severe in the most remote and agriculturally-dependent parts of the sd. • experiences of population decline through the 1990s and 2000s were felt unevenly throughout the sd. • the indigenous population of the sd has grown rapidly over the thirty years to 2016, and more than doubled its share of the population, particularly in the youth and young adult age groups where the demand for schooling and labour market entry support is greatest. acknowledgments the author thanks the reviewer for their incisive and constructive comments on an earlier draft of this paper. all remaining errors and omissions are those of the author alone. 70 argent australian population studies 5 (2) 2021 references abares (australian bureau of agricultural and resource economics and sciences) (2015) agriculture and forestry in the new england and north west region of new south wales, 2015. https://www.awe.gov.au/abares/research-topics/aboutmyregion/nsw-new-england abs (1998) regional statistics – new south wales, commonwealth of australia. abs (2006) deaths, australia, 2006, cat. no. 3302.0, commonwealth of australia. abs (2007a) 2001 census community profile series: time series profile: commonwealth of australia. abs (2007b) births, australia, 2006, cat. no. 3301.0, commonwealth of australia. abs (2020a) births, australia, 2019, cat. no. 3301.0, commonwealth of australia. abs (2020b) deaths, australia, 2019, cat. no. 3302.0, commonwealth of australia. abs (2021) age x sex for select new south wales local government areas, 2016 census, census tablebuilder. accessed 15 july 2021. argent n & walmsley d (2008) rural youth migration trends in australia: an overview of recent trends and two inland case studies. geographical research 46(2): 139-152. https://doi.org/10.1111/j.1745-5871.2008.00505.x argent n (2002) from pillar to post? in search of the post-productivist countryside in australia. australian geographer 33(1): 97-114. https://doi.org/10.1080/00049180220125033 argent n (2016) australia: trap or opportunity? natural resource dependence, scale, and the evolution of new economies in the space/time of new south wales’ northern tablelands. in halseth g. (ed.) transformation of resource towns and peripheries: political economy perspectives. routledge; pp. 18-50. argent n (2016) demographic change: beyond the urban-rural divide. in shucksmith m & brown d (eds.) routledge international handbook of rural studies. routledge; pp. 29-35. argent n, smailes p, & griffin t (2007) the amenity complex: towards a framework for analysing and predicting the emergence of a multifunctional countryside in australia. geographical research 45(3): 217-232. https://doi.org/10.1111/j.1745-5871.2007.00456.x argent n, walmsley d, & sorensen a (2010) something old, something new, something borrowed, something …..? rediscovering the comparative advantage of the ‘new’ pastoral economies of northern new south wales, australia. in halseth g, markey s and bruce d (eds.) the next rural economies: constructing rural place in global economies. cabi publishing; pp. 17-31. atchison j (1977) the evolution of settlement. in lea d, pigram j, & greenwood l (eds.) an atlas of new england, vol. 2 – the commentaries. department of geography, university of new england; pp. 171-181. beale c (1975) the revival of rural population growth in non-metropolitan america, ers report 605. washington, dc: usda-economic research service. biddle n & markham f (2018) indigenous identification change between 2011 and 2016: evidence from the australian census longitudinal dataset. caepr topical issue no. 1/2018, anu. https://caepr.cass.anu.edu.au/research/publications/indigenous-identification-change-between2011-and-2016-evidence-australian brown d & argent n (2016) the impacts of population change on society and environment. in shucksmith m & brown d (eds.) routledge international handbook of rural studies. routledge; pp. 85-95. bryant c & joseph a (2001) canada’s rural population: trends in space and implications in place. canadian geographer 45(1): 132-137. https://doi.org/10.1111/j.1541-0064.2001.tb01177.x burnley i & murphy p (2004) sea change: movement from metropolitan to arcadian australia. unsw press. carmichael g (2019) indigenous fertility in australia: updating alan gray. journal of population research 36(4): 283-317. https://doi.org/10.1007/s12546-019-09233-w https://www.awe.gov.au/abares/research-topics/aboutmyregion/nsw-new-england https://doi.org/10.1111/j.1745-5871.2008.00505.x https://doi.org/10.1080/00049180220125033 https://doi.org/10.1111/j.1745-5871.2007.00456.x https://caepr.cass.anu.edu.au/research/publications/indigenous-identification-change-between-2011-and-2016-evidence-australian https://caepr.cass.anu.edu.au/research/publications/indigenous-identification-change-between-2011-and-2016-evidence-australian https://doi.org/10.1111/j.1541-0064.2001.tb01177.x https://doi.org/10.1007/s12546-019-09233-w australian population studies 5 (2) 2021 argent 71 cawley m (1994) desertification: measuring population decline in rural ireland. journal of rural studies 10(4): 395-407. https://doi.org/10.1016/0743-0167(94)90049-3 dax t & fischer m (2018) an alternative policy approach to rural development in regions facing population decline. european planning studies 26(2): 297-315. https://doi.org/10.1080/09654313.2017.1361596 gray i & lawrence g (2001) a future for regional australia. cambridge university press. holmes j (2006) impulses towards a multifunctional transition in rural australia: gaps in the research agenda. journal of rural studies 22(2): 142-160. https://doi.org/10.1016/j.jrurstud.2005.08.006 hugo g (1994) the turnaround in australia: some first observations from the 1991 census. australian geographer 25(1): 1-17. https://doi.org/10.1080/00049189408703094 ilbery b & bowler i (1998) from agricultural productivism to post-productivism. in ilbery b (ed.) the geography of rural change. addison wesley longman; pp. 57-84. jackson n (2014) sub-national depopulation in search of a theory – towards a diagnostic framework. new zealand population review 40: 3-39. https://population.org.nz/app/uploads/2015/07/nzpr-vol40-_jackson.pdf lowe p & ward n (2009) england's rural futures: a socio-geographical approach to scenarios analysis. regional studies 43(10): 1319-1332. https://doi.org/10.1080/00343400903365169 markham f & biddle n (2018) recent changes to the indigenous population geography of australia: evidence from the 2016 census. australian population studies 2(1): 1-13. https://doi.org/10.37970/aps.v2i1.21 nelson p b (2001) rural restructuring in the american west: land use, family and class discourses. journal of rural studies 17(4): 395-407. https://doi.org/10.1016/s0743-0167(01)00002-x premier’s department (1951) the new england region: a preliminary survey of resources. sydney: government of new south wales. pritchard b & mcmanus p (eds.) (2000) land of discontent. unsw press. ramsey d (2010) rural restructuring and the new rural economy: examples from germany and canada. in halseth g, markey s, & bruce d (eds.) the next rural economies. constructing rural place in global economies. cabi; pp. 115-127. smailes p & molyneux j (1965) the evolution of an australian rural settlement pattern: southern new england, nsw. transactions of the institute of british geographers 36: 31-54. https://doi.org/10.2307/621452 smailes p, griffin t, & argent n (2019) regional cities and city regions in rural australia. springer. https://doi.org/10.1007/978-981-13-1111-6 sorensen a (1990) virtuous cycles of growth and vicious cycles of decline: regional economic change in northern nsw. in walmsley d (ed.) change and adjustment in northern nsw. department of geography and planning, university of new england; pp. 41-59. taylor a, wilson t, temple j, kelaher m, & eades s (2020) the future growth and spatial shift of australia’s aboriginal and torres strait islander population, 2016-2051. population, space and place https://doi.org/10.1002/psp.2401 taylor j (1997) the contemporary demography of the indigenous australians. journal of the australian population association 14: 77-114. https://doi.org/10.1007/bf03029488 united nations (2012) urban and rural areas, 2011. un population division. williams m (1975) more and smaller is better: australian rural settlement 1788-1914. in powell j & williams m (eds.) australian space/australian time: geographical perspectives. oxford university press; pp. 61-103. https://doi.org/10.1016/0743-0167(94)90049-3 https://doi.org/10.1080/09654313.2017.1361596 https://doi.org/10.1016/j.jrurstud.2005.08.006 https://doi.org/10.1080/00049189408703094 https://population.org.nz/app/uploads/2015/07/nzpr-vol-40-_jackson.pdf https://population.org.nz/app/uploads/2015/07/nzpr-vol-40-_jackson.pdf https://doi.org/10.1080/00343400903365169 https://doi.org/10.37970/aps.v2i1.21 https://doi.org/10.1016/s0743-0167(01)00002-x https://doi.org/10.2307/621452 https://doi.org/10.1007/978-981-13-1111-6 https://doi.org/10.1002/psp.2401 https://doi.org/10.1007/bf03029488 72 argent australian population studies 5 (2) 2021 wilson g (2001) from productivism to post-productivism … and back again? exploring the (un)changed natural and mental landscapes of european agriculture. transactions of the institute of british geographers 26(1): 77-102. https://doi.org/10.1111/1475-5661.00007 zelinsky w (1971) the hypothesis of the mobility transition. geographical review 61(2): 219-249. https://doi.org/10.2307/213996 https://doi.org/10.1111/1475-5661.00007 https://doi.org/10.2307/213996 australian population studies 2023 | volume 7 | issue 1 | pages 1–16 © gray & lazzari 2023. published under the creative commons attribution-noncommercial licence 3.0 australia (cc by-nc 3.0 au). journal website: www.australianpopulationstudies.org the continuing decline in cohort fertility and mixed evidence of narrowing educational differences edith gray* the australian national university ester lazzari university of vienna *corresponding author. email: edith.gray@anu.edu.au. address: school of demography, research school of social sciences, college of arts and social sciences, rsss building, 146 ellery crescent, the australian national university, acton act 2601 paper received 16 may 2023; accepted 7 september 2023; published 6 december 2023 abstract background typically, women with higher levels of education have had fewer children and were more likely to be childless compared to those with lower levels of education. however, in recent years, there has been a trend of fertility convergence between education levels in several high-income countries, including australia. new data are needed to determine whether this trend has continued and fertility remains educationally stratified among australian women. aims using the latest available data from the 2021 australian census, the aim of this study is to compare the average completed fertility and number of children ever born of women born between 1952 and 1981, with a specific focus on how these metrics vary by education level. data and methods data on the number of children ever born are sourced from the 2021 australian census to compute population statistics (completed cohort fertility, proportion of women by parity, and parity progression ratios) for six educational categories (postgraduate, graduate, bachelor, diploma, year 12, year 11 and below) and six cohort groups (1952-56, 1957-61, 1962-66, 1967-71, 1972-76, 197781). these statistics are used to describe trends in fertility patterns over time and by education. results completed cohort fertility has continued to decline slowly, from 2.22 for the1952-56 birth cohorts to 1.89 for the 1977-81 birth cohorts, mostly due to an increase in childlessness and a decrease in larger families with more than two children. there has been more divergence than convergence by education level, with those with bachelor's or diploma qualifications experiencing greater declines in fertility than any other education group. although childlessness rates by education level have converged, women with lower education have also become increasingly likely to have larger families. conclusions while there is evidence of convergence in childlessness rates across different levels of education, the gap in fertility rates between education groups continue to increase in australia. key words completed cohort fertility, fertility decline, education differentials. http://www.australianpopulationstudies.org/ mailto:edith.gray@anu.edu.au 2 gray & lazzari australian population studies 7 (1) 2023 1. introduction australia’s total fertility rate (tfr), a measure of fertility in a given year, fell to its lowest point of 1.59 in 2020 (abs 2021), followed by a small upturn to 1.70 in 2021 (abs 2022), and in 2022 returned to the downward trend observed prior to covid-19, with a tfr of 1.63 (2023). since the early 1960s (figure 1), the tfr has shown long periods of decline, followed by shorter periods of upturns. the last peak of 2.02 was recorded in 2008. although a widely used measure, the tfr has some limitations (bongaarts & feeney, 1998), as observed declines may be due to a change in the timing of having children (tempo effect) or an overall decline in the number of children born (quantum effect), or both. in australia, like other high-income countries, there has been an increasing trend of women delaying having children to older ages, which has a negative impact on the tfr (carmichael and mcdonald 2003). when those births are realised at later ages, this can appear as an increase in the tfr. figure 1: total fertility rate, australia, 1960-2021 source: abs (2022). there are alternative measures which are useful for observing fertility delay and overall fertility, especially those that examine cohort change in the number of children born (frejka and calot, 2001; wood et al. 2014). for example, completed cohort fertility (ccf) provides the distribution in the number of children born at the end of a woman’s reproductive lifespan. this information can be used to assess childbearing decline (or increase) over time. parity progression, a related method which uses the number of children born to women, assesses progression from one parity to the next, i.e., from no children to one child, from one child to two, etc. if assessed by age, a delay in childbearing can be observed between cohorts. jain and mcdonald (1997) used cohort measures to effectively show the change in fertility of australian women born in 1932-1936 through to 1948-1952. they observed a decline in completed fertility (from 3.14 to 2.27), and the increase in having smaller numbers of children, or no children, for each successful cohort. changes in parity have also been 1.5 2.0 2.5 3.0 3.5 1960 1965 1970 1975 1980 1985 1990 1995 2000 2005 2010 2015 2020 year t o ta l fe rt il it y r a te australian population studies 7 (1) 2023 gray & lazzari 3 discussed by carmichael and mcdonald (2003), who noted the noticeable decline in women with three, or four or more children. childbearing patterns are usually quite different for women with higher education in high-income countries, with fewer children born to women at higher education levels (d’addio and d’ercole, 2005). using a highly disaggregated measure of education by level, carmichael and mcdonald (2003) demonstrated the negative relationship between average children ever born and education: those with no, or the lowest levels of education had the highest average number of children, and those with the highest levels of education had the fewest. as explained by gray and evans “women with higher levels of education tend to start childbearing later, have fewer children overall and are more likely to be childless” (2019, p.2). higher education is associated with childbearing delay because of the time spent in education, and a prioritisation of careers and financial security following education (ni bhrolchain & beaujouan, 2012). this societal expectation for women to delay starting families until after completing education has become commonplace (blossfeld & huinink, 1991). as a result, there is a reduced window available for having children following education and obtaining workplace experience. as reproductive capacity declines with age, delaying parenthood can contribute to having a smaller family than planned (schmidt, et al. 2012). studies on women’s timing of childbearing in australia confirm that those with higher education start childbearing at later ages (miranti, et al. 2009; lazzari 2019; lazzari 2021a). further, education is also associated with completed family size. gray and evans (2019) showed that the widespread increase in educational attainment was strongly associated with a corresponding decrease in completed fertility. they found that completed fertility was a result of a general decline in fertility among women regardless of education, for those born between 1940 and 1965. but for women born in the late 1960s and 1970s, reductions in fertility were primarily a result of the increasing proportion of women with higher education, a finding supported by lazzari et al. (2021a). despite this difference by education, a fertility convergence between education levels has recently been observed in a number of high-income countries (adserà 2017). in the nordic countries of denmark, norway and sweden, childlessness, which was previously highest among highly-educated women (andersson, et al. 2009), is now higher among the least-educated (sobotka, et al. 2017; jalovaara, et al. 2019), and differences in the educational gradient in completed fertility have ‘vanished’, with higher socioeconomic status being increasingly linked to higher fertility (jalovaara et al. 2019; kolk 2023). the generally low differences by education in the nordic countries has typically been explained by the welfare policies that have assisted families to balance work and family (ronsen and skrede, 2010), but the recent increases in childlessness for the least-educated may be explained by inequality (jalovaara, et al. 2019). adserà (2017) explores why inequality may be associated with low fertility among the leasteducated, writing that periods of economic uncertainty in recent times are associated with a greater need for households to have two incomes. specifically, she noted that “across the oecd countries, fertility has decreased the most among those population groups who were hit hardest by the great recession: namely, young adults, people with low levels of education, and migrants” (adserà, 2017, p.71). a narrowing of educational differences in levels of childlessness over time have been also observed in the united states and germany (kreyenfeld and konietzka 2017; rybińska 2020). 4 gray & lazzari australian population studies 7 (1) 2023 it is also important to consider the number of children born by education (see jain and mcdonald 1997, wood et al. 2014 and brzozowska et al. 2022). in australia, lazzari (2021b) examined changes in fertility by education and found a convergence in childlessness by education and a widening in average completed fertility and in parity progression to three or more children. further, lazzari et al. (2021a) decomposed the overall change in completed cohort fertility into educational composition and education-specific fertility by the number of children for six countries. the research showed that for recent cohorts, changes to overall fertility was driven by a decline in higher-order births (three children or more) for women with the highest level of education, while there was less change by education for those with two children or fewer. in this paper we investigate the most recently available data to compare average completed fertility and number of children born by education of the mother. we compare cohorts born from 1952 through to 1981, allowing us to determine whether completed fertility has continued to decline, and examine changes in parity distributions by education level. 2. data and methods the study uses 2021 australian census microdata obtained from tablebuilder pro (abs 2021) on the number of children ever born, and the educational level, of six cohorts of women born between 1952 and 1981. the cohorts were aged from 40-44 (cohort born 1977-1981) up to 65-69 (cohort born 1952-1956). the census asks the number of children ever born of australians identifying as female who are aged 15 years and over. in this analysis, the information is used to calculate the following cohort measures of lifetime fertility: the completed average number of children ever born, the proportion of childless women, and the parity progression ratios to first and higher-order births. education is classified by the abs using the australian standard classification of education (asced), which is similar, but not perfectly aligned with the international standard classification of education (isced) (abs 2001). the standard has eight categories as follows: 1. year 10 or lower 2. vocational education and training: certificate iii and iv 3. year 11 4. year 12 5. vocational education and training: diploma 6. university: bachelor (3–4 years) 7. university: graduate (0.5–1 year) 8. university: postgraduate (1–5 years) there are benefits to using a more detailed measure of education when considering changes to fertility (gray and evans 2019; lazzari 2021b). for this reason, we use the education measure with a australian population studies 7 (1) 2023 gray & lazzari 5 small modification, combining categories 1 and 3 to one category of year 11 and below. this category represents a measure of incomplete high school with no other qualification obtained. figure 2 provides the percentage distribution of education categories by birth cohort. women’s education has seen dramatic change over the cohorts under observation. the largest change is evident for year 11 and below: in the first cohort of 1952-56, over 40% had incomplete high school education, which had reduced to 10% for the 1977-1981 cohort. the categories which increased the most over time are bachelor’s degree (15% to 31%) and postgraduate degree (5% to 12%). the following methods are applied to understand the overall decline, or otherwise, in fertility over the cohorts, followed by an analysis of fertility change by education. 1. comparison of average completed cohort fertility 2. average completed cohort fertility by education level 3. analysis of distribution of number of children born by education level 4. parity progression ratios by education level. figure 2: percentage distribution of completed education level, australia, cohorts born 1952 to 1981 source: authors’ calculations based on 2021 australian census data retrieved from tablebuilder pro. 6 gray & lazzari australian population studies 7 (1) 2023 3. results 3.1. comparison of completed cohort fertility as shown in figure 1, there has been a substantial decline in the period fertility measure of tfr, more than halving over the period from 1960 to 2021. however, when we examine completed fertility (figure 3), the decline appears to be much smaller, indicating that declines in fertility are gradual, although they clearly trend downwards. figure 3: average completed cohort fertility, cohorts born 1952 to 1981 source: authors’ calculations based on 2021 australian census data retrieved from tablebuilder pro. note: average completed fertility figures available in figure 5, total column. completed fertility can be separated to show the distribution in number of children born by cohort. this provides more information on whether the decline in how many children women have is due to an increase in childlessness, or a decrease in larger families, or both. figure 4 presents information on the number of children born by each cohort. in every cohort, the most common number of children is two, and this has barely changed over the generations. for the 1952-56 cohort, 39% had two children. it has hovered around that level for each generation, dropping slowly over time to just under 38% for the 1966-81 cohort (figure 4). however, over time, there has been a shift in the number of children women are having. specifically, the proportion of women who have no children or one child has been increasing, while the proportion of women who have three children or four or more children has been decreasing. these changes have been consistent across birth cohorts, with more women having two or fewer children and fewer women having more than two children as time goes on. the 1952-56 cohort had a completed fertility of 2.2 children per woman, the highest of all cohorts (figure 3). twelve per cent of women in this cohort had no children, and a further 12% had one child (figure 4). the cohort also had the largest percentages of three children (24%), or four or more children (14%). in comparison, the most recent cohort born in 1977-81 had a completed australian population studies 7 (1) 2023 gray & lazzari 7 fertility of 1.89 children per woman. those with no children increased to 18%, while those with one child increased to 16%. those with three (18%) or four or more children (9%) correspondingly declined. in sum, both increases in permanent childlessness and declines in having large families of three or more children have led to the decline in overall completed fertility by cohort. figure 4: percentage distribution of children ever born, cohorts born 1952 to 1981 source: authors’ calculations based on 2021 australian census data retrieved from tablebuilder pro. 3.2. completed cohort fertility by education level as discussed, there has been a noticeable negative education gradient in completed fertility evident in australia, where those with less education having higher fertility, and those with more education having lower fertility. but in other countries, this generally observed education gradient has disappeared, or is reducing. in recent years, these differences by education have mainly been driven by declines in fertility at lower education levels. the results from the completed cohort fertility by highest level of educational attainment in 2021 do not show any convincing evidence of a convergence in fertility by education in australia. in fact, we see small continuing declines for all education levels except for those with year 11 and below education, which declined for cohorts born 1962-66 and 1967-71, but has since rebounded for later cohorts (figure 5). in all cohorts, women with a post-graduate degree have the lowest completed fertility, and those with year 11 and below report the highest completed fertility. as a percentage change, the largest cohort fertility decline is 13.9% for those with a diploma, followed by a decline of 13.7% for those with a bachelor’s degree (see appendix table a1). 8 gray & lazzari australian population studies 7 (1) 2023 figure 5: average completed cohort fertility by highest level of educational attainment, australia, cohorts 1952 to 1981 source: authors’ calculations based on 2021 australian census data retrieved from tablebuilder pro. note: the cohort born 1977-81 completed fertility is likely to increase slightly in the 2026 round of census data collection as this cohort was aged 40-44 at the 2021 census, and hence their fertility is incomplete. 3.3. number of children born by education level as shown previously, over time there has been a shift in the number of children women are having, where the proportion of women who have no children or one child has increased, while the proportion of women who have three children, or four or more children, has declined. here we investigate these changes to see if there are differences by education level (table 1). overall, the percentage of women with no children has been increasing. this is true for all levels of education except for those with post-graduate degrees, but the growth in having no children is stronger for those without university qualifications, which were coming off a lower starting point. for those with a certificate we see an increase from 9% to 15%, similar to those with year 11 and below increasing from 9% to 14%. for those with year 12 education, having no children increased from 13% to 18%. the only education category that has seen a decline in childlessness is those with a post-graduate qualification, which has declined from 25% for the oldest cohort, to 22% for the youngest cohort. for the oldest cohort, there was a very large gap in the percentage of women who had no children when comparing those with a post-graduate degree and those with other 1.73 1.67 1.61 1.60 1.60 1.52 1.98 1.91 1.85 1.83 1.81 1.74 2.03 1.96 1.88 1.86 1.84 1.75 2.15 2.10 2.01 1.95 1.92 1.85 2.35 2.30 2.24 2.17 2.14 2.07 2.09 2.04 2.00 1.97 1.98 1.94 2.39 2.34 2.27 2.27 2.36 2.43 2.21 2.15 2.06 2.01 1.97 1.88 1.50 1.75 2.00 2.25 1952−1956 1957−1961 1962−1966 1967−1971 1972−1976 1977−1981 birth cohort c o m p le te d c o h o rt f e rt il it y education postgraduate graduate bachelor diploma certificate year 12 year 11 and below total australian population studies 7 (1) 2023 gray & lazzari 9 education levels; a gap which has declined over time. this can be seen as a convergence in having no children by education. table 1: percentage distribution of children ever born by education level, australia, cohorts 1952-56 to 1972-76 birth cohort postgraduate graduate bachelor diploma certificate year 12 year 11 and below total no children 1952-56 25 18 16 13 9 13 9 12 1957-61 25 20 17 14 10 14 10 14 1962-66 25 20 18 15 9 15 12 15 1967-71 24 19 17 16 12 16 13 16 1972-76 22 19 17 16 13 17 13 16 1977-81 22 20 19 18 15 18 14 18 one child 1952-56 14 12 14 13 11 17 10 12 1957-61 16 12 14 14 11 17 11 13 1962-66 17 14 15 14 12 16 12 14 1967-71 18 15 15 15 13 15 13 15 1972-76 19 15 15 16 14 15 14 15 1977-81 21 15 16 17 14 16 14 16 two children 1952-56 35 37 37 38 39 38 39 38 1957-61 35 37 38 38 38 38 38 38 1962-66 37 39 39 39 39 39 37 38 1967-71 38 41 41 40 39 39 36 39 1972-76 41 41 42 40 38 38 32 40 1977-81 42 41 43 39 38 36 29 39 three children 1952-56 18 23 23 24 26 21 27 24 1957-61 17 22 22 23 26 21 26 23 1962-66 16 20 20 21 24 20 24 22 1967-71 15 19 19 20 23 20 23 20 1972-76 14 18 19 19 22 19 22 20 1977-81 12 18 17 18 22 19 21 18 four or more children 1952-56 8 10 11 12 15 12 16 13 1957-61 7 9 9 11 15 11 16 13 1962-66 5 8 8 10 14 10 15 11 1967-71 5 7 7 9 13 10 16 10 1972-76 4 6 6 9 13 10 19 10 1977-81 3 5 5 8 12 11 22 9 source: authors’ calculations based on 2021 australian census data retrieved from tablebuilder pro. 10 gray & lazzari australian population studies 7 (1) 2023 having one child has also increased over time, but there is more variation by education level. those with a post-graduate qualification had the biggest increase (14% to 21%), but smaller increases are also evident for those with a graduate degree, diploma, certificate, and year 11 and below to varying degrees. those with a bachelor’s qualification showed a very small increase over time, and of those with year 12, the percentage having one child hardly changed. having two children has always been the most common response over the cohorts, indicating the continued acceptance of the two-child family norm in australia. in particular, there have been more women with university qualifications having two children over time. those having two children with a post-graduate degree increased from 35% to 42%, similar to those with graduate degrees (37% to 43%), and those with bachelor’s degrees (37% to 43%). the education category that saw a large decline in the percentage of women with two children was those with year 11 and below education, which declined from 39% to 29%. having three children has declined for all education categories over time, with the lowest percentage found among the youngest cohort of women with a postgraduate degree (12%), and the highest percentages found among those with a certificate qualification (22%), or year 11 and below (21%). for the highest parity of four or more children, most educational categories saw declines over time, except women with year 12, or year 11 or below. the most dramatic change is evident for women with year 11 or below, where the percentage with four or more children increased from 16% to 22% over the cohorts. this is evidence of growing divergence rather than convergence. 3.4. parity progression ratios by education level the results of the distributions of children ever born over time showed that there had been some convergence in having no children by education level, with the percentage not having children ranging between 14% and 22% for the youngest cohort. we now consider, of those women who have at least one child, what percentage go on to have higher-order births by cohort. this highlights the pattern of divergence in higher-order parities by education levels seen in the distribution of children ever born. in terms of progression to a second birth, most women who have one child progress to a second child, regardless of education level (figure 6), although we see the usual education gradient with women with higher education having slightly less progression to a second child. for most education categories, in almost all years, the percentage progressing to a second child is 80% or higher. notably, progression to a second child for those women with a postgraduate degree is between 72% and 80% for all but the oldest cohort. it is the parity progressions to third birth or fourth or higher birth where divergence by level of education is the greatest, and the difference is increasing over the cohorts rather than dissipating. overall, we see that while those with year 11 education and below are more likely to remain childless over time, those who do go on to have children are having larger families than in the past. australian population studies 7 (1) 2023 gray & lazzari 11 figure 6: progression to first, second, third, fourth or higher birth by education level (%), cohorts born 1952 to 1981 source: authors’ calculations based on 2021 australian census data retrieved from tablebuilder pro. 3. discussion assessing measures of fertility to understand fertility change can be problematic. the usual period measure of fertility, the total fertility rate, will appear as a decline due to quantum changes, as well as timing changes such as fertility delay (bongaarts & feeney, 1998). as an alternative, completed cohort fertility can be used to see if fertility has declined between birth cohorts, providing a measure of overall fertility regardless of when women have their children. the downside to using this measure is that one must wait for cohorts to reach the end of their childbearing years, usually around age 45 for women, although we note some babies are born to women above this age. in this paper, we assess whether completed fertility has declined for recent cohorts. we then also investigate whether there has been a convergence in fertility by level of education by examining average completed fertility as well as the distribution of children ever born by education. previous research found that women with lower levels of education had higher fertility and were less likely to be childless but, as discussed in the introduction, findings from a number of countries have shown a convergence in fertility as measured by average completed fertility and the percentage having no children (e.g. adsera 2017, jalovaara et al. 2019, kolk 2023). this paper extends this to also look at differences in the distribution of number of children born. ppr 3 ppr 4 ppr 1 ppr 2 1952−56 1957−61 1962−66 1967−71 1972−76 1977−81 1952−56 1957−61 1962−66 1967−71 1972−76 1977−81 0.2 0.4 0.6 0.8 0.2 0.4 0.6 0.8 birth cohort p a ri ty p ro g re s s io n r a ti o s ( p p r ) education postgraduate graduate bachelor diploma certificate year 12 year 11 and below total 12 gray & lazzari australian population studies 7 (1) 2023 we find that average completed cohort fertility continues to decline slowly in australia. for our earliest cohort, born 1952 to 1956, average completed fertility was 2.22. for our last cohort, born 1977 to 1981, average completed fertility was 1.89. we note that not all individuals in this cohort have reached 45 years of age, so fertility will likely increase very slightly. when examining the results by education level, we see that average completed cohort fertility remains lowest for those with the highest education, and there have been greater declines for those with bachelor’s or diploma qualifications. for those whose education was year 11 or below we see an increase, not a decrease in completed cohort fertility for recent cohorts. these results do not provide evidence of a convergence in fertility levels; in fact, they suggest the opposite. unpacking this, we investigate differences in the number of children ever born by education. here we see some very interesting results. first, we see a decline in the percentage with no children for those with postgraduate qualifications, and a slight increase in those with no children for women with a graduate or bachelor’s qualification. we see larger increases in having no children for all other education categories, leading to somewhat of a convergence in the percentage of childless women by education level. there are also increases in the percentage having one child for almost all education categories. but this is where any convergence ends. women with high levels of education see increases in the percentage having two children, but having two children declines for women with the least education, creating a large gap between the highest and lowest education groups. for women having the largest number of children, i.e., three, or four or more children, we see considerable increases for those with year 11 or below education. these results suggest that there is divergence by education level. 4. conclusions taken together, these results suggest that there is mixed evidence for convergence in the australian setting: the gap in average completed fertility is widening between women with the lowest and highest levels of education, yet the percentage of women with no children is narrowing. there is also a narrowing by education level for the percentages who have two children, but not at the higher parities, leading to an increase in the gap by education levels. why do we see these changes in number of children born, particularly for those with low levels of education? for women with year 11 or lower education, we see large changes in both directions: greater percentages of women with this level of education that have no children or one child, but those who have children are more likely to have larger families in recent cohorts. the second area of research is about understanding the characteristics of women by education level for the different cohorts. it may be that women with year 11 or below education are increasingly selective of women from particular backgrounds, and that this selection has narrowed over time, as this education category has reduced. with only around 10% of women having this level of education, we suggest that they hold characteristics which lead them to have larger numbers of children, if they become mothers. one explanation for differences between cohorts relates to changes in the proportion of women who have migrated to australia, and their country of origin. australia has a longstanding history of selective skilled immigration, leading to a notable influx of highly educated individuals. the countries australian population studies 7 (1) 2023 gray & lazzari 13 contributing immigrants have also shifted over time, leading to changes in the nation's overall demographic makeup. while the united kingdom historically dominated as a source of permanent migration to australia, this role has now been assumed by india and china (department of home affairs, 2023). this has led to changes in the relative share of births by country of origin (baffour et al. 2023). australian research highlights substantial diversity in fertility among immigrant groups (abbasishavazi and mcdonald 2000, baffour et al. 2023). research in europe has also revealed that these differences persist in subsequent generations due to ongoing influences from minority subcultures on fertility behaviours (kulu et al. 2019). hence, the changes that we observed in our analysis in cohort fertility based on education are likely influenced by shifts in the immigrant composition. despite the considerable and increasing proportion of immigrants in australia, there has been limited research focusing on the topic of migrant fertility, with scant knowledge about fertility trends in successive generations (see carmichael and mcdonald 2003). future research should address this research gap, given its importance to better understand the country's fertility patterns. the fertility patterns of women by education provide information to better understand overall changes in fertility levels. despite evidence indicating that only a minority of highly educated women express a definite preference for remaining childless (holton et al. 2011; beaujouan and berghammer 2019), delayed family formation among this group has historically resulted in a higher likelihood of experiencing infertility and remaining childless. however, this study reveals a recent decrease in childlessness rates among highly educated women in australia, potentially indicating that these women have recognised the issues associated with delayed childbearing, or they may have been able to achieve pregnancy through the use of assisted reproductive technology, which is increasingly being used in recent years (lazzari et al. 2021b). as educational attainment continues to expand, the fertility decisions of highly educated women will significantly influence future fertility rates (adserà 2017). therefore, it is crucial that future demographic research focuses on understanding the underlying factors driving these changing fertility patterns to comprehend future fertility trends. key messages • completed cohort fertility continues to slowly decline in australia, mostly due to an increase in childlessness and a decrease in large families with more than two children. • having two children remains the most common response over the cohorts, demonstrating a continued desire for the two-child family norm in australia. • a decline in childlessness is only observed in post-graduates, while all other levels of education, particularly those without university qualifications, have seen an increase in childlessness, indicating a growing convergence by education. • the trend of forming large families with more than two children has declined in all education categories, except for those with a year 11 or lower qualification, indicating a growing divergence by education. • these opposing trends provide mixed evidence of convergence of fertility by education in the australian setting. 14 gray & lazzari australian population studies 7 (1) 2023 references abbasi-shavazi, m. j., & mcdonald, p. (2000). fertility and multiculturalism: immigrant fertility in australia, 1977–1991. international migration review 34(1), 215–242. https://doi.org/10.1177/0197918300034001 adserà, a. (2017). education and fertility in the context of rising inequality. vienna yearbook of population research, 2017(15), 63-92. https://doi.org/10.1553/populationyearbook2017s019 andersson, g., rønsen, m., knudsen, l., lappegård, t., neyer, g., skrede, k., teschner, k., & vikat, a. (2009). cohort fertility patterns in the nordic countries, demographic research, 20(14), 313–352. https://doi.org/10.4054/demres.2009.20.14 australian bureau of statistics (2021). tablebuilder, abs, downloaded 31 march 2023, https://www.abs.gov.au/statistics/microdata-tablebuilder/tablebuilder australian bureau of statistics (2022). births, australia, abs website, accessed 8 november 2023, https://www.abs.gov.au/statistics/people/population/births-australia/latest-release australian bureau of statistics (2023). births, australia, abs website, accessed 9 november 2023, https://www.abs.gov.au/statistics/people/population/births-australia/latest-release australian bureau of statistics (abs). (2001). australian standard classification of education (asced), 2001. cat. no. 1272.0. canberra, australia. baffour, b., raymer, j. & evans, a. (2023). recent trends in immigrant fertility in australia. journal of international migration and integration, 24(1), 47-67. https://doi/org/10.1007/s12134-02000767-0 beaujouan, e., & berghammer, c. (2019). the gap between lifetime fertility intentions and completed fertility in europe and the united states: a cohort approach. population research and policy review, 38(4), 507-535. https://doi.org/10.1007/s11113-019-09516-3 blossfeld, h-p., & huinink, j. (1991). human capital investments or norms of role transition? how women’s schooling and career affect the process of family formation. american journal of sociology, 97(1), 143-168. https://doi.org/10.1086/229743 bongaarts, j., & feeney, g. (1998). on the quantum and tempo of fertility. population and development review, 24(2), 271-291. https://doi.org/10.2307/2807974 brzozowska, z., beaujouan, e., & zeman, k. (2022). is tow still best? change in parity-specific fertility across education in low-fertility countries. population research and policy review, 41(5), 20852114. https://doi.org/10.1007/s11113-022-09716-4 carmichael, g. & mcdonald, p. (2003). fertility trends and differentials. pp. 40-76 in khoo, s-e. & mcdonald, p. (eds) the transformation of australia’s population 1970-2030. sydney, unsw press. d’addio, a., & d’ercole, m.m. (2005). trends and determinants of fertility rates: the role of policies. oecd social, employment and migration working papers, no. 27, oecd publishing, paris. https://doi.org/10.1787/880242325663. department of home affairs (2023). migration program statistics. https://www.homeaffairs.gov.au/research-and-statistics/statistics/visa-statistics/live/migrationprogram frejka, t. & calot, g. (2001). cohort reproductive patterns in low-fertility countries. population and development review, 27, 103-132. https://doi.org/10.1111/j.1728-4457.2001.00103.x gray, e., & evans, a. (2019). changing education, changing fertility: a decomposition of completed fertility in australia. australian population studies, 3(2), 1–15. https://doi.org/10.37970/aps.v3i2.42 holton, s., fisher, j., & rowe, h. (2011). to have or not to have? australian women's childbearing desires, expectations and outcomes. journal of population research, 28(4), 353-379. https://doi.org/10.1007/s12546-011-9072-3 https://doi.org/10.1177/0197918300034001 https://doi.org/10.1553/populationyearbook2017s019 https://doi.org/10.4054/demres.2009.20.14 https://www.abs.gov.au/statistics/microdata-tablebuilder/tablebuilder https://doi/org/10.1007/s12134-020-00767-0 https://doi/org/10.1007/s12134-020-00767-0 https://doi.org/10.1007/s11113-019-09516-3 https://doi.org/10.1007/s11113-022-09716-4 https://doi.org/10.1787/880242325663 https://www.homeaffairs.gov.au/research-and-statistics/statistics/visa-statistics/live/migration-program https://www.homeaffairs.gov.au/research-and-statistics/statistics/visa-statistics/live/migration-program https://doi.org/10.1111/j.1728-4457.2001.00103.x https://doi.org/10.37970/aps.v3i2.42 https://doi.org/10.1007/s12546-011-9072-3 australian population studies 7 (1) 2023 gray & lazzari 15 jain, s., & mcdonald, p. (1997). fertility components of australian birth cohorts: components and differentials. journal of the australian population association, 14(1), 31–46. https://doi.org/10.1007/bf03029485 jalovaara, m., neyer, g., andersson, g., dahlberg, j., dommermuth, l., fallesen, p., & lappegård, t. (2019). education, gender, and cohort fertility in the nordic countries. european journal of population, 35(3), 563–586. https://doi.org/10.1007/s10680-018-9492-2 kolk, m. (2023). the relationship between life-course accumulated income and childbearing of swedish men and women born 1940–70. population studies, 77(2), 197-215. https://doi.org/10.1080/00324728.2022.2134578 kreyenfeld, m., & konietzka, d. (2017). childlessness in east and west germany: long-term trends and social disparities. in: kreyenfeld, m., konietzka, d. (eds) childlessness in europe: contexts, causes, and consequences. demographic research monographs. springer, cham. kulu, h., milewski, n., hannemann, t., and mikolai, j. (2019). a decade of life-course research on fertility of immigrants and their descendants in europe. demographic research, 40(46): 1345-1374. https://doi.org/10.4054/demres.2019.40.46 lazzari, e. (2019). socio-economic changes in the age-patterns of childbearing in australia. australian population studies, 3(2), 34-36. https://doi.org/10.37970/aps.v3i2.52 lazzari, e. (2021a). pathways into childbearing delay of men and women in australia. longitudinal and life course studies, 13(2), 307-334. https://doi.org/10.1332/175795921x16197735939121 lazzari, e. (2021b). changing trends between education, childlessness and completed fertility: a cohort analysis of australian women born in 1952–1971. journal of population research, 38, 417–441. https://doi.org/10.1007/s12546-021-09269-x lazzari, e., gray, e., & chambers, g. (2021b). the contribution of assisted reproductive technology to fertility rates and parity transition: an analysis of australian data, demographic research, 45(35), 1081-1096. https://doi.org/10.4054/demres.2021.45.35 lazzari, e., mogi, r. & canudas-romo, v. (2021a). educational composition and parity contribution to completed cohort fertility change in low-fertility settings, population studies, 75(2), 153-167. https://doi.org/10.1080/00324728.2021.1895291 miranti, r., mcnamara, j., tanton, r., & yap, m. (2009). a narrowing gap? trends in the childlessness of professional women in australia 1986–2006. journal of population research, 26(4), 359–379. https://doi.org/10.1007/s12546-009-9022-5 ní bhrolcháin, m., & beaujouan, e. (2012). fertility postponement is largely due to rising educational enrolment. population studies, 66(3), 311–327. https://doi.org/10.1080/00324728.2012.697569 rønsen, m., & skrede, k. (2010). can public policies sustain fertility in the nordic countries? lessons from the past and questions for the future, demographic research, 22(13), 321–346. https://doi.org/10.4054/demres.2010.22.13 rybińska a. (2020). a research note on the convergence of childlessness rates between women with secondary and tertiary education in the united states. european journal of population, 36(5), 827–839. https://doi.org/10.1007/s10680-019-09550-z schmidt l., sobotka t., bentzen j.g., & nyboe andersen a., on behalf of eshre reproduction and society task force. (2012). demographic and medical consequences of the postponement of parenthood. human reproduction update, 18(1), 29-43. https://doi.org/ 10.1093/humupd/dmr040. sobotka, t., beaujouan, e., & van bavel, j. (2017). introduction: education and fertility in low-fertility settings, vienna yearbook of population research, 15, 1–16. https://doi.org/10.1553/populationyearbook2017s001 wood, j., neels, k., & kil t. (2014). the educational gradient of childlessness and cohort parity progression in 14 low fertility countries. demographic research, 31(46), 1365-1416. https://doi.org/10.4054/demres.2014.31.46 https://doi.org/10.1080/00324728.2022.2134578 https://doi.org/10.4054/demres.2019.40.46 https://doi.org/10.37970/aps.v3i2.52 https://doi.org/10.1007/s12546-021-09269-x https://doi.org/10.4054/demres.2021.45.35 https://doi.org/10.1080/00324728.2021.1895291 http://dx.doi.org/10.1007/s12546-009-9022-5 https://doi.org/10.4054/demres.2010.22.13 https://doi.org/10.1007/s10680-019-09550-z https://doi.org/10.1553/populationyearbook2017s001 16 gray & lazzari australian population studies 7 (1) 2023 appendix table a1: percentage change in cohort fertility between cohorts by education level, 1952-56 to 1977-81. postgraduate graduate bachelor diploma certificate year 12 year 11 and below total cohort 1952-56 1.73 1.98 2.03 2.15 2.35 2.09 2.39 2.21 cohort 1977-81 1.52 1.74 1.75 1.85 2.07 1.94 2.43 1.88 percentage change -12.0% -12.2% -13.7% -13.9% -11.8% -7.15% 1.58% -15.3% source: authors’ calculations based on 2021 australian census data retrieved from tablebuilder pro. australian population studies 2021 | volume 5 | issue 2 | pages 1-14 © williamson et al. 2021. published under the creative commons attribution-noncommercial licence 3.0 australia (cc by-nc 3.0 au). journal website: www.australianpopulationstudies.org commentary indigenous peoples and the australian census: value, trust, and participation bhiamie williamson* (euahlayi) the australian national university jacob prehn (worimi) university of tasmania maggie walter (palawa) university of tasmania raymond lovett (wongaibon, ngiyampaa) the australian national university; australian institute of aboriginal and torres strait islander studies gawaian bodkin-andrews (d’harawal) university of technology, sydney bobby maher (yamatji) the australian national university vanessa lee-ah mat (yupungathi, meriam) university of sydney roxanne jones (palawa) the australian national university * corresponding author. email: bhiamie.williamson@anu.edu.au. address: centre for aboriginal economic policy research (caepr), copland building #24, the australian national university, acton, act, 2601, australia. paper received 21 february 2021; accepted 14 september 2021; published 29 november 2021 1. introduction reliable population data are foundational to the functioning of nations. population data are important for making and informing decisions about national policy parameters and supporting resource allocation (lovett 2016). the australian five yearly census of population and housing (the census) is critical in the production of this information and as such, has become a feature in australia’s social and political landscape. the census also provides the data for a national narrative including who or what australia is and how it is changing over time (through categorisation and data compilation). as māori scholar kukutai (2011 p. 33) states, the aotearoa (new zealand) census, similar to the australian census, is “an influential site of inclusion and exclusion where the state selectively acknowledges collective identities”. yet in australia, the census has a complicated and difficult history in relation to aboriginal and torres strait islander (henceforth referred to as indigenous) peoples. because while the australian bureau of statistics (abs) asserts that the purpose of the census is “to measure the number and key characteristics of persons and dwellings in australia” (abs 2017), for most of its history the tabulated number of persons did not include indigenous individuals1 (markham & biddle 2018). similarly, while the motives for the census are stated as “for electoral purposes; and to support activities of governments, business, and other users” indigenous peoples have rarely, if ever, been considered as being among legitimate ‘users’ of census data (walter, 2016). this complicated history and neglect of the unique data needs of indigenous peoples has resulted in a sustained undercount of indigenous peoples in the census (andrews 2018; markham & biddle 2018; taylor & biddle 2010). it is 1 we note that previous to 1971, censuses did count indigenous individuals but for the purposes of removing them from official estimates. http://www.australianpopulationstudies.org/ mailto:bhiamie.williamson@anu.edu.au 2 williamson et al. australian population studies 5 (2) 2021 time that these issues are named, and paths forward explored; a necessary step to better reflect the needs and ambitions of indigenous peoples, while simultaneously addressing the entrenched issue of indigenous participation, or lack thereof, in the census. in this commentary we, as a group of indigenous academics, trace the relationship between the australian census and indigenous peoples, addressing historical and present dimensions through the dual lens of value and trust. although we come from various disciplines including statistics, public health, social science, and human geography, our approach and analysis is firmly rooted in, and builds upon, postcolonial demography that kukutai & taylor (2012 p. 14) state, “highlights the epistemological and methodological shortcomings of applied demographic research on indigenous peoples, and generates calls for more innovative approaches.” firstly, we examine engagements between indigenous peoples and the census. we then discuss the central issue of the value, or lack of value, which census data provides for indigenous communities, and nations, before highlighting the lack of trust indigenous peoples have in government processes such as the census. we argue that the current indigenous ‘origin’ question should be redesigned and draw on examples from other english-speaking settler-colonial nations to support alternates. we finish by drawing attention to indigenous data sovereignty and indigenous data governance and propose that redesigning the census question is equally, if not more so, a question of process and structure as well as the wording of specific census questions. 2. indigenous peoples and the census when the first australian census was conducted in 1911, those indigenous people who were classified as ‘natives’ were deliberately and officially absented from the population count (kukutai & walter 2015; rowse & smith 2010; yu 2012) with section 127 of the australian constitution stating: ‘in reckoning the numbers of the people of the commonwealth, of a state or other part of the commonwealth, aboriginal natives shall not be counted’ (united kingdom house of commons 1900). the lack of debate at constitutional conventions surrounding this exclusion suggests a shared understanding at the time that indigenous peoples were a ‘dying race’ and that they would have no role or place in the future of australia (chesterman & galligan 1997). section 127 of the australian constitution was repealed by the 1967 referendum and in 1971, the ‘race’ question was modified to ask ‘what is the person’s racial origin?’, with tick boxes provided for ‘european origin’, ‘aboriginal origin’ or ‘torres strait islander origin’ (abs 1971). but the question of indigenous identity has evolved and changed since. in 1978, national cabinet (abs 2014) adopted the ‘three-part’ definition of indigenous identity. this was then formalised in the 1980s through the aboriginal affairs’ report on a review of the administration of the working definition of aboriginal and torres strait islanders that stated: an aboriginal or torres strait islander is a person of aboriginal or torres strait islander descent who identifies as an aboriginal or torres strait islander and is accepted as such by the community in which he or she lives (department of parliamentary library 2003). australian population studies 5 (2) 2021 williamson et al. 3 the three-part definition then continued to be adopted across commonwealth, state, and territory government departments as a mechanism for eligibility and access to targeted services for aboriginal and torres strait islander peoples (department of parliamentary library 2003). in subsequent censuses the wording of this question has varied slightly until 1996 when it was replaced with the current question: are you of aboriginal or torres strait islander origin? options respondents are provided with include: (i) no; (ii) yes, aboriginal; and (iii) yes, torres strait islander, with respondents prompted so that persons of both aboriginal and torres strait islander origin, [should] mark both ‘yes’ boxes2 (markham & biddle 2018). the abs states that the: standard indigenous question is based upon the commonwealth working definition but does not include the third element of the commonwealth definition, namely that ‘an aboriginal or torres strait islander is a person who is accepted as such by the community in which he or she lives’ (abs 2014). despite the development and widespread adoption of the three-part definition that demonstrates origin is but one part of a more complex identity, the indigenous identifier question in the census continues to clearly be framed solely in terms of ‘descent’ or ‘origin’. and whilst the abs explicitly acknowledges that the indigenous identifier does not include the third element of community affiliation, we argue that it also does not include the second element, namely that for a person to be indigenous they must identify as such, providing justification later in this paper. inclusion in the national census has also brought with it problems relating to the enumeration of indigenous populations. key among these is the significant and consistent indigenous undercount (andrews 2018; markham & biddle 2018; taylor & biddle 2010). ‘undercount’ refers to both the incorrect compiling of data, and/or the non-completion of the indigenous identifier question, to suggest fewer indigenous people than there actually are. that the census fails to accurately capture indigenous populations compromises the “purpose of the census and what it aims to do” (andrews 2018 p. 45). the reasons given for this consistent undercount vary, but can include historical exclusions (e.g., racism), lack of trust, or relevance of questions asked. to be clear, we do not wish to diminish the efforts to enumerate indigenous populations, especially where indigenous peoples themselves have been engaged in the processes to design the indigenous identifier question. rather, we suggest that an evolution of the indigenous identifier question would create greater value than is currently being generated, and thus would create the conditions for more substantial engagement. but creating more value in data through a redesigned census question may be insufficient in-and-ofitself. there remain serious issues of accessibility (understanding what the census is, and how and when to complete) and availability (some indigenous communities are difficult to reach due to remoteness, varying levels of literacy, and lack of access to services) (jelfs 2016; markham & biddle 2018; walter 2010). the abs, as the agency tasked with conducting the census, aimed to address accessibility and availability issues using methods such as: specific targeting of indigenous peoples; 2 the 2016 census questionnaire can be viewed in full at https://www.abs.gov.au/ausstats/abs@.nsf/lookup/2901.0main%20features802016/$file/2016%20census% 20sample%20household%20form.pdf https://www.abs.gov.au/ausstats/abs@.nsf/lookup/2901.0main%20features802016/$file/2016%20census%20sample%20household%20form.pdf https://www.abs.gov.au/ausstats/abs@.nsf/lookup/2901.0main%20features802016/$file/2016%20census%20sample%20household%20form.pdf 4 williamson et al. australian population studies 5 (2) 2021 the employment of indigenous engagement officers; training (including cultural safe practice) and employment of interviewers in remote indigenous communities; and the establishment of an indigenous statistics centre (jelfs 2016), although much more can and should be done to appropriately engage indigenous peoples (brackertz 2007). however, the more complex issue of deliberate non-participation by indigenous peoples remains unresolved. the exact number of indigenous peoples choosing not to participate in the census is almost impossible to measure. on the basis of the abs post-enumeration survey, where about 50,000 households are approached several weeks after the census in a face-to-face setting, the abs estimates an indigenous undercount of at least 10% since 2001 (markham & biddle 2018) and up to 17.2% in 2011 and 17.5% in 2016 (abs 2017). however, official estimates of the indigenous undercount are likely to be conservative, as they assume that people correctly identify their indigenous ancestry to an abs indigenous or non-indigenous interviewer. this non-participation is largely driven by a perception from indigenous peoples that the census is of little relevance and value (andrews 2018; yap & yu 2016). the emergence of such perceptions is easy to understand. whilst amending the deliberate exclusion of indigenous people from the census corrected a particular historical wrong, it is currently little more than a tick-a-box exercise that does little to recognise and progress the complex identities, voices, needs, and aspirations of indigenous peoples and communities (andrews 2018; crooks et al. 2019; kukutai & walter 2015; taylor 2009; yap & yu 2016). as altman and taylor (1996 p. 193) state, the current census question and the data derived from it are merely the “by-product of including an indigenous identifier”. kukutai and taylor go further, pointing out that: a significant irony is emerging that, at a time when we see an unprecedented volume and range of data on something called the “indigenous population”… there remains a dearth of information on the various socio/cultural entities that make up those populations (kukutai and taylor 2012 p. 18). the census, as a key source of indigenous data without an aligned purpose of meeting indigenous needs, has two main problems; what we call the indigenous data paradox. first, as argued by walter (2016 p. 80), current indigenous data production and analysis tends to be focused around the ‘five ds’ of data. these are: disparity; deprivation; disadvantage; dysfunction; and difference. always pejoratively compared to the non-indigenous population, data contributes to a persistent national data narrative that sees indigenous peoples as the problem. the other side of the paradox is the lack of data value to indigenous ‘users’, that is indigenous peoples, populations, communities, and first nations. as stated by yu (2012), current census output does not support the requirements of indigenous collectives to plan for their people’s current or future needs. the long history of surveillance of, and discrimination against, indigenous peoples by australian governments complicates this even further. what is clear is that unlike mainstream australia whose participation in the census is seen as an extension of their civic duty, not unlike voting, the census is not apolitical. rather, for indigenous peoples, choosing to participate or not is a profoundly political act (andrews 2018). australian population studies 5 (2) 2021 williamson et al. 5 on value and on trust globally, colonised indigenous peoples still largely retain and organise around their own social, cultural, and political groups (alfred & corntassel 2005; anderson et al. 2016; hemming et al. 2019). these groups exist with their own governance systems, legal arrangements, citizenship requirements, and land and water management practices (alfred & corntassel 2005; hemming et al. 2019). prior to colonisation, these distinct indigenous peoples had well-developed processes for recording, producing, controlling, and interpreting their own data3 (cbc radio 2017; pool 2016). these data were a strategic asset that groups could use to make informed decisions and plan for the future of their peoples and management of their lands and waters. today, the vastly different, unique, and complex societies of first peoples who have lived on the australian continent are collectively known as ‘indigenous’ and other allied terms (for example: aboriginal, torres strait islander, and first nations) (peters & mika 2017). although these descriptors are deployed to refer to the first peoples of the australian continent, none of these terms are themselves indigenous (peters & mika 2017). rather, they are constructs of western thought that has aimed to name and define different peoples throughout the ‘new world’ in relation to european colonisers (peters & mika 2017; veracini 2011; wolfe 1999). notwithstanding that there may be instances where indigenous peoples choose to participate in this western discourse, such as adopting the term ‘indigenous’ in the united nations to pursue common goals4, many indigenous polities continue to identify and organise around their own naming conventions (peters & mika 2017; smith 2013; sutton 2003). the now centuries-long suppression of indigenous languages has caused degradation, fragmentation, and in some areas, almost complete eradication of indigenous peoples’ languages (attwood & markus 1999). the suppression of indigenous ways of being, knowing, and doing further degrades indigenous identity and belonging (alfred & corntassel 2005; martin & mirraboopa 2003; wolfe 1999). yet despite these deeply colonial practices, indigenous peoples’ ways of being, knowing, and doing including languages and naming conventions, have both survived and continue to be revived (alfred & corntassel 2005). one of the first observable naming conventions shared by indigenous peoples throughout the world is the naming of their group(s) or nation(s) (peters & mika 2017). identifying with specific language groups or nations is emblematic of indigenous ways of being, knowing, and doing (alfred & corntassel 2005; martin & mirraboopa 2003) but there are added elements of this worldview that are of importance when considering indigenous-specific census questions. the current census question raises concerns because, as we outlined earlier, of its continued emphasis on origin rather than identification (crooks et al. 2019). yes, knowing the total number of people whose origin is aboriginal and/or torres strait islander is important, but it is incomplete in its lack of reference to indigenous communities, nations, and language groups. this framing of being indigenous, as simply being a descendent of an indigenous person or peoples, also does not accord 3 data refers to information or knowledge, in any format or medium, which is about and may affect indigenous peoples both collectively and individually. see maiam nayri wingara & australian indigenous governance institute (2018). 4 see for example the united nations permanent forum on indigenous issues, and the united nations declaration on the rights of indigenous peoples (undrip) (united nations 2008). 6 williamson et al. australian population studies 5 (2) 2021 with citizenship and membership conventions of indigenous communities (crooks et al. 2019). put simply, being of origin in-and-of-itself, whilst necessary, is insufficient (crooks et al. 2019). and whilst the abs suggests that the current indigenous identifier question does account for identification, positing a question around the concept of origin is clearly a matter of fact, not of choice. rather, a more accurate question would reflect the agency of individuals to identify as an indigenous person. crooks et al. (2019) considered the census question with its emphasis on origin against a question on flutracking, a national system for tracking influenza-like illness, which emphasised identity. crooks et al. (2019 p. 2) highlighted that “the origin question was a materially different question to that of identity” and found that “aboriginal stakeholders preferred a question that omitted the concept of origin, and focussed on cultural identity; ‘are you an aboriginal or torres strait islander person?’” in this way we agree with the conclusions drawn by crooks et al., and argue that amending the census question from one focusing on origin to identity would be more reflective of the realities and ambitions of indigenous peoples today (crooks et al. 2019). a focus on identity would also more closely align the census with the ‘three-part definition’ of indigeneity5, which requires ancestry, identification and community recognition6. indeed, for many indigenous peoples, citizenship of their language group or nation are more culturally meaningful than being ‘australian’. the recent high court decision in love v commonwealth of australia (2020) found that an indigenous person of australia who is known and accepted by his, her, or their indigenous group, cannot be considered ‘alien’ under the constitution. this high court decision instructs us that for indigenous peoples, issues of citizenship and group membership carry potentially far-reaching legal consequences. the current census question regarding indigenous origin erases these identities under the colonial category of ‘indigeneity’. a realignment of the census question is needed if the census is to accurately reflect the extant social organisation and observe indigenous peoples’ right to freely determine their own identities (anderson 2008; kukutai & cormack 2020; yap & yu 2016). there also remain serious issues of trust between indigenous peoples and australian governments (andrews 2018). this lack of trust stems from the long history of oppressive settler-indigenous relations. woiwurrung and yorta yorta woman and academic dr julie andrews (2018) investigated census participation in naarm (melbourne) through a qualitative study of twenty indigenous people. she found that the census triggered traumatic encounters between participants (including family members) and governments. one participant commented that some aboriginal people who are in public housing do not participate in the census for fear of being evicted for having too many people in their houses: a lot of our people are very frightened of telling the census how many people lived in their house because if they were in public housing you’re only allowed to have four people and they might have eight because relatives will come and stay, so you can lose your house so they’ll lie to the census (andrews 2018 p. 45). 5 for a critical discussion of the three-part definition, see whittaker (2017). 6 the three-part definition has increasingly come into use through regulation, policy and statutes and continues to be used by governments, organisations, educational institutions, and local indigenous communities, see whittaker (2017). australian population studies 5 (2) 2021 williamson et al. 7 there were other notable encounters that participants highlighted when discussing their nonparticipation. key among these was the forcible removal of indigenous children from their families and communities. as andrews (2018 p. 49) states: several people interviewed shared a common history over the past 40 years – being forcibly removed from their families when they were children…. many of these people appear to have pledged never to participate in any government information-gathering process. whatever the specific reason for non-participation in the census, what is clear is that many indigenous people are suspicious of the census, with many viewing it as another government surveillance tool (andrews 2018; walter 2016). the lack of government engagement with indigenous peoples about the census would indicate that these suspicions will not abate quickly (andrews 2018). it is clear that new arrangements are needed. it is our position that these new arrangements embed indigenous data sovereignty and indigenous data governance to ensure they respond to the trust deficit engendered by governments’ historical and on-going use of data not aligned with the interests of indigenous peoples. 3. time for an indigenous identity census question(s)? we believe that addressing the twin deficits of value and trust may begin by reformulation of the census questions relating to indigeneity. reference to similar questions in other english-speaking settler-colonial nations provides an indication of the shortcomings of the current australian census question. the current australian indigenous identifier question is simplistic when compared to the question sets that apply in nations such as the united states (us) and aotearoa (new zealand). for example, the us 2020 asked a series of questions beginning with ‘what is the person’s race?’ (us census bureau 2020). responses include: white; black or african american; american indian or alaska native; chinese; filipino; asian indian; vietnamese; korean; japanese; other asian; native hawaiian; samoan; chamorro; other pacific islander; or some other race (us census bureau 2020). this is followed by the subsequent question: ‘if american indian and alaska native, print the name of the enrolled or principal tribe’ (us census bureau 2020). the response space is left blank to be filled in by the respondent. similarly, in aotearoa (new zealand), the following series of questions is asked in relation to the person’s ethnicity. first, they are asked ‘to what ethnic group does the person belong to?’ responses are: new zealand european; maori; samoan; cook island maori; tongan; niuean chinese; indian; or other (statistics new zealand 2018). importantly, respondents can tick as many ethnic groups as they choose. there is then a specific question about descent, which asks: ‘are you descended from maori (that is, did you have a maori birth parent, grandparent or great-grandparent etc)?’ responses are: yes; no; and don’t know (statistics new zealand 2018). respondents are then directed to a further two questions on iwi affiliation. the first question asks: ‘do you know the name(s) of your iwi (tribe or tribes)?’; responses here are: yes; or no. respondents are then asked to enter the name(s) and regions of up to five iwi (statistics new zealand 2018). these examples clearly dismiss abs assertion’s that ‘collecting information on the basis of community acceptance is often impractical in a survey or administrative data collection setting and can lead to inaccuracies’ (abs 2014). 8 williamson et al. australian population studies 5 (2) 2021 we ask that governments, agencies, and organisations consider what could be gained for indigenous peoples, our communities, first nations, and australia as a whole, if the focus was switched from origin to identity, asking perhaps: do you identify as an aboriginal and/or a torres strait islander person? critically, this question would need to be accompanied by further identity related questions. these could be: do you identify with an aboriginal or torres strait islander first nation?; and/or; do you identify with an aboriginal or torres strait islander language group? respondents must then be given the opportunity to list their affiliations should they answer yes. such data would allow, for the first time, the beginnings of systematic quantification of the size, composition, and other features of first nations. as per the overall purpose of the census, such population data would, in the hands of first nations, indigenous organisations, and communities, support systems of governance to assess population wellbeing, set appropriate policy parameters, and support effective resource allocation for first nations peoples and communities. we acknowledge that such an evolution of the indigenous identifier question would be complex, but indigenous identity is complex. indeed, the question of identity in settler-colonial nations has been the subject of significant examination by indigenous academics such as carlson (2017), heiss (2018), and internationally by garroutte (2003) to name but a few. the reductive framing of the current indigenous identifier question in the census fails to engage with these complex realities, or the inherent rights of indigenous peoples to freely determine their own identities and express these identities in national forums such as the census (anderson 2008). many nations with identifiable indigenous populations recognise this, with a clear trend emerging of the declining use of ‘racial’, and an increase in ‘ethnicity’, classifications (mullane-ronaki 2017). furthermore, data derived from first nations, language, or community affiliations would have wide utility for indigenous peoples and organisations such as native title prescribed bodies corporate (pbcs) and community-controlled health organisations, among many others (yap & yu 2016). as taylor states: the challenge for social science and for public debate is to move beyond conventional classification and recognise more ethnographically informed geographies of indigenous peoples that are reflective of the demand for customised statistics now being expressed by representative groups (taylor 2011 p. 13)7. indeed, some first nations organisations have gone to extraordinary lengths to produce such data themselves. the yawuru indigenous population survey of broome in 2011, for example, sought to enumerate all indigenous residents of broome and establish their language group affiliation to enable for better community planning (taylor et al. 2014). the yawuru native title holders aboriginal corporation, through its company nyamba buru yawuru (nby), undertook this survey to 7 taylor (2012 p. 13) goes on to name groups that have expressed this view including anindilyakwa land council, thamarrurr incorporated, fitzroy futures forum, nyamba buru yawuru ltd, mirriuwung-gajerong corporation and the murray lower darling rivers indigenous nations. australian population studies 5 (2) 2021 williamson et al. 9 ‘help them to “know” their own community and to engage in strategic decision-making in a postnative title determination environment’ (taylor et al. 2014 p. 14; yap & yu 2016). however, such a survey required a significant investment of resources from nby and the commonwealth government and its expense and technical difficulty has meant that this approach has rarely been replicated elsewhere. there is therefore, a potential role for the census in collecting these data, which would significantly increase the value of the census for indigenous peoples. but there is also potential for harm to be caused by asking specifically about group affiliations. the legacies of colonisation and in particular, the forced removal of indigenous children from their families and communities, has meant that many indigenous people have been disconnected from their country, language, and first nation group. asking about group affiliation when it is not known carries risk of re-exposing people who live with the trauma of separation. conversely, it may also be a useful proxy in identifying settler-colonial process impacting on identity such as knowing how many people live with the trauma of separation (carlson 2016). with such complex and interconnected issues, the real question here is whether indigenous peoples can trust australian governments. issues of trust apply to both the data derived from asking questions about group affiliation, as well as the processes for stewarding and safeguarding such data. there is, which has been documented here and elsewhere (see lovett 2016; walter 2016; yap & yu 2016), a long history of data being weaponised against indigenous peoples as an instrument of maintaining marginalisation and pejorative stereotypes (kukutai & walter 2015; yap & yu 2016). data that provides information on indigenous identity and affiliations could also be used in this way. the continuing lack of indigenous peoples in leadership within the abs heightens the potential for misuse, whether intentional or not. building trust is therefore not just a matter of the abs declaring good intentions, of meeting with indigenous advisory groups or holding community consultations. rather, achieving an actual change in the relationship requires a paradigm shift on how indigenous data are conceptualised. such a shift requires a re-ordering of power in the data ecosystem, prioritising for the first time, the data interests of indigenous peoples across multiple levels and across the varied processes and practices of australian governments. most critically, this shift must be driven by indigenous peoples for indigenous peoples, developing and enriching reciprocal relationships between indigenous peoples and governments to produce systems of data stewardship built on indigenous values and priorities and done in such a way that instils trust (walter et al. 2020). these lessons are not only limited to the specific process that is the census, but it also applies to all government agencies that collect administrative data about indigenous peoples including medicare, centrelink, health, taxation, and education. indigenous data sovereignty and its activation mechanism, indigenous data governance, can provide the framework of such an indigenous data paradigm shift. as now documented extensively, indigenous data sovereignty is a global movement that can be broadly understood as the ability for indigenous peoples to own, control, access, and possess their data (first nations information governance centre 2014; walter 2018). it is a movement which has developed in response to the exclusion of indigenous peoples from data production, analysis, and participation in the wider data ecosystem (kukutai & taylor 2016; walter et al. 2020). in practice, indigenous data governance translates to indigenous oversight and decision-making in relation to indigenous data production. 10 williamson et al. australian population studies 5 (2) 2021 while this might require a paradigm shift to contextualise understanding by government agencies and non-indigenous peoples more generally, many of the technical tools to accomplish indigenous data governance already exist. to return to the example of language group or first nations affiliations in the census, the answers to such questions could be quarantined securely by the abs, and only released to third-parties, including other government agencies, under strict conditions. these conditions might include appropriate data access agreements signed with duly authorised representatives of that language group or first nation. while this approach to indigenous data governance might be novel in terms of indigenous-settler relations, it is already routinely adopted by the abs with respect to other sensitive data collections such as the multi-agency data integration project (madip). inspiration for such an evolution can also be drawn from aotearoa (new zealand) where statistics new zealand (2021) have a formal agreement with the data iwi leaders group, the mana ōrite relationship agreement, that seeks to ‘realise the potential of data to make a sustainable, positive difference to outcomes for iwi, hapū and whānau’. if indigenous data governance processes are in place, such as those in the mana ōrite relationship agreement, then changing the indigenous question in the census would lead to the creation of data that has greater value for indigenous peoples. specifically, in planning for their own community needs and making representations to governments and others such as ngos and private sector interests. to be clear, we do not suggest that amending the census question in the way we propose, can or will address the decades of harm caused by five d data and government statistical surveillance, we merely argue that it is an important and necessary step to create an official statistics system that meets the needs and desires of indigenous peoples. indigenous data governance would then enable trust building to be established in the census process by those to whom such data relates viz. indigenous peoples, our families, communities, organisations, and first nations (lovett et al. 2020). 4. conclusion the ability for indigenous peoples to identify in their own ways builds power and legitimises (in a western legal sense) people’s identities (carlson 2016). it affirms indigenous ways of being, knowing, and doing, and engages with the lived realities of indigenous peoples today, thus creating value (crooks et al. 2019). it also provides critical information, not quantified since colonisation, of group membership, demography, and mobility. however, simply changing the question may in-and-of-itself be insufficient to increase indigenous participation in the census. there remain serious issues of trust between indigenous peoples and the australian nation state (andrews 2018). indigenous data sovereignty enacted through indigenous data governance provides the intellectual thought and practical blueprint to move beyond these historical legacies and promote a more equitable and just future for our communities. in this way, indigenous peoples including our language groups, communities, and first nations can be empowered through the provision of data about our own peoples. such data would provide a basis to build more prosperous futures on our own terms through enhanced decision making and it would establish the foundation for a legitimate and valued place in the future of australia. australian population studies 5 (2) 2021 williamson et al. 11 acknowledgments the authors would like to acknowledge and thank dr francis markham and professor john taylor for their thoughtful contributions and editorial suggestions in the development of this paper. references abs (1971) census of population and housing, 1971. https://www.abs.gov.au/ausstats/abs@.nsf/lookup/2105.0main+features11971?opendocum ent abs (2014) indigenous status standard. https://www.abs.gov.au/statistics/standards/indigenous-statusstandard/latest-release abs (2017) census of population and housing: details of overcount and undercount, australia, 2016. https://www.abs.gov.au/ausstats/abs@.nsf/lookup/by%20subject/2940.0~2016~main%20featu res~indigenous%20status~16 abs (2017) census of population and housing: understanding the census and census data, australia, 2016. https://www.abs.gov.au/ausstats/abs@.nsf/lookup/2900.0main+features42016 abs (2020) aboriginal and torres strait islander peoples. https://www.abs.gov.au/statistics/people/aboriginal-and-torres-strait-islander-peoples alfred t & corntassel j (2005) being indigenous: resurgences against contemporary colonialism. government and opposition 40(4): 597-614. altman j c & taylor j (eds.) (1996) the 1994 national aboriginal and torres strait islander survey: findings and future prospects. canberra: centre for aboriginal economic policy research, australian national university. https://caepr.cass.anu.edu.au/sites/default/files/docs/caepr_mono11_2.pdf andersen c (2008) from nation to population: the racialisation of ‘métis’ in the canadian census. nations and nationalism 14(2): 347-368. anderson i, robson b, connolly m, al-yaman f, bjertness e, king a, tynan m, madden r, bang a, coimbra c e a, et al. (2016). indigenous and tribal peoples’ health (the lancet-lowitja institute global collaboration): a population study. the lancet, 388(10040): 131-157. https://doi.org/10.1016/s0140-6736(16)00345-7 andrews j (2018) blaks and stats in aboriginal victoria: census resistance and participation. australian aboriginal studies, (1), 43–56. attwood b & markus a (1999) the struggle for aboriginal rights. sydney: allen & unwin. brackertz n (2007) who is hard to reach and why? isr working paper, swinburne institute for social research. https://library.bsl.org.au/jspui/bitstream/1/875/1/whois_htr.pdf carlson b (2016) the politics of identity. canberra: aboriginal studies press. cbc radio (2017) harvard undergrad cracks code of knotted inca rope used as “an ancient excel spreadsheet.” https://www.cbc.ca/radio/asithappens/as-it-happens-thursday-edition1.4466385/harvard-undergrad-cracks-code-of-knotted-inca-rope-used-as-an-ancient-excelspreadsheet-1.4466388 chesterman j & galligan b (1997) citizens without rights: aborigines and australian citizenship. cambridge: cambridge university press. crooks k, carlson s & dalton c (2019) defining, controlling and analysing indigenous data: commitment to historical consistency or commitment to australian aboriginal and torres strait islander peoples? public health research & practice 29(4): e2941926. https://doi.org/10.17061/phrp2941926 https://www.abs.gov.au/ausstats/abs@.nsf/lookup/2105.0main+features11971?opendocument https://www.abs.gov.au/ausstats/abs@.nsf/lookup/2105.0main+features11971?opendocument https://www.abs.gov.au/statistics/standards/indigenous-status-standard/latest-release https://www.abs.gov.au/statistics/standards/indigenous-status-standard/latest-release https://www.abs.gov.au/ausstats/abs@.nsf/lookup/by%20subject/2940.0~2016~main%20features~indigenous%20status~16 https://www.abs.gov.au/ausstats/abs@.nsf/lookup/by%20subject/2940.0~2016~main%20features~indigenous%20status~16 https://www.abs.gov.au/ausstats/abs@.nsf/lookup/2900.0main+features42016 https://www.abs.gov.au/statistics/people/aboriginal-and-torres-strait-islander-peoples https://caepr.cass.anu.edu.au/sites/default/files/docs/caepr_mono11_2.pdf https://doi.org/10.1016/s0140-6736(16)00345-7 https://library.bsl.org.au/jspui/bitstream/1/875/1/whois_htr.pdf https://www.cbc.ca/radio/asithappens/as-it-happens-thursday-edition-1.4466385/harvard-undergrad-cracks-code-of-knotted-inca-rope-used-as-an-ancient-excel-spreadsheet-1.4466388 https://www.cbc.ca/radio/asithappens/as-it-happens-thursday-edition-1.4466385/harvard-undergrad-cracks-code-of-knotted-inca-rope-used-as-an-ancient-excel-spreadsheet-1.4466388 https://www.cbc.ca/radio/asithappens/as-it-happens-thursday-edition-1.4466385/harvard-undergrad-cracks-code-of-knotted-inca-rope-used-as-an-ancient-excel-spreadsheet-1.4466388 https://doi.org/10.17061/phrp2941926 12 williamson et al. australian population studies 5 (2) 2021 department of parliamentary library (2003) defining aboriginality in australia. https://www.aph.gov.au/about_parliament/parliamentary_departments/parliamentary_library/ publications_archive/cib/cib0203/03cib10 first nations information governance centre (2014) the first nations principles of ocap. the first nations information governance centre. https://fnigc.ca/ocap-training/ garroutte e (2003) real indians: identity and the survival of native america. university of california press. heiss a (ed.) (2018). growing up aboriginal in australia. melbourne: black inc. hemming s, rigney d, bignall s, berg s, & rigney g (2019) indigenous nation building for environmental futures: murrundi flows through ngarrindjeri country. australasian journal of environmental management 26(3): 216-235. https://doi.org/10.1080/14486563.2019.1651227 jelfs p (2016) the australian bureau of statistics’ aboriginal and torres strait islander enumeration and engagement strategies: challenges and future options. in kukutai t & taylor j (eds.) indigenous data sovereignty: toward an agenda. canberra: anu press; 275-290. http://doi.org/10.22459/caepr38.11.2016 kukutai t & cormack d (2020) “pushing the space”: data sovereignty and self-determination in aotearoa nz. in walter m, kukutai t, russo carroll s, & rodriguez-lonebear d (eds.) indigenous data sovereignty and policy. london: routledge; pp. 21-35. https://doi.org/10.4324/9780429273957 kukutai t & taylor j (2012) postcolonial profiling of indigenous populations. limitations and responses in australia and new zealand. espace populations sociétés 2012/1: 13-27. https://doi.org/10.4000/eps.4812 kukutai t & taylor j (eds.) (2016) indigenous data sovereignty: toward an agenda. canberra: anu press. http://doi.org/10.22459/caepr38.11.2016 kukutai t (2011) building ethnic boundaries in new zealand: representations of maori identity in the census. in axelsson p & skold p (eds.) indigenous peoples and demography: the complex relation between identity and statistics. new york: bergahn books; pp. 33-54. kukutai, t., & walter, m. (2015). recognition and indigenizing official statistics: reflections from aotearoa new zealand and australia. statistical journal of the iaos 31(2): 317-326. https://doi.org/10.3233/sji-150896 love v commonwealth of australia, b43/2018 b64/2018 (high court of australia february 11, 2020). http://eresources.hcourt.gov.au/showcase/2020/hca/3 lovett r (2016) aboriginal and torres strait islander community wellbeing: identified needs for statistical capacity. kukutai t & taylor j (eds.) indigenous data sovereignty: toward an agenda. canberra: anu press; pp. 213-231. http://doi.org/10.22459/caepr38.11.2016 lovett r, jones r, & maher b (2020) the interception of indigenous data sovereignty and closing the gap policy in australia. in walter m, kukutai t, russo carroll s, & rodriguez-lonebear d (eds.) indigenous data sovereignty and policy. london: routledge; pp. 36-51. https://doi.org/10.4324/9780429273957 maiam nayri wingara & australian indigenous governance institute (2018) indigenous data sovereignty communique. indigenous data sovereignty summit, canberra, act. https://www.aigi.com.au/resource/indigenous-data-sovereignty-communique/ markham f & biddle n (2018) recent changes to the indigenous population geography of australia: evidence from the 2016 census. australian population studies 2(1): 1-13. https://doi.org/10.37970/aps.v2i1.21 martin k & mirraboopa b (2003) ways of knowing, being and doing: a theoretical framework and methods for indigenous and indigenist re‐search: journal of australian studies 27(76): 203-214. https://doi.org/10.1080/14443050309387838 https://www.aph.gov.au/about_parliament/parliamentary_departments/parliamentary_library/publications_archive/cib/cib0203/03cib10 https://www.aph.gov.au/about_parliament/parliamentary_departments/parliamentary_library/publications_archive/cib/cib0203/03cib10 https://fnigc.ca/ocap-training/ https://doi.org/10.1080/14486563.2019.1651227 http://doi.org/10.22459/caepr38.11.2016 https://doi.org/10.4324/9780429273957 https://doi.org/10.4000/eps.4812 http://doi.org/10.22459/caepr38.11.2016 https://doi.org/10.3233/sji-150896 http://eresources.hcourt.gov.au/showcase/2020/hca/3 http://doi.org/10.22459/caepr38.11.2016 https://doi.org/10.4324/9780429273957 https://www.aigi.com.au/resource/indigenous-data-sovereignty-communique/ https://doi.org/10.37970/aps.v2i1.21 https://doi.org/10.1080/14443050309387838 australian population studies 5 (2) 2021 williamson et al. 13 mullane-ronaki m-t t k k (2017) indigenising the national census? a global study of the enumeration of indigenous peoples, 1985-2014. master of social sciences thesis. university of waikato, hamilton, new zealand. https://hdl.handle.net/10289/11175 peters m a & mika c t (2017) aborigine, indian, indigenous or first nations? educational philosophy and theory 49(13): 1229-1234. https://doi.org/10.1080/00131857.2017.1279879 pool i (2016) colonialism’s and postcolonialism’s fellow traveller: the collection, use and misuse of data on indigenous people. in kukutai t & taylor j (eds.) indigenous data sovereignty: toward an agenda. canberra: anu press; pp. 57-76. http://doi.org/10.22459/caepr38.11.2016 rowse t & smith l (2010) the limits of ‘elimination’ in the politics of population. australian historical studies 41(1): 90-106. https://doi.org/10.1080/10314610903317598 smith l t (2012) decolonizing methodologies: research and indigenous peoples. london: zed books. statistics new zealand (2018) 2018 census: design of forms. https://secure.livechatinc.com/ statistics new zealand (2021) 2021 mana ōrite relationship agreement. https://www.stats.govt.nz/about-us/what-we-do/mana-orite-relationship-agreement/ sutton p (2003) families of polity. in sutton p (eds.) native title in australia: an ethnographic perspective. cambridge: cambridge university press; 206-231. https://doi.org/10.1017/cbo9780511481635.009 taylor j & biddle n (2010) estimating the accuracy of geographic variation in indigenous population counts. australian geographer 41(4): 469 484. https://doi.org/10.1080/00049182.2010.519695 taylor j (2009) indigenous demography and public policy in australia: population or peoples? journal of population research 26(2): 115–130. https://doi.org/10.1007/s12546-009-9010-9 taylor j (2011) postcolonial transformation of the australian indigenous population. geographical research 49(3): 286-300. taylor j, doran b, parriman m & yu e (2014) statistics for community governance: the yawuru indigenous population survey, western australia. international indigenous policy journal 5(2): article 2. https://doi.org/10.18584/iipj.2014.5.2.2 united kingdom house of commons. (1900). commonwealth of australia constitution act 1900. https://www.legislation.gov.uk/ukpga/vict/63-64/12/enacted us census bureau (2020) questions asked on the form. 2020census.gov. https://www.census.gov/programs-surveys/decennial-census/technicaldocumentation/questionnaires.2020_census.html veracini l (2011) introducing settler colonial studies. settler colonial studies 1(1): 1-12. https://doi.org/10.1080/2201473x.2011.10648799 walter m (2010) the politics of the data: how the australian statistical indigene is constructed. international journal of critical indigenous studies 3(2): 45-56. https://doi.org/10.5204/ijcis.v3i2.51 walter m (2016) data politics and indigenous representation in australian statistics. in kukutai t & taylor j (eds.) indigenous data sovereignty: toward an agenda. canberra: anu press; 79-97. http://doi.org/10.22459/caepr38.11.2016 walter m (2018) the voice of indigenous data. griffith review 60: 256-263. https://www.griffithreview.com/articles/voice-indigenous-data-beyond-disadvantage/ walter m, kukutai t, russo carroll s, & rodriguez-lonebear d (eds.) (2020) indigenous data sovereignty and policy. london: routledge. https://doi.org/10.4324/9780429273957 whittaker a (2017) white law, blak arbiters, grey legal subjects: deep colonisation’s role and impact in defining aboriginality at law. australian indigenous law review 20(1): 4-47. wolfe p (1999) settler colonialism and the transformation of anthropology. london: cassell. https://hdl.handle.net/10289/11175 https://doi.org/10.1080/00131857.2017.1279879 http://doi.org/10.22459/caepr38.11.2016 https://doi.org/10.1080/10314610903317598 https://secure.livechatinc.com/ https://www.stats.govt.nz/about-us/what-we-do/mana-orite-relationship-agreement/ https://doi.org/10.1017/cbo9780511481635.009 https://doi.org/10.1080/00049182.2010.519695 https://doi.org/10.1007/s12546-009-9010-9 https://doi.org/10.18584/iipj.2014.5.2.2 https://www.legislation.gov.uk/ukpga/vict/63-64/12/enacted https://www.census.gov/programs-surveys/decennial-census/technical-documentation/questionnaires.2020_census.html https://www.census.gov/programs-surveys/decennial-census/technical-documentation/questionnaires.2020_census.html https://doi.org/10.1080/2201473x.2011.10648799 https://doi.org/10.5204/ijcis.v3i2.51 http://doi.org/10.22459/caepr38.11.2016 https://www.griffithreview.com/articles/voice-indigenous-data-beyond-disadvantage/ https://doi.org/10.4324/9780429273957 14 williamson et al. australian population studies 5 (2) 2021 yap m & yu e (2016) data sovereignty for the yawuru in western australia. in kukutai t & taylor j (eds.) indigenous data sovereignty: toward an agenda. canberra: anu press; pp. 233-261. http://doi.org/10.22459/caepr38.11.2016 yu p (2012) the power of data in aboriginal hands. caepr topical issue no. 4/2012. centre for aboriginal economic policy research, australian national university. https://caepr.cass.anu.edu.au/research/publications/power-data-aboriginal-hands http://doi.org/10.22459/caepr38.11.2016 https://caepr.cass.anu.edu.au/research/publications/power-data-aboriginal-hands australian population studies 2022 | volume 6 | issue 1 | pages 15–30 © gray, evans & reimondos 2022. published under the creative commons attribution-noncommercial licence 3.0 australia (cc by-nc 3.0 au). journal website: www.australianpopulationstudies.org having babies in times of uncertainty: first results of the impact of covid-19 on the number of babies born in australia edith gray* the australian national university ann evans the australian national university anna reimondos the australian national university *corresponding author. email: edith.gray@anu.edu.au. address: school of demography, research school of social sciences, the australian national university, act 2601. paper received 29 march 2022; accepted 9 july 2022; published 25 july 2022 abstract background there has been considerable speculation on whether the covid-19 pandemic had an effect on childbearing behaviour. based on the experience of other social and economic disruptions, many researchers suggested that births would decline, while others argued that there could be a positive effect. aims this paper considers the uncertainties associated with the impacts of covid-19, particularly the relationship between the timing of covid-19 events and subsequent births. data and methods publicly available birth data from birth registers, perinatal databases, and public hospital data were compiled and analysed to document changes in numbers and patterns of recorded births during 2020 and 2021. results births declined in 2020 but then rebounded in 2021. quarterly birth data from new south wales and western australia suggest that the sharpest drop in conceptions occurred in the january-march 2020 quarter. this coincided with the period when the pandemic was first taking off and when uncertainty about the future was at its highest. conclusions the uncertainty associated with the onset of the covid-19 pandemic had a noticeable impact on births in 2020. it also shows, where data is available, that this impact was relatively short-lived, and births rebounded in 2021. we note that data is still sparse for victoria, a state which was substantially more affected by lockdowns. key words fertility; births; covid-19; australia. http://www.australianpopulationstudies.org/ mailto:edith.gray@anu.edu.au 16 gray, evans & reimondos australian population studies 6 (1) 2022 1. introduction in early 2020 when news of the covid-19 pandemic began spreading around the world there was a general consensus among demographers that this was going to have a major impact on childbearing behaviour. for people of childbearing age, covid-19 would be the first time they would have experienced a global pandemic that caused widespread social and economic disruption. it is well known that during periods of high uncertainty and disruption, people will avoid making major lifechanging decisions such as having a child (comolli & vignoli 2021). however, the exact nature of the impact on childbearing behaviour would remain a matter of speculation until birth data became available. compiling the evidence to demonstrate what effect the pandemic would have on births takes time. a recorded birth involves, an average pregnancy of 40 weeks between conception and birth, then several months or sometimes years, before birth records data are collated and distributed to statistical agencies. this translates to a lag of at least a year before being able to observe the impact of the pandemic on the number of births. before birth data were available, researchers tried novel ways to predict the impact of the pandemic on births, including analysing google search trends for terms related to pregnancies and newborns (wilde et al. 2020). more traditional methods such as asking people directly about their future fertility intentions, and if they had changed their fertility intentions because of the pandemic, were also employed (emery & koops 2022; luppi et al. 2020; naya et al. 2022; qu 2021). in australia, there were also early attempts to predict births based on claim data for items on the medicare benefits schedule (mbs) related to antenatal visits (moaven & brown 2021). birth record data from 2020, and for some states and territories for 2021, is now becoming available in australia. according to the australian bureau of statistics births, australia report (2021) in 2020 there were 294,369 registered births, representing a 3.7% decline from 2019. this represents a record low total fertility rate of 1.58 for 2020 (abs 2021). while most of these births would have been conceived before the pandemic, births in the later part of 2020 would have been conceived during the early stages of the pandemic. the lower number of registered births in 2020 has been attributed to the effect of the pandemic on the number of births themselves, but also on delays in birth registration (abs 2021). in this paper, we use recorded births from states and territories from 2018 to 2021 to analyse the early impact of covid-19 on the pattern of births. we argue that the impact of the pandemic on childbearing decisions and on subsequent number of births, can be divided into two periods. the first period occurred during the early part of 2020 and represented the period of highest uncertainty when knowledge about the pandemic was still very limited. this period should have seen a universal fall in conceptions. the second period was marked by more diverse experiences at both the individual and state level. depending on one’s own work and family circumstances, as well as geographical location there could have been both upwards and downwards pressures on fertility, which are harder to predict. australian population studies 6 (1) 2022 gray, evans & reimondos 17 2. covid-19 and fertility 2.1 first stage of the pandemic: high uncertainty the early stages of the covid-19 pandemic led to an unprecedented level of uncertainty for much of the population in australia, as well as around the world. uncertainty is a ‘forward looking notion’ (guetto et al. 2022) characterised by a lack of information regarding the future. while everything about the future is uncertain, people will generally use past experiences as well as engage in knowledge seeking to help them plan and formulate decisions. as vignoli et al. (2020a) note in terms of decisions, the decision to have a child is one of the most important decisions people make in their lives and one that is always accompanied by a degree of fundamental uncertainty. childless people experience the most uncertainty, or lack of information about how their future with children will be; however, parents also face uncertainty when deciding to have another child because no two children are the same (vignoli et al. 2020b). in periods of uncertainty, people will tend to defer making major life choices which involve long-term commitments, such as starting a family or having more children (vignoli et al. 2020a). increased uncertainty of daily life and lack of job security has been posited as a reason for falling fertility rates in many high-income countries in recent years (comolli & vignoli 2021). additionally, there are many specific historical examples of birth rates falling in times characterised by a high level of uncertainty such as global economic recessions (sobotka et al. 2011; matysiak et al. 2020). during the early months of 2020 there was limited understanding about the virus and how it was spread. people felt a lack of control of the situation and their own lives, and there was also a real fear for the future and a sense of threat for the well-being of oneself and one’s family (maison, et al 2021). as more information was accumulated about the virus and public health safety measures started being imposed, the level of uncertainty decreased. uncertainty is commonly associated with anxiety, worry, and psychological distress (afifi et al. 2012; sweeny et al. 2015). repeated crosssectional as well as longitudinal studies from overseas that tracked mental health during the different stages of the pandemic confirm that poor mental health and anxiety was highest in the first few months of march and april but then, on average recovered (aknin et al. 2021; daly & robinson 2021, 2022). in australia, a longitudinal survey of mental health conducted during three periods between may and october 2020 also found that anxiety and depressive symptoms were highest in the earliest period of may 2020 (o’donnell et al. 2022). 2.2 second stage of the pandemic: adaptation and diverse experiences the timeline in table 1 shows that the first few months of 2020 were the ones where uncertainty was at its highest. this uncertainty would have impacted most of the population, although the impact is correlated to personality traits and level of media consumption (mertens et al. 2020). as the pandemic progressed, people’s experiences became more diverse and would have varied based on their geographic location, due to different state restrictions, and their personal situations. some people may have decided that prolonged delays or reductions in their childbearing plans were unsustainable or unnecessary (lindberg et al. 2021, p. 8). for some, the pandemic meant the loss of a job or reduced work hours, whereas for others their work hours may have increased. large parts of the workforce were able to transition to working from home, whereas this was not possible in all 18 gray, evans & reimondos australian population studies 6 (1) 2022 sectors. working from home may have been a positive experience increasing work-life balance for some, or it may have been a stressful experience (yerkes et al. 2020). similarly, depending on the age and number of children at home, some parents had to navigate home schooling. this diversity of experience makes it particularly difficult to ascribe definitive ‘effects’ of the pandemic. table 1: timeline of covid-19 related events, january 2020 to march 2021 period dates events jan-mar 2020 january 11 china reports first death january 13 thailand reports first case outside china january 23 lockdown of wuhan january 24 first european case confirmed in france january 25 first confirmed case in australia feb 11 officially named coronavirus disease 2019 or covid-19 by who march 1 first covid-19 death in australia march 2 first case of community transmission in australia march 8 italy placed under quarantine measures march 11 who declares a pandemic march 10 australia records its hundredth case march 12 prime minister announces first economic stimulus package march 15 nsw government cancels events of more than 500 people march 18 federal government announced several measures such as social distancing (1.5 meters) and limitation on size of gatherings to be implemented by state governments. march 19 australian borders closed to non-residents/citizens march 23 national cabinet agrees pubs, gyms, cinemas, restaurants and cafes to close (except for delivery) march 27 all returning residents required to spend two weeks in supervised quarantine apr-jun 2020 april 15 jobkeeper payment legislation to support out-of-work australians is passed april 26 covidsafe app, designed to help with contract tracing, is released by the federal government april 20 peak of first wave may 10 nsw government announces easing of some restrictions may 11 victoria government announces easing of some restrictions june 30 victoria announces lockdown in several postcodes jul-sep 2020 july 7 victoria in lockdown july 10 victoria suggests people wear masks outside (not mandated) august 16 victoria announces new lockdown measures including overnight curfew oct-dec 2020 october 26 victoria’s second lockdown ends december 19 nsw imposes lockdown order in northern beaches jan-mar 2021 january 9 lockdown ends for sydney northern beaches february 12-18 victoria goes into snap lockdown. february 22 phase 1a vaccination rollout targeting 678,000 aged care and frontline health workers. march 22 phase 1b vaccination rollout targeting 6 million australians. australian population studies 6 (1) 2022 gray, evans & reimondos 19 table 2 outlines a number of potential upwards and downwards pressures on childbearing that the covid-19 outbreak could have had in high-income countries. an increase in births could result from a variety of reasons including more shared time at home and reassessment of priorities. for example, working from home could encourage a re-evaluation of work-life balance and reduce postponement of having a child, or partners could have more opportunities for intimacy and sexual relations. on the other hand, increased relationship conflict and fewer opportunities to socialise outside the home could reduce sexual intercourse. economic changes could also impact on childbearing decisions. for example, some families may have been able to increase their savings, whereas others suffered a loss of income. reduced access to contraception and abortion services may lead to increased conceptions, whereas reduced access to assisted reproductive technology (art) during lockdowns would decrease births. table 2: potential upwards and downwards pressures on fertility due to the pandemic influence effect on fertility job loss and reduced work hours leading to lower opportunity cost of having children positive less access to contraceptives and abortion clinics positive more time spent at home with partners leading to a strengthening of relationships and increased intimacy and frequency of sexual intercourse positive increased gender equity in the home as men become more involved in household tasks positive working from home increasing work-life balance positive re-evaluation of life priorities and reduction in postponement positive job losses and reduced working hours leading to loss of income and less ability to afford children negative more time spent at home with partners leading to higher relationship tension or domestic violence negative decreased access to assisted-reproduction technology negative reduced ability to socialise outside the home leading to lower levels of dating or new relationship formation negative lockdowns and travel restrictions reducing ability to access informal childcare and social support negative increased stress due to home-schooling or home-childcare negative health concerns about covid-19 and pregnancy as well as visitation restrictions in hospitals negative uncertainty about economic and societal future negative worsening mental health negative source: adapted from berrington et al. (2020) and aasve et al. (2021) 20 gray, evans & reimondos australian population studies 6 (1) 2022 3. data and methods 3.1 data birth record data is now available in australia for 2020 and 2021, allowing us to see what impact the early stages of the pandemic had on the number of births. for each state and territory we sourced publicly available data from birth registers, public health departments, as well as using data prepared by the australian institute for health and welfare (aihw) and australian bureau of statistics (abs). there are three main sources of birth data: (1) registered births as counted by each state/territory’s registrar of births, deaths and marriages, (2) births recorded in each state/territory’s perinatal data collection, (3) health department data on births in the public health system. each source of data is briefly outlined below. birth registrations the registration of a birth in australia follows a two-step process involving notification by a hospital, followed by registration by the parent(s). each state and territory in australia has legislation requiring the responsible person, usually the chief executive of a hospital, a doctor or midwife, to notify the registrar of births, deaths and marriages of the birth of a child within a certain time frame. the time frame varies by jurisdiction. for example, in new south wales it is within 21 days after the birth, whereas in queensland it is after two working days (victorian law reform commission 2013). the parents are then provided with a birth registration form which must be signed and submitted to the registrar within a certain time period. in most jurisdictions, the parent(s) have 60 days to register the birth, except the act where they have 6 months (act legislation register 2014). data on number of births sourced from the state and territory registers are subject to delays which impacts on its usefulness in assessing trends over short periods of time. these delays may be due to parents not registering births within the required time frame, as well as processing delays by the registrar offices. for example, the abs (2021) notes that processing delays following the second covid-19 wave in victoria led to lower numbers of birth registrations being processed and reported in december 2020. similarly, in nsw a backlog of birth registrations was processed in 2018 leading to an unusually high number of births being reported for 2018 compared to 2017 and 2019 (abs 2021). perinatal data the other major source of data on births in australia is the perinatal data collection. when a birth occurs, a midwife or attending health professional collects information about the maternal characteristics, birth and neonatal outcomes and forwards this to the relevant health department and this forms the basis of the state/territory’s perinatal dataset. perinatal data is usually collected and reported for the purpose of monitoring public health rather than measuring fertility (wilson 2017). public hospital data in an effort to provide greater accountability on the performance of their public hospitals, a number of states and territories also publish regular reports outlining key metrics on the quality and effectiveness of their public health systems, including measures such as average waiting times at australian population studies 6 (1) 2022 gray, evans & reimondos 21 emergency departments. in some cases, the number of babies born in public hospitals is also reported. in australia, 96.7% of women give birth in hospitals, and most of those do so in public hospitals (aihw 2022a), so data on births in public hospitals encompasses a large proportion of all births. for example, in new south wales in 2019, 73% of all women giving birth did so in a public hospital, and in 2020 this was largely unchanged at 74% (aihw 2022a). individual state/territory data and national datasets while each state/territory has its own vital statistics registrar, and its own perinatal data collection, they vary in their approach to making this data publicly available. for example, every month the registry of births deaths and marriages victoria publishes how many births were registered in the previous month. in some other states no such data is published in a timely manner. however, for both registry and perinatal data all states/territories forward this to the australian bureau of statistics (abs) and the australian institute of health and welfare (aihw), respectively. these organisations then collate the data into national datasets. each state and territory registrar provides birth register data electronically to the abs on a monthly basis. the abs then publishes these data on a preliminary basis five to six months after the reference period in its national, state and territory population report, and publishes a revised version 21 months after the end of the financial year (abs 2021). registered birth data are also published around a year after the reference year in the births, australia publication which is the main source also of fertility estimates such as the total fertility rate. similarly, perinatal data from each state/territory is collated into the national perinatal data collection by the aihw (aihw 2022b). these data are published in the aihw mothers and babies report two years after the reference period. for example, the mothers and babies report published in 2022 refers to births that occurred in 2020. 2.2 methods for each state and territory we compiled annual data on number of births recorded from 2018-2021, or the latest available year, using this time-series to examine the impact of covid-19. we then used quarterly data from new south wales and western australia, as these are available up to and including the first quarter of 2022, to examine in more detail how the number of conceptions and subsequent births changed during the early stages of the pandemic. it is important to note that these data relate to live births but they contain no information on the characteristics of mothers. it is therefore not possible to calculate additional indicators such as age-specific fertility rates or the total fertility rate. 3. results table 3 shows the number of births reported in 2018 to 2021 or the latest available year for each state/territory based on different sources. for each source of data the table indicates whether the information is from perinatal data or public hospital data (based on year of occurrence), or registry data (based on year of registration). 22 gray, evans & reimondos australian population studies 6 (1) 2022 table 3: number of births by year, state and data source, 2019-2021 state/ no. of births change in births (%) territory data source type 2018 2019 2020 2021 2018-19 2019-20 2020-21 nsw aihw (2020,2021,2022a) occurrence perinatal 94,942 95,096 92,539 0.2 -2.7 abs (2021) registration 105,463 96,909 93,579 -8.1 -3.4 abs (2022) registration 98,626 98,678 94,822 101,333 0.1 -3.9 6.9 nsw health (2022) occurrence perinatal 95,552 95,133 92,541 -0.4 -2.8 nsw, bureau of health information (2022) occurrence public hosp. 72,497 72,123 70,252 74,185 -0.5 -2.6 5.6 vic births, deaths and marriages victoria (2022) registration 79,726 79,597 74,620 76,410 -0.2 -6.3 2.4 births, deaths and marriages vic (2022)* registration 79,726 79,597 75,018 76,822 -0.2 -5.8 2.4 ccopmm (2022)+ occurrence perinatal 78,521 78,954 76,990 0.5 -2.5 aihw (2020,2021,2022a) occurrence perinatal 78,233 79,344 77,380p 1.4 -2.5 abs (2021) registration 78,488 77,220 73,543 -1.6 -4.8 abs (2022) registration 77,512 77,052 72,159 75,365 -0.6 -6.4 4.4 qld queensland government (2022) registration 62,248 62,184 59,914 64,673 -0.1 -3.7 7.9 queensland health (2021) occurrence– perinatal 60,503 60,443 59,603 -0.1 -1.4 aihw (2020,2021,2022a) occurrence perinatal 60,120 60,431 59,584 0.5 -1.4 abs (2021) registration 61,931 61,735 59,490 -0.3 -3.6 abs (2022) registration 61,076 61,054 59,297 64,112 -0.03 -2.9 8.2 wa department of justice (2022) registration 33,459 33,754 32,677 34,300 0.9 -3.2 5.0 department of health (2022) occurrence perinatal 33,204 33,147 32,027 34,476 -0.2 -3.4 7.6 aihw (2020,2021,2022a) occurrence perinatal 33,204 33,368 32,255 0.5 -3.3 abs (2021) registration 33,257 33,539 32,426 0.8 -3.3 abs (2022) registration 33,442 33,427 32,563 34,039 0.0 -2.6 4.5 sa aihw (2020,2020,2022a) occurrence perinatal 19,199 19,173 18,738 -0.1 -2.3 abs (2021) registration 19,113 19,490 18,526 2.0 -4.9 abs (2022) registration 18,968 18,957 18,450 19,757 -0.1 -2.7 7.1 tas tasmanian government (2022) registration 5,570 5,835 5,850 6,365 4.8 0.3 8.8 aihw (2020,2021,2022a) occurrence perinatal 5,480 5,736 5,637 4.7 -1.7 abs (2021) registration 5,547 5,741 5,780 3.5 0.7 abs (2022) registration 5,412 5,648 5,606 6,044 4.4 -0.7 7.8 australian population studies 6 (1) 2022 gray, evans & reimondos 23 table 3 continued state/ no. of births change in births (%) territory data source type 2018 2019 2020 2021 2018-19 2019-20 2020-21 act aihw (2020,2021,2022a) occurrence perinatal 5,994 6,314 6,147 5.3 -2.6 abs (2021) registration 5,374 5,520 5,368 2.7 -2.8 abs (2022) registration 5,300 5,522 5,240 5,543 4.2 -5.1 5.8 act health (2022) occurrence public hosp. 5,055 5,151 5,110 5,293 1.9 -0.8 3.6 nt aihw (2020,2021,2022a) occurrence -perinatal 3,730 3,592 3,688 -3.7 2.7 abs (2021) registration 4,050 3,658 3,752 -9.7 2.6 abs (2022) registration 3,763 3,595 3,690 3,785 -4.5 2.6 2.3 note: + data refer to adjusted live births and excludes terminations of pregnancy for congenital anomalies or for maternal psychosocial indications. p data for victoria in 2021 published by aihw (2022a) is preliminary and may be subject to revision. * discrepancy in victorian registration data when registered births by month are summed compared to their data annual data. for new south wales (nsw), the trend for 2018-2019 differs substantially across the sources. aihw and abs population estimates suggest a small increase in births between 2018 and 2019, whereas abs (2021) births data based on registered births shows a substantial decline. this is due to a substantial backlog of births being processed by the nsw register in 2018 (abs 2021). for 2019-2020, there is a decline of between -2.6% to -3.9% in births. both birth register data and public hospital data suggest a rebound in 2021 of 6.9 to 5.6% respectively. for victoria (vic), between 2018 and 2019 according to most sources there was a small trend downwards. the drop in births in victoria in 2020, was between -2.5% and -6.4% depending on the source. the largest drops are those recorded for registered births so this indicates that it may be due in large part to delays in registration due to the victorian lockdowns. in 2020-2021 registered births increased between 2.4% to 4.4%. for queensland (qld), again we see a small trend downwards in the number of births between 2018 and 2019, followed by a decline between 2019-2020 of between -1.4% and -3.7%. however registered births in 2021 increased by 7.9-8.2%. western australia (wa) data suggests that births may have been increasing already between 2018-19, but 2020 saw a decline of between -2.6% to 3.4%. register and midwife data shows a rebound of 4.5% to 7.6% in 2021. for south australia (sa), the decline in 2020 is estimated at -2.7% to -4.9%, again followed by a rebound. for the smaller states and territories, the pattern is less clear. tasmanian (tas) and australian capital territory (act) both had increasing births in 2019. for tasmania, the birth registration data as provided by the tasmanian government suggests that births continued to increase in 2020, however, at a slower pace. in 2021, registered births increased by 7.8-8.8%. the australian capital territory registered declines of -0.5% to -5.1% in 2020, whereas in the northern territory (nt), births increased in 2020. to better understand how patterns of childbearing decision making may have changed across the different stages of the pandemic, we use quarterly birth data. figure 1 presents the actual estimated quarterly conceptions (leading to births) for nsw and wa. nsw data come from the nsw bureau of 24 gray, evans & reimondos australian population studies 6 (1) 2022 health information, which publishes quarterly data regarding babies born in nsw public hospitals using admitted patient data. the latest available quarter is january-march 2022 quarter. western australia data come from the midwives notification system as published by the department of health wa. the number of conceptions leading to live births are based on backdating births. for example, conceptions in january-march lead to births in october-december of that year. the data is indexed on the first quarter available for each state. while quarterly data is also available for registered births, we prefer to use perinatal and public hospital data due to the registration delays in the register data which may lead to an inaccurate portrayal of the trends over time (wilson 2017). with register data we are unable to tell if a decline in registered births in one quarter is due to a decline in births, or to decline in registrations (for example due to lockdowns). with the perinatal data and public hospital data which record births at time of occurrence, we have more confidence that any temporal trends observed represent actual trends in the number of births. both states display a similar pattern. conceptions in 2019 were generally lower than conceptions in 2018. due to the seasonal pattern of births in australia, births tend to be lower in december overall (wilson, et al. 2020). however, in both the nsw public hospital data and wa perinatal data there is a distinct sharp drop in conceptions in january-march 2020. following this, there is an increase in conceptions. figure 1: conceptions by quarter, nsw and wa, 2018-2021 source: nsw bureau of health information 2022, wa department of health 2022 australian population studies 6 (1) 2022 gray, evans & reimondos 25 4. discussion in 2020, the number of babies born in australia dropped significantly compared to 2019 in nearly every state and territory. this pattern can be seen across a number of different sources of data, including birth registrations as well as perinatal data and public hospital data. while the number of births had been trending downward in recent years, 2020 was a clear disjunction. while most of the births in 2020 would have been conceived in pre-pandemic times, babies born in the last few months of 2020 would have been conceived in the early months of 2020. our analysis of quarterly data available for new south wales and western australia, confirms that the large drop in 2020 can be attributed in particular to a reduction of births in the last quarter of 2020. in 2021, the data indicated a significant recuperation starting as early as the first quarter of 2021. based on the literature on uncertainty and fertility, as well as evidence from surveys tracking the pattern of mental health during the pandemic, we suggest that the drop in births in the octoberdecember quarter of 2020, corresponding to conceptions in january-march 2020 was related to the unprecedented uncertainty about the future that people felt during this time. as more became known about the virus, the immediate level of anxiety lessened and childbearing plans were revisited again. while we suggest that the decline in the number of births in 2020 was due to the uncertainty of the early stages of the pandemic, another possibility could be that there were fewer women to give birth in the population at that time due to a decline in immigration. every quarter the abs publishes estimates of the resident population of australia. estimates for 2019 to 2021 have recently been rebased based on the 2021 census of the population. at 30th june 2019, there were an estimated 6.026 million women aged 15-49 in australia. by 30th june 2020, this had increased slightly to 6.048, and a year later it had fallen to 5.977 (abs 2022). however, there were large differences in female population numbers by age group as well as by state. in particular, the number of women aged 20-24 age is estimated to have declined from 846,908 in 2019 to 786,274 in 2021. the early 20s is a key age for students and travellers to migrate temporarily to australia for study or work. while the estimated resident population numbers exclude overseas visitors who are in australia for less than 12 months over a 16-month period, many young people on temporary visas for studying or working holidays stay for 12 months or longer. so the decline in the number of women in this age group could perhaps be explained by the travel restrictions preventing the usual temporary migration of young women. women on temporary visas have relatively low fertility rates (mcdonald 2021) therefore we do not believe that the drop in births in 2020 was due to any large degree to fewer women in the population. while the focus on this paper has been on number of births, it is important to understand how fertility measures such as total fertility rate (tfr) may have been affected by changes in the number of births as well as the number of women, or the population at risk of giving birth. the total fertility rate is calculated by the australian bureau of statistics and is reported by calendar year (abs 2021) as well as by financial year (abs 2022). the most recently published tfr of 1.58 for 2020 was calculated by the australian bureau of statistics using registered births and the estimated number of 26 gray, evans & reimondos australian population studies 6 (1) 2022 resident women projected in 2020 based on the 2016 census. if the tfr was calculated using the same number of registered births but using the more recently available estimates of resident women rebased on 2021 census the tfr is slightly higher at around 1.59. in any average year, the tfr of australia can be thought of as being slightly underestimated due to the fact that included in the denominator is young women on extended temporary visits to australia for work or study. with some of this population removed due to the pandemic, this had a small effect on increasing the tfr. this effect can also be observed to some degree by comparing the tfr of calendar and financial years. in most years the calendar year tfr and the financial year tfr are closely aligned and in pre-pandemic years they followed the same trend. however, while the calendar year tfr for 2020 was 1.58 (abs 2021), the tfr for 2019-20 (covering 1st july 2019 to 30th june 2020), which would include babies conceived pre-pandemic was 1.62 (abs 2022). the tfr for 2020-21, which would be primarily babies conceived after the pandemic began, was higher at 1.63 (abs 2022). this is likely because the 2020-21 includes both the large dip of births but then the large increase in mid-2021, and because there were fewer resident women as a result of border closures leading to a lower denominator. 5. conclusions using recorded birth data from birth registers as well as perinatal and public hospital data we observe that in australia births declined significantly in most states and territories in 2020, followed by a rebound in 2021. this pattern was found across a number of different types of birth data, and therefore not due simply to delays in registration. analysis of public health data, and perinatal data for western australia and new south wales indicates that the 2020 decline appears to have been primarily due to a significant reduction in births in the later part of the year. this corresponds to births conceived during the first stage of the pandemic. this was a period of high uncertainty when, as predicted by the literature on uncertainty and fertility, people avoided making important childbearing decisions. as the pandemic progressed, people’s experiences became more diverse and it becomes more difficult to predict the direction that fertility would take. the situation across different parts of australia became more heterogeneous. different states imposed lockdown orders of varying severity, and as case numbers peaked unevenly across the country, people’s experiences of the pandemic, therefore, became more diverse in the later stages and depended on where they lived as well as what their job and family situation was. there are a number of possible upwards and downwards pressures that could have occurred, but based on the quarterly data available for new south wales and western australia it appears that for a significant part of the population, the possible upwards pressures outweighed any negative pressures as the pandemic progressed, leading to at least a resumption in previously postponed childbearing plans. was the drop in births in late 2020, followed by a rebound a universal experience across australia, as well as in other high-income countries? it is likely that the uncertainty of the first few months of the pandemic would be mediated by individual factors such as level of media consumption, personal characteristics, including traits such as intolerance of uncertainty (mertens et al. 2020), but also by contextual factors including the response of the different state and national governments to the australian population studies 6 (1) 2022 gray, evans & reimondos 27 pandemic. unfortunately, apart from register data no other sources of birth record data are available yet for victoria which arguably experienced the greatest level of disruption of all the jurisdictions as a result of the pandemic. it is possible that the pattern of conceptions and births in victoria will reveal a different trend to that observed in new south wales and western australia. for the act, quarterly data available up to july-september 2021 did not show a significant dip. the pattern observed in australia is very similar to that observed in many other countries. for example, in the united states there is a similar pattern of a sharp drop in births in late 2020, corresponding to conceptions in early 2020 (morse 2021). analysis of monthly birth data tracked between november 2020 and january 2021 across 22 highly developed countries by sobotka et al. (2021) also showed a significant decline in births in most countries in the later part of 2020. however, other countries appeared not to have followed this pattern. in europe, birth rates declined from november 2020 in southern europe, austria and belgium, but no effect was found in the nordic countries (aassve et al. 2021). this points to the importance of the social context in mediating the effect of pandemic uncertainty on childbearing. countries with strong welfare systems to act as safety nets may have led to lower levels of uncertainty for their populations (gassen et al. 2022). while the birth data allows us to see the change in childbearing behaviour at the population level, the currently available data also does not allow us to analyse how childbearing decisions and behaviour may have changed during the pandemic for different groups of the population. we will have to wait for more data to become available, including monthly birth data, as well as survey data to understand how childbearing patterns were affected according to people’s geographic location, age, parity, occupation and work experience. the current data may hide counterbalancing effects with some segments of the population more likely to have children, whereas others may have abandoned all childbearing plans. as more data become available, it will be interesting to observe whether the widespread lockdowns experienced during the 2021 had further negative impact on births. questions abound on whether australia’s trending decline in fertility rates will continue and how the changing nature of work, including working from home, will affect childbearing in the future. key messages • birth data from a variety of sources, including registry data, perinatal and public hospital data were analysed to see the impact of the covid-19 on number of births. • births declined in 2020 but then rebounded in 2021. • quarterly birth data from nsw and wa suggest that january-march 2020 was the period with the sharpest drop in conceptions. this coincides with the highest period of uncertainty. references aassve, a., cavalli, n., mencarini, l., plach, s., & livi bacci, m. (2020). the covid-19 pandemic and human fertility. science, 369(6502), 370-371. https://doi.org/10.1126/science.abc9520. aassve, a., cavalli, n., mencarini, l., plach, s., & sanders, s. (2021). early assessment of the relationship between the covid-19 pandemic and births in high-income countries. proceedings of the national academy of sciences, 118(36). https://doi.org/10.1073/pnas.2105709118. https://doi.org/10.1126/science.abc9520 https://doi.org/10.1073/pnas.2105709118 28 gray, evans & reimondos australian population studies 6 (1) 2022 act health (2022). act public health services quarterly performance report. https://health.act.gov.au/about-our-health-system/act-public-health-services-quarterlyperformance-report act legislation register (2014). births, deaths and marriages registration amendment act 2014. act government. https://www.legislation.act.gov.au/a/2014-8/ afifi, w. a., felix, e. d., & afifi, t. d. (2012). the impact of uncertainty and communal coping on mental health following natural disasters. anxiety, stress & coping, 25(3), 329-347. https://doi.org/10.1080/10615806.2011.603048 aknin, l. b., de neve, j. e., dunn, e. w., fancourt, d. e., goldberg, e., helliwell, j. f., ... & ben amor, y. (2021). mental health during the first year of the covid-19 pandemic: a review and recommendations for moving forward. perspectives on psychological science, 17456916211029964. https://doi.org/10.1177/17456916211029964 australian bureau of statistics (2021). births, australia 2020. australian bureau of statistics, canberra. https://www.abs.gov.au/statistics/people/population/births-australia/2020. australian bureau of statistics (2022). national, state and territory population, december 2021. australian bureau of statistics, canberra. https://www.abs.gov.au/statistics/people/population/national-state-and-territorypopulation/latest-release#states-and-territories. accessed 29 june 2022. australian institute of health and welfare (2020). australia’s mothers and babies 2018 – in brief. data tables: australia’s mothers and babies 2018in brief supplementary tables. table 1.1. https://www.aihw.gov.au/reports/mothers-babies/australias-mothers-and-babies-2018-inbrief/data australian institute of health and welfare (2021). australia’s mothers and babies 2019. data tables: national perinatal data collection annual update 2019. table 1.1. https://www.aihw.gov.au/reports/mothers-babies/australias-mothers-babies/data australian institute of health and welfare (2022a). australia’s mothers and babies 2020. data tables: national perinatal data collection annual update 2020. table 1.1. https://www.aihw.gov.au/reports/mothers-babies/australias-mothers-babies/data australian institute of health and welfare (2022b). national perinatal data collection https://www.aihw.gov.au/about-our-data/our-data-collections/national-perinatal-data-collection berrington, a., ellison, j., kuang, b., vasireddy, s., & kulu, h. (2021). scenario‐based fertility projections incorporating impacts of covid‐19. population, space and place, e2546. https://doi.org/10.1002/psp.2546. berrington, a., ellison, j., kuang, b., vasireddy, s., & kulu, h. (2022). what is the likely impact of covid-19 on fertility in the uk?. policy briefing 66. esrc centre for population change. https://www.cpc.ac.uk/docs/2022_pb66_what_is_the_likely_impact_of_covid19_on_fertility_i n_the_uk.pdf. ccopmm (consultative council on obstetric and paediatric mortality and morbidity) (2021). victoria’s mothers babies and children 2020 report and presentations. supplementary tablesbirths https://www.safercare.vic.gov.au/publications/victorias-mothers-babies-and-children-2020report-and-presentations. accessed 28 june 2022 comolli, c. l., & vignoli, d. (2021). spreading uncertainty, shrinking birth rates: a natural experiment for italy. european sociological review. https://doi.org/10.1093/esr/jcab001. daly, m., & robinson, e. (2021). psychological distress and adaptation to the covid-19 crisis in the united states. journal of psychiatric research, 136, 603-609. https://doi.org/10.1016/j.jpsychires.2020.10.035. daly, m., & robinson, e. (2022). depression and anxiety during covid-19. the lancet, 399(10324), 518. https://doi.org/10.1016/s0140-6736(22)00187-8 https://health.act.gov.au/about-our-health-system/act-public-health-services-quarterly-performance-report https://health.act.gov.au/about-our-health-system/act-public-health-services-quarterly-performance-report https://www.legislation.act.gov.au/a/2014-8/ https://doi.org/10.1080/10615806.2011.603048 https://doi.org/10.1177/17456916211029964 https://www.abs.gov.au/statistics/people/population/births-australia/2020 https://www.abs.gov.au/statistics/people/population/national-state-and-territory-population/latest-release#states-and-territories https://www.abs.gov.au/statistics/people/population/national-state-and-territory-population/latest-release#states-and-territories https://www.aihw.gov.au/reports/mothers-babies/australias-mothers-and-babies-2018-in-brief/data https://www.aihw.gov.au/reports/mothers-babies/australias-mothers-and-babies-2018-in-brief/data https://www.aihw.gov.au/reports/mothers-babies/australias-mothers-babies/data https://www.aihw.gov.au/reports/mothers-babies/australias-mothers-babies/data https://www.aihw.gov.au/about-our-data/our-data-collections/national-perinatal-data-collection https://doi.org/10.1002/psp.2546 https://www.cpc.ac.uk/docs/2022_pb66_what_is_the_likely_impact_of_covid19_on_fertility_in_the_uk.pdf https://www.cpc.ac.uk/docs/2022_pb66_what_is_the_likely_impact_of_covid19_on_fertility_in_the_uk.pdf https://www.safercare.vic.gov.au/publications/victorias-mothers-babies-and-children-2020-report-and-presentations https://www.safercare.vic.gov.au/publications/victorias-mothers-babies-and-children-2020-report-and-presentations https://doi.org/10.1093/esr/jcab001 https://doi.org/10.1016/j.jpsychires.2020.10.035 https://doi.org/10.1016/s0140-6736(22)00187-8 australian population studies 6 (1) 2022 gray, evans & reimondos 29 department of health, western australia (2022). mothers and babies, western australia. summary birth data: graphs and tables for last 3 months. https://ww2.health.wa.gov.au/reports-andpublications/western-australias-mothers-and-babies-summary-information. data extracted 01 february 2022 department of justice, western australia. statistics births, deaths and marriages registered 1981-2021. https://www.wa.gov.au/organisation/department-of-justice/the-registry-of-births-deaths-andmarriages/statistics-births-deaths-and-marriages-registered#:~:text=statistics%20%20births%2c%20deaths%20and%20marriages%20registered%20,of%20name%2086%2c010%2 0%2019%20more%20rows%20. accessed 2 february 2022 emery, t., & koops, j. c. (2022). the impact of covid-19 on fertility behaviour and intentions in a middle income country. plos one, 17(1), e0261509. https://doi.org/10.1371/journal.pone.0261509. gassen, n.s, jokinen, j.c and cuadrado, a. (2022). marriage, divorce, and birth trends. in norlén, g., randall, l., sánchez gassen, n., tapia, c., bogason, á., cuadrado, a., heleniak, t., jokinen, j.c., karlsdóttir, a., lundgren, a. and moodie, j. (eds.) state of the nordic region. nordic council of ministers, nordregio. http://doi.org/10.6027/r2022:2.1403-2503. guetto, r., bazzani, g., & vignoli, d. (2022). narratives of the future and fertility decision-making in uncertain times. an application to the covid-19 pandemic. vienna yearbook of population research, 20, 1-38. https://doi.org/10.1553/populationyearbook2022.res1.6. lindberg, l. d., mueller, j., kirstein, m., & vandevusse, a. (2021). the continuing impacts of the covid-19 pandemic in the united states: findings from the 2021 guttmacher survey of reproductive health experiences. new york: guttmacher institute. https://www.guttmacher.org/report/continuingimpacts-covid-19-pandemic-findings-2021-guttmacher-survey-reproductive-health. luppi, f., arpino, b., & rosina, a. (2020). the impact of covid-19 on fertility plans in italy, germany, france, spain, and the united kingdom. demographic research, 43, 1399-1412. https://doi.org/10.4054/demres.2020.43.47. maison, d., jaworska, d., adamczyk, d., & affeltowicz, d. (2021). the challenges arising from the covid19 pandemic and the way people deal with them. a qualitative longitudinal study. plos o ne, 16(10), e0258133. https://doi.org/10.1371/journal.pone.0258133. matysiak, a., sobotka, t., & vignoli, d. (2020). the great recession and fertility in europe: a sub-national analysis. european journal of population. https://doi.org/10.1007/s10680-020-09556-y. mcdonald, p. (2020). a projection of australia’s future fertility rates. centre for population research paper. canberra: australian government. https://population.gov.au/research/research-fertility. mertens, g., gerritsen, l., duijndam, s., salemink, e., & engelhard, i. m. (2020). fear of the coronavirus (covid-19): predictors in an online study conducted in march 2020. journal of anxiety disorders, 74, 102258. https://doi.org/10.1016/j.janxdis.2020.102258. moaven, l., & brown, j. (2021). covid‐19 “baby boom”. the medical journal of australia, 214(8), 386386. https://doi.org/10.5694/mja2.51010. morse, a. (2021). fewer babies born in december and january but number started to rise in march. united states census bureau. https://www.census.gov/library/stories/2021/09/united-statesbirths-declined-during-the-pandemic.html naya, c. h., saxbe, d. e., & dunton, g. f. (2021). early effects of the covid-19 pandemic on fertility preferences in the united states: an exploratory study. fertility and sterility, 116(4), 1128-1138. https://doi.org/10.1016/j.fertnstert.2021.05.092. nsw health (2022). mothers and babies 2020. https://www.health.nsw.gov.au/hsnsw/pages/mothersand-babies-2020.aspx. accessed 28 june 2022 o'donnell, j., cárdenas, d., orazani, n., evans, a., & reynolds, k. j. (2022). the longitudinal effect of covid-19 infections and lockdown on mental health and the protective effect of neighbourhood https://ww2.health.wa.gov.au/reports-and-publications/western-australias-mothers-and-babies-summary-information https://ww2.health.wa.gov.au/reports-and-publications/western-australias-mothers-and-babies-summary-information https://www.wa.gov.au/organisation/department-of-justice/the-registry-of-births-deaths-and-marriages/statistics-births-deaths-and-marriages-registered#:~:text=statistics%20-%20births%2c%20deaths%20and%20marriages%20registered%20,of%20name%2086%2c010%20%2019%20more%20rows%20 https://www.wa.gov.au/organisation/department-of-justice/the-registry-of-births-deaths-and-marriages/statistics-births-deaths-and-marriages-registered#:~:text=statistics%20-%20births%2c%20deaths%20and%20marriages%20registered%20,of%20name%2086%2c010%20%2019%20more%20rows%20 https://www.wa.gov.au/organisation/department-of-justice/the-registry-of-births-deaths-and-marriages/statistics-births-deaths-and-marriages-registered#:~:text=statistics%20-%20births%2c%20deaths%20and%20marriages%20registered%20,of%20name%2086%2c010%20%2019%20more%20rows%20 https://www.wa.gov.au/organisation/department-of-justice/the-registry-of-births-deaths-and-marriages/statistics-births-deaths-and-marriages-registered#:~:text=statistics%20-%20births%2c%20deaths%20and%20marriages%20registered%20,of%20name%2086%2c010%20%2019%20more%20rows%20 https://doi.org/10.1371/journal.pone.0261509 http://doi.org/10.6027/r2022:2.1403-2503 https://doi.org/10.1553/populationyearbook2022.res1.6 https://www.guttmacher.org/report/continuing-impacts-covid-19-pandemic-findings-2021-guttmacher-survey-reproductive-health https://www.guttmacher.org/report/continuing-impacts-covid-19-pandemic-findings-2021-guttmacher-survey-reproductive-health https://doi.org/10.4054/demres.2020.43.47 https://doi.org/10.1371/journal.pone.0258133 https://doi.org/10.1007/s10680-020-09556-y https://population.gov.au/research/research-fertility https://doi.org/10.1016/j.janxdis.2020.102258 https://doi.org/10.5694/mja2.51010 https://www.census.gov/library/stories/2021/09/united-states-births-declined-during-the-pandemic.html https://www.census.gov/library/stories/2021/09/united-states-births-declined-during-the-pandemic.html https://doi.org/10.1016/j.fertnstert.2021.05.092 https://www.health.nsw.gov.au/hsnsw/pages/mothers-and-babies-2020.aspx https://www.health.nsw.gov.au/hsnsw/pages/mothers-and-babies-2020.aspx 30 gray, evans & reimondos australian population studies 6 (1) 2022 social relations. social science & medicine, 297, 114821. https://doi.org/10.1016/j.socscimed.2022.114821. qu, l. (2021). report no. 4: impacts of covid-19 on pregnancy and fertility intentions. australian institute of family studies. https://aifs.gov.au/publications/impacts-covid-19-pregnancy-and-fertilityintentions. queensland government (2022). life event statistics. https://www.qld.gov.au/law/births-deathsmarriages-and-divorces/data/life-event-statistics queensland health (2021). queensland perinatal statistics 2020. https://www.health.qld.gov.au/hsu/peri/peri_2020/queensland-perinatal-statistics-2020 satici, b., saricali, m., satici, s. a., & griffiths, m. d. (2020). intolerance of uncertainty and mental wellbeing: serial mediation by rumination and fear of covid-19. international journal of mental health and addiction. https://doi.org/10.1007/s11469-020-00305-0 sobotka, t., jasilioniene, a., galarza, a. a., zeman, k., nemeth, l., & jdanov, d. (2021). baby bust in the wake of the covid-19 pandemic? first results from the new stff data series. https://osf.io/preprints/socarxiv/mvy62/ sobotka, t., skirbekk, v., & philipov, d. (2011). economic recession and fertility in the developed world. population and development review, 37(2), 267–306. https://doi.org/10.1111/j.17284457.2011.00411.x. sweeny, k., andrews, s. e., nelson, s. k., & robbins, m. l. (2015). waiting for a baby: navigating uncertainty in recollections of trying to conceive. social science & medicine, 141, 123-132. https://doi.org/10.1016/j.socscimed.2015.07.031. tasmanian government (2022). births, deaths, and marriage. https://www.justice.tas.gov.au/bdm/statistics. retrieved 16 march 2022 vignoli, d., bazzani, g., guetto, r., minello, a., & pirani, e. (2020a). uncertainty and narratives of the future: a theoretical framework for contemporary fertility. in schoen, r (ed.) analyzing contemporary fertility (pp. 25-47). springer, cham. https://doi.org/10.1007/978-3-030-485191_3. vignoli, d., guetto, r., bazzani, g., pirani, e., & minello, a. (2020b). a reflection on economic uncertainty and fertility in europe: the narrative framework. genus, 76(1), 1-27. https://doi.org/10.1186/s41118-020-00094-3. wilde, j., chen, w., & lohmann, s. (2020). covid-19 and the future of us fertility: what can we learn from google? iza dp no. 13776 https://www.iza.org/publications/dp/13776/covid-19-and-the-futureof-us-fertility-what-can-we-learn-from-google wilson, t. (2017). comparing alternative statistics on recent fertility trends in australia. journal of population research, 34(2), 119-133. https://doi.org/10.1007/s12546-016-9176-x. wilson, t., mcdonald, p., & temple, j. (2020). the geographical patterns of birth seasonality in australia. demographic research, 43, 1185-1198. https://doi.org/10.4054/demres.2020.43.40. yerkes, m. a., andré, s. c., besamusca, j. w., kruyen, p. m., remery, c. l., van der zwan, r., ... & geurts, s. a. (2020). ‘intelligent’ lockdown, intelligent effects? results from a survey on gender (in)equality in paid work, the division of childcare and household work, and quality of life among parents in the netherlands during the covid-19 lockdown. plos one, 15(11), e0242249. https://doi.org/10.1371/journal.pone.0242249. https://doi.org/10.1016/j.socscimed.2022.114821 https://aifs.gov.au/publications/impacts-covid-19-pregnancy-and-fertility-intentions https://aifs.gov.au/publications/impacts-covid-19-pregnancy-and-fertility-intentions https://www.qld.gov.au/law/births-deaths-marriages-and-divorces/data/life-event-statistics https://www.qld.gov.au/law/births-deaths-marriages-and-divorces/data/life-event-statistics https://www.health.qld.gov.au/hsu/peri/peri_2020/queensland-perinatal-statistics-2020 https://doi.org/10.1007/s11469-020-00305-0 https://osf.io/preprints/socarxiv/mvy62/ https://doi.org/10.1111/j.1728-4457.2011.00411.x https://doi.org/10.1111/j.1728-4457.2011.00411.x https://doi.org/10.1016/j.socscimed.2015.07.031 https://www.justice.tas.gov.au/bdm/statistics https://doi.org/10.1007/978-3-030-48519-1_3 https://doi.org/10.1007/978-3-030-48519-1_3 https://doi.org/10.1186/s41118-020-00094-3 https://www.iza.org/publications/dp/13776/covid-19-and-the-future-of-us-fertility-what-can-we-learn-from-google https://www.iza.org/publications/dp/13776/covid-19-and-the-future-of-us-fertility-what-can-we-learn-from-google https://doi.org/10.1007/s12546-016-9176-x https://doi.org/10.4054/demres.2020.43.40 https://doi.org/10.1371/journal.pone.0242249 australian population studies 2022 | volume 6 | issue 1 | pages 31-36 © kippen 2022. published under the creative commons attribution-noncommercial licence 3.0 australia (cc by-nc 3.0 au). journal website: www.australianpopulationstudies.org by what age do people experience familial death, and how has that changed over time? rebecca kippen* monash university * email: rebecca.kippen@monash.edu. address: po box 666, bendigo, vic, 3552. paper received 10 may 2022; accepted 20 june 2022; published 25 july 2022 introduction at the beginning of the twentieth century, death was a familiar and frequent occurrence. average period life expectancy at birth in australia in 1901 was 54 years, and based on contemporary mortality rates (smith 2007), less than three-quarters of the population survived to age 40, with 10% of babies dying before their first birthday, and another 10% before age 30. consequently, it was common to experience the death of close family members from a young age. in early twenty-first-century australia, male and female life expectancy at birth are now both well above 80 years. based on 2018–20 australian life tables, more than 98% of people survive to their 40th birthday, and death is concentrated at very old ages, with more than half of deaths occurring over the age of 85 years (abs 2021a). as noted by jalland (2006 p. 4) “old age [has now] replaced infancy as the most likely time of death…parents today expect their children to survive at least to adulthood, and many people do not experience death until their elderly parents die.” the ages at which people experience the death of family members has changed enormously over the past century. this study aims to quantify this experience by considering two ‘average’ australians born 100 years apart, in 1901 and 2001, and employing relevant demographic measures to trace the ages by which they are likely to experience familial death, given their own survival. here, ‘familial death’ (or, death of a family member) is defined as the death of a parent, sibling, partner, or child. data and methods basic demographic assumptions are set out in table 1. for our average australian born in 1901 (hereafter referred to as ‘1901’), their mother was born 30 years earlier in 1871 and their father three years before that in 1868 (based on likely age patterns of partnership and fertility). it is assumed that 1901’s mother survived at least to their birth, and father at least to their conception (nine months before birth). australian wives born in 1871 who had children, and who survived their childbearing years, had an average of five children each (knibbs 1911; wickens 1921). if this average is assumed for 1901’s d e m o g ra p h ic http://www.australianpopulationstudies.org/ mailto:rebecca.kippen@monash.edu 32 kippen australian population studies 6 (1) 2022 table 1: demographic assumptions for family members (self, parents, siblings, partner, children), average australians born in 1901 and 2001 family member assumptions for 1901’s family assumptions for 2001’s family common assumptions self born in 1901. born in 2001. assumed to survive well into old age. for the demography of partner and children, 1901 and 2001 are assumed to be female. mother born in 1871. born in 1971. survives at least to 1901/2001’s birth. subsequent partnerships/marriages are not accounted for. father born in 1868. born in 1969. survives at least to 1901/2001’s conception. subsequent partnerships/marriages are not accounted for. siblings 4 siblings born 1897, 1899, 1903, 1905. 1 sibling, born in either 1999 or 2003. no multiple births. birth intervals of 2 years. each sibling is at risk of death from their birth. the birth of siblings after 1901/2001 is predicated on the survival of their mother to birth, and their father to conception. the sex ratio at birth for siblings (to calculate mortality) is assumed to be 1.05 males to 1 female. partner born in 1899. partnership/ marriage in 1925. born in 1999. partnership/ marriage in 2029. partner is male. survives at least to partnership/marriage. children 3 children born in 1927, 1929 and 1931. 2 children born in 2031 and 2033. the birth of children is predicated on the survival of their father to conception. subsequent partnerships/marriages are not accounted for. the sex ratio at birth for children (to calculate mortality) is assumed to be 1.05 males to 1 female. for parents, siblings, partner and children, annual age-sex-year-specific probabilities of dying are calculated from relevant australian life tables, 1901–2019, and projected probabilities assuming that life expectancy at birth increases to 91 years for males and 93 years for females in 2101. mortality in 1897–1900 is assumed to be the same as in 1901. mother, with no multiple births, birth intervals of two years, and 1901 being the middle child, then 1901’s four siblings were born in 1897, 1899, 1903 and 1905. each sibling is at risk of death from the time of birth. it is further assumed that 1901 is female, and that she marries in 1925 when she is aged 24 years to a man two years older (based on marriage patterns in carmichael 1988). assuming she has children, she has three—the median for her cohort (abs 2005; abs 1999)—born in 1927, 1929 and 1931. 1901’s husband is at risk of death from marriage, and her children, from birth. the sex ratio at birth for 1901’s siblings and children is assumed to be 105 males per 100 females. the birth of siblings after 1901 is predicated on survival of 1901’s mother to each birth, and 1901’s father to each conception. second and subsequent marriages/partnerships are not accounted for. a similar method is followed for the ‘average australian’ born in 2001 (hereafter known as ‘2001’). based on average ages at marriage and birth (carmichael 1988; kippen 2006; abs 2021c), her mother was born in 1971, and her father in 1969, and she is one of two children. future partnering and birth australian population studies 6 (1) 2022 kippen 33 rates by age are not known, but based on likely trends, 2001 is assumed to partner in 2029 with a man two years older than her, and have two children born in 2031 and 2033. both 1901 and 2001 are assumed to survive well into old age. mortality rates for family members are assumed to be independent. relevant national age-sex-year-specific mortality rates are applied to 1901 and 2001’s parents, siblings, partner, and children, from the relevant starting point for each, for each year over time. annual period single-year-of-age life tables were available from 1901 to 2019 (smith 2007; abs 2021a; abs 2021b). it was assumed that 1901 rates applied for the period 1897–99. for 2020 onwards, it was assumed that period life expectancy at birth will continue to increase to 91 years for males and 93 years for females in 2101, implying an average increase of around 1.2 years per decade for males, and 0.9 years per decade for females. projected male life expectancy is between the ‘high’ and ‘low’ male life expectancy assumptions made in the most recent australian population projections (abs 2018), and similar to projected australian male life expectancy in the most recent united nations population projections (united nations 2019). projected female life expectancy is in line with the ‘high’ australian projection (abs 2018), and slightly lower than the united nations projection (united nations 2019). male and female life tables for the given life expectancies in 2101 were sourced from the united nations extended model life tables, general pattern (united nations 2011). annual age-sex-specific mortality rates were interpolated between 2019 and 2101. period rates by single year of age were rearranged into cohort rates by age to trace cohort survival. all calculations were conducted using microsoft excel. key features figure 1 shows, for our typical australian born in 1901, the probability across age of having experienced the death of a parent, sibling, partner, or child. figure 2 shows the same results for our 2001-born australian. the differences are striking. by the age of 40, 1901 had a three-in-four chance that at least one parent was deceased—57% for her father, and 37% for her mother. she also had a 59% likelihood that a sibling had died, 5% for her husband, and 20% for a child. assuming independence of outcomes, 1901 had a 92% probability that a family member (parent, sibling, spouse, child) died before her 40th birthday, and an 83% probability that such a death occurred when she was between the ages of 5 and 40. the early ‘steps’ evident in the probabilities of death for siblings and children are due to the high risk of infant mortality for 1901’s siblings born in 1903 and 1905, and her children born in 1927, 1929 and 1931 (figure 1). in contrast, by age 40, 2001 has a three-in-four chance that both her parents are still alive. the likelihood that 2001’s father dies before 2001 reaches age 40 is 17%, and her mother, 9%. by age 40, 2001 has only a 2% probability that her sibling dies, 1% for her partner, and less than 1% for a child. the risk is only around one-quarter that, before the age of 40 (or between 5 and 40), 2001 will suffer the death of a parent, sibling, spouse or child (figure 2). 34 kippen australian population studies 6 (1) 2022 figure 1: cumulative probability of familial death (parent, sibling, partner, child) by age, average australian born in 1901 source: cohort age-sex probabilities of death calculated from smith (2007). notes: mother born in 1871; father in 1868; 4 siblings in 1897, 1899, 1903 and 1905; husband in 1899; children in 1927, 1929 and 1931. ≥ 1 family member: death of at least 1 family member (father, mother, sibling, partner, child); ≥ 1 family member after age 5: death of at least 1 family member (father, mother, sibling, partner, child) after 1901 reaches age 5 years; ≥ 1 parent: death of at least 1 parent; ≥ 1 sibling: death of at least 1 sibling; ≥ 1 child: death of at least 1 child. we can also consider the median ages by which 1901 and 2001 first experience the death of a family member; that is, the age at which the cumulative probability of death is 50%. on average, 1901 was almost 20 years younger than 2001 when she lost a parent (age 31 years versus 50 years) and her partner (70 years versus 88 years). the median age at which 1901 first lost a sibling was 23 years, compared to 93 years for 2001. 1901 could expect to experience the death of a child by age 88 years, whereas 2001’s children will likely all outlive her, even if she survives to her 100th birthday (figures 1–2). by age 50, the probability that 1901 experienced the death of a family member is almost 100%, whereas the probability for 2001 is around 50% (figures 1–2). in line with jalland’s statement in the introduction, this analysis indicates that the continuing postponement of death to older ages means that the majority of australians born at the turn of the millennium will not suffer a familial death—parent, sibling, partner, or child—until they themselves are well into middle age. for the cohort born a century earlier, death of family members from a young age was a much more common experience. australian population studies 6 (1) 2022 kippen 35 figure 2: cumulative probability of familial death (parent, sibling, partner, child) by age, average australian born in 2001 source: cohort age-sex probabilities of death calculated from smith (2007); abs (2021a); abs (2021b); united nations (2011). notes: mother born in 1971; father in 1969; 1 sibling in 1999 or 2003; partner in 1999; children in 2031 and 2033. ≥ 1 family member: death of at least 1 family member (father, mother, sibling, partner, child); ≥ 1 family member after age 5: death of at least 1 family member (father, mother, sibling, partner, child) after 2001 reaches age 5 years; ≥ 1 parent: death of at least 1 parent; ≥ 1 child: death of at least 1 child. limitations since single-year-of-age mortality rates by sex were only available annually for australia from 1901 to 2019, it was assumed that 1901 rates applied for the period before 1901 (to 1901’s siblings born in 1897 and 1899). mortality rates after 2019 were projected assuming continuing decline. if the actual future mortality decline is not as great, or death rates increase in future, then 2001 will be more likely to experience the death of close family members at a younger age. the calculations assume independence of mortality risk, whereas mortality is likely to cluster within families and time periods. these scenarios are a national average only. the demography of different sub-populations (e.g., mccalman et al. 2009) is not accounted for. acknowledgements the author thanks two anonymous reviewers whose expert and insightful suggestions improved this work. 36 kippen australian population studies 6 (1) 2022 references australian bureau of statistics. (1999). single-year-of-age fertility rates, australia, 1921–1997 [customised dataset]. australian bureau of statistics. (2005). women by age by children ever born, 1981 census [customised dataset]. australian bureau of statistics. (2018). population projections, australia, 2017 (base) – 2066. https://www.abs.gov.au/statistics/people/population/population-projections-australia australian bureau of statistics. (2021a). life tables, states, territories and australia, 2018–2020. https://www.abs.gov.au/statistics/people/population/life-tables australian bureau of statistics. (2021b). life tables, states, territories and australia, 2005–2007 to 2017– 2019. https://www.abs.gov.au/statistics/people/population/life-tables australian bureau of statistics. (2021c). births australia. https://www.abs.gov.au/statistics/people/population/births-australia carmichael, g. a. (1988). with this ring: first marriage patterns, trends and prospects in australia. australian family formation project monograph no.ll, department of demography, the australian national university, and australian institute of family studies. jalland, p. (2006). changing ways of death in twentieth-century australia: war, medicine, and the funeral business. university of new south wales press. kippen, r. (2006). the rise of the older mother. people and place, 14(3), 1–11. knibbs, g. h. (1917). the first commonwealth census, 1911. j. kemp, government printer. https://www.abs.gov.au/ausstats/abs@.nsf/detailspage/2112.01911 mccalman, j., smith, l., anderson, i., morley, r., mishra, g. (2009). colonialism and the health transition: aboriginal australians and poor whites compared, victoria, 1850–1985. history of the family, 14, 253-265. https://doi.org/10.1016/j.hisfam.2009.04.005 smith, l. (2007). australian demographic databank, 1901‒2003: deaths and population [dataset]. the australian national university. united nations. (2011). extended model life tables. https://www.un.org/en/development/desa/population/publications/mortality/model-lifetables.asp united nations. (2019). world population prospects 2019, life expectancy, australia, 1950–55 to 2095– 2100. https://population.un.org/wpp/dataquery/ wickens, c. h. (1927). census of the commonwealth of australia, 1921. h. j. green, government printer. https://www.abs.gov.au/ausstats/abs@.nsf/detailspage/2111.01921. https://www.abs.gov.au/statistics/people/population/population-projections-australia https://www.abs.gov.au/statistics/people/population/life-tables https://www.abs.gov.au/statistics/people/population/life-tables https://www.abs.gov.au/statistics/people/population/births-australia https://www.abs.gov.au/ausstats/abs@.nsf/detailspage/2112.01911 https://doi.org/10.1016/j.hisfam.2009.04.005 https://www.un.org/en/development/desa/population/publications/mortality/model-life-tables.asp https://www.un.org/en/development/desa/population/publications/mortality/model-life-tables.asp https://population.un.org/wpp/dataquery/ https://www.abs.gov.au/ausstats/abs@.nsf/detailspage/2111.01921 australian population studies 2020 | volume 4 | issue 1 | pages 37-56 © adair & lopez 2020. published under the creative commons attribution-noncommercial licence 3.0 australia (cc by-nc 3.0 au). journal website: www.australianpopulationstudies.org widening inequalities in premature mortality in australia, 2006-16 tim adair* the university of melbourne alan d lopez the university of melbourne * corresponding author. email: timothy.adair@unimelb.edu.au. address: melbourne school of population and global health, the university of melbourne, level 5, building 379, 207 bouverie st, carlton 3053 vic, australia. paper received 13 december 2019; accepted 11 may 2020; published 25 may 2020. abstract background increases in australia’s life expectancy have slowed since 2003. within this context, it is important to analyse whether socio-economic and geographic inequalities in premature mortality have changed in recent years. aim the aim is to measure the extent of and change in inequalities in premature mortality (under 75 years) by area socio-economic status and remoteness in australia from 2006-16. data and methods this study uses australian death registration data from 2006-16. we identify area socio-economic quintile (q1 lowest to q5 highest) and remoteness (major cities, inner regional, outer regional/ remote/very remote) for each decedent. age-standardised death rates (asdr) at ages 35-74 and 0-34 years are calculated and smoothed over time. results asdr decline at ages 35-74 years has slowed markedly, and from 2011-16 q1 asdrs stagnated. the q1:q5 asdr ratio for females rose from 2011-16 for males (1.97 to 2.11) and females (1.78 to 1.98), and also widened in the q2:q5 and q3:q5 comparisons for females. asdrs outside major cities stagnated from 2011-16 and inequalities compared with major cities widened to be 40% higher in 2016. asdr declines at 0-34 years were quicker than for 35-74 years and inequalities widened slightly. conclusions slowing declines in premature mortality in australia are accompanied by substantial and widening socioeconomic and geographic inequalities. without reductions in mortality in lower socio-economic groups and outside capital cities, future life expectancy growth in australia will likely be hampered. key words mortality, premature mortality, inequalities, socio-economic, australia. http://www.australianpopulationstudies.org/ mailto:timothy.adair@unimelb.edu.au 38 adair & lopez australian population studies 4 (1) 2020 1. introduction from 1970 to the early 2000s, australian life expectancy increased more rapidly than most other highincome countries to become among the highest in the world (lopez and adair 2019a). this trend was primarily driven by large declines in mortality from cardiovascular diseases and other causes strongly related to smoking (taylor et al. 2006, adair et al. 2011, taylor et al. 2012). however, since 2003 australia’s life expectancy gains have slowed, both compared with preceding decades but also to other high-income countries, largely due to slowing mortality decline among males aged 45-64 and from deaths due to cardiovascular diseases (o’flaherty et al. 2012, hayes et al. 2017, gbd 2017 mortality collaborators 2018, lopez and adair 2019a, lopez and adair 2019b). indeed, since 2013, life expectancy for both males and females in australia has stagnated. investigation of the reasons for slowing mortality decline typically focuses on identifying the contributions from specific age groups, birth cohorts, causes of death, and risk factors. while such analyses can be very informative, it is also possible that inequalities in mortality risk according to socioeconomic status or place of residence have widened, with slowing declines or even increases in mortality rates in some population sub-groups being sufficient to prevent continued improvements in overall life expectancy. for example, in the usa, where the slowdown in life expectancy has been particularly acute, and has not increased for males since 2010, inequalities in life expectancy by county and education have widened; indeed, life expectancy at birth has declined in a number of counties since 1980 and adult mortality rates have increased among low-educated whites (case and deaton 2015, dwyer-lindgren et al. 2017, gbd 2017 mortality collaborators 2018). in australia, previous analyses have focussed on mortality inequalities by socio-economic status and place of residence. analysis across all ages in 2015 revealed that people living in the lowest socioeconomic quintile areas had age-standardised mortality rates almost 50% higher than those living in the richest quintile areas (australian institute of health and welfare 2018). another study of mortality at 20 years and over from 2001-07 showed the relative risk of mortality to be 88% higher in the lowest versus highest individual income quintile, compared with 32% higher in the equivalent area socioeconomic quintile. the same study also showed 63% higher mortality risk for those with 12 or less years of education versus those with more than 12 years (clarke and leigh 2011). mortality differences according to individual education level were even higher. based on registered deaths at ages 25-84 years in 2011-12 linked to 2011 census data, mortality rates of males who had not completed year 12 nor post-secondary education were 120% higher than those who had completed a bachelor’s degree or higher, and for females were 64% higher (korda et al. 2019). by region of residence, people living in very remote areas had mortality rates 40% higher than those living in capital cities, 30% higher than those living in remote areas and 20% higher than people living in outer regional areas (australian institute of health and welfare 2018). earlier studies had shown that both area socio-economic and regional inequalities in mortality were higher under age 65 years than at older ages (draper et al. 2004, australian institute of health and welfare 2014, korda et al. 2019). inequalities in life expectancy between indigenous and non-indigenous australians were particularly large, being 10.6 years for males and 9.5 years for females in 2010-12 (australian institute of health and welfare 2018). less has been written about trends in inequalities, despite the obvious policy relevance and concern about the impact of social and economic policies and structural changes in society. from 1986 to 2002, australian population studies 4 (1) 2020 adair & lopez 39 annual declines in what has been defined as ‘avoidable mortality’ were greater in higher (5.0% per annum) than lower socio-economic groups (3.5% per annum), contributing to widening inequality in survival (korda et al. 2007). cardiovascular disease mortality declines from 1979-2006 were slightly quicker in higher than lower area socio-economic quintiles (page et al. 2012). not all evidence points to widening health inequalities, however: a study of life expectancy differences by area socio-economic quintile in new south wales over the period 2001-2012 found no increase in the relative advantage of the better off (stephens et al. 2017). trends in health and mortality inequalities in australia in recent years are of considerable interest given the widening inequalities in wealth, and to a lesser extent, income, that have been reported since the early 2000s (abs 2019). notably, multi-country analysis has revealed that country-level cardiovascular disease mortality increases with greater income inequality, controlling for socio-economic factors (dewan et al. 2019). in the context of this slowdown in life expectancy increases, and widening inequalities in wealth and income, it is informative to investigate whether inequalities in mortality rates have widened or not over the past decade or so since 2006. this study uses australian death registration data from 2006 to 2016 to analyse trends in inequalities in premature mortality, taken here to refer to deaths under the age of 75 years, given that the literature review showed that these are the ages where slowing mortality decline is most pronounced and where inequalities are greatest. inequalities are measured by area socio-economic status, remoteness, area socio-economic status within major cities, and remoteness within states. 2. data and methods data for this study were extracted from the australian cause of death unit record file (cod urf). the cod urf is a de-identified dataset of all deaths registered in australia (australian coordinating registry 2019). the dataset comprises death records collected by each state and territory registry of births, deaths and marriages and the state and chief coroners through the national coronial information system (ncis) and compiled by the australian bureau of statistics (abs) (australian coordinating registry 2019). the cod urf is maintained by the australian coordinating registry (acr) based at the queensland registry of births, deaths and marriages. ethics approval for the study was provided by the university of melbourne medicine and dentistry human ethics subcommittee (ethics id 1953560). we extracted cod urf data for deaths registered between 2006 and 2017, and which occurred between 2006 and 2016; all analyses were conducted according to year of death occurrence. we did not analyse deaths that occurred in 2017 nor deaths by year of registration because of an unusually high number of late registrations in late 2017; that is, there was a sharp decline in the number of deaths reported to have occurred during october to december 2017 compared with what would be expected based on the number of deaths in the preceding nine months of that year. as a result, there was a much lower number of deaths registered in 2017 than in preceding years. the cod urf contains information about the characteristics of the death and decedent (australian coordinating registry 2019). this includes the age at death, sex, place of usual residence and cause of death, but no socio-economic information about the decedent. place of usual residence was reported according to statistical local area (sla) for deaths with registration years 2006-10 and for 40 adair & lopez australian population studies 4 (1) 2020 statistical area level 2 (sa2) for deaths with registration years 2011-17 (abs 2006, abs 2016). deaths can be linked using sla or sa2 level to area-level socio-economic indicators and geographic indicators. we used the abs’s socio-economic indexes for areas (seifa) index of relative socioeconomic advantage and disadvantage (irsad) as the area-level socio-economic indicator (seifa) (abs 2008, abs 2013b) the irsad is constructed from census variables covering income, education, housing and employment status. the abs produce other measures of area-level socio-economic indicators, including the index of relative socio-economic disadvantage (irsd), but it is recommended to use the irsad in studies which seek to take into account residents experiencing advantage as well as those experiencing disadvantage, rather than just disadvantage (abs 2013b). given that we are measuring mortality experience of all decedents in each area (i.e. those experiencing advantage and disadvantage), we believe that irsad is an appropriate measure. we used the irsad scores to derive irsad quintiles of equal total population size. quintile 1 (q1) refers to the population with the lowest socioeconomic status as defined by irsad, while quintile 5 (q5) denotes the population with the highest socioeconomic status as defined by this measure. the abs releases a new version of the irsad after each census. recent irsads include 2006, which is measured at the sla level, and 2011, which is measured at the sa2 level (abs 2008, abs 2013b). we hence use irsad 2006 for deaths with registration years 2006-10 where usual residence of the decedent was available for slas and irsad 2011 for deaths with registration years 2011-17 where usual residence was available for sa2s. the irsad 2006 and 2011 are derived from different censuses and are also based on different variables (21 variables in 2006, 25 in 2011). additionally, there were approximately 1,400 slas in australia during 2006-10 and approximately 2,100 sa2s in 2011-17, so it is highly likely that an increase in the number of areas, and corresponding reduction in population size per area, would increase the likelihood that the irsad quintile would be the same as irsad if it theoretically was measured at the individual level. therefore, it is very possible that inequalities in mortality measured according to irsad would widen from the sla-derived (2006) to the sa2-derived (2011) measure, even if true inequalities did not. hence, irsad 2006 and irsad 2011 are not directly comparable and so trends in mortality by irsad can only be analysed separately for the period 200610 and 2011-16. (norman et al. 2016, abs 2013b). however, these analyses will still provide valuable insights into trends in inequalities in mortality from 2006-16. a new statistical geography was released by the abs in 2016 which includes additional sa2s. we used abs correspondence files linking 2011 and 2016 sa2s to determine the irsad 2011 for each new sa2 (abs 2016). many new sa2s were wholly derived from a single original sa2. however other new sa2s were derived from multiple original sa2s. for these new sa2s the irsad decile was that which comprised the highest proportion of the population based on a calculation of the irsad decile of each original sa2 weighted by the fraction of the new sa2’s population that was derived from each original sa2. irsad 2016 is also available however it is measured based on the new sa2s, and because 2011-15 mortality data are based on the original sa2s then irsad 2011 is more appropriate to measure area socio-economic status over the period of analysis (abs 2018a). we also analysed geographic inequalities in mortality using the abs remoteness classification, which classifies areas as major cities, inner regional, outer regional, remote or very remote. the 2006 version of remoteness is measured for each sla; however, the 2011 and 2016 versions of remoteness are measured at the sa1 level (i.e. smaller areas than sa2) (abs 2011, 2013a). the abs australian population studies 4 (1) 2020 adair & lopez 41 has not published populations for sa1s, so we cannot calculate the 2011 and 2016 versions of remoteness at the sa2 level. we hence used the 2006 version of remoteness for all years of analysis. we used abs correspondence files to convert remoteness from sla to sa2 by calculating the percentage of each sa2 population classified as each remoteness category (weighted by the populations of the slas from which each sa2 was derived from), and categorising remoteness for each sa2 as its largest remoteness category (abs 2011). as with irsad, we used sa2 2011-2016 correspondences to categorise remoteness for new sa2s. we conduct our analyses for the following groups: major cities (70% of population during 2006-16), inner regional (19%), and outer regional, remote and very remote combined (11%). additional analyses were conducted by irsad within major cities. three irsad groups were used: the four lowest irsad deciles (grouped based on all australia), the 5th to 7th, and 8th to 10th. to provide more geographic specificity, we also analysed mortality by state/territory and remoteness. in wa and sa outer regional, remote and very remote were combined, and we did not disaggregate tasmania, act and nt. population data were obtained from the abs (2018b). death rates were analysed for two age groups – 0-34 years and 35-74 years – in order to distinguish between ages where non-communicable diseases are particularly prominent and those where injuries, infectious diseases and childhood diseases are relatively more common. deaths within each age group were age-standardised using the global burden of disease population age standard (gbd mortality collaborators 2017). age-standardised death rates (asdr) for data from 2006-16 (including for all mortality) were smoothed over time using localised polynomial smoothing with stata 15.0 (fan & gijbels 1996, statacorp lp 2017). localised polynomial smoothing of asdr trends for the periods 2006-10 and 2011-16 provided unstable results, so a linear regression of asdrs with covariate of year was used instead. all results are presented with 95% confidence intervals and all charts present asdr point estimates. ratios of smoothed asdrs were calculated with the reference category being that with the lowest asdr, again with 95% confidence intervals. for the state and remoteness analysis, the ratios were calculated compared to the asdr for all australia for that sex. for each category, we also measured the annualised rate of change in smoothed asdr during the period 2006 to 2011 (or 2006-10 where irsad is used) and 2011 to 2016. 3. results 3.1. 35-74 years: area socio-economic quintiles the decline in asdrs at ages 35-74 years in australia has slowed markedly, from 1.9% per annum in 2006-11 to 1.0% in 2011-16 for males, and from 1.6% to 1.0% for females (table a1 – see appendix). during 2014-16, there was even a stagnation in male asdrs. over the more recent period 2011-16, the asdr at these ages did not change at all for males nor females in q1, the lowest socioeconomic group (figure 1, table a2). from 2011-16, inequalities in mortality have widened for the lowest socioeconomic group compared with the highest socioeconomic group for males (1.97 to 2.11, or 14%) and females (1.78 to 1.98, or 26%); that is, for each sex, people living in the lowest area socioeconomic quintile had premature mortality rates twice as high as those in the highest quintile. there 42 adair & lopez australian population studies 4 (1) 2020 was also a widening in the q2:q5 and q3:q5 comparisons for females, and this caused them to reach approximately the same level in 2016 as the respective ratios for males. during 2006-10, inequalities based on irsad 2006 were much narrower than for 2011-16 that was based on irsad 2011, and there was no clear increase in asdr ratios for males while increases for q1:q5, q2:q5 and q4:q5 for females. it should again be emphasised that mortality by area socio-economic quintile in 2006-10 and 2011-16 are not directly comparable. figure 1: asdr (per 100,000) by sex and area socio-economic quintile: trends and ratios (versus q5), 35-74 years, australia, 2006-10 and 2011-16 source: abs; author’s calculations notes. shaded area is 95% confidence interval. dots are point estimates. 2006 and 2010 measured using irsad 2006, 2011 and 2016 measured using irsad 2016. 2006 and 2010 results cannot be directly compared with 2011 and 2016 results australian population studies 4 (1) 2020 adair & lopez 43 3.2. 35-74 years: remoteness asdrs for the population living in outer regional, remote and very remote areas of australia were 40% higher in 2016 than for the population living in major cities, and this excess mortality has been rising since 2006, more so for women (figure 2, table a1). this is due to declines in mortality at ages 35-74 years being more evident in the major cities (2011-16 1.3% per annum decline for males and 1.4% per annum decline for females) than inner regional, outer regional, remote and very remote areas in recent years, where there has been no clear decline. figure 2: asdr (per 100,000) by sex and remoteness: trends and ratios (versus q5), 35-74 years, australia, 2006-16 source: abs; author’s calculations notes. shaded area is 95% confidence interval. dots are point estimates. outer reg.: outer regional. v. remote: very remote. 44 adair & lopez australian population studies 4 (1) 2020 3.3. 35-74 years: major cities area socio-economic deciles for the population living in capital cities, inequalities in mortality have increased slightly during 201116, after being steady or declining slightly in 206-10 (figure 3, table a2). the asdr ratio was 79% higher for males and 70% higher for females in the 1-4 (lowest socio-economic status) versus 8-10 (highest) deciles, and about one-third higher for both sexes in the 5-7 (middle) versus 8-10 deciles. notably, the annualised decline in asdrs for the population in deciles 1-4 was very slow during 201116 for both males and females, while there was no slowdown in the rate of mortality decline in the highest deciles. figure 3: asdr (per 100,000) by sex and major cities area socio-economic decile group (1-4, 5-7, 8-10): trends and ratios (versus 8-10), 35-74 years, australia, 2006-10 and 2011-16 source: abs; author’s calculations notes. shaded area is 95% confidence interval. dots are point estimates. 2006 and 2010 measured using irsad 2006, 2011 and 2016 measured using irsad 2016. 2006 and 2010 results cannot be directly compared with 2011 and 2016 results. australian population studies 4 (1) 2020 adair & lopez 45 3.4. 35-74 years: state & remoteness significant inequalities in premature mortality also exist by state and remoteness. major cities in nsw, vic, qld and wa, as well as the act, had the lowest asdrs (figure 4). conversely, the highest excess mortality compared with the national average was reported for the nt (ratio compared to australia in 2016: 1.58 for men and 2.01 for women), hobart, and nsw and qld outer regional, remote and very remote areas (table a1). over the decade from 2006-16, there was a noticeable widening in the asdr ratio in many locations in australia, notably nsw outside major cities (males and females), vic outside major cities (females), qld inner regional (males and females), sa outside major cities (males), wa outside major cities (males), and tas (males), while there was a narrowing for nt males. figure 4: ratio of state & remoteness asdr to australia asdr by sex, 35-74 years, australia, 2016 source: abs; author’s calculations notes. capped vertical line is 95% confidence interval. or: outer regional. r: remote. vr: very remote. 3.5. 0-34 years: area socio-economic quintile while the vast majority of deaths in australia now occur beyond age 35, with comparatively low levels of child mortality, inequalities in survival chances for children and young adults should be a policy priority. in australia, declines in asdrs at ages 0-34 years were quicker than at 35-74 years, being 2.2% per annum for males from 2011-16 and 2.9% per annum for females (figure 5, table a4). asdr ratios q1:q5 and q2:q5 at ages 0-34 years in 2016 were only slightly wider than for adults aged 35-74 years after being somewhat higher in 2011, because there had not been the clear widening in inequalities as found for 35-74 years. 0.6 0.8 1.0 1.2 1.4 1.6 1.8 2.0 2.2 2.4 r at io n sw m aj or c iti es n sw in ne r r eg io na l n sw o r / r / vr vi c m aj or c iti es vi c in ne r r eg io na l vi c o r / r / vr q ld m aj or c iti es q ld in ne r r eg io na l q ld o r / r / vr sa m aj or c iti es sa o th er w a m aj or c iti es w a o th er ta s n t ac t state & remoteness male 0.6 0.8 1.0 1.2 1.4 1.6 1.8 2.0 2.2 2.4 r at io n sw m aj or c iti es n sw in ne r r eg io na l n sw o r / r / vr vi c m aj or c iti es vi c in ne r r eg io na l vi c o r / r / vr q ld m aj or c iti es q ld in ne r r eg io na l q ld o r / r / vr sa m aj or c iti es sa o th er w a m aj or c iti es w a o th er ta s n t ac t state & remoteness female 46 adair & lopez australian population studies 4 (1) 2020 figure 5: asdr (per 100,000) by sex and area socio-economic quintile: trends and ratios, 0-34 years, australia, 2006-10 and 2011-16 source: abs; author’s calculations notes. shaded area is 95% confidence interval. dots are point estimates. 2006 and 2010 measured using irsad 2006, 2011 and 2016 measured using irsad 2016. ref.: reference category. 3.6. 0-34 years: remoteness outer regional, remote and very remote areas have a much higher asdr ratio versus major cities at ages 0-34 (1.79 males, 2.04 females) when compared with 35-74 years (1.41 males, 1.40 females) (figure 6, table a3). a similar differential is also found for inner regional areas. for females, the outer regional, remote and very remote to major cities ratio widened from 1.65 to 2.04 and inner regional to major cities ratio from 1.38 to 1.58 over the period, due to very quick declines in mortality in major cities. male inequalities did not exhibit a consistent trend during the period and the net change was small, especially when accounting for the wide 95% confidence intervals. australian population studies 4 (1) 2020 adair & lopez 47 figure 6: asdr (per 100,000) by sex and remoteness: trends and ratios, 0-34 years, australia, 2006-16 source: abs; author’s calculations notes. shaded area is 95% confidence interval. dots are point estimates. outer reg.: outer regional. v. remote: very remote. 4. discussion australia has enjoyed very substantial declines in mortality since the early 1970s, largely due to the impact of public health policies to control major cardiovascular diseases, chronic respiratory diseases and other smoking-related causes (lopez and adair 2019a). however, in recent years these declines have slowed substantially and, as this study shows, are accompanied by substantial and widening socioeconomic and geographical inequalities in risk of death. at ages 35-74, death rates in the lowest area socio-economic quintile are double those in the highest quintile, with the gap widening by over 26% for females and 14% for males from 2011-16. a similar trend for females is apparent for other quintiles of socioeconomic status, if less extreme. populations living in outer regional, remote and 48 adair & lopez australian population studies 4 (1) 2020 very remote areas have death rates that are about 40% higher than in major cities, and the gap has been increasing. notably, death rates at ages 35-74 among the lowest area socio-economic quintile and outside of major cities have not fallen since 2011. the increase in asdr ratios by state and remoteness from 2006-16, compared with the national level, has only occurred outside major cities. even within major cities, there has been a widening in inequalities in death rates between lower and higher area socio-economic deciles, but notably no slowdown in the rate of mortality decline in the highest deciles. importantly, there was a stagnation in overall mortality rates between 2014-16 at ages 35-74 years, although longer-term trends suggest it is the lowest quintiles that are most affected. at ages less than 35 years, inequalities did not widen to the extent of 35-74 years; they were substantially higher earlier in the period but were only slightly higher in 2016. the recent widening of mortality inequalities in australia are a continuation of a similar trend in inequalities in avoidable mortality from 1986 to 2002, however the period covered by this study is typified by vastly slower, and even very recently stagnating, mortality decline than the earlier period (korda et al. 2007). in this context, the widening of mortality differentials in australia by socioeconomic group, and by area of residence, should be a matter for immediate policy action. while the causes of death contributing to these unfavourable trends are beyond the scope of this analysis, it is likely that much of the explanation, at least for the socioeconomic differentials, lies with differential risk factor exposure and health behaviours in the various socioeconomic quintile groups. in particular, risk factors such as smoking, alcohol, obesity, diet and fasting plasma glucose levels are the primary drivers of ischaemic heart diseases, diabetes, stroke, cancers, chronic respiratory disease, road traffic accidents and suicide, all of which are among the leading causes of death in australia (gbd 2017 causes of death collaborators 2018). there is consistent evidence to suggest that these adverse health behaviours are much more prevalent among lower than higher socioeconomic groups of australian society, with the latter having been the most responsive to health promotion messages about their hazards more than others. for example, smoking prevalence is 2.7 times higher and obesity is 1.6 times higher in the lowest compared with highest socioeconomic group (australian institute of health and welfare 2018). improving the efficacy of health promotional campaigns among the lower socioeconomic groups in australia is clearly important if these inequalities are to be effectively addressed and overall levels of mortality reduced. for geographic discrepancies in mortality risk, inequalities in access to health care, as well as the availability of essential trauma and acute care services, are likely to also play an important role in the excess mortality risks observed in rural and regional populations, in addition to higher exposure to key risk factors. for example, levels of overweight/obesity, no or low level of physical exercise and lifetime risky alcohol consumption, all important determinants of cardiovascular disease, diabetes, liver cirrhosis and certain forms of cancer, were significantly higher in outer regional and remote populations than in capital cities (australian institute of health and welfare 2018). a limitation of this study is that it uses an area-level rather than individual level measure of socioeconomic status. hence, it is likely that it underestimates the true extent of socioeconomic inequalities in mortality risk. a study of mortality at 20 years and over from 2001-07 showed the relative risk of mortality to be 88% higher in the lowest versus highest individual income quintile; this is lower than what our analysis found in 2016. however, inequalities have widened since that period and that analysis also included deaths over 75 years, an age where inequalities are not as great as at australian population studies 4 (1) 2020 adair & lopez 49 younger ages (clarke and leigh 2011, australian institute of health and welfare 2014). a study comparing mortality at ages 25-84 years by individual education level (not completed year 12 nor post-secondary education versus bachelor’s degree or higher) showed inequalities to be twice as high among males than females, a large contrast to the similar inequalities by sex found in our study (korda et al. 2019). this demonstrates how individual measures of inequality can reveal more granular information that area-level measures, in particular the benefits of increased education on mortality risk for males compared with females. another issue was there was no consistent measure of area socio-economic quintile from 2006 to 2016. the irsad used from 2011-16 showed much larger and increasing inequalities in mortality than the irsad used during 2006-10. these differences very likely reflect an improved measure of socioeconomic status, although the increased speed of asdr ratios in the later period are also highly likely to reflect genuine trends. 5. conclusion the widening inequalities presented in this study, if they continue, are likely to be a key factor in hampering future life expectancy growth in australia. these trends are particularly concerning given the significant slowdown in mortality declines at ages 35-74 years in recent years, and the concurrent widening of wealth and income inequalities in australia. reducing inequalities in mortality is likely to require a substantial shift in public policy orientation, with much more emphasis on adapting interventions to the socioeconomic context and stressors that are common among the lower socioeconomic groups. these derive from occupation, income, housing, transport, and financial pressures, to name a few, but it is unlikely that reducing their impact on survival will happen quickly. reducing risk factor exposure among lower socio-economic groups and improved access to secondary prevention and treatment in these groups, and outside of capital cities, is key, but is unlikely to happen unless these broader structural determinants of health that drive individual health behaviour are also addressed. key messages • mortality decline in australia at 35-74 years has slowed substantially in recent years. • in the poorest area socio-economic quintile, mortality rates stagnated from 2011 to 2016 and inequalities widened compared with the highest quintile. • mortality rates outside major cities also were steady from 2011 to 2016 and inequalities also increased compared with major cities. • australia’s future life expectancy growth will be hindered without further reductions in mortality in the lowest socio-economic groups and outside major cities. acknowledgements the authors wish to thank the registries of births, deaths and marriages, the coroners and the national coronial information system for enabling cod urf data to be used for this publication. the authors also wish to acknowledge the assistance of surender pandey in assisting with the literature review for this study. this research received no specific grant from any funding agency. 50 adair & lopez australian population studies 4 (1) 2020 references adair t, hoy d, dettrick z, and lopez a d (2011) reconstruction of long-term tobacco consumption trends in australia and their relationship to lung cancer mortality. cancer causes & control 22(7): 10471053. abs (2006) australian standard geographic classification (asgc), 2006. catalogue no. 1216.0. canberra: abs. abs (2008) census of population and housing: socio-economic indexes for areas (seifa), australia, 2006. catalogue no. 2033.0.55.001. canberra: abs. abs (2011) australian standard geographical classification (asgc) remoteness area correspondences, 2006. catalogue no. 1216.0.15.003. canberra: abs. abs (2013a) australian statistical geography standard (asgs): volume 5 remoteness structure. catalogue no. 2033.0.55.005. canberra: abs. abs (2013b) census of population and housing: socio-economic indexes for areas (seifa), australia, 2011. catalogue no. 2033.0.55.001. canberra: abs. abs (2016) australian statistical geography standard (asgs): volume 1 main structure and greater capital city statistical areas, july 2016. catalogue no. 1270.0.55.001. canberra: abs. abs (2018a) census of population and housing: socio-economic indexes for areas (seifa), australia, 2016. catalogue no. 2033.0.55.001. canberra: abs. abs (2018b) population by age and sex, regions of australia. catalogue no. 3235.0. canberra: abs. abs (2019) household income and wealth, australia, 2017-18. catalogue no. 6523.0. canberra: abs. australian coordinating registry (2019) cause of death unit record file, 2006-17. brisbane: australian coordinating registry. australian institute of health and welfare (2014) mortality inequalities in australia, 2009-11. catalogue no. aus 184. canberra: australian institute of health and welfare. australian institute of health and welfare (2018) australia's health 2018. catalogue no. aus 221. canberra: australian institute of health and welfare. case a and deaton a (2015) rising morbidity and mortality in midlife among white non-hispanic americans in the 21st century. proceedings of the national academy of sciences of the united states of america 112(49): 15078-15083. clarke p and leigh a (2011) death, dollars and degrees: socio-economic status and longevity in australia. economic papers 30(3): 348-355. dewan p, rorth r, jhund p s, et al. (2019) income inequality and outcomes in heart failure: a global between-country analysis. jacc heart failure 7(4): 336-346. draper g, turrell g, and oldenburg b (2004) health inequalities in australia: mortality. category no. phe 55. canberra: queensland university of technology and the australian institute of health and welfare. dwyer-lindgren l, bertozzi-villa a, stubbs r, et al. (2017) inequalities in life expectancy among us counties, 1980 to 2014: temporal trends and key drivers. jama internal medicine 177(7): 10031011. fan j and gijbels i (1996) local polynomial modelling and its applications. london: chapman & hall. gbd 2017 causes of death collaborators (2018) global, regional, and national age-sex-specific mortality for 282 causes of death in 195 countries and territories, 1980–2017: a systematic analysis for the global burden of disease study 2017. lancet 392: 1736–1788. gbd 2017 mortality collaborators (2018) global, regional, and national age-sex-specific mortality and life expectancy, 1950–2017: a systematic analysis for the global burden of disease study 2017. lancet 392: 1684–1735. australian population studies 4 (1) 2020 adair & lopez 51 gbd mortality collaborators (2017) global, regional, and national under-5 mortality, adult mortality, agespecific mortality, and life expectancy, 1970-2016: a systematic analysis for the global burden of disease study 2016. lancet 390(10100): 1084-1150. hayes, a j, lung t w, bauman a ,and howard k (2017) modelling obesity trends in australia: unravelling the past and predicting the future. international journal of obesity 41(1): 178-185. korda r j, biddle n, lynch j, et al. (2019) education inequalities in adult all-cause mortality: first national data for australia using linked census and mortality data. international journal of epidemiology dyz191, https://doi.org/10.1093/ije/dyz191. korda r j, butler j r, clements m s, and kunitz s j (2007) differential impacts of health care in australia: trend analysis of socioeconomic inequalities in avoidable mortality. international journal of epidemiology 36(1): 157-165. lopez a d and adair t (2019a) slower increase in life expectancy in australia than in other high income countries: the contributions of age and cause of death. medical journal of australia 210(9): 403408. lopez a d and adair t (2019b) is the long-term decline in cardiovascular-disease mortality in high-income countries over? evidence from national vital statistics. international journal of epidemiology 48(6): 1815-1823. norman p, charles-edwards e, and wilson t (2016) relationships between population change, deprivation change and health change at small area level: australia 2001–2011. in: wilson t, charles-edwards e, and bell m (eds) demography for planning and policy: australian case studies. basel: springer. o'flaherty, m, allender s, taylor r, et al. (2012) the decline in coronary heart disease mortality is slowing in young adults (australia 1976-2006): a time trend analysis. international journal of cardiology 158(2): 193-198. page a, lane a, taylor r, and dobson a (2012) trends in socioeconomic inequalities in mortality from ischaemic heart disease and stroke in australia, 1979-2006. european journal preventive cardiology 19(6): 1281-1289. statacorp lp (2017) stata/se 15.0. college station tx: statacorp lp. stephens a s, gupta l, thackway s, and broome r a (2017) socioeconomic, remoteness and sex differences in life expectancy in new south wales, australia, 2001-2012: a population-based study. bmj open 7(1): e013227. taylor r, dobson a, and mirzaei m (2006) contribution of changes in risk factors to the decline of coronary heart disease mortality in australia over three decades. european journal of cardiovascular prevention and rehabilitation 13(5): 760-768. taylor r, page a, and danquah j (2012) the australian epidemic of cardiovascular mortality 1935-2005: effects of period and birth cohort. journal of epidemiology and community health 66(7): e18. https://doi.org/10.1093/ije/dyz191 52 adair & lopez australian population studies 4 (1) 2020 appendix table a1: asdr by sex, remoteness and state and remoteness: annualised change (%) and ratios, 35-74 years, australia, 2006-16 males annualised change (%) ratio to reference category 2006-11 2011-16 2006 2011 2016 australia -1.9 (-2.3 – -1.6) -1.0 (-1.4 – -0.6) remoteness major cities -1.9 (-2.4 – -1.3) -1.3 (-1.8 – -0.7) ref. ref. ref. inner regional -1.8 (-2.3 – -1.2) -0.2 (-0.8 – +0.4) 1.12 (1.08 – 1.16) 1.12 (1.10 – 1.15) 1.19 (1.14 – 1.23) outer regional/ remote/ very remote -1.4 (-1.7 – -1.0) -0.2 (-0.6 – +0.1) 1.31 (1.27 – 1.34) 1.34 (1.32 – 1.36) 1.41 (1.37 – 1.45) state and remoteness nsw major cities -1.8 (-2.3 – -1.2) -1.2 (-1.8 – -0.6) 0.94 (0.91 – 0.97) 0.95 (0.93 – 0.96) 0.94 (0.91 – 0.97) nsw inner regional -1.4 (-2.4 – -0.4) +0.1 (-1.0 – +1.2) 1.06 (1.01 – 1.11) 1.08 (1.06 – 1.11) 1.15 (1.09 – 1.21) nsw outer regional/ remote/ very remote -0.8 (-1.5 – -0.2) +0.3 (-0.4 – 1.0) 1.25 (1.20 – 1.29) 1.32 (1.29 – 1.34) 1.40 (1.35 – 1.46) vic major cities -2.2 (-2.8 – -1.7) -1.7 (-2.3 – -1.0) 0.90 (0.87 – 0.92) 0.88 (0.87 – 0.90) 0.85 (0.82 – 0.88) vic inner regional -2.6 (-3.8 – -1.4) -0.9 (-2.2 – +0.5) 1.06 (1.00 – 1.12) 1.02 (0.99 – 1.06) 1.03 (0.96 – 1.10) vic outer regional/ remote/ very remote +0.9 (-1.1 – +3.0) -2.2 (-4.2 – -0.1) 1.04 (0.94 – 1.14) 1.20 (1.14 – 1.26) 1.13 (1.01 – 1.25) qld major cities -1.6 (-2.7 – -0.4) -1.3 (-2.6 – -0.1) 0.93 (0.88 – 0.98) 0.95 (0.92 – 0.98) 0.93 (0.88 – 0.99) qld inner regional -0.9 (-1.7 – -0.2) -0.2 (-1.0 – +0.6) 0.99 (0.95 – 1.03) 1.04 (1.02 – 1.06) 1.08 (1.04 – 1.13) qld outer regional/ remote/ very remote -1.7 (-2.6 – -0.8) -0.8 (-1.7 – +0.2) 1.19 (1.14 – 1.24) 1.20 (1.17 – 1.23) 1.22 (1.16 – 1.28) sa major cities -1.9 (-2.8 – -0.9) -1.5 (-2.5 – -0.4) 1.01 (0.96 – 1.05) 1.01 (0.98 – 1.03) 0.98 (0.93 – 1.04) sa other -1.2 (-2.3 – -0.2) +1.0 (0.0 – 2.1) 1.03 (0.98 – 1.08) 1.07 (1.04 – 1.10) 1.18 (1.12 – 1.25) wa major cities -1.4 (-2.1 – -0.7) -0.5 (-1.2 – +0.3) 0.89 (0.86 – 0.92) 0.91 (0.89 – 0.93) 0.93 (0.9 – 0.97) wa other -1.6 (-2.7 – -0.4) +0.4 (-0.8 – +1.7) 1.04 (0.98 – 1.10) 1.06 (1.03 – 1.09) 1.14 (1.07 – 1.21) tas -1.9 (-3.1 – -0.7) +0.4 (-0.8 – +1.7) 1.14 (1.08 – 1.21) 1.14 (1.11 – 1.18) 1.23 (1.15 – 1.30) nt -5.0 (-6.3 – -3.6) -1.9 (-3.6 – -0.3) 1.94 (1.79 – 2.09) 1.65 (1.58 – 1.73) 1.58 (1.43 – 1.74) act -2.7 (-4.6 – -0.8) 0.2 (-1.9 – +2.4) 0.85 (0.78 – 0.92) 0.81 (0.78 – 0.85) 0.86 (0.79 – 0.95) australian population studies 4 (1) 2020 adair & lopez 53 table a1 continued females annualised change (%) ratio to reference category australia -1.6 (-2.0 – -1.2) -1.0 (-1.4 – -0.6) remoteness major cities -1.8 (-2.1 – -1.5) -1.4 (-1.7 – -1.1) ref. ref. inner regional -0.8 (-1.8 – +0.1) +0.1 (-0.9 – +1.1) 1.07 (1.02 – 1.12) 1.13 (1.10 – 1.16) 1.22 (1.16 – 1.28) outer regional/ remote/ very remote -1.0 (-1.9 – -0.1) -0.4 (-1.3 – +0.6) 1.28 (1.22 – 1.34) 1.33 (1.30 – 1.36) 1.40 (1.33 – 1.48) state and remoteness nsw major cities -1.9 (-2.3 – -1.5) -1.8 (-2.2 – -1.3) 0.96 (0.94 – 0.98) 0.95 (0.94 – 0.96) 0.91 (0.89 – 0.94) nsw inner regional -0.3 (-2 – 1.5) -0.5 (-2.2 – +1.3) 1.02 (0.94 – 1.10) 1.09 (1.05 – 1.14) 1.12 (1.03 – 1.22) nsw outer regional/ remote/ very remote -1.5 (-3.4 – +0.3) 0.1 (-1.8 – +2.0) 1.25 (1.14 – 1.37) 1.26 (1.20 – 1.32) 1.33 (1.20 – 1.47) vic major cities -2.7 (-3.3 – -2.0) -0.9 (-1.6 – -0.3) 0.93 (0.90 – 0.96) 0.89 (0.87 – 0.90) 0.89 (0.86 – 0.92) vic inner regional -0.4 (-1.5 – +0.8) -0.1 (-1.3 – +1.0) 1.00 (0.95 – 1.06) 1.07 (1.04 – 1.10) 1.12 (1.05 – 1.18) vic outer regional/ remote/ very remote -0.8 (-1.9 – +0.3) +0.5 (-0.6 – +1.7) 1.03 (0.98 – 1.08) 1.07 (1.04 – 1.10) 1.16 (1.10 – 1.23) qld major cities -1.4 (-1.7 – -1.1) -1.7 (-2.1 – -1.4) 0.95 (0.93 – 0.97) 0.96 (0.95 – 0.97) 0.92 (0.9 – 0.95) qld inner regional -1.0 (-2.4 – +0.3) +1.2 (-0.2 – +2.5) 0.96 (0.91 – 1.02) 0.99 (0.96 – 1.02) 1.10 (1.04 – 1.18) qld outer regional/ remote/ very remote -0.5 (-2.4 – 1.6) -1.5 (-3.5 – +0.6) 1.13 (1.03 – 1.25) 1.20 (1.14 – 1.27) 1.18 (1.06 – 1.31) sa major cities +0.1 (-0.7 – +1.0) -1.8 (-2.7 – -1.0) 0.95 (0.92 – 0.99) 1.04 (1.02 – 1.06) 1.00 (0.95 – 1.05) sa other -2.3 (-4.2 – -0.4) -0.6 (-2.6 – +1.5) 1.08 (0.99 – 1.17) 1.04 (0.99 – 1.09) 1.06 (0.96 – 1.17) wa major cities -1.9 (-3.3 – -0.4) -0.9 (-2.4 – 0.7) 0.91 (0.85 – 0.97) 0.90 (0.87 – 0.93) 0.90 (0.84 – 0.97) wa other -2.7 (-5.0 – -0.3) +0.8 (-1.7 – +3.5) 1.11 (1.00 – 1.24) 1.05 (0.99 – 1.12) 1.16 (1.03 – 1.31) tas -0.2 (-1.8 – +1.4) -0.4 (-2.0 – +1.2) 1.17 (1.08 – 1.27) 1.26 (1.21 – 1.31) 1.30 (1.19 – 1.41) nt -0.8 (-3.5 – +1.9) 0.0 (-2.8 – +2.8) 1.84 (1.57 – 2.15) 1.91 (1.76 – 2.09) 2.01 (1.69 – 2.39) act -2.2 (-4.0 – -0.4) 0.0 (-1.9 – +2.0) 0.90 (0.84 – 0.97) 0.88 (0.84 – 0.92) 0.92 (0.85 – 1.01) notes. ref.: reference category. 54 adair & lopez australian population studies 4 (1) 2020 table a2: asdr by sex, area socio-economic quintile and major cities area socio-economic decile group: annualised change (%) and ratios, 35-74 years, australia, 2006-10 and 2011-16 males annualised change (%) ratio to reference category 2006-10 2011-16 2006 2010 2011 2016 area socio-economic quintile q1 (lowest) -1.5 (-2.4 – -0.6) +0.2 (-0.3 – 0.8) 1.62 (1.56 – 1.68) 1.62 (1.56 – 1.69) 1.97 (1.91 – 2.03) 2.11 (2.04 – 2.17) q2 -2.6 (-3.7 – -1.6) -1.0 (-1.5 – -0.4) 1.48 (1.43 – 1.54) 1.42 (1.36 – 1.48) 1.69 (1.64 – 1.74) 1.70 (1.65 – 1.75) q3 -3.2 (-4.0 – -2.4) -1.1 (-2.4 – +0.2) 1.41 (1.36 – 1.45) 1.31 (1.27 – 1.36) 1.46 (1.38 – 1.55) 1.46 (1.38 – 1.55) q4 -0.1 (-3.0 – 2.9) -1.5 (-2.1 – -0.9) 1.12 (1.02 – 1.22) 1.19 (1.08 – 1.3) 1.27 (1.24 – 1.31) 1.25 (1.21 – 1.29) q5 (highest) -1.6 (-2.3 – -0.8) -1.1 (-1.6 – -0.5) ref. ref. ref. ref. major cities area socio-economic decile group 1-4 -1.9 (-3.0 – -0.8) -0.5 (-1.3 – +0.3) 1.49 (1.42 – 1.57) 1.45 (1.37 – 1.53) 1.70 (1.63 – 1.77) 1.79 (1.71 – 1.87) 5-7 -2.5 (-2.8 – -2.2) -1.3 (-2.4 – -0.3) 1.31 (1.26 – 1.36) 1.24 (1.19 – 1.30) 1.33 (1.27 – 1.39) 1.34 (1.27 – 1.40) 8-10 -1.2 (-2.8 – +0.3) -1.5 (-2.2 – -0.8) ref. ref. ref. ref. females annualised change (%) ratio to reference category 2006-10 2011-16 2006 2010 2011 2016 area socio-economic quintile q1 (lowest) -1.8 (-3 – -0.6) 0.0 (-0.5 – +0.5) 1.49 (1.42 – 1.57) 1.57 (1.49 – 1.66) 1.78 (1.74 – 1.83) 1.98 (1.93 – 2.04) q2 -1.3 (-2.7 –+ 0.2) -0.4 (-1.1 – +0.3) 1.35 (1.28 – 1.43) 1.46 (1.37 – 1.55) 1.51 (1.46 – 1.57) 1.65 (1.59 – 1.71) q3 -3.2 (-4.3 – -2.2) -1.1 (-1.8 – -0.4) 1.31 (1.25 – 1.36) 1.3 (1.24 – 1.37) 1.34 (1.30 – 1.38) 1.41 (1.36 – 1.46) q4 -0.3 (-2.6 – +2.1) -2.0 (-2.7 – -1.2) 1.07 (0.99 – 1.16) 1.20 (1.11 – 1.31) 1.21 (1.17 – 1.25) 1.22 (1.18 – 1.27) q5 (highest) -3.1 (-4.3 – -2.0) -2.1 (-2.7 – -1.6) ref. ref. ref. ref. major cities area socio-economic decile group 1-4 -1.3 (-2.5 – 0.0) -0.6 (-1.2 – 0.0) 1.40 (1.33 – 1.47) 1.41 (1.34 – 1.49) 1.58 (1.53 – 1.62) 1.72 (1.67 – 1.77) 5-7 -3.3 (-4.2 – -2.4) -1.2 (-2.0 – -0.4) 1.26 (1.22 – 1.32) 1.18 (1.13 – 1.23) 1.25 (1.21 – 1.29) 1.32 (1.27 – 1.37) 8-10 -1.6 (-2.7 – -0.4) -2.3 (-2.7 – -1.9) ref. ref. ref. ref. notes. 2006 and 2010 measured using irsad 2006, 2011 and 2016 measured using irsad 2016. 2006 and 2010 results cannot be directly compared with 2011 and 2016 results. ref.: reference category. australian population studies 4 (1) 2020 adair & lopez 55 table a3: asdr by sex and remoteness: annualised change (%) and ratios, 0-34 years, australia, 2006-16 males annualised change (%) ratio to reference category 2006-11 2011-16 2006 2011 2016 australia -3.6 (-4.1 – -3.1) -2.2 (-2.8 – -1.6) major cities -3.1 (-4.0 – -2.3) -1.8 (-2.8 – -0.9) ref. ref. ref. inner regional -3.4 (-5.2 – -1.5) -1.1 (-3.2 – +1.0) 1.46 (1.33 – 1.61) 1.45 (1.36 – 1.53) 1.50 (1.32 – 1.70) outer regional/ remote/ very remote -5.0 (-6.9 – -3.0) -2.1 (-4.5 – +0.3) 2.00 (1.79 – 2.24) 1.82 (1.69 – 1.95) 1.79 (1.53 – 2.09) females annualised change (%) ratio to reference category australia -3.2 (-4.1 – -2.4) -2.9 (-3.9 – -2.0) major cities -3.9 (-4.7 – -3.1) -3.3 (-4.2 – -2.4) ref. ref. inner regional -1.9 (-4.0 – +0.2) -0.7 (-3.0 – +1.6) 1.25 (1.12 – 1.39) 1.38 (1.3 – 1.47) 1.58 (1.38 – 1.80) outer regional/ remote/ very remote -1.3 (-3.0 – +0.5) -1.8 (-3.7 – +0.1) 1.65 (1.49 – 1.82) 1.89 (1.78 – 2.00) 2.04 (1.80 – 2.32) notes. ref.: reference category. 56 adair & lopez australian population studies 4 (1) 2020 table a4: asdr by sex and area socio-economic quintile: annualised change (%) and ratios, 0-34 years, australia, 2006-10 and 2011-16 males annualised change (%) ratio to reference category 2006-10 2011-16 2006 2010 2011 2016 q1 (lowest) -4.8 (-5.6 – -4.0) -2.3 (-4.2 – -0.4) 2.06 (1.94 – 2.18) 1.89 (1.77 – 2.02) 2.20 (1.96 – 2.48) 2.02 (1.77 – 2.29) q2 -2.9 (-4.4 – -1.3) -2.4 (-4.7 – -0.1) 1.57 (1.46 – 1.7) 1.57 (1.43 – 1.71) 1.88 (1.65 – 2.14) 1.71 (1.49 – 1.97) q3 -3.3 (-5.1 – -1.5) -3.1 (-4.8 – -1.3) 1.46 (1.35 – 1.59) 1.43 (1.30 – 1.58) 1.49 (1.33 – 1.68) 1.32 (1.16 – 1.49) q4 -3.0 (-8.3 – 2.5) -1.8 (-4.8 – +1.2) 1.17 (0.98 – 1.39) 1.15 (0.95 – 1.40) 1.25 (1.07 – 1.45) 1.17 (0.99 – 1.38) q5 (highest) -2.8 (-5.4 – -0.1) -0.6 (-3.7 – +2.6) ref. ref. ref. ref. females annualised change (%) ratio to reference category 2006-10 2011-16 2006 2010 2011 2016 q1 (lowest) -1.6 (-3.7 – +0.5) -2.7 (-5.3 – -0.1) 1.85 (1.68 – 2.04) 1.89 (1.7 – 2.10) 2.11 (1.86 – 2.39) 2.18 (1.88 – 2.53) q2 -3.3 (-4.3 – -2.2) -2.3 (-3.9 – -0.7) 1.61 (1.5 – 1.73) 1.54 (1.43 – 1.67) 1.66 (1.54 – 1.78) 1.75 (1.61 – 1.91) q3 -7.0 (-8.6 – -5.5) -4.5 (-7.4 – -1.6) 1.56 (1.44 – 1.69) 1.28 (1.16 – 1.41) 1.55 (1.37 – 1.75) 1.46 (1.25 – 1.69) q4 -6.7 (-11.7 – -1.3) -0.8 (-3.7 – +2.2) 1.26 (1.06 – 1.49) 1.04 (0.85 – 1.28) 1.14 (1.02 – 1.28) 1.30 (1.15 – 1.48) q5 (highest) -2.2 (-5.0 – +0.7) -3.4 (-4.3 – -2.4) ref. ref. ref. ref. notes. 2006 and 2010 measured using irsad 2006, 2011 and 2016 measured using irsad 2016. 2006 and 2010 results cannot be directly compared with 2011 and 2016 results. ref.: reference category. abstract background aim data and methods this study uses australian death registration data from 2006-16. we identify area socio-economic quintile (q1 lowest to q5 highest) and remoteness (major cities, inner regional, outer regional/ remote/very remote) for each decedent. age-standardised d... results asdr decline at ages 35-74 years has slowed markedly, and from 2011-16 q1 asdrs stagnated. the q1:q5 asdr ratio for females rose from 2011-16 for males (1.97 to 2.11) and females (1.78 to 1.98), and also widened in the q2:q5 and q3:q5 comparisons for ... conclusions slowing declines in premature mortality in australia are accompanied by substantial and widening socioeconomic and geographic inequalities. without reductions in mortality in lower socio-economic groups and outside capital cities, future life expectan... key words mortality, premature mortality, inequalities, socio-economic, australia. 2. data and methods key messages acknowledgements references australian population studies 2023 | volume 7 | issue 1 | pages 34-40 © gray & reimondos 2023. published under the creative commons attribution-noncommercial licence 3.0 australia (cc by-nc 3.0 au). journal website: www.australianpopulationstudies.org reporting of housework hours in the australian census edith gray* the australian national university anna reimondos the australian national university *corresponding author. email: edith.gray@anu.edu.au. address: school of demography, research school of social sciences, college of arts and social sciences, rsss building, 146 ellery crescent, the australian national university, acton act 2601 paper received 16 may 2023; accepted 3 august 2023; published 6 december 2023 introduction ‘housework’, or ‘unpaid domestic work’ includes activities such as cooking, cleaning and maintaining a household. these activities contribute significantly to the economy and wellbeing of individuals and families, but are not recognised as ‘labour’ in national accounts (alonso, et al. 2019). although women’s labour force participation has increased and traditional gender role attitudes have diminished over time, women still do the bulk of unpaid domestic work (mills, et al. 2014). time devoted to unpaid domestic work can affect a person’s ability to participate in other activities, such as paid work (wgea 2016). hence it is recognised that having an accurate understanding of the patterns of unpaid domestic work is crucial for achieving gender equality, understanding the economy, and informing policy decisions that affect individuals and families. the australian bureau of statistics (abs) has now completed the second stage of the 2026 census topic consultation. one of the topics under review for removal was that of unpaid domestic work (abs 2023). this recommendation is based on the availability of alternative data or that the data need is not as strong compared to the public value criteria, which includes that there is a need to collect the data for small population groups and/or small geographic areas. the inclusion of unpaid work in the census was the subject of debate for at least a decade before it was included for the first time in 2006. many submissions called for its inclusion prior to the 1996 and the 2001 censuses, but it was ultimately excluded as there were already other sources of highquality data on the topic. it was also noted that national data existed and that there was no specific need identified to have this topic collected for small areas or for small population groups (abs 1996, 2001a; 2001b). in june 2003 a consultative committee on unpaid work comprised of representatives from abs and key external organisations was established to determine the need for questions on unpaid work in the 2006 census. the committee identified a need for data on unpaid work at the small area level, and as a result four questions on unpaid work were included, on: (1) unpaid domestic work, (2) volunteering, (3) caring for children, and (4) caring for elderly or disabled persons. the question has been repeated in all subsequent censuses. d e m o g ra p h ic http://www.australianpopulationstudies.org/ mailto:edith.gray@anu.edu.au australian population studies 7 (1) 2023 gray & reimondos 35 however, high-quality alternative data on unpaid domestic work is available in australia. the abs collects rich data on primary and secondary activities through the national time use surveys (ntus) (1992, 1997, 2006 and 2020-21). a question on housework is also included in the household income and labour dynamics in australia (hilda) survey. although domestic work is collected in the census, the results differ quite substantially to that from other data collections. the census data records a large proportion of the population not participating in any domestic work: at the 2021 census, 29% of men, and 22% of women aged 15+ were recorded as not doing any domestic work in the previous week. in contrast, the 2020 hilda survey records 7% of men and 3% for women aged 15 + who stated that they did not do any housework in a typical week. in the household impacts of covid-19 survey, fewer than 5% of men aged 18 +, and fewer than 2% of women did no indoor housework in the previous week (abs 2021), and in the ntus 2020-21, 18% of men and 7% of women did not engage in ‘domestic activities’ in the last two days (abs 2022). this paper compares the stated hours of domestic labour in the census with data from the hilda survey to better understand the patterns behind the large proportion of australians that record doing no domestic work in the census. data and methods we use the australian census (2006, 2011, 2016 and 2021) and approximately corresponding years from the hilda survey. the census question asks: “in the last week did the person spend time doing unpaid domestic work for their household?”. further instructions note to include all housework, food/drink preparation and clean-up, laundry, gardening, home maintenance and repairs, household shopping and finance management. the answer categories are: 1) no, did not do any unpaid domestic work in the last week, 2) yes, less than 5 hours 3) yes, 5-14 hours 4) yes, 15-29 hours and 5) yes, 30 hours or more the hilda survey is a longitudinal national household panel survey. one component of the interview involves a self-completion questionnaire, which asks: “how much time would you spend on each of the following activities in a typical week?” a list of different activities is given including “housework, such as preparing meals, washing dishes, cleaning house, washing clothes, ironing and sewing” and respondents are asked to answer in terms of hours and or minutes. we compare the percentage who do no housework over time from both sources of data for people aged 15 and over, before examining in further detail the distribution of answers by age group and sex for those aged 18-64 in the latest census and hilda data. all hilda data presented is weighted with the cross-section self-completion questionnaire responding person population weight (summerfield, et al. 2021). 36 gray & reimondos australian population studies 7 (1) 2023 key features figure 1 shows the percentage of the population aged 15 and over who reported doing no housework in the census and in hilda. for comparability, we created figures for both the overall census data and a subset that only includes individuals residing in private dwellings, and excluding those living in remote or very remote areas. this adjustment aligns with the hilda data's exclusion of non-private dwelling residents and those in remote or sparsely populated regions. figure 1: percentage of the population aged 15+ who did zero hours of unpaid domestic work, by sex and year (census and hilda) source: abs census and hilda survey data. the census figures are substantially higher than the hilda proportions, even when the census data is restricted. in the census, approximately one quarter of men are recorded as performing no domestic work, with the proportion increasing over time. conversely, the hilda data shows less than 10% of men indicating no involvement in household chores, with the proportion decreasing over time . a similar pattern is evident among women in both datasets. the distribution of responses by gender and age group for the 2021 census and 2020 hilda data, was then analysed and is illustrated in figure 2. the disparity in response patterns is apparent across all age groups. when we compare those who completed 5 hours of domestic work or less, the differences become slightly less pronounced. it is likely that hilda is capturing people who do small amounts of hours, while they are likely being reported as ‘none’ in the census. the census data on domestic work shows a different distribution to other sources of data, particularly when it comes to those who do no housework at all. a potential explanation for reporting no hours may be proxy reporting. in the census, one household member can respond for everyone, which could lead to less accurate reporting. however, if proxy reporting were a australian population studies 7 (1) 2023 gray & reimondos 37 (a) men (b) women figure 2: distribution of hours of unpaid domestic work in census 2021 and hilda 2020 by age group and gender source: abs census and hilda survey data. 38 gray & reimondos australian population studies 7 (1) 2023 contributing factor, restricting both data sets to those living alone should narrow the gap. our analysis shows the contrary, as the gap is larger. in the 2021 census, of those aged 18-64, 36% of men living alone reported not engaging in housework, compared to just 3% in the hilda survey. for women, the respective figures were 29% and 1%. another explanation is the different question wording. both questions refer to a one-week period but in hilda the responses are given as hours/minutes rather than categories. the census question asks if the person spent any time, whereas the hilda question asks how much time was spent. further, the census wording of ‘unpaid domestic work’ may sound unfamiliar, as it is colloquially referred to as ‘housework’. most papers on domestic labour in australia have relied on hilda (craig & siminksi 2010, 2011; baxter & hewitt 2013; foster & stratton 2018,2019; mooi-reci & craig 2020; ervin, et al. 2023), or ntus (bittman, et al. 2003, 2004; craig & mullan 2010; craig, et al. 2016; craig & powell 2018). we could find no published research using the census for small-area or small population groups, although we acknowledge that community organisations may have used it for their own purposes. this demographic shows that reported hours spent on housework are substantially lower in the census than any other data source. other data collections have greater space for specific instructions, or more detailed time recording, providing high-quality data. we question whether the current question is currently overstating zero participation in housework, and believe that this is an additional reason for removing the current question from the census. acknowledgements this paper uses unit record data from household, income and labour dynamics in australia survey [hilda] conducted by the australian government department of social services (dss). the findings and views reported in this paper, however, are those of the authors and should not be attributed to the australian government, dss, or any of dss’ contractors or partners. doi: http://dx.doi.org/10.26193/yp7mnu. references alonso, c., brussevich, m., dabla-norris, m. e., kinoshita, y., & kochhar, m. k. (2019). reducing and redistributing unpaid work: stronger policies to support gender equality. imw working papers, 2019(225): international monetary fund. https://doi.org/10.5089/9781513514536.001 abs (1996). 2007.0 information paper: census of population and housing, abs views on content and procedures, 1996. https://www.abs.gov.au/ausstats/abs@.nsf/detailspage/2007.01996?opendocument abs (2001a). unpaid work and the census. abs. https://www.abs.gov.au/websitedbs/censushome.nsf/home/2001unpaidwork?opendocument& navpos=320 abs (2001b). 2007.0 information paper: census of population and housing, abs views on content and procedures, 2001. https://www.abs.gov.au/ausstats/abs@.nsf/detailspage/2007.02001?opendocument http://dx.doi.org/10.26193/yp7mnu https://doi.org/10.5089/9781513514536.001 https://www.abs.gov.au/ausstats/abs@.nsf/detailspage/2007.01996?opendocument https://www.abs.gov.au/websitedbs/censushome.nsf/home/2001unpaidwork?opendocument&navpos=320 https://www.abs.gov.au/websitedbs/censushome.nsf/home/2001unpaidwork?opendocument&navpos=320 australian population studies 7 (1) 2023 gray & reimondos 39 abs (2021). household impacts of covid-19 survey. abs. https://www.abs.gov.au/statistics/people/people-and-communities/household-impacts-covid19-survey/dec-2020. abs (2022). how australians use their time. abs. https://www.abs.gov.au/statistics/people/people-andcommunities/how-australians-use-their-time/latest-release. abs (2023). 2026 census topic review: phase one directions. https://www.abs.gov.au/statistics/research/2026-census-topic-review-phase-onedirections#summary-of-topic-outcomes baxter, j., hewitt, b., & haynes, m. (2008). life course transitions and housework: marriage, parenthood, and time on housework. journal of marriage and family, 70(2), 259-272. https://doi.org/10.1111/j.1741-3737.2008.00479.x baxter, j., & hewitt, b. (2013). negotiating domestic labor: women's earnings and housework time in australia. feminist economics, 19(1), 29-53. https://doi.org/10.1080/13545701.2012.744138 bittman, m., england, p., sayer, l., folbre, n., & matheson, g. (2003). when does gender trump money? bargaining and time in household work. american journal of sociology, 109(1), 186-214. https://doi.org/10.1086/378341 bittman, m., rice, j. m., & wajcman, j. (2004). appliances and their impact: the ownership of domestic technology and time spent on household work. the british journal of sociology, 55(3), 401-423. https://doi.org/10.1111/j.1468-4446.2004.00026.x craig, l., & mullan, k. (2010). parenthood, gender and work‐family time in the united states, australia, italy, france, and denmark. journal of marriage and family, 72(5), 1344-1361. https://doi.org/10.1111/j.1741-3737.2010.00769.x craig, l., & powell, a. (2018). shares of housework between mothers, fathers and young people: routine and non-routine housework, doing housework for oneself and others. social indicators research, 136(1), 269-281. https://doi.org/10.1007/s11205-016-1539-3 craig, l., powell, a., & brown, j. e. (2016). gender patterns in domestic labour among young adults in different living arrangements in australia. journal of sociology, 52(4), 772-788. https://doi.org/10.1177/1440783315593181 craig, l., & siminski, p. (2010). men's housework, women's housework, and second births in australia. social politics, 17(2), 235-266. https://doi.org/10.1093/sp/jxq004 craig, l., & siminski, p. (2011). if men do more housework, do their wives have more babies? social indicators research, 101, 255-258. https://doi.org/10.1007/s11205-010-9644-1 ervin, j., taouk, y., hewitt, b., & king, t. (2023). the association between unpaid labour and mental health in working-age adults in australia from 2002 to 2020: a longitudinal population-based cohort study. the lancet public health, 8(4), e276-e285. https://doi.org/10.1016/s2468-2667(23)000300 foster, g., & stratton, l. s. (2018). do significant labor market events change who does the chores? paid work, housework, and power in mixed-gender australian households. journal of population economics, 31, 483-519. https://doi.org/10.1007/s00148-017-0667-7 foster, g., & stratton, l. s. (2019). what women want (their men to do): housework and satisfaction in australian households. feminist economics, 25(3), 23-47. https://doi.org/10.1080/13545701.2019.1609692 mooi-reci, i., & craig, l. (2020). dual-parent joblessness, household work and its moderating role on children’s joblessness as young adults. journal of family issues, 41(9), 1569-1596. https://doi.org/10.1177/0192513x19894353 https://www.abs.gov.au/statistics/people/people-and-communities/household-impacts-covid-19-survey/dec-2020 https://www.abs.gov.au/statistics/people/people-and-communities/household-impacts-covid-19-survey/dec-2020 https://www.abs.gov.au/statistics/people/people-and-communities/how-australians-use-their-time/latest-release https://www.abs.gov.au/statistics/people/people-and-communities/how-australians-use-their-time/latest-release https://www.abs.gov.au/statistics/research/2026-census-topic-review-phase-one-directions#summary-of-topic-outcomes https://www.abs.gov.au/statistics/research/2026-census-topic-review-phase-one-directions#summary-of-topic-outcomes https://doi.org/10.1111/j.1741-3737.2008.00479.x https://doi.org/10.1080/13545701.2012.744138 https://doi.org/10.1086/378341 https://doi.org/10.1111/j.1468-4446.2004.00026.x https://doi.org/10.1111/j.1741-3737.2010.00769.x https://doi.org/10.1007/s11205-016-1539-3 https://doi.org/10.1177/1440783315593181 https://doi.org/10.1093/sp/jxq004 https://doi.org/10.1007/s11205-010-9644-1 https://doi.org/10.1016/s2468-2667(23)00030-0 https://doi.org/10.1016/s2468-2667(23)00030-0 https://doi.org/10.1007/s00148-017-0667-7 https://doi.org/10.1080/13545701.2019.1609692 https://doi.org/10.1177/0192513x19894353 40 gray & reimondos australian population studies 7 (1) 2023 mills, m. c., tsang, f., präg, p., ruggeri, k., miani, c., & hoorens, s. (2014). gender equality in the workforce reconciling work, private and family life in europe. santa monica, ca: rand corporation. https://doi.org/10.7249/rr462 summerfield, m., garrard, b., nesa, m.k, kamath, r., macalalad, n., watson, n., wilkins, r. and wooden, m. (2021). hilda user manual – release 21. melbourne institute: applied economic and social research, university of melbourne. https://melbourneinstitute.unimelb.edu.au/__data/assets/pdf_file/0009/4384386/hilda-usermanual-release-21.0.pdf workplace gender equality agency [wgea] (2016). unpaid care work and the labour market. https://www.wgea.gov.au/publications/unpaid-care-work-and-the-labour-market. https://doi.org/10.7249/rr462 https://melbourneinstitute.unimelb.edu.au/__data/assets/pdf_file/0009/4384386/hilda-user-manual-release-21.0.pdf https://melbourneinstitute.unimelb.edu.au/__data/assets/pdf_file/0009/4384386/hilda-user-manual-release-21.0.pdf https://www.wgea.gov.au/publications/unpaid-care-work-and-the-labour-market australian population studies 2023 | volume 7 | issue 1 | pages 17-29 © leigh 2023. published under the creative commons attribution-noncommercial licence 3.0 australia (cc by-nc 3.0 au). journal website: www.australianpopulationstudies.org commentary how covid changed the world of data andrew leigh parliament of australia email: andrew.leigh.mp@aph.gov.au. address: parliament house, canberra act 2600, australia paper received 7 november 2023; accepted 30 november 2023; published 6 december 2023 this is an edited version of a speech delivered to the population symposium at the australian national university school of demography on 6th october 2023. 1. introduction covid might have shrunk our worlds to frequent zoom calls in which we took it in turns to remind each other ‘you’re on mute’. but covid also opened up a whole new world of data. amidst the most severe health pandemic in a century, the australian bureau of statistics (abs) produced new data products, shortened some of its release timelines, and adapted data series. as a consequence of the pandemic, australians have access to more timely and frequent updates, new sources of data and greater integration. this article explores how the pandemic changed australia’s data landscape, and how new innovations in ‘nowcasting’ might shape the nation’s data infrastructure in the future. the covid pandemic was not only a rapidly evolving health crisis but an economic crisis too – the most severe economic downturn since the great depression (kennedy 2022). restrictions to limit the spread of covid saw a reduction in spending, business turnover, losses in jobs and hours worked, and supply chain disruptions (abs 2022a). by june 2023, gdp had returned to its pre-pandemic trajectory (figure 1). however, over the course of the pandemic, the cumulative gdp loss was $116 billion – representing the area between the two lines in figure 1. the covid pandemic rocked the demographic boat. australia’s population growth slowed to 0.1 per cent in 2020–21 – the lowest rate in more than 100 years (figure 2). australia’s net overseas migration fell into negative territory for the first time since the end of wwii, with a net loss of 85,000 people in 2020-21 (abs 2023c). births fluctuated in interesting ways. in the december 2020 quarter, nine months after the pandemic hit, births fell. as gray et al. (2022) point out, crises can make couples cautious about starting a family, and this drop likely reflected the uncertainty that many couples felt in the early months of the covid lockdowns. but then things turned around. in the march 2021 quarter, nine months after mid-2020, births hit an all-time high (abs 2023c). we can’t be sure as to why this occurred, but it may be that couples felt a little less anxious about the future as the year unfolded. lockdown boredom may also have been a factor. http://www.australianpopulationstudies.org/ mailto:andrew.leigh.mp@aph.gov.au 18 leigh australian population studies 7 (1) 2023 figure 1: australia’s gdp, actual and pre-covid trajectory source: abs 2023a and treasury figure 2: australia’s population growth and components of growth, historical and projected source: abs 2023b and treasury australian population studies 7 (1) 2023 leigh 19 border closures between states and territories reduced internal mobility. the number of interstate moves in the year to march 2021 was 30.2 per cent lower than in 2018-19 (abs 2023c). the pandemic also influenced where people wanted to live with an increase in net moves from urban areas towards suburban and regional areas (figure 3) (centre for population 2020). covid doubled the net number of people moving to the regions. the break in the data series in figure 3 is due to the pandemic’s impact on medicare address information this series relied on, discussed below. figure 3: net internal migration to regions outside capital cities, quarterly source: abs 2023c and abs 2021d. note: the break in the data series is due to the pandemic’s impact on medicare address information. in 2020-21, regional areas grew faster than capital cities for the first time since 1993-94 (abs 2023e) (figure 4). the popular television show seachange debuted in 1998, but it would take two more decades and a pandemic before the data showed more rapid population growth in regional areas than capital cities. 2. putting the data spotlights on the data challenge for government agencies during the pandemic was enormous. data had to be timely to respond to a rapidly changing environment. it had to be delivered at a higher frequency. and data had to be continuous and granular to help us understand the different things happening in different states and territories. treasury recognised early on that existing data assets weren’t going to be sufficient for handling the speed and scale of dual health and economic shocks. a treasury paper (hambur et al. 2022) likened using existing data during covid to the ‘streetlight effect’ where a drunk only searches for their lost keys under the lamp post, because that’s where the light is best. 20 leigh australian population studies 7 (1) 2023 figure 4: combined capital city and rest of state population growth, annual source: abs 2023e and abs 2019. when it came to forecasting during the pandemic, historical data – traditionally the foundation for models – suddenly seemed inadequate. the usual patterns and trends, which underpinned models, were disrupted, making usual methodologies less reliable. then there was the unpredictable nature of the virus. new strains emerged, with different levels of transmissibility and severity, complicating predictive models. the only plausible option involved shining new light on the rapidly evolving economic and demographic situation using new sources of data. by mid-march 2020 – about the same time that toilet paper madness was unrolling – the abs already had introduced a range of covid related products. this included preliminary data as well as smaller, rapid-response surveys to deliver more timely insights (abs 2021a, abs n.d). the abs introduced the business impacts of covid-19 (now named business conditions and sentiments) and the household impacts of covid-19 rapid-response surveys. these surveys fast-tracked the time between the field and publication. the business impacts of covid-19 survey provided information with a delay of 7 days from interviewing to publication, while the household impacts of covid-19 was released around 17 days after the end of the collection period. the abs also made greater use of alternate data sources, such as non-survey administrative data and transactions data to develop indicators (abs 2021a, abs n.d). in one example, the abs partnered with other agencies to provide new data through two integrated longitudinal assets – the person level integrated data asset (plida, formerly known as madip) and the business longitudinal analysis data environment (blade). these data assets proved valuable because they could be updated frequently, making quicker and more sophisticated analysis possible. the plida and blade core data was linked australian population studies 7 (1) 2023 leigh 21 providing new insights into the interactions between businesses and their employees (gruen 2022a). these insights allowed governments and policymakers to determine covid’s impact on the business community and examine the economic recovery including employment patterns. data from the australian immunisation register (air) are also linked to plida each week and provisional death registrations data is linked and updated monthly. during the pandemic, the department of health and aged care used this data to generate insights for australia’s covid vaccine and treatment strategy. state health departments and primary health networks are also data users. the pandemic also saw the emergence of new official data sources. it made practical sense for the abs to draw on data already generated in our increasingly digital world (gruen 2022a). it makes even more sense when that source provided previously unavailable statistics, faster and with greater coverage than before. the single-touch payroll is a prime example. from april 2020, the abs worked with the australian taxation office to receive expedited access to the high-frequency single-touch payroll data. this access was significant for a few reasons. first, it provided new and almost real-time insights into the labour market based on more than 10 million employees in australia (gruen 2022a). the teamwork resulted in the weekly payroll jobs publication and the more recent monthly employee earnings indicator publication. second, the access enabled detailed disaggregation including geospatial analysis which just wasn’t possible from the existing monthly labour force survey. and third, it filled a major void. treasury’s data experts observed that during the covid pandemic the ‘unemployment rate was a less informative measure of labour market health than usual (because many workers had been stood down temporarily or were working reduced hours) and other indicators such as hours worked gained more prominence’ (hambur et al. 2022 p. 4). they noted, ‘the timeliness and broad coverage of single touch payroll made it one of the most valuable datasets in assessing the health of the labour market in real time… and in modelling fiscal policy response options’ (p. 4). the abs was not the only government agency stepping up to the challenges of measuring covid’s impact. the australian institute of family studies introduced the families in australia survey series (https://aifs.gov.au/research_programs/families-australia-survey) which ran from may 2020 to december 2021. the survey series explored how families’ lives changed during the pandemic, covering topics such as changes to workforce participation, work patterns, child care, household forms and family supports. the australian institute of health and welfare established the covid-19 register – a national and linked data set (aihw 2023). the register combines de-identified covid case data with other data sets, including immunisation, hospitals and deaths data. importantly, it will be updated as more data become available, allowing a long-term view of the effects of covid on health outcomes and health service usage. combining all the data and research, there are many more lamp posts in the data park to help people find their lost keys. 3. the realm of demographic data in australia prior to covid, summary mortality information was reported with a 6-month delay from when the death was registered to publication (abs 2023j). the abs expedited its mortality data reporting to capture the immediate and detailed covid impacts. the reporting became essential for https://aifs.gov.au/research_programs/families-australia-survey 22 leigh australian population studies 7 (1) 2023 understanding covid’s direct impact and its indirect consequences on the population. in june 2020, the abs released the first provisional mortality statistics report to account for the beginning of the pandemic (abs 2020). the abs continued to produce the report monthly as a way of tracking mortality through the pandemic. this meant that deaths were now reported 3 months after they occurred and on a more frequent and detailed basis. so the consequence of covid has been to reduce the delay in deaths reporting and support the tracking of mortality during the pandemic and beyond. these reports showed that the pandemic could impact mortality in unexpected ways. in 2022, we saw deaths due to other causes rise, highlighting some of the direct and indirect impacts of the pandemic on our health (abs 2023f). the abs released an article in july on measuring australia’s excess mortality during the covid pandemic (abs 2023d). the article provides excess mortality estimates for australia and the states and territories from 2020 to march 2023. one can see the variations across the states over the omicron period especially, with western australia recording the lowest and tasmania the highest in this map (figure 5). figure 5: excess mortality (%), omicron period (january 2022 – march 2023) source: abs 2023d the abs found that mortality was 9.1 per cent above the expected number of deaths in the first quarter of 2023. in 2022, deaths were 10.9 per cent above expected. this contrasted with the years 2020 and 2021, when australia had relatively few covid cases, and significant mobility restrictions. analysing abs mortality data for those years, oxford’s philip clarke and i published research in bmj: global health that studied the impact of lockdowns on mortality (clarke and leigh 2022). we found australian population studies 7 (1) 2023 leigh 23 the mortality rate in australia was 5.9 per cent lower in 2020–2021 than in 2015–2019. contrary to those who claimed lockdowns would increase the death rate, we found lockdowns and social distancing reduced short‑term mortality. the abs has published other articles on covid mortality so we can see how it has been distributed across population groups (abs 2023h). the articles have presented covid mortality data by demographic characteristics. this breakdown helps us track the incidence of direct covid mortality and indirect mortality, and offered insights into demographics variations. delivering the w.d. borrie lecture last year, i explored some of the health inequalities created by the covid pandemic. it turns out that although britain had a higher level of covid mortality than australia, the socioeconomic patterns of covid mortality in the two countries are very similar (leigh 2022). collecting some kinds of demographic data became more difficult during covid. registration delays in the reporting of births data were especially pronounced in some places during the pandemic as a result of activity restrictions and lockdowns (centre for population 2021). during the pandemic, the centre for population – housed in the australian treasury – was able to use medicare newborn entrants data from services australia (centre for population 2021, pp. 42-43). this data provided a proxy for births to inform fertility assumptions and improve population projections during the pandemic and beyond. migration trends were a major focus during the pandemic. with international travel restrictions, the abs provided more detailed and timely data on returning australians, international students, and the implications for population growth. for example, additional information was included in overseas arrivals and departures (oad) to provide more detail on the numbers of arrivals and departures to and from australia (gruen 2020). additional information included analysis of short-term visitor arrivals and a breakdown of international students for each state and territory. in response to heightened interest in traveller data, the abs also created the temporary overseas travel statistics provisional publication (abs 2021b). this data was effective in giving governments earlier monthly preliminary information on international travel using border crossing data. as a result, information for the reference month was available four weeks earlier than what was available in the overseas arrivals and departures release (abs 2021a). given the value of this information, provisional travel data was added to the existing overseas arrivals and departures release from august 2021. the data give us a dramatic picture (figure 6). as shown in this chart, from 20 march 2020, almost all overseas travel was banned, as the who declared the covid outbreak a public health emergency. the pandemic disrupted international travel with volumes at a record low in 2021. the government began to ease travel restrictions in late 2021. international travel in australia has seen a resurgence since early 2022, with over 19 million arrivals and departures combined across 2022. the abs also introduced more frequent revisions of net overseas migration estimates (gruen 2022b). this improvement was about providing the most up-to-date information and enabled adjustments to the preliminary model to account for changed traveller behaviour. 24 leigh australian population studies 7 (1) 2023 figure 6: total overseas arrivals and departures, monthly source: abs 2023g as noted above, internal movements were fascinating from a demographic point of view. with borders closing and movement restrictions in place, the abs adjusted its approach to monitoring internal population dynamics, tracking movements between states as well as urban and rural divides. covid challenged traditional understandings of internal migration, not to mention forecasting the distribution of the population and the flows of people between areas. remember the demographic boat was rocking – things were still happening fast when the abs introduced the provisional regional internal migration estimates (prime) in november 2020 (presented in figure 3). these estimates were a way to provide timely insights into internal migration in response to the pandemic and the heightened interest in internal migration data. however, the publication was suspended after the march 2021 release due to the impact of the covid vaccination program on medicare data, which is the primary source for internal migration estimates. the vaccination program saw the number of people updating their medicare address data go through the roof – reflected in the number of interstate moves jumping to a series record 164,000 in september 2021 while there were lockdowns and border restrictions in place (figure 7). the pandemic also posed some challenges to the approach taken for finalising internal migration estimates from the 2021 census. the issue with the medicare data led the abs to declare that movements over the second half of 2021 were not reflective of real world internal migration. in addition, the census data was also affected by lockdown policies in some jurisdictions. the abs responded to these challenges by giving greater weight to the census and medicare data that was unaffected by these impacts. this included discounting the medicare data over 2020–21 in favour of the census-based estimates and giving more weight to the earlier (2016) census data than usual. australian population studies 7 (1) 2023 leigh 25 figure 7: interstate movements (based on medicare address changes), quarterly source: abs 2023c note: a large number of medicare address updates occurred during the rollout of the vaccination program in 2021. 4. looking ahead: demographic data post-covid covid demonstrated the potential for more data integration. the life course data initiative, announced in this year’s budget, will connect administrative datasets into a linked longitudinal data asset to support evidence-based policy making at the community level. linking data across different subject-matter areas and levels of government will bring many benefits. it will enable more comprehensive analysis of the characteristics, programs and service interactions of individuals, households and families (gruen 2023). the abs is exploring new ways to enhance the census using administrative data. the 2021 census was supported by integrated administrative data that helped in assuring its quality and led to more accurate area population counts (abs 2023i). more recently, new income and benefits data added to the 2021 census for its final data release supported new analysis and insights on australian pensioners. work to understand administrative data has led to the release of census-like population and housing data snapshots (abs 2023k; abs 2023l). these experimental datasets have been released as stand-alone products and over time are expected to provide additional information on sub-national population and internal migration patterns. 26 leigh australian population studies 7 (1) 2023 5. conclusion the covid pandemic delivered lasting change to australia’s statistical systems. • before the pandemic, deaths were reported 6 months after the end of the reporting period. now, they are reported with a 3 month delay. • before the pandemic, the delay in reporting overseas arrivals and departures was 6 weeks. now, provisional data is available in only 2 weeks. • before the pandemic, australia was one of only two oecd countries (the other being new zealand) to report inflation on a quarterly basis. now, we have a monthly cpi indicator. • as a result of the pandemic, the abs created the weekly payroll jobs publication and the monthly employee earnings indicator publication. yet there may be more to be done. right now, the reporting delay is 2 weeks for unemployment, 3½ weeks for quarterly inflation, and just over 2 months for national income accounts and house price data. by contrast, share market data are available in real time. and in the september 2023 issue of the australian economic review, a series of papers explored innovations in ‘nowcasting’. michael anthonisz of the queensland treasury corporation (qtc) uses data from bloomberg and macrobond, as well as the citi australian data change index and the qtc-constructed financial conditions index to provide daily insights into key indicators of the australian economy including inflation, gdp growth and the unemployment rate (anthonisz 2023). looking locally, ashton de silva, maria yanotti, sarah sinclair and sveta angelopoulos find unconventional real-time data such as pedestrian counts could be good high-frequency predictors of local economies, such as suburbs or central business districts (cbds) (de silva et al. 2023). and finally, in response to rising house prices, shuping shi and peter c. b. phillips created a real-time “thermometer” of “housing fever” in australia and new zealand. using data on mortgage interest rates, rents, disposable income, employment and housing supply the authors diagnose in real-time bubbles as they begin to emerge (shi and phillips 2023). these intriguing examples show what is possible using novel data and machine learning algorithms. it may also be possible to use additional data sources to improve estimates of state and territory populations, which are used for a range of important purposes, such as allocating goods and services tax revenue across jurisdictions. it is vital that government continues to watch developments in this space. in producing statistics, there is always a trade-off between quality and timeliness. in another crisis, policymakers may want data more quickly, even if it comes at the cost of perfection. australian population studies 7 (1) 2023 leigh 27 acknowledgments my thanks to officials at the australian bureau of statistics for valuable assistance in preparing this text, and to editor tom wilson and two anonymous referees for feedback on an earlier draft. references abs (2019). historical population, 2016. https://www.abs.gov.au/statistics/people/population/historicalpopulation/2016 abs (2020). provisional mortality statistics, jan-mar 2020. https://www.abs.gov.au/statistics/health/causes-death/provisional-mortality-statistics/jan-mar2020 abs (2021a). changes to abs statistical work program. https://www.abs.gov.au/media-centre/mediareleases/changes-abs-statistical-work-program abs (2021b). overseas travel statistics, provisional. https://www.abs.gov.au/statistics/industry/tourismand-transport/overseas-travel-statistics-provisional/jun-2021 abs (2021c). investigation into application-based positioning data for covid-19 1504.0 methodological news, mar 2021. https://www.abs.gov.au/ausstats/abs@.nsf/mf/1504.0 abs (2021d). regional internal migration estimates, provisional, march 2021. https://www.abs.gov.au/statistics/people/population/regional-internal-migration-estimatesprovisional/mar-2021 abs (2022a). effects of covid-19 strains on the australian economy. https://www.abs.gov.au/articles/effects-covid-19-strains-australian-economy abs (2022b). economic gains and losses over the covid-19 pandemic. https://www.abs.gov.au/articles/economic-gains-and-losses-over-covid-19-pandemic abs (2023a). australian national accounts: national income, expenditure and product, june 2023. https://www.abs.gov.au/statistics/economy/national-accounts/australian-national-accountsnational-income-expenditure-and-product/jun-2023 abs (2023b). national, state and territory population, september 2022. https://www.abs.gov.au/statistics/people/population/national-state-and-territorypopulation/sep-2022 abs (2023c). national, state and territory population, march 2023. https://www.abs.gov.au/statistics/people/population/national-state-and-territorypopulation/mar-2023 abs (2023d). measuring australia's excess mortality during the covid-19 pandemic until the first quarter 2023. https://www.abs.gov.au/articles/measuring-australias-excess-mortality-during-covid-19pandemic-until-first-quarter-2023 abs (2023e). regional population, 2021-22. https://www.abs.gov.au/statistics/people/population/regional-population/2021-22 abs (2023f). provisional mortality statistics, jan-may 2023. https://www.abs.gov.au/statistics/health/causes-death/provisional-mortality-statistics/jan-may2023 abs (2023g). overseas arrivals and departures, australia, july 2023. https://www.abs.gov.au/statistics/industry/tourism-and-transport/overseas-arrivals-anddepartures-australia/jul-2023 abs (2023h). covid-19 mortality in australia: deaths registered until 31 july 2023. https://www.abs.gov.au/articles/covid-19-mortality-australia-deaths-registered-until-31-july2023 https://www.abs.gov.au/statistics/people/population/historical-population/2016 https://www.abs.gov.au/statistics/people/population/historical-population/2016 https://www.abs.gov.au/statistics/health/causes-death/provisional-mortality-statistics/jan-mar-2020 https://www.abs.gov.au/statistics/health/causes-death/provisional-mortality-statistics/jan-mar-2020 https://www.abs.gov.au/media-centre/media-releases/changes-abs-statistical-work-program https://www.abs.gov.au/media-centre/media-releases/changes-abs-statistical-work-program https://www.abs.gov.au/statistics/industry/tourism-and-transport/overseas-travel-statistics-provisional/jun-2021 https://www.abs.gov.au/statistics/industry/tourism-and-transport/overseas-travel-statistics-provisional/jun-2021 https://www.abs.gov.au/ausstats/abs@.nsf/mf/1504.0 https://www.abs.gov.au/statistics/people/population/regional-internal-migration-estimates-provisional/mar-2021 https://www.abs.gov.au/statistics/people/population/regional-internal-migration-estimates-provisional/mar-2021 https://www.abs.gov.au/articles/effects-covid-19-strains-australian-economy https://www.abs.gov.au/articles/economic-gains-and-losses-over-covid-19-pandemic https://www.abs.gov.au/statistics/economy/national-accounts/australian-national-accounts-national-income-expenditure-and-product/jun-2023 https://www.abs.gov.au/statistics/economy/national-accounts/australian-national-accounts-national-income-expenditure-and-product/jun-2023 https://www.abs.gov.au/statistics/people/population/national-state-and-territory-population/sep-2022 https://www.abs.gov.au/statistics/people/population/national-state-and-territory-population/sep-2022 https://www.abs.gov.au/statistics/people/population/national-state-and-territory-population/mar-2023 https://www.abs.gov.au/statistics/people/population/national-state-and-territory-population/mar-2023 https://www.abs.gov.au/articles/measuring-australias-excess-mortality-during-covid-19-pandemic-until-first-quarter-2023 https://www.abs.gov.au/articles/measuring-australias-excess-mortality-during-covid-19-pandemic-until-first-quarter-2023 https://www.abs.gov.au/statistics/people/population/regional-population/2021-22 https://www.abs.gov.au/statistics/health/causes-death/provisional-mortality-statistics/jan-may-2023 https://www.abs.gov.au/statistics/health/causes-death/provisional-mortality-statistics/jan-may-2023 https://www.abs.gov.au/statistics/industry/tourism-and-transport/overseas-arrivals-and-departures-australia/jul-2023 https://www.abs.gov.au/statistics/industry/tourism-and-transport/overseas-arrivals-and-departures-australia/jul-2023 https://www.abs.gov.au/articles/covid-19-mortality-australia-deaths-registered-until-31-july-2023 https://www.abs.gov.au/articles/covid-19-mortality-australia-deaths-registered-until-31-july-2023 28 leigh australian population studies 7 (1) 2023 abs (2023i). how administrative data improved the quality of the 2021 census. https://www.abs.gov.au/articles/how-administrative-data-improved-quality-2021-census. abs (2023j). national, state and territory population methodology, march 2023. https://www.abs.gov.au/methodologies/national-state-and-territory-populationmethodology/mar-2023 abs (2023k). administrative data snapshot of population, 30 june 2021. https://www.abs.gov.au/statistics/people/population/administrative-data-snapshot-populationand-housing-experimental-population-data/30-june-2021 abs (2023l). administrative data snapshot of housing, 30 june 2021. https://www.abs.gov.au/statistics/people/housing/administrative-data-snapshot-populationand-housing-experimental-housing-data/30-june-2021 abs (n.d). abs responds to covid-19. https://www.abs.gov.au/abs-responds-covid-19 aihw (2023). covid-19 linked data set: linkage results. cat. no. phe 328. https://www.aihw.gov.au/reports/covid-19/covid-19-linked-data-set-linkageresults/contents/about anthonisz, m. (2023). nowcasting key australian macroeconomic variables. australian economic review, 56(3), 371-380. https://doi.org/10.1111/1467-8462.12524 centre for population (2020). migration between cities and regions: a quick guide to covid-19 impacts. commonwealth treasury. https://population.gov.au/research/research-migration-betweencities-and-regions-a-quick-guide-to-covid-19-impacts centre for population (2021). 2021 population statement. commonwealth treasury. https://population.gov.au/publications/statements/2021-population-statement clarke, p. & leigh, a. (2022). understanding the impact of lockdowns on short‑term excess mortality in australia. bmj global health, 7 article e009032. http://dx.doi.org/10.1136/bmjgh-2022-009032 de silva, a., yanotti, m., sinclair, s. & angelopoulos, s. (2023) place-based policies and nowcasting. australian economic review 56(3), 363-370. https://doi.org/10.1111/1467-8462.12526 gray, e., evans, a., & reimondos, a. (2022). having babies in times of uncertainty: first results of the impact of covid-19 on the number of babies born in australia. australian population studies, 6(1), 15-30. https://doi.org/10.37970/aps.v6i1.101 gruen, d. (2020). measuring the impacts of covid-19. speech to the australian business economists, 30th april 2020. https://www.abs.gov.au/websitedbs/d3310114.nsf/home/australian%20statististian%20%20speeches%20-%20measuring%20the%20impacts%20of%20covid19:%20briefing%20to%20the%20australian%20business%20economists gruen, d. (2022a). realising the potential of data in government. speech to the institute of public administration australia, 15th june 2022. https://www.abs.gov.au/about/ourorganisation/australian-statistician/speeches/realising-potential-data-government gruen, d. (2022b). a tale of two pandemics in australia: 1918-19 influenza and covid-19. speech to the 20th australian population association conference, 23rd november 2022. https://www.abs.gov.au/about/our-organisation/australian-statistician/speeches/20thaustralian-population-association-conference-opening-address gruen, d. (2023). supporting analysis of the life course. speech to the life course centre data for policy summit, 15th august 2023. https://www.abs.gov.au/about/our-organisation/australianstatistician/speeches/supporting-analysis-life-course hambur, j., montaigne, m., parsons, s. & whalan, e. (2022). looking under the lamppost or shining a new light: new data for unseen challenges. commonwealth treasury. https://www.abs.gov.au/articles/how-administrative-data-improved-quality-2021-census https://www.abs.gov.au/methodologies/national-state-and-territory-population-methodology/mar-2023 https://www.abs.gov.au/methodologies/national-state-and-territory-population-methodology/mar-2023 https://www.abs.gov.au/statistics/people/population/administrative-data-snapshot-population-and-housing-experimental-population-data/30-june-2021 https://www.abs.gov.au/statistics/people/population/administrative-data-snapshot-population-and-housing-experimental-population-data/30-june-2021 https://www.abs.gov.au/statistics/people/housing/administrative-data-snapshot-population-and-housing-experimental-housing-data/30-june-2021 https://www.abs.gov.au/statistics/people/housing/administrative-data-snapshot-population-and-housing-experimental-housing-data/30-june-2021 https://www.abs.gov.au/abs-responds-covid-19 https://www.aihw.gov.au/reports/covid-19/covid-19-linked-data-set-linkage-results/contents/about https://www.aihw.gov.au/reports/covid-19/covid-19-linked-data-set-linkage-results/contents/about https://doi.org/10.1111/1467-8462.12524 https://population.gov.au/research/research-migration-between-cities-and-regions-a-quick-guide-to-covid-19-impacts https://population.gov.au/research/research-migration-between-cities-and-regions-a-quick-guide-to-covid-19-impacts https://population.gov.au/publications/statements/2021-population-statement http://dx.doi.org/10.1136/bmjgh-2022-009032 https://doi.org/10.1111/1467-8462.12526 https://doi.org/10.37970/aps.v6i1.101 https://www.abs.gov.au/websitedbs/d3310114.nsf/home/australian%20statististian%20-%20speeches%20-%20measuring%20the%20impacts%20of%20covid-19:%20briefing%20to%20the%20australian%20business%20economists https://www.abs.gov.au/websitedbs/d3310114.nsf/home/australian%20statististian%20-%20speeches%20-%20measuring%20the%20impacts%20of%20covid-19:%20briefing%20to%20the%20australian%20business%20economists https://www.abs.gov.au/websitedbs/d3310114.nsf/home/australian%20statististian%20-%20speeches%20-%20measuring%20the%20impacts%20of%20covid-19:%20briefing%20to%20the%20australian%20business%20economists https://www.abs.gov.au/about/our-organisation/australian-statistician/speeches/realising-potential-data-government https://www.abs.gov.au/about/our-organisation/australian-statistician/speeches/realising-potential-data-government https://www.abs.gov.au/about/our-organisation/australian-statistician/speeches/20th-australian-population-association-conference-opening-address https://www.abs.gov.au/about/our-organisation/australian-statistician/speeches/20th-australian-population-association-conference-opening-address https://www.abs.gov.au/about/our-organisation/australian-statistician/speeches/supporting-analysis-life-course https://www.abs.gov.au/about/our-organisation/australian-statistician/speeches/supporting-analysis-life-course australian population studies 7 (1) 2023 leigh 29 https://www.rba.gov.au/publications/other-confs/abs-and-rba-joint-conferences/2022/pdf/absrba-conference-2022-hambur-montaigne-parsons-whalan.pdf kennedy, s. (2022). the 2022 sir leslie melville lecture. speech delivered to the australian national university, 27th july 2022. https://treasury.gov.au/speech/2022-sir-leslie-melville-lecture leigh, a. (2022). health inequalities in the covid pandemic: evidence from australia. australian population studies, 6(2), 33-44. https://doi.org/10.37970/aps.v6i2.108 shi, s., & phillips, p. c. b. (2023). housing fever in australia 2020-23: insights from an econometric thermometer. australian economic review, 56(3), 357-362. https://doi.org/10.1111/14678462.12523 https://www.rba.gov.au/publications/other-confs/abs-and-rba-joint-conferences/2022/pdf/abs-rba-conference-2022-hambur-montaigne-parsons-whalan.pdf https://www.rba.gov.au/publications/other-confs/abs-and-rba-joint-conferences/2022/pdf/abs-rba-conference-2022-hambur-montaigne-parsons-whalan.pdf https://treasury.gov.au/speech/2022-sir-leslie-melville-lecture https://doi.org/10.37970/aps.v6i2.108 https://doi.org/10.1111/1467-8462.12523 https://doi.org/10.1111/1467-8462.12523 australian population studies 2022 | volume 6 | issue 2 | pages 33-44 © andrew leigh 2022. published under the creative commons attribution-noncommercial licence 3.0 australia (cc by-nc 3.0 au). journal website: www.australianpopulationstudies.org apa conference health inequalities in the covid pandemic: evidence from australia andrew leigh parliament of australia email: andrew.leigh.mp@aph.gov.au. address: parliament house, canberra act 2600, australia received 25 november 2022; accepted 28 november 2022; published 19 december 2022 this is an edited version of the w.d. borrie lecture given by the hon. dr andrew leigh mp, assistant minister for competition, charities and treasury, at the 20th australian population association in canberra on 23 november 2022 1. introduction i acknowledge the ngunnawal people, traditional custodians of the land on which we gather today, pay my respects to their elders past and present, and commit myself to the full implementation of the uluru statement from the heart. it is my pleasure to be invited to talk at the australian population association’s 2022 conference, and an honour to be giving the w. d. borrie lecture. born in new zealand in 1913, wilfred david ‘mick’ borrie was at the vanguard of population studies in the 1950s. at a time when australia’s post-war migration was booming, it made sense for the nation to be studying demography. in 1952, four years after borrie joined the australian national university’s demography group, it was formally designated as a department. this made it the first demography department in the world. five years later, borrie was promoted to chair. that made him the world’s first professor of demography (foster and varghese 2009). borrie was prodigious, authoring or co-authoring 15 books and 163 articles or reports (price 2000). when i joined the research school of social sciences in 2004, a quarter century after borrie’s retirement, his spirit remained strong in the demography department. demographers held regular seminars in the borrie seminar room, a location guaranteed to produce spirited engagement from the attendees. due to the uniquely complicated layout of the coombs building, it was all but off-limits to outsiders. when occasionally a non-coombs dweller stumbled into a seminar, they invariably bore the expression of someone who has just found their way through an especially complicated hedge maze. i learned a great deal from demography colleagues in the research school of social sciences, including doyens of the discipline peter mcdonald, ann evans, terry hull and edith gray. http://www.australianpopulationstudies.org/ mailto:andrew.leigh.mp@aph.gov.au 34 leigh australian population studies 6 (2) 2022 it has been said that demography tells the story of our lives and the last few years have been quite a story. today i will analyse covid mortality, and in particular how the health burden of the pandemic has fallen hardest on the most disadvantaged people in the australian community. 2. provisional mortality statistics it is possible to undertake this analysis only because of new data collection efforts and fresh research that has been built off these data. more timely data in recent years has helped australians see what has been happening during the pandemic. and this in turn has allowed policy makers to respond in an informed way, faster than ever before. the agility of the australian bureau of statistics (abs) has greatly aided this effort. provisional mortality statistics were first released on 24 june 2020 to account for the first 3 months of the pandemic, and continued to be produced each month thereafter as a way of tracking mortality through the covid pandemic. specifically, the report aimed to identify deaths due to covid and non-covid causes – recognising that the disruption of the pandemic could potentially impact mortality in unexpected ways. legislation in australia mandates that a death can only be registered after a burial or cremation has occurred, hence there can be a delay in the abs receiving death registrations. to minimise these delays, the abs’s preliminary mortality reports initially produced information on doctor-certified deaths only, which are finalised more quickly than coroner-certified deaths. the benefit of producing information on doctor-certified deaths was that detailed cause of death information for a range of causes could be produced. cause-of-death information can indicate if non direct effects of the pandemic such as access to healthcare have an impact on mortality. as the pandemic progressed, questions arose around mortality regarding mental health, suicide, substance use and risk-taking behaviours – all deaths that are more likely to be certified by a coroner. while causes of death for a coroner-certified death are not available quickly, the decision was made to include coroner certified deaths at the ‘all cause’ level into the publication from april this year. this has then allowed for any changes in coroner certified deaths to be tracked at a broad level, providing early indications into the patterns of deaths that may be considered preventable. the abs also introduced reports specifically on covid mortality in australia, which provide provisional detail about covid mortality and how it has been distributed across australian population groups. 3. covid mortality the abs covid-19 mortality in australia publication gives us an insight into what happened in the first 2 years of the pandemic (abs 2022a, 2022b, 2022c) (figure 1). in 2020 there were 906 deaths due to covid, most of which occurred in victoria during the first 2 waves of the pandemic. in 2021 there were 1,345 covid deaths, with the majority occurring in new south wales and victoria during the delta wave. as you can see from figure 2, covid mortality remained very low in australia compared to other countries (johns hopkins 2022). australian population studies 6 (2) 2022 leigh 35 figure 1: weekly covid deaths in australia, august 2021-july 2022 source: australian bureau of statistics figure 2: covid mortality: international comparison, february 2020-november 2022 source: johns hopkins university note: daily new covid deaths per million people, 7 day rolling average 36 leigh australian population studies 6 (2) 2022 on a technical note, i should say that these data from johns hopkins university are measured on a different basis to official abs statistics, as well as being different across countries. the united kingdom and united states suffered significant losses of life. using this johns hopkins data, across 2020 and 2021 combined, australia suffered 86 deaths per million people. in the uk and us there were around 2,500 deaths per million people, around 30 times as high as experienced in australia. estimates of global excess deaths tell a grimmer story. excess deaths are the number of observed deaths above or below the number of deaths that would otherwise have been expected in a given time period. during 2020 and 2021, excess deaths have been estimated at a staggering 15 million worldwide (who 2021). the largest proportion of excess deaths is estimated to have occurred in lower-middle income countries, such as india, egypt and indonesia. even this is likely to be a significant underestimate, with data on deaths in 41 of 54 countries in africa not sufficiently reliable to prepare estimates. remarkably, australia went in the opposite direction. as philip clarke and i documented in a recent paper published in bmj global health, the mortality rate in australia was 5.9 per cent lower in 20202021 than in 2015-2019 (clarke and leigh 2022). contrary to the claims of those who suggested that lockdowns would increase the death rate, we find that lockdowns and social distancing reduced short-term mortality. as you might expect, we observe a reduction in mortality from infectious diseases. for example, influenza typically claims around 600 australian lives per year – yet in 2021 it claimed fewer than 5 lives. however, we also see a reduction in mortality due to non infectious causes of death, including cancer and heart attacks. we speculate that this may be because infectious diseases predispose people to other causes of death, but it is also possible that working from home had a positive health effect. our research contributes to the global discussion around the short-term health effects of lockdowns. it would not have been possible without the abs’s publication of preliminary mortality statistics. in 2022, covid infection rates and mortality in australia climbed considerably. in the first 9 months of 2022, covid accounted for 8,028 doctor certified deaths. • across all waves of the pandemic, deaths from covid were highest among those aged 80 89 years. • the median age of those who died from covid was 87.4 years for females and 83.6 years for males. • males had a higher number of registered covid deaths than females. for every 100 female covid deaths, there were 126 male covid deaths. • around 3-quarters of all covid deaths occurred in victoria and new south wales. • deaths due to covid were more prevalent in areas with greater socio-economic disadvantage. in 2022, unlike 2020 and 2021, australia had a similar covid-19 death rate to other advanced countries. however, australian death rates remain well below the peaks seen in other countries in the early parts of the pandemic. australian population studies 6 (2) 2022 leigh 37 using johns hopkins data on covid deaths, from 1 january to 4 november 2022, there were more than 500 deaths per million australians. this was a little higher than in canada and the uk, which saw 441 and 476 deaths per million people respectively. it’s lower than the us, which saw 732 deaths per million people over the same period. by now, a great many australians have been touched by covid. there have been more than 10 million confirmed cases of covid, and because reporting of rapid antigen tests is no longer mandatory, it is likely that a majority of australians have contracted the disease. according to official abs statistics, there have been 10,279 deaths from covid since the start of the pandemic until 30 september 2022. these are people whose underlying cause of death was covid. a few additional points on covid mortality. in addition to deaths from covid, the abs also classifies 2,266 deaths as covid-related. these are people who died from other causes (such as cancer) but for whom covid contributed to their death. including these deaths brings deaths from or with covid to 12,545. another key data source is the national notifiable diseases surveillance system. according to this dataset, there were 14,718 deaths associated with covid from the start of the pandemic until 30 september 2022. statistics derived from surveillance systems are not directly comparable with figures from the abs. 4. inequality in covid mortality in australia one of the harshest impacts of economic inequality is the differences in health outcomes across socio-economic groups. in terms of life expectancy, we know that those in the highest income quintile live six years longer than those in the lowest income quintile (clarke and leigh 2011). health gaps show up in other areas too. the 2017-18 national study of adult oral health allows us to compare dental outcomes for low income australians (with a household income below $40,000) and high-income australians (with a household income above $100,000). in the richest households, 0.5 per cent of adults are missing all their teeth. in the poorest households, 10.3 per cent of adults are missing all their teeth (peres and lalloo 2020). among dentate respondents, the average number of missing teeth is 4.4 in richer households and 8.0 in poorer households. these are gaps that show with every smile. yet in general, too little is known about the distribution of health burdens across the population. in the case of the mortality study referenced above, we used a longitudinal study (the household income and labour dynamics in australia survey, or hilda), and matched it with deaths data. this allowed us to identify those who had died, and study their last reported survey responses to questions about income. more typically, studies of socio-economic differences in mortality do not have this kind of information about the deceased. instead, they focus on the average income in the geographic area where the deceased last resided. this can be informative – though as our research shows, area-level metrics are no longer statistically significant after controlling for individual-level metrics (clarke and leigh 2011). 38 leigh australian population studies 6 (2) 2022 in the case of covid deaths, the available data presently only allows us to look at regional average incomes. it is useful to explore these patterns, but also worth bearing in mind that the true socioeconomic gaps – if we could measure individual incomes – are likely larger still. table 1 shows covid deaths by seifa quintile. socio-economic indexes for areas (seifa) is a product developed by the abs that ranks areas in australia according to relative socio-economic advantage and disadvantage. the table shows that covid mortality was higher in areas of greater socioeconomic disadvantage. the number of people who died due to covid was over 3 times higher among those living in the most disadvantaged areas when compared to those living in the least disadvantaged areas. around one-third of all deaths from and with covid (covid associated deaths) in 2020 to 2022 were in the most disadvantaged quintile, compared to 11 per cent in the least disadvantaged quintile. table 1: socio-economic status of australians who died from covid, january 2020-september 2022 source: australian bureau of statistics note: index of relative social disadvantage (seifa/irsd) quintile of those who died from covid, deaths registered by 30 september 2022 figure 3 breaks down these patterns over time. adjusting for age, covid death rates were 2.3 times higher in 2020 for the most disadvantaged quintile compared to the least disadvantaged quintile. in 2021, they were 4 times higher. in 2022, despite widespread increases in covid infections, people in the most disadvantaged quintile suffered death rates 1.7 times higher than the least disadvantaged quintile. i hope that future researchers will be able to shed light on why it is that the largest socio-economic differences in covid deaths were in the 2021 delta wave. it may be relevant that other disparities were also larger. for example, the male-female mortality gap was bigger in the delta wave than it was in the omicron wave or in the two 2020 waves. notably, the socioeconomic patterns are not specific to younger or older australians. in each age group, people in the most disadvantaged seifa quintile had a higher chance of dying of covid. let’s turn now to country of birth. before the pandemic, the mortality of residents born overseas was lower than those born in australia. this was particularly clear between the ages of 10 and 60. one likely explanation may be the presence of a selection effect for those moving to australia, known as australian population studies 6 (2) 2022 leigh 39 the ‘healthy migrant effect’ (centre for population and australia government actuary 2021). however, during the pandemic, australian residents born overseas have had consistently higher death rates from and with covid than the australian-born population. figure 3: covid mortality by socioeconomic quintile (age-standardised) source: australian bureau of statistics note: covid associated mortality for deaths registered and received by the abs until the end of september 2022 figure 4 shows that in the first year of the pandemic, the age standardised covid death rate for people born overseas was more than twice as high as that for those born in australia. in 2021, this gap widened to become 3.6 times higher for those born overseas. from january to september 2022, the covid mortality rate for people born overseas was 1.3 times higher than for people born in australia. again, the socio-economic gap in covid mortality was largest in 2021. there are also considerable differences among the overseas-born population. from the beginning of the pandemic until the end of september this year, people born in the middle east had the highest age-standardised death rate, at 41 covid deaths per 100,000 people. this is more than 4 times higher than the rate experienced by those born in australia. by contrast, at 9 deaths per 100,000, people born in sub saharan africa had a lower age-standardised covid death rate than people born in australia. first nations australians are at heightened risk of more severe outcomes from covid than non-first nations australians. there are several reasons for this, including higher rates of socioeconomic disadvantage, higher prevalence of chronic diseases and limited access to culturally safe healthcare. 40 leigh australian population studies 6 (2) 2022 figure 4: covid mortality by country of birth source: australian bureau of statistics note: covid associated deaths by place of birth, age-standardised death rate, deaths registered by 30 september 2022 figure 5 shows covid mortality broken down by indigenous status. unlike previous charts, this analysis includes deaths with covid and deaths from covid – in other words, it includes deaths where covid was not the main cause of death. accordingly, the mortality rates shown in figure 5 are higher than in previous charts. the comparison shows that the rate of death from or with covid for first nations australians is 1.7 times higher than for non-indigenous australians. for deaths with covid – that is, where covid was a contributory cause but not the underlying cause – this difference was particularly stark. among first nations women, the rate of mortality with covid is close to 3 times higher than that of non-indigenous females. among first nations men, the ratio is twice as high. this difference has emerged later in the pandemic. there were no covid associated deaths of first nations people during wave 1 and wave 2 of the pandemic. there were 20 deaths during the delta wave, with the remaining 158 deaths occurring during the omicron wave. during the omicron wave the largest number of deaths have so far been recorded as occurring in february (34), followed by january with 24 deaths. the number of deaths for later months is expected to increase as additional information on deaths is received. here are some statistics to give you an idea of how different covid mortality outcomes are for first nations australians. • a higher proportion of first nations people died with covid as a contributing factor compared with non-indigenous people (33.1 per cent compared with 22.1 per cent). • the rate of covid mortality in first nations people aged between 55 64 is 4.5 times higher than non-indigenous people of the same age. australian population studies 6 (2) 2022 leigh 41 • first nations people who died from covid had higher rates of diabetes, chronic kidney disease, chronic respiratory diseases and hypertension listed as pre-existing chronic conditions compared to non-indigenous people. • first nations people who died from covid had an average of 3.9 conditions listed on the death certificate. this compares to an average of 3.4 conditions on the death certificate among nonindigenous people. according to the latest closing the gap report, the life expectancy gap between first nations and non-indigenous australians is 8.6 years for males and 7.8 years for females (australian government 2022). the disproportionate mortality impact of covid on first nations australians has served to widen this disparity. figure 5: covid mortality by indigenous status (includes deaths with and from covid) source: australian bureau of statistics note: deaths from or with covid among first nations people, age-standardised death rate, deaths registered by 30 september 2022 5. inequality in covid mortality – how australia compares an unequal distribution of covid mortality is not unique to australia. in the us, demographers from the university of california, berkeley, found that in the first wave, between march and may of 2020, covid mortality was highest in the most advantaged counties (dukhovnov and barbieri 2022). this pattern may reflect the significantly higher spread of the virus in the north-east of the united states, particularly new york. over the second half of 2020 this pattern reversed, such that covid mortality rates were 2.6 times higher in the most disadvantaged quintile than in the least. this is close to the australian figure of 2.3 for the entirety of 2020. across the entire period from 2020 to 2022, the mortality burden in the united states has fallen most heavily on the most disadvantaged. 42 leigh australian population studies 6 (2) 2022 in the uk, those born overseas had higher death rates (ons 2022). during the first wave of the pandemic, people from ethnic minority groups generally had higher rates of death involving covid compared with the white british population. in the first wave, the rate of death for the black african group was 2.7 times greater than the white british group for males, and 2.6 greater for females. this ratio is similar to the difference observed in australia between overseas born and australian born people during the first 2 years of the pandemic. in the second covid wave, british people of bangladeshi ancestry had the highest rates, 5.0 and 4.5 times greater than for white british males and females respectively. after adjusting for location – measures of disadvantage, occupation, living arrangements, and pre-existing health conditions accounted for a large proportion of the excess covid mortality risk in most ethnic groups. although even after accounting for these factors, most black and south asian groups remained at higher risk. in canada, some neighbourhoods had noticeably higher age standardised covid death rates in 2020 (subedi and aitken 2022). the densely populated urban neighbourhoods, characterised by very high proportions of migrants, lone-parent families, and low-income families had significantly higher covid mortality rates than other neighbourhoods. death rates in these neighbourhoods were almost twice as high as in high socio-economic status urban neighbourhoods. what might account for these differences? us researchers have hypothesised that mortality rates may be higher among low-income populations because they are over represented in the essential workforce and economic sectors in which it is difficult to work remotely, thereby increasing their exposure to covid (dukhovnov and barbieri 2022). in addition, low-income people rely more on public transport and are more likely to live in large and multigenerational households. 6. conclusion as we have seen, the mortality impact of the covid pandemic has been felt unevenly across socioeconomic groups in australia. this should not have been a surprise, since there were uneven mortality outcomes across socio economic groups in australia well before covid. across a range of indicators, it appears that the disparity was worse in 2021 than in 2020 or 2022. the mortality ratios from covid in australia are quite similar to those estimated in other advanced nations. as a share of the population, fewer people died from covid in australia than in most other affluent nations. yet among those who died, the same health inequalities can be seen in australia as in other advanced countries. what might have driven the socioeconomic disparities in covid mortality? and why might many of those disparities have been largest in the delta wave? as i have noted, disadvantaged people may be less able to work remotely, more reliant on public transport, and more likely to live in crowded households. uptake of vaccination and antiviral treatments have varied across society as vaccines and treatment became increasingly available. another factor is that successive covid waves have had varying degrees of severity. a final factor is that in the years since covid began, population immunity has steadily risen. although geographic-level measures of inequality have typically been used in research looking at mortality and health outcomes, they are statistically insignificant after controlling for individual-level australian population studies 6 (2) 2022 leigh 43 measures of socio-economic status (clarke and leigh 2011). this means that individual-level measures like household income and educational attainment should be the preferred measure we use in health research. however, this data is not always available, which makes it necessary to look at other measures, such as neighbourhood seifa indices. as the minister with responsibility for the abs, i am proud to see it producing the type of data that enables researchers like yourselves (and occasionally myself) to glean novel insights. the abs has made great strides in this regard and now provides access to linked datasets such as the multiagency data integration project (madip). the australian government also wants to provide timely data. the department of health and aged care has funded the abs to continue the provisional mortality statistics, as well as reports on covid mortality and excess deaths for this financial year. reducing health inequality is a priority for our government, and we are acutely aware of the value of high-quality research in helping us understand the problem. only by shining a spotlight onto the health gaps that exist between advantaged and disadvantaged australians can we begin to develop policies to narrow these disparities. acknowledgements my thanks to officials in the australian treasury’s centre for population for invaluable assistance in preparing this talk. references abs (2022a). covid-19 mortality in australia: deaths registered until 30 september 2022, release date 27 october 2022, australian government, accessed november 2022. https://www.abs.gov.au/articles/covid-19-mortality-australia-deaths-registered-until-30september-2022 abs (2022b). provisional mortality statistics, january – july 2022, release date 27 october 2022, australian government, accessed november 2022. https://www.abs.gov.au/statistics/health/causes-death/provisional-mortality-statistics/jan-jul2022 abs (2022c). covid mortality by wave, release date 16 november 2022, accessed november 2022. https://www.abs.gov.au/articles/covid-19-mortality-wave australian government (2022). commonwealth closing the gap annual report 2022. australian government. https://www.niaa.gov.au/node/127616 australian institute of health and welfare (aihw) (2022a). deaths in australia. australian government, accessed november 2022. https://www.aihw.gov.au/reports/life-expectancy-death/deaths-inaustralia/contents/summary australian institute of health and welfare (aihw) (2022b). health across socioeconomic groups, release date 7 july 2022, australian government, accessed november 2022. https://www.aihw.gov.au/reports/australias-health/health-across-socioeconomic-groups australian government actuary (aga) and centre for population (cpop) (2021). sub-group mortality: new insights using australian microdata, accessed november 2022. https://population.gov.au/publications/research/sub-group-mortality-microdata-approachresident-sub-group-life-tables https://www.abs.gov.au/articles/covid-19-mortality-australia-deaths-registered-until-30-september-2022 https://www.abs.gov.au/articles/covid-19-mortality-australia-deaths-registered-until-30-september-2022 https://www.abs.gov.au/statistics/health/causes-death/provisional-mortality-statistics/jan-jul-2022 https://www.abs.gov.au/statistics/health/causes-death/provisional-mortality-statistics/jan-jul-2022 https://www.abs.gov.au/articles/covid-19-mortality-wave https://www.niaa.gov.au/node/127616 https://www.aihw.gov.au/reports/life-expectancy-death/deaths-in-australia/contents/summary https://www.aihw.gov.au/reports/life-expectancy-death/deaths-in-australia/contents/summary https://www.aihw.gov.au/reports/australias-health/health-across-socioeconomic-groups https://population.gov.au/publications/research/sub-group-mortality-microdata-approach-resident-sub-group-life-tables https://population.gov.au/publications/research/sub-group-mortality-microdata-approach-resident-sub-group-life-tables 44 leigh australian population studies 6 (2) 2022 clarke, p., & leigh, a. (2011). death, dollars and degrees: socio‐economic status and longevity in australia, economic papers, 30(3), 348-355. https://doi.org/10.1111/j.1759-3441.2011.00127.x clarke, p. & leigh, a. (2022). understanding the impact of lockdowns on short-term excess mortality in australia, bmj global health, 7, article e009032. http://dx.doi.org/10.1136/bmjgh-2022-009032 dukhovnov, d., & barbieri, m. (2022). county-level socio-economic disparities in covid mortality in the usa, international journal of epidemiology, 51(2), 418-428. https://doi.org/10.1093/ije/dyab267 foster, s. & varghese, m. (2009). the making of the australian national university, 1946-1996. anu press, canberra. http://doi.org/10.22459/manu.08.2009 johns hopkins university (johns hopkins) (2022). mortality analyses, coronavirus resource centre, last updated 9 november 2022, johns hopkins university & medicine. https://coronavirus.jhu.edu/data/mortality office for national statistics (ons) (2022). updating ethnic contrasts in deaths involving the coronavirus (covid), england: 8 december 2020 to 1 december 2021, release date 26 january 2022, united kingdom government, accessed november 2022. https://www.ons.gov.uk/peoplepopulationandcommunity/birthsdeathsandmarriages/deaths/arti cles/updatingethniccontrastsindeathsinvolvingthecoronaviruscovid19englandandwales/8decemb er2020to1december2021 peres, m., & lalloo, r. (2020). tooth loss, denture wearing and implants: findings from the national study of adult oral health 2017–18, australian dental journal, 65(s1), s23-s31. https://doi.org/10.1111/adj.12761 price, c. a. (2000). wilfred david (mick) borrie, cbe 1913-2000. journal of population research, 17, v-vi. https://doi.org/10.1007/bf03029444 subedi, r. & aitken, n. (2022). inequalities in covid mortality rates by neighbourhood types in canada, release date 9 may 2022, statistics canada, accessed november 2022. https://www150.statcan.gc.ca/n1/pub/45-28-0001/2022001/article/00006-eng.htm world health organization (who) (2021). global excess deaths associated with covid (modelled estimates), latest update 5 may 2022; first published 20 may 2021, world health organization. https://www.who.int/data/sets/global-excess-deaths-associated-with-covid-19-modelledestimates https://doi.org/10.1111/j.1759-3441.2011.00127.x http://dx.doi.org/10.1136/bmjgh-2022-009032 https://doi.org/10.1093/ije/dyab267 http://doi.org/10.22459/manu.08.2009 https://coronavirus.jhu.edu/data/mortality https://www.ons.gov.uk/peoplepopulationandcommunity/birthsdeathsandmarriages/deaths/articles/updatingethniccontrastsindeathsinvolvingthecoronaviruscovid19englandandwales/8december2020to1december2021 https://www.ons.gov.uk/peoplepopulationandcommunity/birthsdeathsandmarriages/deaths/articles/updatingethniccontrastsindeathsinvolvingthecoronaviruscovid19englandandwales/8december2020to1december2021 https://www.ons.gov.uk/peoplepopulationandcommunity/birthsdeathsandmarriages/deaths/articles/updatingethniccontrastsindeathsinvolvingthecoronaviruscovid19englandandwales/8december2020to1december2021 https://doi.org/10.1111/adj.12761 https://doi.org/10.1007/bf03029444 https://www150.statcan.gc.ca/n1/pub/45-28-0001/2022001/article/00006-eng.htm https://www.who.int/data/sets/global-excess-deaths-associated-with-covid-19-modelled-estimates https://www.who.int/data/sets/global-excess-deaths-associated-with-covid-19-modelled-estimates australian population studies 2021 | volume 5 | issue 1 | pages 77-100 © wilson & rees 2021. published under the creative commons attribution-noncommercial licence 3.0 australia (cc by-nc 3.0 au). journal website: www.australianpopulationstudies.org introductory guide a brief guide to producing a national population projection tom wilson* the university of melbourne philip rees the university of leeds * corresponding author. email: wilson.t1@unimelb.edu.au. melbourne school of population and global health, the university of melbourne, melbourne, vic 3010, australia. paper received 10 february 2021; accepted 23 march 2021; published 31 may 2021 abstract background there are surprisingly few resources available which offer an introductory guide to preparing a national population projection using a cohort-component model. many demography textbooks cover projections quite briefly, and many academic papers on projections focus on advanced technical issues. aims the aim of this paper is to provide a short and accessible guide to producing a national-scale population projection using the cohort-component model. data and methods the paper describes the cohort-component model from a population accounting perspective, presents all the necessary projection calculations, and covers the key steps which form part of the projections preparation process – from gathering input data to validating outputs. an accompanying excel workbook implements the model and contains example projections for australia. conclusions calculating a national population projection using a cohort-component model involves fairly simple algebra, but the broader projections preparation process is more complex, and requires careful consideration and judgement. key words population projections; cohort-component model; population accounting; australia http://www.australianpopulationstudies.org/ mailto:wilson.t1@unimelb.edu.au 78 wilson & rees australian population studies 5 (1) 2021 1. introduction national statistical offices and some international agencies, such as the united nations, produce national population projections on a regular basis. projections are used for many purposes, including calculating the amount of additional housing needed in coming years, estimating future aged care demand, assessing the fiscal implications for government of future population growth and ageing (e.g., australian government 2015), and contributing to national debates about population change. researchers also produce national population projections from time to time in order to understand demographic change, explore alternative demographic scenarios, or demonstrate methodological refinements, among other reasons (e.g., lomax et al. 2019; wilson et al. 2021). surprisingly, the guidance available on how to create national population projections is quite limited. many academic papers and books on projections, as would be expected, report the latest research and usually focus on advanced technical aspects of projections. there are relatively few handbooks or websites which provide introductory advice on how to create population projections. examples of resources which provide guidance include a manual on projections published by the united nations (1989), the unfpa/iussp web-based demography course which includes a section on projections (timaeus 2015), and manuals for specific population projection software products (e.g., shorter et al. 1995; stover and kirmeyer 2007; us census bureau 2019); and the practitioner’s guide by smith et al. (2013) is a good example for subnational projections. most demography textbooks do include a section on the principal model used to create national population projections – the cohortcomponent model – though the description is often brief. and the model usually includes only net international migration (which has some limitations) rather than separate immigration and emigration flows. the aim of this paper is to present a short practical guide to producing a national population projection. our goal is to offer an accessible introduction to projection modelling which avoids complex algebra but includes sufficient mathematical detail so that readers can produce their own projections if desired. we present the cohort-component model from a population accounting perspective because it offers an intuitive and transparent way of expressing the model. this approach is based on the well-known population accounting (or balancing) equation, and involves projecting the demographic components of change as the product of occurrence/exposure rates and populations-at-risk. our description is less directly connected to the life table than some other presentations of the cohort-component model, and we do not express the model in matrix algebra as many researchers do1. our projection equations also differ somewhat from those used by the australian bureau of statistics for their population projections (abs 1999). primarily, the description in this paper is most relevant to countries with good statistical systems and reasonable quality demographic data. we begin with a few preliminary terms and concepts (section 2) and then give an overview of the projections process (section 3). the cohort-component model forms the focus of section 4, while section 5 covers the preparation of necessary input data and assumptions. in section 6 we briefly describe how projection outputs can be checked for plausibility and consistency, before a few 1 readers interested in matrix arrangements of the cohort-component model are recommended to consult keyfitz and caswell (2005: chapter 3) or rogers (1995: chapter 3). australian population studies 5 (1) 2021 wilson & rees 79 concluding remarks are given in section 7. accompanying this paper is an excel workbook which incorporates a cohort-component model for national population projections, illustrated with example projections for australia. it is available from https://doi.org/10.6084/m9.figshare.13565222.v3. 2. preliminaries before preparing any national-scale population projections, it is very useful to have a reasonable knowledge of: • projections terminology, • life tables, • the lexis diagram, and • population accounting. we also assume knowledge of basic demographic concepts and terms, which are usually covered in the initial chapters of introductory demography textbooks (e.g., preston et al. 2001; rowland 2003). 2.1. projections terminology box 1 defines terms population projections terms used in this paper. box 1: definitions jump-off year: the year from which the projections ‘jump off’ into the demographic future. it is sometimes also known as the launch year. population estimate: the population for a past point in time based on information available for that time. population projection: a calculation of population beyond the jump-off year based on selected assumptions about the future drivers of population change (commonly, fertility, mortality and migration). population forecast: a population projection regarded as the most likely outcome; a prediction. projection horizon: the time period between the jump-off year and the final year of the projections. projection interval: the time interval between each projected population (usually 1 or 5 years). projection assumptions: data inputs to the projection calculations which usually include the assumed future paths of fertility, mortality, and migration (and sometimes other variables). more broadly defined, projection assumptions also implicitly or explicitly include qualitative elements, such as expectations about future immigration policy. population-at-risk: the number of person-years lived in a projection interval by a particular population exposed to the ‘risk’ of a demographic event. it is often approximated as the mean of the start and end populations of a projection interval. (occurrence/exposure) rate: the ratio of the number of demographic events occurring divided by the population exposed to the risk of that event (population-at-risk). source: wilson (2019); authors’ definitions https://doi.org/10.6084/m9.figshare.13565222.v3 80 wilson & rees australian population studies 5 (1) 2021 2.2. the life table a life table is a model of a cohort’s mortality experience over its lifetime, but it is often calculated for a particular year or group of years and therefore describes a synthetic cohort. although we do not make extensive use of life table measures in this paper, it is helpful to understand the calculations behind, and meaning of, nlx, the life table stationary population for each age group. in life table notation, 𝑥 is the exact starting age of an age group and 𝑛 is the width of the age group in years. the life table is covered comprehensively in demography textbooks, so is not described here. readers unfamiliar with the life table are recommended to consult an appropriate source (e.g. kinter 2004; preston et al. 2001: chapter 3; rowland 2003: chapter 8; sloggett 2015). we return to life tables in section 5 on input data preparation. 2.3. the lexis diagram (or age-time plan) the lexis diagram consists of a graph with age increasing up the vertical axis and time progressing along the horizontal axis. it provides a very useful way of precisely locating the age, time and cohort extent of demographic stocks and flows in population projections (and, more generally, any demographic analysis). longer discussions on the lexis diagram can be found in calot (2015), pressat (2008: chapter 2), and rau (2018). figure 1 illustrates some key demographic variables. if the lexis diagram is used to illustrate the demographic history of just a handful of individuals then individual lifelines can be drawn, and demographic events (e.g. death, migration, giving birth) marked at points on those lines (figure 1, upper diagram). an individual’s lifeline starts on the horizontal axis and proceeds at a 45º angle because as each year passes, the person ages by 1 year. symbols are placed on lifelines to denote demographic events, such as giving birth, migration, and eventually, death. however, once there are more than just a few lifelines the diagram becomes cluttered, and it becomes necessary to use thick lines to denote all the lifelines crossing a point in time (or exact age) and shapes to represent all demographic events occurring within a particular age-time space. for example, in the lower diagram in figure 1 the thick vertical lines represent the population of the cohort aged 1 year old last birthday at time 𝑡 and aged 2 years at time 𝑡 + 1. the parallelogram shaded grey defines the period-cohort age-time space which refers to demographic events experienced by the cohort over the one year period. the shaded square depicts demographic events defined by period and age. the thick lines at the left and right sides of the square represent populations of that age at the start and end of the time interval. the horizontal thick blue line denotes babies born during a year, and the shaded triangle is the age-period-cohort space representing demographic events occurring to this cohort of newly-born babies in that year. 2.4. population accounting population accounts connect population stocks at two points in time with the demographic components which add to, and deplete, the population (rees 1985; rees and willekens 1986). at the national scale, the population at time 𝑡 + 1 equals the population at time 𝑡 plus births, minus deaths, plus immigration, minus emigration: 𝑃(𝑡 + 1) = 𝑃(𝑡) + 𝐵(𝑡, 𝑡 + 1) − 𝐷(𝑡, 𝑡 + 1) + 𝐼(𝑡, 𝑡 + 1) − 𝐸(𝑡, 𝑡 + 1) (1) australian population studies 5 (1) 2021 wilson & rees 81 3 2  age 1 0 time → 3 2  age 1 0 time → figure 1: the age-time location of lifelines and demographic stocks and flows where: 𝑃 = population 𝑡 = a point in time 𝑡 + 1= a point in time one year later 𝑡, 𝑡 + 1= the one year interval starting at t and ending at t+1 𝐵 = births 𝐷 = deaths 𝐼 = immigration 𝐸 = emigration migration period-age period-cohort age-period-cohort death population population t t+1 births population population population 82 wilson & rees australian population studies 5 (1) 2021 this is the demographic accounting, or balancing, equation for a national population. strictly, it refers to what are known as movement population accounts in which the migration flows are counts of moves across the national border. an alternative population accounting equation in which the migration components refer to counts of migrants (not moves) uses data on people living in different countries at time 𝑡 and time 𝑡 + 1 (as measured by a census). this transition accounts perspective is more complex and is not considered in this paper2. as an example, consider the population accounts for australia over the 12 month period from mid2019 to mid-2020 (abs 2020a). the start-of-interval population 𝑃(𝑡) was 25,365,745, the number of births 304,101, deaths 166,974, immigration 486,757, and emigration 302,588. the end-of-interval population 𝑃(𝑡 + 1) was therefore: = 25,365,745 + 304,101 − 166,974 + 486,757 − 302,588 = 25,687,041. 3. the process for producing a national population projection the population projections process involves a number of steps, starting with gathering data, checking data and making adjustments if necessary, preparing projection assumptions, entering data into your projection program, calculating the projections, and then checking outputs. step 1: obtain the jump-off populations and all data needed to prepare fertility, mortality and migration assumptions all projections begin with the jump-off population, which is usually the most up-to-date set of population estimates published by the national statistical office. at the national scale, projections of fertility and mortality are often created by models which work best with many decades of data. the extent of historical time series required will depend on the precise fertility and mortality projection approaches chosen (discussed in section 5 later). immigration and emigration data are also ideally obtained for the last few decades – or at the very least total immigration and emigration flows for several decades, and by age and sex for the most recent years. all projections begin with the jumpoff population, which is usually the most up-to-date set of population estimates published by the national statistical office. at the national scale, projections of fertility and mortality are often created by models which work best with many decades of data. the extent of historical time series required will depend on the precise fertility and mortality projection approaches chosen (discussed in section 5 later). immigration and emigration data are also ideally obtained for the last few decades – or at the very least total immigration and emigration flows for several decades, and by age and sex for the most recent years. step 2: check that the input data are good quality and that recent population accounts are consistent, and carry out adjustments if necessary past fertility and mortality rates by age are usefully graphed to check for any obvious data problems. the national statistical office which published the data may also have accompanying notes or reports on data quality. in most countries with well-developed statistical systems, the quality and coverage 2 interested readers are recommended to consult rees (1986) or rees and wilson (1977). australian population studies 5 (1) 2021 wilson & rees 83 of births and deaths data are normally good (though not perfect). in australia, the quality of international migration data is very good, but in some countries it is less reliable. a high-level check of population accounting consistency can be made by comparing past • recorded net international migration (calculated as immigration minus emigration), with • residual net migration (calculated as population growth minus recorded natural change). a more detailed check of accounting consistency involves comparing residual net migration with recorded net migration by sex and cohort for a recent period to ensure that the age profiles of immigration and emigration are reasonable. does period-cohort • recorded net migration (immigration minus emigration) equal • residual net migration (calculated as cohort population change plus period-cohort deaths)? if there are large discrepancies, then there is a problem. if the population estimates are deemed more reliable than the migration data, then either or both immigration and emigration values should be adjusted. step 3: prepare assumptions on the expected future direction of fertility, mortality and migration once data quality checks have been completed, and data adjustments made if necessary, then assumptions on the future of fertility, mortality and migration can be prepared. section 5 provides an overview of various approaches and methods which can be used. step 4: enter all input data into the projection program/spreadsheet the next stage involves loading all the input data into computer code or a workbook for producing population projections. even with highly automated processes there is still potential for errors to creep in (e.g. using old assumptions; male and female populations being mixed up), so it is useful to double check all the input data has been entered, and entered in the correct locations. step 5: calculate the projections then the population projections are calculated using the cohort-component model projection equations described next in section 4. the excel workbook accompanying this paper can be used, or you may wish to write your own code. step 6: check outputs for plausibility and consistency, and revise assumptions if necessary an important part of the projection process is checking that the output projections are plausible and internally consistent. unfortunately, it is quite possible, even with carefully prepared projection assumptions, for the cohort-component model to generate implausible or suboptimal outcomes. these outcomes can be caused by data quality issues, random noise in input data, errors in projection assumptions, mistakes in loading assumptions data into the projection program/spreadsheet, projection model coding errors, and the lack of mechanisms in the cohortcomponent model to maintain sensible relationships between certain aspects of the projections. for example, it might be the case that sex ratios become increasingly unlikely over the projection horizon. section 6 covers some of the key checks which should be made to projections output. a further check on the input assumptions and projection model can be made by creating a test run set 84 wilson & rees australian population studies 5 (1) 2021 of projections from a few years ago up to the latest population estimate, and comparing the ‘projected’ populations and components with recorded data. 4. the cohort-component model 4.1. overview the cohort-component model projects the population broken down by cohort by adding and subtracting the demographic components of change. it is usually also disaggregated by sex. the lexis diagram in figure 2 illustrates the age-time spaces of the cohort-specific population accounting equation which lies at the heart of the model. the population of a cohort at the start of the time interval, 𝑡, is depleted by deaths and emigration during the interval, but supplemented by immigration. after taking into account these component changes, we have the cohort population size at the end of the time interval, 𝑡 + 1. final cohort: immigration end-of-interval population aged 101+ start-of-interval population aged 100+ emigration & deaths  age immigration most cohorts: end-of-interval cohort population aged a+1 start-of-interval cohort population aged a emigration & deaths immigration first cohort: end-of-interval cohort population aged 0 births time → emigration & deaths figure 2: defining features of the cohort-component population projection model notes. population stocks are shown by thick black lines, and births by the thick blue line. the age-time extent of immigration, emigration and deaths for the cohorts shown is depicted by the grey shaded areas. age ∞ t t+1 australian population studies 5 (1) 2021 wilson & rees 85 the middle part of the diagram shows the age-time spaces of stocks and flows for all cohorts except the very oldest and youngest. the cohort aged 𝑎 at the start of the projection interval ages to become aged 𝑎 + 1 by the end, and during the year experiences population losses through deaths and emigration, and gains from immigration. the lower part of the diagram illustrates the newlyborn infant cohort where the ‘starting’ population is the number of births which occur during the year. the upper section of the diagram illustrates the slightly different situation at the highest ages. the elongated extent of this part of the diagram indicates the absence of an upper age limit. so, for example, if the oldest open-ended age group in the projections is 100+, then this part of the diagram illustrates the population aged 100+ at time 𝑡 ageing to become the 101+ population at time 𝑡 + 1. there are three main stages to the cohort-component model calculations: • stage 1: project cohorts already alive at the start of the projection interval • stage 2: project the number of births • stage 3: project newly-born babies to age 0 at the end of the projection interval which we describe in turn below. 4.2. stage 1: project existing cohorts in the first stage, all cohort populations alive at the start of the projection interval are projected. this refers to all cohorts except for the cohort of babies born during the interval. the cohort-specific population accounting equation may be written as: 𝑃𝑠,𝑎+1(𝑡 + 1) = 𝑃𝑠,𝑎(𝑡) − 𝐷𝑠,𝑎→𝑎+1 − 𝐸𝑠,𝑎→𝑎+1 + 𝐼𝑠,𝑎→𝑎+1 (2) where: 𝑠 = sex 𝑎 = age group 𝑎 + 1 = age group one year older than 𝑎 𝑎 → 𝑎 + 1 = the period-cohort aged 𝑎 at the start of the time interval which shifts to age 𝑎 + 1 by the end (as shown by the shaded parallelogram in the middle of figure 2). time labels 𝑡, 𝑡 + 1 have been omitted from the components in the equation to avoid clutter. the accounting equation above is converted to a projection equation by substituting deaths and emigration flows for occurrence/exposure rates multiplied by populations-at-risk (rees 1984; willekens & drewe 1984): 𝑃𝑠,𝑎+1(𝑡 + 1) = 𝑃𝑠,𝑎(𝑡) − 𝑑𝑠,𝑎→𝑎+1 1 2 (𝑃𝑠,𝑎(𝑡) + 𝑃𝑠,𝑎+1(𝑡 + 1)) −𝑒𝑠,𝑎→𝑎+1 1 2 (𝑃𝑠,𝑎(𝑡) + 𝑃𝑠,𝑎+1(𝑡 + 1)) + 𝐼𝑠,𝑎→𝑎+1 (3) where: 𝑑 = death rate 𝑒 = emigration rate. 86 wilson & rees australian population studies 5 (1) 2021 the population-at-risk is the number of person-years the cohort population is exposed to the risk of death or emigration over the projection interval. it is approximated by the mean of the startand end-of-interval populations. for the moment, we leave aside the issue of preparing the various assumed age-specific rates and return to this in section 5 on input data and assumptions. in the cohort-component model described here, immigration is projected directly as immigration numbers3. it would be possible in theory to project immigration as the product of immigration rates and the rest of the world’s population, but we would then have to include the rest of the world in the projections to calculate the required population-at-risk for sending migrants to the country we are producing projections for. in any case, immigration to many countries is strongly shaped by the migration policies and regulations of that country. for this reason, it is often projected directly as immigration flows. some cohort-component models use net international migration, but we advise against this wherever possible. see box 2 for a brief note on this issue. by re-arranging the projection equation, the end-of-interval populations can be removed from the right-hand side (willekens and drewe 1984). first, multiplying out the terms in brackets gives: 𝑃𝑠,𝑎+1(𝑡 + 1) = 𝑃𝑠,𝑎(𝑡) − 1 2 𝑑𝑠,𝑎→𝑎+1 𝑃𝑠,𝑎(𝑡) − 1 2 𝑑𝑠,𝑎→𝑎+1 𝑃𝑠,𝑎+1(𝑡 + 1) − 1 2 𝑒𝑠,𝑎→𝑎+1 𝑃𝑠,𝑎(𝑡) − 1 2 𝑒𝑠,𝑎→𝑎+1 𝑃𝑠,𝑎+1(𝑡 + 1) + 𝐼𝑠,𝑎→𝑎+1. then those terms involving the end-of-interval populations are all moved to the left-hand side of the equation: 𝑃𝑠,𝑎(𝑡 + 1) + 1 2 𝑑𝑠,𝑎→𝑎+1 𝑃𝑠,𝑎+1(𝑡 + 1) + 1 2 𝑒𝑠,𝑎→𝑎+1 𝑃𝑠,𝑎+1(𝑡 + 1) = 𝑃𝑠,𝑎(𝑡) − 1 2 𝑑𝑠,𝑎→𝑎+1 𝑃𝑠,𝑎(𝑡) − 1 2 𝑒𝑠,𝑎→𝑎+1 𝑃𝑠,𝑎(𝑡) + 𝐼𝑠,𝑎→𝑎+1. this can be simplified to: (1 + 1 2 𝑑𝑠,𝑎→𝑎+1 + 1 2 𝑒𝑠,𝑎→𝑎+1) 𝑃𝑠,𝑎+1(𝑡 + 1) = (1 − 1 2 𝑑𝑠,𝑎→𝑎+1 − 1 2 𝑒𝑠,𝑎→𝑎+1) 𝑃𝑠,𝑎(𝑡) + 𝐼𝑠,𝑎→𝑎+1 and then re-arranged so that just the end-of-interval population remains on the left-hand side of the equation: 𝑃𝑠,𝑎+1(𝑡 + 1) = (1− 1 2 𝑑𝑠,𝑎→𝑎+1−1 2 𝑒𝑠,𝑎→𝑎+1) (1+1 2 𝑑𝑠,𝑎→𝑎+1+1 2 𝑒𝑠,𝑎→𝑎+1) 𝑃𝑠,𝑎(𝑡) + 𝐼𝑠,𝑎→𝑎+1 (1+1 2 𝑑𝑠,𝑎→𝑎+1+1 2 𝑒𝑠,𝑎→𝑎+1) (4) to illustrate the calculations, we use the example of the female cohort aged 45 years in mid-2019 and 46 in mid-2020. if the population in 2019 was 164,940, the period-cohort death rate 0.001127, the emigration rate 0.005916, and the immigration flow 1,693.3, then the projected population in 2020 is: 3 alternative ways of handling international migration in the cohort-component are often applied. see also bijak (2012), rees (2015), and vandresse (2015). australian population studies 5 (1) 2021 wilson & rees 87 = (1 − 1 2 0.001127 − 1 2 0.005916) (1 + 1 2 0.001127 + 1 2 0.005916) 164,940 + 1,693.3 (1 + 1 2 0.001127 + 1 2 0.005916) = 165,469.84 4.3. stage 2: project the number of births in the second stage, births are projected as the product of age-specific fertility rates and female populations-at-risk, and then summed over all childbearing age groups (usually ages 15 to 49). the total number of projected births is calculated as: 𝐵 = ∑ (𝑏𝑎 1 2 [𝑃𝑓,𝑎(𝑡) + 𝑃𝑓,𝑎(𝑡 + 1)])𝑎 (5) where: 𝑏 = age-specific fertility rate 𝑓 = female population. note how the calculations make use of end-of-interval female populations which were calculated in the first stage. age-specific fertility rates are period-age (shown by the shaded square in figure 1) and not period-cohort because this is the form in which age-specific births and fertility statistics are often provided5. as an example, consider females aged 30. if the age-specific fertility rate for women aged 30 is 0.112436 and the startand end-of-interval populations of women aged 30 are 193,557 and 198,656 respectively, then the projected number of births born to women in this age group is: = 0.112436 1 2 [193,557 + 198,656] = 22,049.4 projected births from each age group of women are summed to obtain total births, and they are then divided into male and female births using the assumed sex ratio at birth: 𝐵𝑓 = 𝐵 100 100 + 𝑆𝑅𝐵 𝐵𝑚 = 𝐵 𝑆𝑅𝐵 100 + 𝑆𝑅𝐵 where: 𝑆𝑅𝐵 = sex ratio at birth, the number of males born for every 100 females. typically, this lies between 103 and 107 male births per 100 female births, with a mean of about 105 (ritchie 2019). 4 in reality, there cannot be fractional population numbers, of course. but during the calculations, fractions are usually retained, and rounding applied to the final projection outputs at the end of the calculation process. 5 it would be possible to project births using period-cohort fertility rates and populations-at-risk, but it isn’t necessary for projecting total numbers of births. 88 wilson & rees australian population studies 5 (1) 2021 as an example, assume the total number of projected births is 304,365 and the srb is 105.7 males per 100 females. then the number of female babies is: = 304,365 100 100+105.7 = 147,965.5 and the number of male babies: = 304,365 105.7 100+105.7 = 156,399.5 4.4. stage 3: project the infant cohort the third stage involves projecting the cohort of newly-born babies forwards to obtain the population aged 0 at the end of the projection interval. the population accounting equation includes the number of births as the ‘initial’ population: 𝑃𝑠,0(𝑡 + 1) = 𝐵𝑠 − 𝐷𝑠,𝑏→0 − 𝐸𝑠,𝑏→0 + 𝐼𝑠,𝑏→0 (6) where: 𝑏 → 0 = the infant cohort born during the projection interval and aged 0 at the end (illustrated by the shaded triangle at the bottom of figure 2). for the projection equation, deaths and emigration flows are replaced by rates multiplied by the population-at-risk: 𝑃𝑠,0(𝑡 + 1) = 𝐵𝑠 − 𝑑𝑠,𝑏→0 1 2 𝑃𝑠,0(𝑡 + 1) − 𝑒𝑠,𝑏→0 1 2 𝑃𝑠,0(𝑡 + 1) + 𝐼𝑠,𝑏→0 (7) the population-at-risk is approximated as half the end-of-interval population. this provides a reasonable estimate of the person-years lived by the newly-born cohort where there is low infant mortality because the end-of-interval population consists of nearly all members of the new cohort who were born, and they were part of the population for an average of half the time interval. with re-arranging, this projection equation can be simplified to: 𝑃𝑠,0(𝑡 + 1) + 1 2 𝑑𝑠,𝑏→0 𝑃𝑠,0(𝑡 + 1) + 1 2 𝑒𝑠,𝑏→0 𝑃𝑠,0(𝑡 + 1) = 𝐵𝑠 + 𝐼𝑠,𝑏→0 then: (1 + 1 2 𝑑𝑠,𝑏→0 + 1 2 𝑒𝑠,𝑏→0)𝑃𝑠,0(𝑡 + 1) = 𝐵𝑠 + 𝐼𝑠,𝑏→0 and: 𝑃𝑠,0(𝑡 + 1) = 1 (1+1 2 𝑑𝑠,𝑏→0+1 2 𝑒𝑠,𝑏→0) 𝐵𝑠 + 𝐼𝑠,𝑏→0 (1+1 2 𝑑𝑠,𝑏→0+1 2 𝑒𝑠,𝑏→0) (8) to illustrate the calculations, we take the female cohort born during 2019-20 which ends the projection interval in mid-2020 aged 0. if there are 147,965.5 births, a death rate of 0.005207, an australian population studies 5 (1) 2021 wilson & rees 89 emigration rate of 0.012508, and an immigration flow of 1,865.8, then the end-of-interval population aged 0 is: = 1 (1 + 1 2 0.005207 + 1 2 0.012508) 147,965.5 + 1,865.8 (1 + 1 2 0.005207 + 1 2 0.012508) = 148,515.8 4.5. prepare for the next projection interval the end-of-interval populations then become the start-of-interval populations for the next projection interval. this requires summing the end-of-interval populations for the two oldest cohorts. for example, if the oldest age group in the projections output is 100+, then the projected end-of-interval populations at age 100 and age 101+ must be summed to obtain the 100+ start-of-interval population for the next projection interval. 𝑃𝑠,100+(𝑡 + 1) = 𝑃𝑠,100(𝑡 + 1) + 𝑃𝑠,101+(𝑡 + 1) (9) the projection calculations are then repeated for as many years ahead as required. box 2: why is it best to avoid projecting international migration as net migration? projecting international migration as net migration numbers is appealing because the calculations are simple and there are minimal data requirements. in fact, net migration can be calculated from a recent year as the remaining cohort population change once deaths have been taken into account, so no actual migration data are needed. however, there are both conceptual and practical limitations of modelling net migration. conceptually, there is no actual demographic process of net migration, only immigration flows into a country and emigration flows out. in addition, by using net migration numbers, we remove the direct relationship between the age pattern of migration and the population age structure. by using emigration rates, emigration is projected as a function of the population-at-risk in each period-cohort, so that the age pattern of emigration is influenced by population age structure. from a practical perspective, net migration age patterns are much more variable over time than immigration and emigration age patterns, and are shaped by the relative amounts of immigration and emigration. it is therefore difficult to adjust the age profile of net migration over the course of the projection horizon (in response to migration policy changes or global pandemics, for example). simply scaling the net migration age profile up and down is unlikely to give a realistic outcome. in addition, in some extreme circumstances, negative net migration held constant over long periods can yield ‘negative’ populations. 4.6. strengths and weaknesses the cohort-component model is widely seen as the ‘gold standard’ method of population projection, and it possesses many excellent features. but it also has some weaknesses which are worth keeping in mind. some of the strengths and weaknesses are listed in table 1 below (see also burch 2018). 90 wilson & rees australian population studies 5 (1) 2021 table 1: some of the strengths and weaknesses of the cohort-component model strengths weaknesses good representation of actual population processes. it includes the effects of population age structure and age-specific rates assumes a homogenous population, with no allowance for the changing internal composition can be accurate if good assumptions about fertility, mortality and migration are chosen can be highly inaccurate if poor assumptions about fertility, mortality and migration are chosen provides age-disaggregated projections which are useful for a wide variety of planning, policy, budgeting, and service delivery purposes requires a non-trivial amount of input data; and assumptions about future fertility, mortality and migration are not intrinsic to the model and have to be prepared separately provides the starting point for many derived projections (e.g., labour force; living arrangement; school enrolments) lack of in-built mechanism for ensuring consistency between male and female projections useful for understanding population change (e.g., by creating alternative future scenarios; or applying to past periods) immigrants to the population are assumed to take on the fertility, mortality and emigration rates assumed for the whole population simple mathematics. basic versions can be implemented in a spreadsheet births are produced by women only, rather than men and women. 4.6. the excel workbook the excel workbook accompanying this paper (national projection model example.xlsx) contains an example set of projections for australia using the cohort-component model equations presented above. it includes the following worksheets. notes: a short introduction to the layout and workings of the excel workbook. jumpoff: the jump-off populations should be supplied in this sheet. fertility: this is where the fertility assumptions are placed. to aid assumption-setting, fertility assumptions are provided separately for the level (tfr) and age profiles (age-specific rates) of fertility. the age-specific fertility rates are scaled to match the tfr further down the sheet. this approach allows you to make slight adjustments to the tfr assumptions without having to recalculate the age-specific rates. mortality: this is where the projected 1lx life table values, used to calculate death rates by periodcohort, are placed. migration: like fertility, migration assumptions are split into level and age profile components to aid assumption-setting. for immigration, both the annual total immigration intake and immigration numbers by period-cohort are required. for emigration, the assumed gross migraproduction rate (gmr) (rogers et al. 1978) for the overall level of emigration, and emigration rates by period-cohort, are needed. the gmr is simply the sum of all age-specific rates, much like the total fertility rate. further down the sheet the supplied immigration numbers and emigration rates are scaled to sum to, respectively, the total immigration and emigration gmr assumptions. australian population studies 5 (1) 2021 wilson & rees 91 projections: the projection calculations are carried out in this sheet using equations 4, 5 and 8. scroll down to find projected births, deaths, and emigration flows by sex and period-cohort. population: the final projected populations are set out in this sheet. accountssummary: summary population accounts for each single year projection interval are provided in this sheet (start-of-interval populations, births, deaths, immigration, emigration, and end-of-interval populations). 5. input data and projection assumptions the cohort-component model requires the following input data: • jump-off populations by sex and age group • future fertility rates by age of mother • future mortality rates by sex and period-cohort • future emigration rates by sex and period-cohort • future immigration numbers by sex and period-cohort 5.1. jump-off populations jump-off population estimates by sex and single years of age are needed to provide the starting point of the projections. ideally the age disaggregation would extend up to an age high enough so that there are relatively few people in the highest open-ended age group. usually the jump-off populations would be the most recently published population estimates available from the national statistical office. 5.2. fertility age-specific fertility rates are required for each year of the projection horizon. sometimes agespecific rates are projected separately from the total fertility rate, and then adjusted to be consistent with the tfr if necessary. three broad approaches are suggested here: 1) assume the average tfr and age-specific fertility rates of recent years remain constant into the future. if fertility has remained approximately constant in recent years, and there are no obvious reasons for assuming change, this could be a reasonable assumption. 2) use fertility assumptions prepared by others, such as the national statistical office (e.g. abs 2018), the united nations population division (2019), or researchers who have prepared projections and make the data available (e.g. mcdonald 2020). 3) prepare your own projections. this option does come with some challenges, however. there is no single fertility projection method widely regarded as the best option. and, despite considerable amounts of research over many decades to understand fertility better, it remains notoriously hard to predict. some fertility projections are produced from extrapolative models, some from explanatory models linked to variables such as unemployment rates, some projections are created by trending from recent fertility to a long-run target, while others are based on expert opinions. an overview of fertility projection methods used in practice can be found in gleditsch and syse (2020), while reviews 92 wilson & rees australian population studies 5 (1) 2021 of the academic literature are included in booth (2006), shang and booth (2020), hilton et al. (2019), and bohk-ewald et al. (2018) ,though this last paper focuses largely on cohort fertility projections. in practice, fertility assumptions are often created by drawing on a variety of models and approaches. the tfr assumptions for australia in the example excel workbook accompanying this paper were prepared qualitatively, but were informed by other projections (mcdonald 2020), research on the impacts of major economic shocks to fertility (e.g. matysiak et al. 2020; sobotka et al. 2011), and the declining fertility trend observed in australia in recent years (abs 2020b). a long-run tfr of 1.70 was assumed, with a short-run dip to allow for the economic impacts of covid-19. the age profile of fertility rates was projected by extrapolating the parameters of the peristera and kostaki (2007) model of fertility, which fits well to the australian fertility rate age profile. 5.3. mortality death rates by sex and period-cohort are required for every year of the projection horizon. unlike fertility, the assumption of holding rates constant into the future is rarely a good choice for mortality because it tends to decline over time. therefore, two broad approaches are recommended here. 1) use mortality assumptions prepared by others, such a national statistical office. 2) prepare your own mortality projections. national mortality forecasting has received the greatest amount of attention in demographic forecasting research and there is no shortage of methods (booth and tickle 2008; ediev 2020; terblanche 2015) and software (e.g., hyndman et al. 2019; sevcikova et al. 2019) to choose from. generally, mortality trends are fairly stable over time, making it easier to forecast than fertility and migration (though in some countries covid-19 has now interrupted those trends). many of the mortality forecasting methods and software packages devised over the last couple of decades have been shown to produce quite accurate forecasts (e.g. terblanche 2016). the mortality assumptions in the excel workbook accompanying this paper were created using ediev’s (2008) extrapolative mortality model. age-specific death rates (in period-age form) were projected, and then life tables created. the excel workbook contains projected 1lx values which are used to calculate period-cohort death rates. because these values comprise the life table stationary population, the difference between populations at adjacent ages can be interpreted as the number of period-cohort deaths, shown by the shaded parallelogram in the lexis diagram in figure 3. note that the life table populations and deaths exactly match the age-time locations shown for the cohortcomponent model earlier (middle of figure 2). death rates are calculated with these deaths in the numerator and the mean of the two populations as the population-at-risk: 𝑑𝑎→𝑎+1 = 𝐿𝑥1 − 𝐿𝑥+11 1 2 ( 𝐿𝑥1 + 𝐿𝑥+11 ) where 𝑥 is the exact starting age of age group 𝑎. the sex subscript is omitted in the equation to prevent clutter. australian population studies 5 (1) 2021 wilson & rees 93  age 1lx+1 1lx time → figure 3: life table data used to calculate period-cohort death rates for newly-born infants and at the very highest ages, death rates are calculated slightly differently: 𝑑𝑏→0 = 𝑙0− 𝐿01 1 2 𝐿01 𝑑100+→101+ = 𝐿100∞ − 𝐿101∞ 1 2 ( 𝐿100∞ + 𝐿101∞ ) where 𝑙0 = the radix (or number of births) in the life table population. these deaths and populations exactly match the age-time locations as indicated for the youngest and oldest cohorts in figure 2. 5.4. international migration international migration is hard to forecast. it tends to exhibit quite volatile trends in volume over time in response to changes of policy (at least for immigration), macroeconomic conditions, and major world events. and for some countries (though not australia), international migration data suffers from quality and coverage problems. fortunately, the age profiles of immigration and emigration rates tend to be far more stable. our suggested approaches to migration forecasting are like those for fertility: 1) assume the trends of recent years remain constant into the future. in the absence of clear changes in trends in immigration and emigration, this can be a reasonable assumption. 2) use migration assumptions prepared by others (e.g. abs 2018). 3) prepare your own migration projections. this can be a challenging task. unlike fertility and mortality, there are relatively few available projection methods and software packages to choose from and, unfortunately, no single method has been found to produce consistently accurate migration projections. projection approaches include extrapolative models, explanatory models, trending to a target, expert panels, and what might best be described as a “forecaster’s judgement” approach that is not based on any formal model or process. this last approach qualitatively takes into account recent migration trends, perceptions of factors driving those trends, broad economic trends, policy influences (including immigration program plans), and expert opinions, among other things. it is only loosely connected to theory and modelling – but that is not necessarily a weakness given that theory and modelling have not yet delivered a robust migration projection method. 94 wilson & rees australian population studies 5 (1) 2021 if a more formal process or model is to be used, what are the options? a few examples are given here. one option is to apply time series or econometric models. examples of these types of models can be found in bijak et al. (2019), de beer (2008), and raymer and wiśniowski (2018). statistics norway uses a model based on neoclassical economic theory to project various immigration streams (cappelen et al. 2015). explanatory variables include relative income, norwegian unemployment rates, and the size of the migrant population. a similar explanatory model of migration has been devised for use in belgium’s population projections (vandresse 2015). another approach is to engage expert panels who are actively involved in the assumption formulation process (e.g. dion et al. 2020; lutz 2009). the expert panel considers and discusses a wide range of factors which might increase or decrease migration (and other demographic variables) and then suggests future migration levels or distributions. whichever methods and approaches are selected, many authors have recommended dividing international migration flows into different streams (e.g. permanent immigrants, temporary workers, foreign students, etc.) and creating projections by focusing on the different drivers of those flows (e.g. de beer 2008; wilson 2017a). for the excel workbook, immigration and emigration assumptions for australia were prepared separately for the headline indicators of total immigration flows and the emigration gross migraproduction rate (gmr), and the age profiles of immigration and emigration rates. given that the shapes of immigration and emigration rate age profiles tend to change slowly over time, they were assumed to remain unchanged in the projections. in contrast, the total volumes of immigration and emigration can change markedly from year to year. given the disruption to international migration created by the covid-19 pandemic and the closure of the australian border, we took note of the projected migration flows from the federal government’s centre for population in formulating our assumptions over the next few years (centre for population, 2020). a large drop in both immigration and emigration was assumed in the short term, reflecting the border restrictions introduced in response to the pandemic. for further into the future, we assumed total immigration would resume its long-run gradual upward trend and that the emigration gmr would return to its previous level and remain constant. 6. checking outputs the final part of the projections process involves checking the projections output for plausibility, consistency, and any possible errors with input data preparation or projections calculations. this is a form of exploratory data analysis for projections data. what constitutes ‘plausible’ is not easy to define, but projections outputs should look roughly in line with expectations given the chosen input assumptions. checks of plausibility may not provide a precise answer. instead, they may only suggest that assumptions would benefit from some refinement, or that certain data inputs may need attention. unfortunately, just because projections seem plausible does not mean that they will turn out to be accurate, but it should at least mean they will be free of obvious errors or unbelievable assumptions. total population. check that the projected total population trend looks plausible in light of historical data by creating a graph of past and projected total populations (and/or total population growth rates). figure 4 shows an example. in most circumstances there should be no discontinuity or sudden australian population studies 5 (1) 2021 wilson & rees 95 changes in trend, unless there is clearly good reason for it (e.g., the impact of a pandemic; a sudden immigration policy change). figure 4: example graph of total population to check projection outputs source: abs (2020a); authors’ projections demographic components. do the projected demographic components of change look plausible in light of historical trends? again, it is useful to create a graph showing past and projected births, deaths, immigration and emigration. births and deaths normally show relatively smooth trends over time, while immigration and emigration are often more variable. population age-sex structure. does the evolution of the age-sex structure of the population look plausible? it is useful to create population pyramids, or at least line graphs of the population by age, over time. if a population has been ageing in the past, it is likely to continue to age in the future. does the number of 0 year olds seem consistent with the fertility assumptions? does the sex ratio of the total population (males per 100 females) change slowly over time and not differ from the initial value too much? does the sex ratio by age show a pattern of values above 100 in the childhood and younger adult ages followed by declines with increasing age? does it change slowly over time? if male and female life expectancy is projected to converge in the future, does the sex ratio at the older ages increase? does the relative size of the young adult population seem about right when considering the migration assumptions? others’ projections. it can be useful to compare your projections with those produced by others, such as projections produced by a national statistical office. unless your projections have radically different assumptions, the broad shape of the outputs is likely to be similar, at least in the initial years of the projection horizon. but differences with other projections should not necessarily be interpreted as problems with your projection. if you have a higher fertility assumption, for example, you would expect a larger projected childhood population. for a more detailed checklist see ‘a checklist for reviewing draft population projections’ (wilson 2017b). 96 wilson & rees australian population studies 5 (1) 2021 7. conclusions this paper has presented a brief guide to producing a national population projection with a cohortcomponent model. several stages in the projections preparation process were described, starting with data gathering and assumptions preparation, and finishing with projection calculations and output checking. we described the cohort-component model from a population accounting perspective, which derives projection equations directly from simple cohort-specific population accounting equations. and we created an excel workbook to accompany this paper which implements the cohort-component model for australia. the paper covered many of the key elements of the projections preparation process, but it has not been possible to discuss all relevant issues in one short paper. we said little about the smoothing of age profiles of fertility, mortality and migration rates to remove noise, though noise is generally minimal at the national scale. this topic is covered by preston et al. (2001: chapter 9), de beer (2011, 2012), and wilson (2020). we didn’t go into much detail about population accounting reconciliation because accounting discrepancies are usually not a major problem at the national scale, nor did we consider forecast uncertainty and error (on this, see keilman 2018, 2020, and for an australian example of probabilistic forecasts, see bell et al. 2011). we also did not cover any topics which become relevant once the projections have been finalised, such as the best ways of communicating projections to stakeholders (johnstone 2015). it is important to note that while preparing a population projection may initially seem like a mostly quantitative and objective exercise, there are actually many judgements which must be made as part of the process. while some may seem like minor technical decisions, they require careful consideration because they might substantially impact the projection outputs. for example, decisions are made about input data adjustments, about which fertility, mortality, and migration projection methods and assumptions to use, and whether the outputs seem plausible and reasonable. mastering the calculations for population projections is relatively easy; making good decisions throughout the projections process is more challenging and is developed with practice. excel workbook the excel workbook containing a cohort-component model and example population projections for australia may be downloaded from https://doi.org/10.6084/m9.figshare.13565222.v3. acknowledgements funding support for tw from the australian research council centre of excellence in population ageing research is gratefully acknowledged (project number ce1101029). helpful comments on an earlier draft of this paper were kindly provided by ahana banerjee sen, rosabella borsellino, dagmara laukova, and the anonymous reviewers. https://doi.org/10.6084/m9.figshare.13565222.v3 australian population studies 5 (1) 2021 wilson & rees 97 references abs (1999) demographic estimates and projections: concepts, sources and methods, 1999. canberra: abs. https://www.abs.gov.au/ausstats/abs@.nsf/mf/3228.0. accessed 16 march 2021. abs (2018) population projections, australia. canberra: abs. https://www.abs.gov.au/statistics/people/population/population-projections-australia/latestrelease. accessed 16 december 2020. abs (2020a) national, state and territory population. https://www.abs.gov.au/statistics/people/population/national-state-and-territorypopulation/jun-2020. accessed 17 december 2020. abs (2020b) births, australia. canberra: abs. https://www.abs.gov.au/statistics/people/population/births-australia/latest-release. accessed 16 december 2020. australian government (2015) 2015 intergenerational report: australia in 2055. canberra: the treasury. https://treasury.gov.au/publication/2015-igr bell m, wilson t and charles-edwards e (2011) australia’s population future: probabilistic forecasts incorporating expert judgement. geographical research 49(3): 261-275. https://doi.org/10.1111/j.1745-5871.2011.00702.x bijak j (2012) migration assumptions in the uk national population projections: methodology review. university of southampton, uk. https://www.ons.gov.uk/peoplepopulationandcommunity/populationandmigration/population projections/methodologies/nationalpopulationprojectionsmigrationassumptionsmethodologyr eview. accessed 22 august 2013. bijak j, disney g, findlay a m, forster j j, smith p w f, and wiśniowski a (2019) assessing time series models for forecasting international migration: lessons from the united kingdom. journal of forecasting 38(5): 470-487. https://doi.org/10.1002/for.2576 bohk-ewald c, li p, and myrskylä m (2018) forecast accuracy hardly improves with method complexity when completing cohort fertility. proceedings of the national academy of sciences 115(37): 9187-9192. https://doi.org/10.1073/pnas.1722364115 booth h (2006) demographic forecasting: 1980 to 2005 in review. international journal of forecasting 22(3): 547-581. https://doi.org/10.1016/j.ijforecast.2006.04.001 booth h and tickle l (2008) mortality modelling and forecasting: a review of methods. annals of actuarial science 3(1-1): 3-43. https://doi.org/10.1017/s1748499500000440 burch t k (2018) the cohort-component population projection: a strange attractor for demographers. in model-based demography: essays on integrating data, technique and theory. cham: springer; 135-151. calot g (2015) demographic techniques: lexis diagram. in wright j d (ed.) international encyclopedia of the social & behavioral sciences. second edition, volume 6. amsterdam: elsevier; pp. 154-161. cappelen a, skjerpen t, and tønnessen m (2015) forecasting immigration in official population projections using an econometric model. international migration review 49(4): 945-980. https://doi.org/10.1111/imre.12092 centre for population (2020) population statement: december 2020. canberra: the treasury. https://population.gov.au/publications/publications-population-statement.html de beer j (2008) forecasting international migration: time series projections vs argument-based forecasts. in raymer j and willekens f (eds) international migration in europe: data, models and estimates. chichester: wiley; pp. 283-306. de beer j (2011) a new relational method for smoothing and projecting age specific fertility rates: topals. demographic research 24(18): 409-454. https://doi.org/10.4054/demres.2011.24.18 https://www.abs.gov.au/ausstats/abs@.nsf/mf/3228.0 https://www.abs.gov.au/statistics/people/population/population-projections-australia/latest-release https://www.abs.gov.au/statistics/people/population/population-projections-australia/latest-release https://www.abs.gov.au/statistics/people/population/national-state-and-territory-population/jun-2020 https://www.abs.gov.au/statistics/people/population/national-state-and-territory-population/jun-2020 https://www.abs.gov.au/statistics/people/population/births-australia/latest-release https://treasury.gov.au/publication/2015-igr https://doi.org/10.1111/j.1745-5871.2011.00702.x https://www.ons.gov.uk/peoplepopulationandcommunity/populationandmigration/populationprojections/methodologies/nationalpopulationprojectionsmigrationassumptionsmethodologyreview https://www.ons.gov.uk/peoplepopulationandcommunity/populationandmigration/populationprojections/methodologies/nationalpopulationprojectionsmigrationassumptionsmethodologyreview https://www.ons.gov.uk/peoplepopulationandcommunity/populationandmigration/populationprojections/methodologies/nationalpopulationprojectionsmigrationassumptionsmethodologyreview https://doi.org/10.1002/for.2576 https://doi.org/10.1073/pnas.1722364115 https://doi.org/10.1016/j.ijforecast.2006.04.001 https://doi.org/10.1017/s1748499500000440 https://doi.org/10.1111/imre.12092 https://population.gov.au/publications/publications-population-statement.html https://doi.org/10.4054/demres.2011.24.18 98 wilson & rees australian population studies 5 (1) 2021 de beer j (2012) smoothing and projecting age-specific probabilities of death by topals. demographic research 27(20): 543-592. https://doi.org/10.4054/demres.2012.27.20 dion p, galbraith n, and sirag e (2020) using expert elicitation to build long-term projection assumptions. in mazzuco s and keilman n (eds) developments in demographic forecasting. cham: springer; pp. 43-62. ediev d m (2008) extrapolative projections of mortality: towards a more consistent method. part i: the central scenario. vienna institute for demography working paper 3/2008. https://www.oeaw.ac.at/fileadmin/subsites/institute/vid/pdf/publications/working_papers/ wp2008_03.pdf ediev d m (2020) mortality projection. in gu d and dupre m (eds) encyclopedia of gerontology and population aging. cham: springer; pp 1-9. https://doi.org/10.1007/978-3-319-69892-2_557-1 gleditsch r f and syse a (2020) ways to project fertility in europe: perceptions of current practices and outcomes. discussion papers no. 929, statistics norway. https://www.ssb.no/en/forskning/discussion-papers/ways-to-project-fertility-in-europe. accessed 16 december 2020. hilton j, dodd e, forster j j, smith p w f, and bijak j (2019) forecasting fertility using parametric mixture models. https://arxiv.org/pdf/1909.09545.pdf. accessed 16 december 2020. hyndman r j, booth h, tickle l and maindonald j (2019) demography r package. https://cran.rproject.org/package=demography johnstone k (2015) communicating population projections to stakeholders: a case study from new south wales. in wilson t, charles-edwards e, and bell m (eds) demography for planning and policy: australian case studies. cham: springer; pp. 71-89. keilman n (2018) probabilistic demographic forecasts. vienna yearbook of population research 16: 111. https://austriaca.at/0xc1aa5576%200x003a2757.pdf keilman n (2020) uncertainty in population forecasts for the twenty-first century. annual review of resource economics 12: 449-470. https://doi.org/10.1146/annurev-resource-110319-114841 kinter h j (2004) the life table. in swanson d a (ed.) the methods and materials of demography. second edition. san diego: elsevier academic press; pp. 301-340. keyfitz n and caswell h (2005) applied mathematical demography. third edition. new york: springer. lomax n, wohland p, rees p, and norman p (2020) the impacts of international migration on the uk’s ethnic populations. journal of ethnic and migration studies 46(1): 177-199. https://doi.org/10.1080/1369183x.2019.1577726 lutz w (2009) toward a systematic, argument-based approach to defining assumptions for population projections. interim report ir-09-037, international institute for applied systems analysis, laxenburg, austria. http://pure.iiasa.ac.at/id/eprint/9115/ matysiak a, sobotka t, and vignoli d (2020) the great recession and fertility in europe: a sub-national analysis. european journal of population. https://doi.org/10.1007/s10680-020-09556-y mcdonald p (2020) a projection of australia’s future fertility rates. centre for population research paper. canberra: abs. https://population.gov.au/research/research-fertility.html. accessed 16 december 2020. peristera p and kostaki a (2007) modeling fertility in modern populations. demographic research 16(6): 141-194. https://doi.org/10.4054/demres.2007.16.6 pressat r (2008) demographic analysis: projections on natality, fertility and replacement. new brunswick: aldine transaction. preston s h, heuveline p and guillot m (2001) demography: measuring and modeling population processes. oxford: blackwell. rau r (2018) the lexis diagram. in rau r, bohk-ewald c, muszyńska m m, and vaupel j w visualizing mortality dynamics in the lexis diagram. cham: springer; pp. 5-10. https://doi.org/10.4054/demres.2012.27.20 https://www.oeaw.ac.at/fileadmin/subsites/institute/vid/pdf/publications/working_papers/wp2008_03.pdf https://www.oeaw.ac.at/fileadmin/subsites/institute/vid/pdf/publications/working_papers/wp2008_03.pdf https://doi.org/10.1007/978-3-319-69892-2_557-1 https://www.ssb.no/en/forskning/discussion-papers/ways-to-project-fertility-in-europe https://arxiv.org/pdf/1909.09545.pdf https://cran.r-project.org/package=demography https://cran.r-project.org/package=demography https://austriaca.at/0xc1aa5576%200x003a2757.pdf https://doi.org/10.1146/annurev-resource-110319-114841 https://doi.org/10.1080/1369183x.2019.1577726 http://pure.iiasa.ac.at/id/eprint/9115/ https://doi.org/10.1007/s10680-020-09556-y https://population.gov.au/research/research-fertility.html https://doi.org/10.4054/demres.2007.16.6 australian population studies 5 (1) 2021 wilson & rees 99 raymer j and wiśniowski a (2018) applying and testing a forecasting model for age and sex patterns of immigration and emigration. population studies 72(3): 339-355. https://doi.org/10.1080/00324728.2018.1469784 rees p (1984) spatial population analysis using movement data and accounting methods: theory, models, the ‘move’ program and examples. working paper 404, school of geography, university of leeds, uk. https://doi.org/10.13140/rg.2.2.14084.76169 rees p (1985) does it really matter which migration data you use in a population model? in white p and van der knaap b (eds.) contemporary studies of migration. norwich: geobooks; pp 55-77. https://www.researchgate.net/publication/349059242_rees_1985_chapter_8_in_white_and_ van_der_knaap. rees p (1986) choices in the construction of regional population projections. in woods r and rees p (eds) population structures and models. london: allen & unwin; pp. 126-159. rees p h and willekens f (1986) data and accounts. in rogers a and willekens f (eds.) migration and settlement: a multiregional comparative study. dordrecht: reidel press; pp. 19-58. rees p h and wilson a g (1977) spatial population analysis. london: edward arnold. rees p (2015) which migration models should be used in population projections? report for edge analytics ltd. school of geography, university of leeds, uk. ritchie h (2019) how does the sex ratio at birth vary across the world? our world in data. global change data lab. oxford: oxford martin school. https://ourworldindata.org/sex-ratio-at-birth. accessed 7 february 2021. rogers a, racquillet r, and castro l j (1978) model migration schedules and their applications. environment and planning a 10(5): 475-502. https://doi.org/10.1068/a100475 rogers a (1995) multiregional demography: principles, methods and extensions. chichester: wiley. rowland d t (2003) demographic methods and concepts. oxford: oxford university press. sevcikova h, raftery a, and chunn j (2019) bayeslife r package. https://cran.rproject.org/package=bayeslife shang h l and booth h (2020) synergy in fertility forecasting: improving forecast accuracy through model averaging. genus 76:27. https://doi.org/10.1186/s41118-020-00099-y shorter f c, sendek r, and bayoumy y (1995) computational methods for population projections. second edition. new york: population council. sloggett a (2015) life tables (i) (papp101 s07). in population analysis for policy and programmes [online course]. paris: international union for the scientific study of population. https://papp.iussp.org/sessions/papp101_s07/papp101_s07_010_010.html. accessed 16 december 2020. smith s k, tayman j, and swanson d a (2013) a practitioner’s guide to state and local population projections. dordrecht: springer. sobotka t, skirbekk v, and philipov d (2011) economic recession and fertility in the developed world. population and development review 37(2): 267-306. https://doi.org/10.1111/j.17284457.2011.00411.x stover j and kirmeyer s (2007) demproj version 4: a computer program for making population projections. https://data.unaids.org/pub/manual/2007/demproj_2007_en.pdf terblanche w (2015) population estimates and projections for australia’s very elderly population at state and national level. phd thesis, the university of queensland, australia. https://doi.org/10.14264/uql.2015.979 terblanche w (2016) retrospective testing of mortality forecasting methods for the projection of very elderly populations in australia. journal of forecasting 35(8): 703-717. https://doi.org/10.1002/for.2404 https://doi.org/10.1080/00324728.2018.1469784 https://doi.org/10.13140/rg.2.2.14084.76169 https://www.researchgate.net/publication/349059242_rees_1985_chapter_8_in_white_and_van_der_knaap https://www.researchgate.net/publication/349059242_rees_1985_chapter_8_in_white_and_van_der_knaap https://ourworldindata.org/sex-ratio-at-birth https://doi.org/10.1068/a100475 https://cran.r-project.org/package=bayeslife https://cran.r-project.org/package=bayeslife https://doi.org/10.1186/s41118-020-00099-y https://papp.iussp.org/sessions/papp101_s07/papp101_s07_010_010.html https://doi.org/10.1111/j.1728-4457.2011.00411.x https://doi.org/10.1111/j.1728-4457.2011.00411.x https://data.unaids.org/pub/manual/2007/demproj_2007_en.pdf https://doi.org/10.14264/uql.2015.979 https://doi.org/10.1002/for.2404 100 wilson & rees australian population studies 5 (1) 2021 timaeus i (2015) population projections: concepts and methods (papp101 s10). in population analysis for policy and programmes [online course]. paris: international union for the scientific study of population. https://papp.iussp.org/sessions/papp101_s10/papp101_s10_010_010.html united nations (1989) projection methods for integrating population variables into development planning. volume 1: methods for comprehensive planning. module 1: conceptual issues and methods for preparing demographic projections. new york: united nations. https://www.un.org/development/desa/pd/sites/www.un.org.development.desa.pd/files/files /documents/2020/jan/un_1989_projection_methods_for_integrating_pop_variables_into_dev elopment_planning_-_module_1.pdf. accessed 2 january 2021. united nations population division (2019) world population prospects 2019. data available from https://population.un.org/wpp/download/standard/population/. accessed 2 january 2021. us census bureau (2019) demographic analysis & population projection system (dapps) software. https://www.census.gov/data/software/dapps.html. accessed 16 march 2021. vandresse (2015) modelling the future evolution of international migration for belgium. working paper 02-15, federal planning bureau, belgium. https://www.plan.be/publications/publication-1435en-modelling_the_future_evolution_of_international_migration_for_belgium. accessed 16 december 2020. willekens f j and drewe p (1984) a multiregional model for regional demographic projection. in heide h & willekens f j (eds.) demographic research and spatial policy: the dutch experience. london: academic press; pp. 309-334. https://pure.knaw.nl/portal/en/publications/amultiregional-model-for-regional-demographic-projection-2. accessed 2 january 2021. wilson t (2017a) can international migration forecasting be improved? the case of australia. migration letters 14(2): 277-291. https://doi.org/10.33182/ml.v14i2.333 wilson t (2017b) a checklist for reviewing draft population projections. technical paper. northern institute, charles darwin university, darwin. https://doi.org/10.13140/rg.2.2.35291.80161. wilson t (2019) an introduction to population projections for australia. australian population studies 3(1): 40-56. https://doi.org/10.37970/aps.v3i1.46 wilson t (2020) modelling age patterns of internal migration at the highest ages. spatial demography 8(2): 175-192. https://doi.org/10.1007/s40980-020-00062-7 wilson t, temple j, and charles-edwards e (2021) will the covid-19 pandemic affect population ageing in australia? journal of population research https://doi.org/10.1007/s12546-021-09255-3 https://papp.iussp.org/sessions/papp101_s10/papp101_s10_010_010.html https://www.un.org/development/desa/pd/sites/www.un.org.development.desa.pd/files/files/documents/2020/jan/un_1989_projection_methods_for_integrating_pop_variables_into_development_planning_-_module_1.pdf https://www.un.org/development/desa/pd/sites/www.un.org.development.desa.pd/files/files/documents/2020/jan/un_1989_projection_methods_for_integrating_pop_variables_into_development_planning_-_module_1.pdf https://www.un.org/development/desa/pd/sites/www.un.org.development.desa.pd/files/files/documents/2020/jan/un_1989_projection_methods_for_integrating_pop_variables_into_development_planning_-_module_1.pdf https://population.un.org/wpp/download/standard/population/ https://www.census.gov/data/software/dapps.html https://www.plan.be/publications/publication-1435-en-modelling_the_future_evolution_of_international_migration_for_belgium https://www.plan.be/publications/publication-1435-en-modelling_the_future_evolution_of_international_migration_for_belgium https://pure.knaw.nl/portal/en/publications/a-multiregional-model-for-regional-demographic-projection-2 https://pure.knaw.nl/portal/en/publications/a-multiregional-model-for-regional-demographic-projection-2 https://doi.org/10.33182/ml.v14i2.333 https://doi.org/10.13140/rg.2.2.35291.80161 https://doi.org/10.37970/aps.v3i1.46 https://doi.org/10.1007/s40980-020-00062-7 https://doi.org/10.1007/s12546-021-09255-3 australian population studies 2021 | volume 5 | issue 2 | pages 29–42 © fry et al. 2021. published under the creative commons attribution-noncommercial licence 3.0 australia (cc by-nc 3.0 au). journal website: www.australianpopulationstudies.org covid-19 and the australian labour market: how did older australians fare during 2020? jane fry the university of melbourne jeromey b. temple* the university of melbourne peter mcdonald the university of melbourne alysia blackham the university of melbourne *corresponding author. email: jeromey.temple@unimelb.edu.au. address: melbourne school of population and global health, the university of melbourne, melbourne, vic, 3010, australia paper received 29 june 2021; accepted 25 october 2021; published 29 november 2021 abstract background in analysing the effects of the covid-19 pandemic on the labour market, attention has focussed on younger people, leaving a research gap when it comes to outcomes for older australians aged 50 years or over, in terms of employment, unemployment, underemployment and hours worked. aims to describe levels of labour force participation, unemployment, underemployment, and hours worked by older workers and job seekers during 2020. data and methods using australian bureau of statistics data, we perform descriptive analyses of variations in labour market outcomes by geographic areas, public and private sector employment, industry of employment and demographic characteristics. results older employment fell in april but recovered by december. as the full-time share initially increased, average hours worked decreased due to reductions in hours offered to workers, increasing the underemployment rate. there was little recovery of employment in metropolitan melbourne due to prolonged lockdown conditions. of the largest industries, retail trade and manufacturing were worst affected. conclusions by december 2020, employment levels for older workers in some sectors had recovered from the initial downturn caused by the pandemic. however, for older workers in some industries, there is a major concern about their potential for future employment. key words older workers; covid-19; labour market; retirement. http://www.australianpopulationstudies.org/ mailto:jeromey.temple@unimelb.edu.au 30 fry et al. australian population studies 5 (2) 2021 1. introduction as of march 2021, approximately one third of all covid-19 cases in australia were among people aged 50 years and over, with this same group accounting for 99% of covid-19 deaths, the majority of which were aged over 70 years (department of health 2021). in terms of individual health and mortality, covid-19 has heavily impacted older australians (typically aged 50 years or over), while younger australians (defined here as those aged 15–24 years) have faced particularly deleterious effects on their mental health and general wellbeing (varma et al. 2021). although younger australians have faced considerable increases in unemployment and underemployment (churchill 2020), little attention has been given to how older workers and job seekers have fared in 2020 during the covid-19 pandemic. ordinarily, we would expect older workers to fare worse, as they do in other recessions, with reduced employment opportunities (chomik 2020), increased age discrimination in recruitment, and steep wage losses (bui et al. 2020), especially given they are more vulnerable to the effects of covid19. that said, this may not be the case, because younger workers are more casualised and therefore more easily laid off (borland & charlton 2020). there remains uncertainty as to the long-term effects on particular groups of workers, particularly on the basis of gender, age, and in particular sectors (birch & preston 2020) as shutdowns have led to job loss, unemployment and inactivity (chomik 2020). in this paper we address this research gap through a detailed examination of labour force participation, unemployment, underemployment, and hours worked by older workers (aged 50 years or over) and job seekers during the covid-19 pandemic in 2020. at the urging of unions and some business groups, the australian government introduced the jobkeeper wage subsidy scheme to help keep more people employed in private businesses and not-for-profit organisations. a number of oecd countries have introduced similar job retention (wage subsidy) schemes, such as new zealand, canada, estonia, poland and the netherlands (oecd 2020). under jobkeeper, eligible employers were required to pay a minimum of $1,500 per fortnight to the employee (forsyth 2020). jobkeeper objectives were to support business and job survival, preserve employment relationships, and provide needed income support (neil et al. 2021). it continued throughout the period of observation used in this paper, march to december 2020. the effects of the jobkeeper payments on employment are an important backdrop to the paper given that jobkeeper essentially disguises unemployment (coates et al. 2020). this research complements the detailed evidence base on older labour force participation, including, but not limited to, the determinants, barriers and enablers of older employment, the role of workplaces and the efficiency of government policy in enabling participation and futures of labour supply (oecd 1998; adair & temple 2012; yeatts et al. 2000). apart from providing a complementary view to previous studies, our analysis underscores the impacts of covid-19 on older workers. australian population studies 5 (2) 2021 fry et al. 31 2. data and methods in this paper, we use population-weighted australian bureau of statistics (abs) data tables based on the labour force survey. this survey of approximately 50,000 individuals in 26,000 dwellings is conducted monthly, with households interviewed for eight consecutive months. the nationally representative sample rotates with one eighth of the households replaced each month. in scope individuals are civilians aged 15 years and over. in addition to socio-demographic information, questions relate to labour market activities in a ‘reference week’ just prior to the survey. more information on the sample and survey methodology are available elsewhere (abs 2021c), as are the underlying data tables (abs 2021b). within the abs labour force framework there are seven key concepts for analysing labour market performance: 1. population is identified as the number of usually resident civilians in the relevant age/sex group. 2. employment is defined as engaged in full-time or part-time work. 3. hours worked (in our context) is hours actually worked in all jobs. 4. underemployment is defined as either: a) employed full-time but worked less than 35 hours in the reference week due to economic reasons and available to work more hours, or b) employed part-time and would prefer more hours and available to work more hours. 5. unemployment is defined as not employed but waiting to start a new job or actively looking for work. 6. the labour force comprises individuals who are employed or unemployed. 7. not in the labour force (nilf) comprises individuals who are neither employed nor unemployed. further details are available elsewhere (abs 2018). these concepts are then used to calculate six summary measures: 1. employment to population ratios divide the number employed (full-time or part-time) by the population. 2. labour force participation rates divide the labour force by population. 3. unemployment rates divide the number unemployed by the labour force. 4. underemployment rates divide the number underemployed by the labour force. 5. hours per worker divides total hours of actual work by the number employed. 6. full-time share divides the number of full-time employed by the total number employed. we refine our aggregate analysis to consider geographic areas, public and private sector employment, industry of employment and demographic characteristics. summary measures are provided for march 2020 (or the february quarter), the initial period, which compares march and april (or the may quarter) to gauge the initial effect; and the longer period, which compares march with december (or the november quarter) to gauge the 2020 effect. these effects are percentage changes for selected concepts (such as the number employed) or percentage point changes for summary measures (except hours per worker, where we use percentages). ethics approval for this project was granted by the university of melbourne human ethics committee lnr 2a — ethics id 2021-21421-15592-2. 32 fry et al. australian population studies 5 (2) 2021 3. results 3.1. aggregate effects initially, young people fared the worst: their employment rate fell by more than six points (table 1). most who lost their jobs withdrew from the labour force, cushioning the effect on the unemployment rate. job losses occurred despite the jobkeeper program, for which there were 900,000 applicants per month up to september, then 500,000 applicants per month through the remainder of the year. a recovery by december left the employment rate at march levels, although an increase in participation raised the unemployment rate as more individuals searched for work. initially, hours worked dropped due to the combined effects of a loss of part-time jobs and reduced hours for remaining workers. this served to increase underemployment among the youngest workers. in the longer period, hours worked more than recovered, and desired and actual hours worked were in balance. table 1: effect of covid-19 on employment and participation by age group, australia, 2019 and 2020 march 2020 march-april 2020 change march-april 2019 change marchdecember 2020 change marchdecember 2019 change age 15-24 years participation rate 68.9% -6.4 points -0.3 points +1.8 points +2 points employment rate 60.2% -6.5 points +0.3 points +0.6 points +2.6 points unemployment rate 12.7% +1.5 points -0.9 points +1.3 points -1.2 points full-time share 43.7% +1.5 points -1.3 points -0.2 points +0.2 points hours per worker 24.9 hours -16.4% -2.0% +4.6% +1.7% underemployment rate 18.1% +4.5 points +0.8 points +0.4 points +3.5 points age 25-49 years participation rate 85.7% -2 points +0.3 points +0.4 points +0.6 points employment rate 82.1% -2.8 points +0.3 points -0.2 points +1 point unemployment rate 4.3% +1.1 points 0 points +0.6 points -0.5 points full-time share 75.5% +0.6 points -0.2 points +0.1 points +0.2 points hours per worker 34.3 hours -11.4% -1.8% +3.4% +2.2% underemployment rate 6.7% +5.5 points +0.3 points +0.7 points -0.1 points age 50+ years participation rate 45.0% -1.5 points 0 points +0.4 points 0 points employment rate 43.1% -1.6 points 0 points +0.1 points +0.1 points unemployment rate 4.1% +0.4 points +0.1 points +0.6 points -0.2 points full-time share 66.2% +1.3 points -0.3 points +0.8 points -0.2 points hours per worker 32.5 hours -9.7% -0.3% +2.7% +2.2% underemployment rate (aged 55+ years) 6.9% +4.3 points +0.2 points -0.4 points +0.5 points source: abs labour force survey. note: the age group used for underemployment differ from those of the other measures because of the availability of data. points refers to percentage points. among individuals aged 25–49 years there was also a marked reduction in participation and employment in the initial period. hours per worker fell substantially (11%) and underemployment rose. however, by december labour market conditions were almost the same as in march, although australian population studies 5 (2) 2021 fry et al. 33 hours worked were higher. at first, the fall in the employment rate among older workers was matched almost exactly by a fall in participation. initially, therefore, workers losing their jobs defined themselves as being out of the labour force. this meant the rise in the unemployment rate was much smaller than may have been expected. in the longer period, the fall in employment for older workers was fully compensated, but the participation rate rose more than the employment rate. thus, in the longer period, the unemployment effect was worse than it had been in the initial period. participation was higher in december than in march but does not appear to be seasonal based on a comparison with the march–december 2019 change. a reduction in part-time jobs (many held on a casual basis) led to a small increase in the full-time/ part-time job share. however, average hours per worker declined as workers initially faced reduced hours. in particular, fewer casual workers in food, hospitality and other service industries were required due to sector-specific closures and limits on social gatherings. some employers reduced the hours offered to employees rather than laying them off. indeed, for employees aged 55 years and over, a comparison between august 2019 and 2020 shows a large increase in employees citing not enough work as a reason for working fewer hours than usual: a three-fold increase for employees with paid leave entitlements and a 50% increase for employees without paid leave (casual) (abs 2021a). 3.2. distributional results: geographical areas the initial effect of covid-19 on the participation of older workers was similar across all the geographies shown in table 2, but was a little higher in western australia where the prevalence of covid-19 was lower. this suggests that people who lost jobs initially reacted to the pandemic announcement by leaving the labour force. the same effects are evident for employment at the state level and for regional australia, with the fall in employment roughly matching the fall in participation — implying minimal impact overall on unemployment. however, regional victoria differs from other areas in that, while participation also fell, employment fell much less than elsewhere. the implied reduction in unemployment is consistent with the reduction in the participation rate. table 2: effect of covid-19 on the employment and participation of workers aged 50 and over, selected states, metropolitan melbourne and regional victoria, 2020 participation rate employment rate march 2020 march-april change marchdecember change march 2020 march-april change marchdecember change new south wales 43.4% -1.3 points +1.1 points 41.6% -1.5 points +0.5 points western australia 49.0% -1.6 points +0.2 points 46.4% -1.5 points +0.1 points victoria 45.7% -1.3 points 0 points 44.0% -1.1 points -0.3 points metropolitan melbourne 46.2% -1.2 points -0.5 points 44.6% -1.3 points -1.0 points regional victoria 44.3% -1.4 points +1.1 points 42.7% -0.5 points +1.4 points regional australia 42.8% -1.3 points +0.4 points 41.1% -1.2 points +0.6 points source: abs labour force survey. note: points refers to percentage points. 34 fry et al. australian population studies 5 (2) 2021 in the longer period, as might have been expected given the duration of the second lockdown in metropolitan melbourne, there was little recovery of employment in metropolitan melbourne while there was strong recovery everywhere else, especially in regional victoria. while population levels remained stable between metropolitan melbourne and regional victoria, metropolitan melbourne’s share of employment declined, suggesting jobs may have shifted to regional victoria . 3.3. employment by age group and sector considering individuals aged 45 years or over, initially the numbers employed fell for all of the older age groups in both the public and the private sectors, but the largest percentage falls tended to be in the older ages 60-64 and 65+ years, especially in the public sector (table 3). in the longer period, in the public sector, recovery was very strong for all age groups, especially among 50–59 year olds. in the longer period, in the private sector, there was recovery for the two oldest age groups. this may mean that, unlike in previous recessions, 2020 may not lead to a growth in early retirement (rutledge & coe 2012). table 3: numbers employed and percentage changes in numbers employed by age group and sector, australia, 2020 public sector private sector no. employed % change no. employed % change age group march 2020 (‘000) marchapril marchdecember march 2020 (‘000) marchapril marchdecember 45-49 255.9 -2.9 +2.2 1137.0 -3.9 -2.8 50-54 220.5 -5.9 +5.8 1029.9 -3.1 -0.5 55-59 204.4 -3.1 +8.4 923.9 -3.8 -2.8 60-64 150.6 -8.1 +1.7 645.2 -1.7 +2.6 65+ 80.6 -14.3 +1.4 517.8 -9.3 +3.1 source: abs labour force survey. 3.4. employment by industry between february and may, employment of older workers fell by between 4 and 9 per cent for the six largest industries of employment (table 4). however, in the longer period, the outcomes for these six industries varied considerably, rising by 10.4 per cent for education and training but falling by 11.2 per cent for manufacturing. retail also showed a heavy fall in the longer period. for the other three largest industries, employment returned to its february level by november. for the remaining industries, there were some for which the february to may effect was positive (construction (perhaps boosted by the ‘homebuilder’ program); agriculture, forestry and fishing; wholesale trade; financial and insurance services; rental, hiring and real estate services; electricity, gas, water and waste services). covid-19 gave rise to issues of distribution which may account for the strong positive result for wholesale trade. there were strong negative initial outcomes for arts and recreation services and for accommodation and food services, due in part to spatial distancing measures (coates et al. 2020). between february and november, sectors outside the top six that had positive initial outcomes had — not unexpectedly — positive longer-term outcomes as well. that said, there were some australian population studies 5 (2) 2021 fry et al. 35 strong reversals for a few industries that had negative initial outcomes. for example, employment in accommodation and food services swung from an initial fall to a longer period gain. administrative and support services, accommodation and food services and mining also showed a large turnaround from negative to positive from the initial period to the longer period table 4: numbers employed and percentage change in numbers employed by industry, workers aged 50+ years, australia, 2020 industry (in rank order of employment) number employed february 2020 (‘000s) percentage change in number employed february to may 2020 february to november 2020 health care and social assistance 597.8 -4.0 -2.2 education and training 345.9 -3.6 +10.4 professional, scientific and technical services 305.1 -5.3 -0.4 retail trade 300.3 -8.9 -6.7 manufacturing 295.4 -6.9 -11.2 public administration and safety 292.1 -4.6 +0.1 construction 288.3 +1.1 +5.5 transport, postal and warehousing 260.6 -15.9 -6.3 agriculture, forestry and fishing 174.9 +6.3 0 other services 139.5 -16.5 -3.2 wholesale trade 126.8 +10.9 +7.8 administrative and support services 125.3 -3.5 +12.6 accommodation and food services 123.5 -11.4 +12.7 financial and insurance services 115.0 +12.5 +7.5 rental, hiring and real estate services 69.7 +2.1 +8.2 arts and recreation services 62.6 -20.2 -3.6 mining 60.6 -9.0 +12.7 information, media and telecommunications 53.4 -23.5 -6.7 electricity, gas, water and waste services 36.1 +36.0 +41.2 source: abs labour force survey. note: the periods used in this table differ from those of other tables because of the availability of data. 3.5. employment by demographic characteristics initially, individuals aged 50–64 years from north-east asia were worst hit by declining employment rates, although they recovered well (table 5). individuals aged 50–64 years from the americas did not fare well, with an initial reduction of 3 percentage points followed by further decline in the longer period. among those aged 65 years or over, highest employment rates were for individuals from oceania/antarctica and sub-saharan africa. alongside north africa/middle east and the americas, they have the largest initial reductions in employment. however, rates for oceania and the americas recovered well, increasing by 2 and 6 percentage points respectively over the longer period. 36 fry et al. australian population studies 5 (2) 2021 table 5: percentage employed and changes in percentage employed by age group by country of birth, sex, partner status, relationship status and education, australia 2020 characteristic 50-64 65+ mar 2020 % empl’d mar-apr change mar-dec change mar 2020 % empl’d mar-apr change mar-dec change country of birth australia 71.3 -1.5 pts -0.1 pts 15.9 -1.1 pts 0.3 pts oceania and antarctica 72.6 -2.9 pts 3.1 pts 21.9 -3 pts 2.1 pts north-west europe 74.3 -2.9 pts -0.1 pts 13.3 -1.6 pts -0.1 pts southern and eastern europe 64.7 -2.2 pts -2.8 pts 8.2 -1.2 pts 0.2 pts north africa and the middle east 41.2 -1.4 pts 2.4 pts 9.8 -3.5 pts -0.2 pts south-east asia 64.4 -2.7 pts -0.6 pts 16.4 0.9 pts -0.1 pts north-east asia 54.0 -6.8 pts 3.3 pts 8.3 -1.6 pts 3.9 pts southern and central asia 72.9 -1.1 pts -2.7 pts 17.8 1.1 pts 2 pts americas 71.1 -3 pts -4.1 pts 15.8 -3.1 pts 6 pts sub-saharan africa 79.1 -1.7 pts -5.4 pts 19.3 -2.6 pts 0.3 pts sex males 74.9 -1.7 pts -0.4 pts 17.7 -1.3 pts 1.3 pts females 64.8 -2.2 pts 0.3 pts 10.9 -0.9 pts 0.1 pts partner status partnered 73.3 -2.3 pts 0.1 pts 16.5 -1.3 pts +1.9 pts not partnered 60.8 -1.2 pts -0.5 pts 10.4 -0.9 pts -1.5 pts relationship status has spouse only 69.1 -2.5 pts -0.4 pts 16.1 -1.6 pts 1.4 pts has dependent children 76.7 -2.5 pts 0.3 pts 18.0 -0.5 pts 2.2 pts lone person 61.2 -0.5 pts 1 pts 12.1 -0.9 pts -2.5 pts other 51.3 -0.6 pts -2.7 pts 6.0 -0.3 pts 0.6 pts education bachelor and above 78.3 -2.5pts 0.5 pts 24.6 -3.1 pts +0.8 pts diplomas, advanced diplomas 76.8 -3.3pts -2.4 pts 19.2 +1.1 pts -0.5 pts certificate iii/iv 75.9 -2.4pts -0.1 pts 17.7 -2.5 pts -0.7 pts year 12 68.2 -4.3pts -1.3 pts 14.2 -2.1 pts +1.3 pts below year 12 56.2 -2.1pts -0.3 pts 10.2 -1.2 pts -1.1 pts source: abs labour force survey. note: other relationships include non-dependent child, other related individual, nonfamily members and visitors. pts refers to percentage points. the initial reduction in the employment rate fell disproportionately for females and partnered individuals aged 50–64 years (birch & preston 2021), although these older individuals were not disadvantaged in the longer period (however, wood et al. (2021) find women’s overall employment remains below pre-covid-19 levels). hidden behind the overall recovery in employment rates in the longer term was a gain for males and partnered individuals aged 65 years or over. hardest hit by the initial reduction in employment were individuals aged 50–64 years with a spouse and/or dependent children. however, their prospects improved over the longer period. by december, the largest losses of employment were among lone persons aged 65 years or over (-2.5 points), but older individuals with dependent children increased their employment by 2.2 points. australian population studies 5 (2) 2021 fry et al. 37 among those with identified education levels and aged 50–64 years, all experienced an initial decline in employment, particularly individuals with year 12 (which concords with isced 3, upper secondary) or diplomas and advanced diplomas. individuals with diplomas and advanced diplomas were particularly disadvantaged as their employment rates did not recover by december. this is an occupational effect associated with the decline in roles in the health care and social assistance sector. among those aged 65 years or over, the largest initial reduction in employment is for holders of bachelor’s degrees and above, associated with sectoral contractions in health care and social assistance, education and training, and professional, scientific and technical services. however, driven by growth in education and training, this group recovers over the longer period. 4. discussion the impacts of covid-19 on australian employment in 2020 exacerbated and exposed inequalities in the labour market (birch & preston 2021). initially, employment rates fell between march and april 2020 for almost all groups, regions and industries included in the study. young people were hardest hit due to their overrepresentation in casual employment and in service industries hardest hit by lockdown restrictions. however, older workers also suffered. overall, there were initial period negative effects as older workers exited the labour force, but the economic recovery and policy responses such as jobkeeper (with uptake of 3.8% in december) and working from home appear to have ameliorated the longer period effects. the pandemic was initially broadly perceived to be an event that would have a strong negative impact on business and, hence, on employment. later, when businesses and organisations obtained access to jobkeeper, and the pandemic’s impact in some geographic areas was not large, employment recovered steadily. thus, the employment impacts of covid in australia were buffered in regions where case numbers were low. the introduction of government policies like jobkeeper (which favoured those in non-casualised employment) reduced the impact for some groups, such as older workers. at a national level, then, the long-term impact of 2020 on the employment outcomes for older workers is likely to be different to other recessions because the economic costs are likely to persist (coates et al. 2020). in terms of geography, we concentrated primarily upon victoria, but comparison was made with new south wales, a similar state in economic terms (gross state product per capita); and with western australia, where covid-19 was least prevalent among the states. initially, covid-19 was slightly more prevalent in new south wales than in victoria. however, victoria had a very large second wave not experienced by new south wales, and that experience was confined largely to metropolitan melbourne. as the prevalence of covid-19 was heavily concentrated in cities, we compare with regional australia and regional victoria. these results reflect a natural experiment: victoria was subject to the same employment support policies as the rest of australia because those policies were mainly federal (the exception being paid pandemic leave for covid-19 testing and associated selfisolation). the different effects between victoria, new south wales and western australia were therefore due (almost) purely to the different experience of covid-19 (besides any changes that would have taken place in the absence of covid-19, which can be expected to be very small in the short term). the same argument can be made about metropolitan melbourne and regional victoria. the reason victoria (particularly regional victoria) had a smaller contraction in employment is that it is less intensive in mining, and, to a lesser extent, accommodation and food services and other services sectors, which all suffered large reductions in employment. 38 fry et al. australian population studies 5 (2) 2021 by december 2020, there were few instances where employment loss remained considerable. the main exceptions where employment effects were still negative at the end of 2020 were people living in metropolitan melbourne, private sector workers aged 50–59 years, people living alone aged 65 years and over, and those with the lowest education qualification. in the private sector, for those aged 45–60 years, the initial negative impact on employment numbers was largely sustained in the longer period. this may indicate that private sector employers took the opportunity to retrench some older, lower skilled workers. data not shown here reveal that outcomes were worse for occupations at the lower end of the skills spectrum, although this was not always the case. for example, strongly negative outcomes in the longer period were evident for food preparation assistants, farm forestry and garden workers, sales support workers, and clerical and office support workers. on the other hand, there was a private sector recovery for those aged 60 years or over and this could be because these people tend to be self-employed or employed as managers/professionals and had more control over their own employment. in terms of occupations, for storepersons, a low skilled job, there were strong positive results in the longer period, probably because covid-19 stimulated jobs in this occupation. overall, while the national picture for older workers appeared positive by the end of 2020, lowerskilled workers in some sectors experienced significant deleterious effects. long-term job loss for older workers was concentrated in the following industries: manufacturing; retail; health care and social administration; transport, postal, warehouse; arts and recreation; and information, media and telecommunications. the immediate effects of covid-19 on these industries were obvious as entertainment venues, cafes and restaurants were closed, and hotels were affected by travel restrictions. lockdowns probably also prevented information, media and telecommunications industries operating as normal. the large initial negative impact for transport, postal and warehousing was probably mainly in transport, as lockdowns affected airlines and public transport. some industries recovered well. for example, distribution probably accounts for the large increase in employment in accommodation and food services as, beyond the initial period, there was a massive expansion in home delivery of food. gains in administrative and support services may be due to the time taken to develop alternative approaches to on-site working. the turnaround in mining was related to increases in demand from outside of australia. in the case of iron ore, where exports and prices have expanded greatly, this was the result of the disruption that covid-19 caused in other countries, especially brazil (ker 2020). jobkeeper was clearly successful at maintaining attachment to the workplace, especially for higher level workers. however, in some industries, the data imply that employers were either ineligible for payments (with an insufficient reduction in turnover, or being an excluded organisation), chose not to enrol staff for jobkeeper, or employees were not eligible for payments. this particularly affected sectors with high casualisation or many migrant workers: jobkeeper was structured in favour of permanent employees and long-term casuals, and those with australian residency. although there was little impact on the unemployment rate, job search duration increased over time, suggesting no churn in the unemployed and an increase in long-term unemployment. initially, males were more affected by unemployment (particularly in arts and recreation services and accommodation and food services). australian population studies 5 (2) 2021 fry et al. 39 there were industries where the pandemic itself stimulated employment, even in the initial period. this includes financial and insurance services, wholesale trade and construction. businesses wishing to take advantage of various government schemes to support employment (especially jobkeeper) and people making personal decisions about their financial circumstances probably led to the early surge in employment in the financial and insurance services sector. growth in construction was likely also related to the later ‘homebuilder’ stimulus program, which provided grants for homeowners to build, renovate and buy new-build homes. the longer-term impacts by industry were very diverse, with some sectors benefitting substantially while some others suffered considerably. there remain major concerns about the employment of older workers in the industries where employment has remained low in the longer term, especially those with lower skill levels. some sectoral shifts effected by covid-19 — such as the growth of online shopping rather than retail, and the growth in food delivery — will particularly affect older workers’ employment going forward. from past experience in the manufacturing industry, older workers who lose their jobs likely have great difficulty ever gaining work again and retire completely from the labour force (callan & bowman 2015; chomik 2020; oecd 2016). while this has negative consequences for income and superannuation, it can also affect mental health (which can also affect physical health), thereby further worsening financial circumstances through increased medical expenditure (mandal & roe 2007). there may also be negative effects on the wellbeing of others in the household (siegel et al. 2004). those who do find work can have lower earnings due to shorter hours or lower hourly pay rates and often find themselves in precarious forms of employment. our analysis has used cross-sectional data tabulations from the abs labour force survey. this means our employment measures do not necessarily capture the same individuals. rather, there could be churn with other labour market states hidden in these figures. moreover, take-up of jobkeeper may have influenced employment results, with individuals classified as employed although they did not work. future research could explore casualisation of the workforce more fully, potentially using longitudinal data from the household, income and labour dynamics in australia (hilda) survey or the abs longitudinal labour force dataset. linked census data may also help to flesh out the longitudinal aspects. space considerations meant we were unable to explore redundancy; the relationship between employment outcomes and working arrangements would also be of interest, as would exploring the gendered aspects of jobs that have been created post-covid-19. it would also be useful to examine job search duration post-retrenchment. future research could also consider the wider array of government stimulus measures, and how they have impacted job prospects in specific sectors. 5. conclusions noting these limitations and extensions, our study provides evidence of the impact of the covid-19 pandemic on labour force behaviour and outcomes for older australians. usually in recessions older workers fare worse than younger workers. however, in the recession induced by covid-19 this did not always occur, as younger workers are more casualised and concentrated in affected industries. the longer-term impacts by industry were very diverse, with some sectors benefitting substantially, while some others suffered considerably. this is the most important finding of the paper. by 40 fry et al. australian population studies 5 (2) 2021 december 2020, employment levels for older workers in some sectors had recovered from the initial downturn caused by the pandemic. however, for older workers in some industries hard hit by the pandemic, there is a major concern about their potential for future employment. there remain major concerns about the employment of older workers in the industries where employment has remained low in the longer term, namely manufacturing; retail trade and information; media and telecommunications; and transport, postal and warehousing. future government stimulus measures could be particularly aimed at supporting workers from these industries. with the phase out of jobkeeper in 2021, and continued shutdowns in victoria and new south wales, it is important to track the continued impact of the covid-19 pandemic on older workers and job seekers. with covid now affecting employment in australia’s two most populous states, it is essential that research continues to consider the ongoing impacts of covid on particular groups in the labour market in the medium and longer-term. key messages • overall employment rates for older workers fell in april 2020 but recovered by december. • the longer-term impacts by industry were very diverse, with some sectors benefitting substantially, while others suffered considerably. • there remain major concerns about the future employment prospects of older workers (particularly those with lower skill levels) in sectors where employment levels had not recovered by december 2020. acknowledgments funding was provided by covid19 impacts on society scheme, university of melbourne. fry, temple and mcdonald are funded by the australian research council’s centre of excellence in population ageing research (ce1101029). references abs (2018) labour statistics: concepts, sources and methods, feb 2018. canberra: abs. https://www.abs.gov.au/ausstats/abs@.nsf/lookup/by%20subject/6102.0.55.001~feb%202018~ main%20features~overview%20of%20labour%20statistics~1. accessed on 11 march 2021. abs (2021a) dataset: characteristics of employment, 2014–2020, canberra: abs. https://www.abs.gov.au/statistics/labour/earnings-and-work-hours/characteristics-employmentaustralia/latest-release. accessed on 11 march 2021. abs (2021b) dataset: labour force australia, detailed. canberra: abs. https://www.abs.gov.au/statistics/labour/employment-and-unemployment/labour-forceaustralia-detailed/latest-release. accessed on 11 march 2021. abs (2021c) labour force, australia methodology. canberra: abs. https://www.abs.gov.au/methodologies/labour-force-australia-methodology/jan-2021. accessed on 11 march 2021. adair t & temple j (2012) barriers to mature age labour force engagement in australia: canberra: department of education, employment and workplace relations. https://nationalseniors.com.au/uploads/201208_pacreport_research_barriersmatureageemplo yment_full_1.pdf https://www.abs.gov.au/ausstats/abs@.nsf/lookup/by%20subject/6102.0.55.001~feb%202018~main%20features~overview%20of%20labour%20statistics~1 https://www.abs.gov.au/ausstats/abs@.nsf/lookup/by%20subject/6102.0.55.001~feb%202018~main%20features~overview%20of%20labour%20statistics~1 https://www.abs.gov.au/statistics/labour/earnings-and-work-hours/characteristics-employment-australia/latest-release https://www.abs.gov.au/statistics/labour/earnings-and-work-hours/characteristics-employment-australia/latest-release https://www.abs.gov.au/statistics/labour/employment-and-unemployment/labour-force-australia-detailed/latest-release https://www.abs.gov.au/statistics/labour/employment-and-unemployment/labour-force-australia-detailed/latest-release https://www.abs.gov.au/methodologies/labour-force-australia-methodology/jan-2021 https://nationalseniors.com.au/uploads/201208_pacreport_research_barriersmatureageemployment_full_1.pdf https://nationalseniors.com.au/uploads/201208_pacreport_research_barriersmatureageemployment_full_1.pdf australian population studies 5 (2) 2021 fry et al. 41 birch e & preston a (2021) the australian labour market in 2020. journal of industrial relations 63(3): 303-320. https://doi.org/10.1177/00221856211002224 borland j & charlton a (2020) the australian labour market and the early impact of covid-19: an assessment. australian economic review 53(3): 297-324. https://doi.org/10.1111/14678462.12386 bui t t m, button p, & picciotti e g (2020) early evidence on the impact of covid-19 and the recession on older workers. nber working paper no. w27448. cambridge, ma: national bureau of economic research. https://doi.org/10.3386/w27448 callan v j & bowman k (2015) industry restructuring and job loss: helping older workers get back into employment. adelaide: national centre for vocational education research. https://www.ncver.edu.au/research-and-statistics/publications/all-publications/industryrestructuring-and-job-loss-helping-older-workers-get-back-into-employment chomik r (2020) covid-19 and the demographic distribution of health and economic risks. arc centre of excellence in population ageing research factsheet. sydney: cepar. https://cepar.edu.au/sites/default/files/covid-19-and-populations-at-risk.pdf churchill b (2020) covid‐19 and the immediate impact on young people and employment in australia: a gendered analysis. gender, work & organization 28(2): 783-794. https://doi.org/10.1111/gwao.12563 coates b, cowgill m, chen t, & mackey w (2020) shutdown: estimating the covid-19 employment shock. melbourne: grattan institute. https://grattan.edu.au/report/shutdown-estimating-the-covid-19employment-shock/ department of health (2021) coronavirus (covid-19) current situation and case numbers. https://www.health.gov.au/news/health-alerts/novel-coronavirus-2019-ncov-healthalert/coronavirus-covid-19-current-situation-and-case-numbers#cases-and-deaths-by-age-andsex. accessed on 11 march 2021. forsyth a (2020) covid-19 and labour law: australia. italian labour law e-journal 13(1s). https://doi.org/10.6092/issn.1561-8048/10812 ker p (2020) brazil’s virus pain to extend australia’s iron ore gain. australian financial review, 28 september 2020. mandal b & roe b e (2007) job loss, retirement and the mental health of older americans. ssrn. http://dx.doi.org/10.2139/ssrn.991134 neil i, chin d, & parkin c (2021) the jobkeeper scheme. australian journal of labour law 34: 24-42. https://advance.lexis.com/api/permalink/aebe5bcb-6da7-4e76-928cfc1fae2fcc93/?context=1201008&federationidp=f2d4kj52509. oecd (1998) work force ageing: consequences and policy responses, working paper awp 4.1. paris: oecd publishing. https://www.oecd.org/els/public-pensions/2429096.pdf oecd (2016) back to work: australia. improving the re-employment prospects of displaced workers. paris: oecd publishing. https://www.oecd.org/employment/emp/back-to-work-australia-ar.pdf oecd (2020) job retention schemes during the covid-19 lockdown and beyond. paris: oecd publishing. https://www.oecd.org/coronavirus/policy-responses/job-retention-schemes-during-the-covid19-lockdown-and-beyond-0853ba1d/ rutledge m s & coe n b (2012) great recession-induced early claimers: who are they? how much do they lose? center for retirement research working paper 2012-12. boston ma: boston college. https://crr.bc.edu/wp-content/uploads/2012/04/wp_2012-12-508.pdf siegel m j, bradley e h, gallo w t, & kasl s v (2004) the effect of spousal mental and physical health on husbands’ and wives’ depressive symptoms, among older adults: longitudinal evidence from the health and retirement survey. journal of aging and health 16: 398-425. https://doi.org/10.1177/0898264304264208 https://doi.org/10.1177/00221856211002224 https://doi.org/10.1111/1467-8462.12386 https://doi.org/10.1111/1467-8462.12386 https://doi.org/10.3386/w27448 https://www.ncver.edu.au/research-and-statistics/publications/all-publications/industry-restructuring-and-job-loss-helping-older-workers-get-back-into-employment https://www.ncver.edu.au/research-and-statistics/publications/all-publications/industry-restructuring-and-job-loss-helping-older-workers-get-back-into-employment https://cepar.edu.au/sites/default/files/covid-19-and-populations-at-risk.pdf https://doi.org/10.1111/gwao.12563 https://grattan.edu.au/report/shutdown-estimating-the-covid-19-employment-shock/ https://grattan.edu.au/report/shutdown-estimating-the-covid-19-employment-shock/ https://www.health.gov.au/news/health-alerts/novel-coronavirus-2019-ncov-health-alert/coronavirus-covid-19-current-situation-and-case-numbers#cases-and-deaths-by-age-and-sex https://www.health.gov.au/news/health-alerts/novel-coronavirus-2019-ncov-health-alert/coronavirus-covid-19-current-situation-and-case-numbers#cases-and-deaths-by-age-and-sex https://www.health.gov.au/news/health-alerts/novel-coronavirus-2019-ncov-health-alert/coronavirus-covid-19-current-situation-and-case-numbers#cases-and-deaths-by-age-and-sex https://doi.org/10.6092/issn.1561-8048/10812 http://dx.doi.org/10.2139/ssrn.991134 https://advance.lexis.com/api/permalink/aebe5bcb-6da7-4e76-928c-fc1fae2fcc93/?context=1201008&federationidp=f2d4kj52509 https://advance.lexis.com/api/permalink/aebe5bcb-6da7-4e76-928c-fc1fae2fcc93/?context=1201008&federationidp=f2d4kj52509 https://www.oecd.org/els/public-pensions/2429096.pdf https://www.oecd.org/employment/emp/back-to-work-australia-ar.pdf https://www.oecd.org/coronavirus/policy-responses/job-retention-schemes-during-the-covid-19-lockdown-and-beyond-0853ba1d/ https://www.oecd.org/coronavirus/policy-responses/job-retention-schemes-during-the-covid-19-lockdown-and-beyond-0853ba1d/ https://crr.bc.edu/wp-content/uploads/2012/04/wp_2012-12-508.pdf https://doi.org/10.1177/0898264304264208 42 fry et al. australian population studies 5 (2) 2021 varma p, junge m, meaklim h, & jackson m (2021) younger people are more vulnerable to stress, anxiety and depression during covid-19 pandemic: a global cross-sectional survey. progress in neuropsychopharmacology and biological psychiatry 109: 110236. https://doi.org/10.1016/j.pnpbp.2020.110236 wood d, griffiths k, & crowley t (2021) women’s work: the impact of the covid crisis on australian women. melbourne: grattan institute. https://grattan.edu.au/report/womens-work/ yeatts d e, folts w e, & knapp j (2000) older workers adaptation to a changing workplace: employment issues for the 21st century. educational gerontology 26(6): 565-582. https://doi.org/10.1080/03601270050133900 https://doi.org/10.1016/j.pnpbp.2020.110236 https://grattan.edu.au/report/womens-work/ https://doi.org/10.1080/03601270050133900 australian population studies 2022 | volume 6 | issue 1 | pages 1–14 © heley & hewitt 2022. published under the creative commons attribution-noncommercial licence 3.0 australia (cc by-nc 3.0 au). journal website: www.australianpopulationstudies.org same-sex marriage and remarriage in australia, 2018-2020 emma heley* the university of melbourne belinda hewitt the university of melbourne *corresponding author. email: ekheley@gmail.com. address: school of social and political sciences, level 4, john medley building, the university of melbourne, parkville vic 3010 paper received 15 february 2022; accepted 9 may 2022; published 25 july 2022 abstract background since same-sex marriage was legalised in australia in december 2017, thousands of same-sex couples have married. little australian research to date has examined same-sex marriage trends and how they are similar or different to different-sex marriages. aims to describe trends in same-sex marriages and remarriages in australia for the years 2018–2020, comparing and contrasting with different-sex marriages and remarriages by age and sex. data and methods data come from the australian bureau of statistics, marriage and divorces, australia. we use available data on different-sex and same-sex marriages and remarriages for age and sex. results the proportion of same-sex marriages declined from 5.5% of total marriages to 3.7% between 2018 and 2020. same-sex marriages tended to be at older ages, but the age gap between same-sex and different-sex marriages for males and females reduced over time. similar trends were observed for remarriage. for different-sex remarriages, there was a slightly higher proportion of males remarrying than females. in contrast, for same-sex remarriages, the proportion of females remarrying was double that of same-sex males. conclusions early trends suggest same-sex marriages and remarriages occurred at older ages than different-sex. this age gap reduced over time, suggesting that early adopters of same-sex marriage may be a different group. interestingly, same-sex remarriage was much more common for females. given that same-sex marriage was not legal in australia until late 2017, it is likely that many of their first marriages were different-sex. this has received little attention in the research literature to date and requires further investigation. key words marriage; remarriage; same-sex; australia. http://www.australianpopulationstudies.org/ mailto:ekheley@gmail.com 2 heley & hewitt australian population studies 6 (1) 2022 1. introduction in december 2017, same-sex marriage was legalised in australia. gay rights movements had been pushing to reform discriminatory legislation since the 1970s (greenwich & robinson, 2018). however, it wasn’t until 1997 that homosexuality was decriminalised in australia (willett, 2010). since 2000, a number of bills supporting same-sex marriage had been introduced in parliament (mcallister & snagovsky, 2018), but none were passed until 2017 when the commonwealth government held a postal vote administered by the australian bureau of statistics (abs) (mcallister & snagovsky, 2018). the australian public was asked: ‘should the law be changed to allow same-sex couples to marry?’, to which respondents could voluntarily answer “yes” or “no” (mcallister & snagovsky, 2018). the majority of the australian public voted “yes” in the same-sex marriage postal survey (wilson, shalley & perales, 2020), and shortly after legislation, same-sex marriage was passed by the australian parliament (mcallister & snagovsky, 2018). since then, many queer australian couples have legally married, with thousands of same-sex marriages registering from 2018 to 2020 (abs, 2021a; abs, 2020a; abs, 2019). in this study, ‘queer’ is used as an umbrella term inclusive of all lgbtqia+ identities (australian institute of family studies, 2022a). in this paper we describe and document marriage and remarriage trends for same-sex and differentsex marriages since the law change. we also investigate if this varies according to age and sex. we use the term different-sex, rather than heterosexual, to refer to non-same-sex marriages because the abs includes people who don’t identify as male or female in the total marriage counts (abs, 2021b). therefore, even though the number of marriages impacted is likely to be small, the term differentsex is more accurate and inclusive of marriages which are not same-sex but involve people who are queer and just don't fit the "same-sex" definition. an example of a queer couple who are not “samesex” would be a lesbian couple where one partner is a nonbinary lesbian (hord 2020), and the other partner is a cisgender female lesbian. using data from the abs (abs, 2021a; abs, 2020a; abs, 2019), we investigate the following questions: in the years since same-sex marriage was legalised (i) what proportion of total marriages were same-sex marriages?; (ii) are there differences in proportions of remarriages between different-sex and same-sex couples?; (iii) are there age differences between different-sex and same-sex marriage and remarriage patterns?; (iv) are there sex differences between different-sex and same-sex marriage and remarriage patterns? from a life course perspective, we would expect differences in the timing and patterning of marriage and remarriage for those in same-sex marriages compared to different-sex marriages (carpenter, 2010). the life course perspective emphasises interactions between individual life biographies and historical time and place (elder, 1994), where an individual’s biography and life course transitions are shaped by the broader social context including legal, cultural, and moral constraints. marriage and sexual identity have been particularly influenced by societal shifts in many western developed countries over recent decades (cohler, 2005), with opportunities for marriage over time differing according to sexuality. while we have seen changes in trends of marriage and remarriage for different-sex couples over time, such as increases in cohabitation, delayed marriage, and declining rates of marriage overall, legal marriage has always been a life course option (de vaus, 2004). for same-sex couples in australia, the 2017 change in marriage laws has shifted marriage opportunities. australian population studies 6 (1) 2022 heley & hewitt 3 marriage for sexual minorities has only recently become a normative, and legal, life event in a handful of countries globally (cohler, 2005; hewitt & churchill, 2020; digregorio, 2016). in contrast to older queer generations in those countries (cohler, 2005), queer young people will grow up in a context where same-sex marriage is available to them. historically, many same-sex attracted people had entered into different-sex marriages for a range of reasons (higgins 2002; pearcey 2005). for example, in a qualitative australian study of 26 gay or bisexual men in victoria, the most common reasons why these men had previously married a woman was because they wanted children and a family, and that it ‘seemed natural’ in the social context at the time (higgins 2002). the literature on why lesbian and bisexual women marry opposite sex partners is limited compared to the literature available on men. little research has been undertaken on same-sex remarriages, with most australian and international studies being conducted on re-partnering in different-sex relationships (hughes, 2000; raley & sweeney, 2020; skew, evans, & gray, 2009). in australia, remarriages of one or both partners after divorce or widowhood accounted for 27.42% to 28.56% of total marriages from 2018 to 2020 (abs, 2020a). however, it should be noted that remarriage in australia has declined from around 34% since the year 2000 (abs, 2021a; abs, 2020a; abs, 2019). nowadays, many couples re-partner by cohabiting rather than remarrying (skew, evans, & gray, 2009). research on re-partnering and remarriage amongst different-sex couples suggests that sex and age are important factors. men are more likely to remarry and re-partner than women (hughes, 2000; skew, evans, & gray, 2009). this is possibly because men have more options on the remarriage market (skew, evans, & gray, 2009), as it is more socially acceptable for them to marry younger women (england & mcclintock, 2009). age is negatively associated with re-partnering, where people who have separated or divorced at younger ages are more likely to re-partner (skew, evans, & gray 2009). the current study provides some of the first australian information on same-sex remarriages between men and same-sex remarriages between women. despite gay male and lesbian female unions being qualitatively different (ducharme & kollar, 2012), discussion of sex differences in samesex remarriage has been limited in previous studies. 2. data and methods 2.1 data the data come from the abs which collates data on all marriages registered in a calendar year for each of the states and territories in australia (abs, 2021b). all data and variables used in this study were pre-determined by the abs. even though the same-sex marriage legislation took effect from the 9th december 2017, data were not reported by the abs for same-sex marriages in 2017 due to privacy considerations, so our study examines the years 2018, 2019 and 2020 (abs, 2021a; abs, 2020a; abs, 2019). 2.2 definitions and measures marriage refers to all marriages which are registered with births, deaths and marriages registrars in each australian state and territory which are then recorded by the abs (abs, 2021b). same-sex marriage, as defined by the abs, refers to either the number of marriages in which both partners 4 heley & hewitt australian population studies 6 (1) 2022 identify as female, or where both partners identify as male (abs 2021b). we define any marriages not classified as same-sex marriages as different-sex marriages, and note that the abs includes people who don’t identify as male or female in the total marriage counts (abs, 2021b). therefore, it is likely that a small number of marriages which are not same-sex might include people who are queer, for example they could be male-nonbinary or female-nonbinary marriages, but don't fit this "same-sex" definition. remarriages refers to registered marriages where one or both persons were previously married (abs, 2019). this is inclusive of couples where one spouse has previously been married, but the other could be entering their first or second marriage (abs, 2021a; abs, 2020a; abs, 2019). information on the total number and proportion of same-sex and different-sex remarriages were derived by summing the number of marriages for males and females whose prior marital status was widowed or divorced. sex was defined by the abs as either ‘male’, ‘female’, or ‘x’ (abs, 2021b). both individuals and samesex marriages are categorised by sex in this study, as determined by the abs (abs, 2021b). marriages where one or both partners identified outside of the male-female binary were excluded from abs data presented by sex (abs, 2021b). however, marriages and individuals were included in counts of total marriages and total people married (abs, 2021b). as such, the sex categories in this study are male and female. all age categories in this study were pre-determined by the abs. age was measured in two ways in the study: age (with categories ranging from age 16 to 50 and above) and median age at marriage. age refers to the number of individuals in a given age group (in years) who were married in that specific year. this is represented as the following age ranges: 16-19, 20-24, 25-29, 30-34, 35-39, 4044, 45-49, and 50+. the abs grouped individuals aged 50+ into one category and we were therefore unable to differentiate between smaller age categories within the 50+ age group in this study. median age at marriage refers to the median age of individuals who got married in that year, according to sex. this was calculated by the abs. 2.3 approach this study compares and contrasts trends in same-sex and different-sex marriages in australia. data provided by the abs was represented as a series of cross-tabulations with numerical data. these were then re-arranged in a series of tables and graphs, which address the research questions. percentages of each variable, at either the individual or marriage level, were calculated to allow for comparison across and between variables. 3. results 3.1 marriage table 1 shows the proportion of total marriages in each year that were same-sex compared to different-sex. the results show that out of total marriages, same-sex marriages accounted for 5.49% in 2018, 4.84% in 2019, and 3.67% in 2020. we note that the number of total marriages for 2020 was low because of the impact of covid-19 restrictions on being able to marry and host weddings in australia (australian institute of family studies, 2022b). australian population studies 6 (1) 2022 heley & hewitt 5 table 1: same-sex marriages as a proportion of total marriages same-sex marriages different-sex marriages total marriages year % % n 2018 5.49 94.51 119,188 2019 4.84 95.16 113,815 2020 3.67 96.33 78,989 source: abs 2021a; abs 2020a; abs 2019. in table 2, we present the proportion of same-sex marriages for females and males. we do not present these numbers separately for different-sex marriages because the overall numbers of males and females entering different-sex marriages are virtually equal. the results indicate that a greater proportion of same-sex marriages were between females (57-61%) compared to males (38-42%), demonstrating that a larger number of female same-sex couples got married each year than male same-sex couples. table 2: same-sex marriages by sex female male total year % % n 2018 57.83 42.17 6,538 2019 58.89 41.07 5,507 2020 61.41 38.49 2,902 source: abs 2021a; abs 2020a; abs 2019. figures 1 and 2 present the age distributions of total marriages and same-sex marriages separately for females and males. overall, the results indicate that same-sex marriages were more evenly distributed across all age groups. in 2018, same sex marriages were most common in the 50 and over age category, particularly for males, likely reflecting the pent-up demand for marriages amongst older couples in long term relationships (greenwich & robinson, 2018). in 2019 and 2020, the proportion of same-sex marriages in this age group reduced and the proportion of marriages in the younger age groups of 25-29, 30-34, and 35-39 increased. in comparison, the distribution across age groups for different-sex marriages was remarkably consistent. figure 1: age at marriage for females, same-sex and different-sex marriages, 2018-2020 source: abs 2021a; abs 2020a; abs 2019. 6 heley & hewitt australian population studies 6 (1) 2022 figure 2: age at marriage for males, same-sex and different-sex marriages, 2018-2020 source: abs 2021a; abs 2020a; abs 2019. 3.2 remarriage table 3 shows the results for the proportion of same-sex and different-sex marriages that were remarriages. the results show that about one-quarter of total marriages, both different-sex and same-sex, were remarriages. while the proportion of different-sex remarriages is slightly higher across all years, there is not a large difference. table 3: remarriage proportions of same-sex and different-sex marriages same-sex remarriages different-sex remarriages year % of total same-sex marriages % of total different-sex marriages 2018 26.38 27.97 2019 23.77 27.42 2020 25.09 28.56 source: abs 2021a; abs 2020a; abs 2019. table 4 presents the most interesting finding in this study. we present same-sex and different-sex remarriages as a proportion of total marriages for females and males. the literature on differentsex remarriage indicates that men are more likely than women to remarry (de vaus, 2004; skew, evans & gray, 2009). this is supported by the data for different-sex remarriages, where a slightly higher proportion of males were remarried each year (20% – 21%) than females (18% – 19%) across all time points. we find very different results for same-sex remarriages. a much higher proportion of females entered same-sex remarriages than males. this difference was large. while male same-sex remarriages accounted for around 10% of all male same-sex marriages in 2018, 2019 and 2020, for females, same-sex remarriages accounted for between 16% and 20% of samesex marriages, around double that for same-sex males. australian population studies 6 (1) 2022 heley & hewitt 7 table 4: people in same-sex and different-sex remarriages as a proportion of total people married, by sex same-sex remarriages different-sex remarriages female male female male year % % % % 2018 19.35 10.41 19.00 20.77 2019 16.52 09.64 18.75 20.39 2020 18.08 10.39 19.74 20.93 source: abs 2021a; abs 2020a; abs 2019. to investigate this finding further, we examined the previous marital status of males and females in same-sex marriages. while females were more likely to have been widowed than males, the main differences were in those who had been divorced. of total male same-sex marriages, the proportion of men who had been previously divorced was 10.12% in 2018, 9.31% in 2019, and 10.12% in 2020. women who had previously been divorced accounted for 18.88% of total female same-sex marriages in 2018, 15.97% in 2019, and 17.55% in 2020. this suggests that the much higher proportion of same-sex remarriages for females is due to women having previously been married and subsequently divorced. table 5: previous marital status of same-sex marriages male female year previous marital status no. % no. % 2018 never married 4,940 89.59 6,104 80.69 widowed 16 0.29 36 0.48 divorced 558 10.12 1,428 18.88 total 5,514 100.00 7,565 100.00 2019 never married 4,081 90.25 5,421 83.55 widowed 15 0.33 36 0.56 divorced 421 9.31 1,036 15.97 total 4,522 100.00 6,488 100.00 2020 never married 1999 89.52 2,929 82.12 widowed 6 0.26 19 0.53 divorced 226 10.12 626 17.55 total 2,233 100.00 3,567 100.00 source: abs 2021a; abs 2020a; abs 2019. the abs does not publish specific age category data on remarriages, but rather provides the median ages for females and males marrying for the first time, or remarrying after becoming widowed or divorced. the abs distinguishes between same-sex marriages and total marriages (which is all marriages including same-sex marriages), according to sex. due to median age being calculated by the abs, we are unable to comment on different-sex marriages specifically. in figures 3 and 4 we present information for females and males in same-sex unions compared to total marriages in australia. for females, the results presented in figure 3 are consistent with the information provide in figure 1, where the median age for same-sex marriages was older than the median age for different-sex marriages. in general, this age gap was largest in 2018 and narrowed over the subsequent years. for those in first marriages, the median age gap dropped from 7.1 years in 2018 to 4.3 years in 2020. the smallest gap in median age at marriage between total remarriages and same-sex remarriages was for widowed females at approximately 2.5 years. for 8 heley & hewitt australian population studies 6 (1) 2022 divorced females who remarried, the age gap between total marriages and same-sex marriages was 6.5 years in 2018 and halved to around 3 years by 2020. figure 3: median age at marriage for females, same-sex and total marriages, 2018-2020 source: abs 2021a; abs 2020a; abs 2019. note: total marriages includes same-sex marriages. figure 4: median age at marriage for males, same-sex and total marriages, 2018-2020 source: abs 2021a; abs 2020a; abs 2019. note: total marriages includes same-sex marriages. 29.2 36.3 29.3 34.6 29.2 33.5 55.8 57.6 56.3 57.5 57.7 62.0 44.8 51.3 44.7 48.4 43.8 46.8 0 10 20 30 40 50 60 70 total marriages same-sex marriages total marriages same-sex marriages total marriages same-sex marriages 2 0 1 8 2 0 1 8 2 0 1 9 2 0 1 9 2 0 2 0 2 0 2 0 divorced widowed never married 30.7 43.3 30.7 38.1 30.6 35.9 64.1 68.0 65.1 58.8 65.3 63.5 48.3 55.7 48.5 53.8 48.0 52.5 0.0 10.0 20.0 30.0 40.0 50.0 60.0 70.0 80.0 total marriages same-sex marriages total marriages same-sex marriages total marriages same-sex marriages 2 0 1 8 2 0 1 8 2 0 1 9 2 0 1 9 2 0 2 0 2 0 2 0 divorced widowed never married australian population studies 6 (1) 2022 heley & hewitt 9 4. discussion with the legalisation of same-sex marriage in australia in december 2017, same-sex couples were able to marry for the first time. this legal change brought australia in line with around 22 other countries across the world at the time (hewitt & churchill, 2020), with another seven countries introducing marriage equality laws since. with three years of abs data now available on same-sex marriages in australia, the main purpose of this paper was to describe some early trends and identify similarities and differences with different-sex marriages and remarriages. 4.1 marriage first, we described the proportion of total marriages that were same-sex marriages. given that australia’s sexual minority population aged over 18 is estimated to be around 3.5% of the total adult population (wilson et al., 2020), there was a relatively high proportion of total marriages which were same-sex marriages in the first two years after the legalisation of same-sex marriage; 5.49% in 2018 and 4.84% in 2019. international data suggests this proportion of same-sex marriages is higher compared to other countries, including belgium, the netherlands, norway, spain, and sweden (chamie & mirkin, 2011). rather than necessarily reflecting a higher propensity for those in the nonheterosexual population to marry, it is more likely due to a backlog of same-sex couples wanting to be married prior to 2017 (greenwich & shirleene, 2018). the proportion of same-sex marriages in 2020 (3.67%) is more closely aligned to the population estimates. this trend of declining numbers of same-sex marriages is similar to that observed in other countries over the years after same-sex marriage was introduced (hewitt & churchill 2020; chamie and mirkin 2011). similar trends have been observed in other countries where legalisation of same-sex marriage was highly politicised. for example, in france, scotland, and ireland, a high proportion of same-sex marriages were observed immediately after legalisation, but in the following years, the proportion declined (hewitt & churchill, 2020). table 2 shows that females account for a higher proportion of people in same-sex marriages compared to men. this aligns with australian and international literature on same-sex marriages, including in france, norway, sweden, and some states in the us, where more women got married compared to men (chamie & mirkin, 2011; meslay, 2019). prior research has suggested that lesbian women and gay men relate to marriage differently. gay men are more sceptical of marriage, have a lower desire to marry, and get married less compared to women (baiocco et al., 2014; chamie & mirkin, 2011, philpot et al., 2016). whereas, lesbian women are more likely to be sexually monogamous, have greater financial needs, and, importantly, are more likely to have children (chamie & mirkin, 2011; dempsey, 2015; farr & goldberg, 2019; gates et al., 2007; gates & ost, 2004). marriage is linked to parenthood and childbearing, especially for lesbian women who are increasingly choosing to have children after they marry (kolk & andersson, 2020). this intersection of gender and parenthood means that marriage and divorce play different roles for same-sex married men compared to women (farr & goldberg, 2019). these differences may in part explain gender differences in same-sex marriage. figures 1 and 2 show that men tended to marry at slightly older ages compared to women, in both same-sex and different-sex marriages. this is consistent with different-sex marriage literature (hewitt & churchill, 2020; skew, evans, & gray 2009). we also found relatively large age differences 10 heley & hewitt australian population studies 6 (1) 2022 between people in same-sex and different-sex marriages, where people in same-sex marriages tended to be older than those in different-sex marriages. people in different-sex marriages tended to be in the 20-24 and 25-29 age categories, while people in same-sex marriages were more likely to be in the 25-29, 30-34 and the 50+ age categories. of note is the consistent trend for a larger proportion of men to enter same-sex marriages over the age of 50, compared to women. this is broadly consistent with trends in other countries such as the netherlands and spain, where a much higher proportion of men enter same-sex marriages in their 50s compared to women (hewitt & churchill, 2020). the age differences between people in same-sex and different-sex marriages reduced over the 3 years since the legalisation of same-sex marriage. given that same-sex marriage as a life transition could not take place in australia until legalised in 2017, the first cohorts of gay and lesbian couples who are marrying may have been together for a long time before having access to legal marriage (ducharme & kollar, 2012), thus resulting in an older same-sex marriage cohort. consistent with life course literature (cohler, 2005; reczek, 2020), this change in socio-political and legal context could have shaped the trajectories and therefore life courses of same-sex couples. 4.2 remarriage in terms of the proportion of total marriages that were remarriages for different-sex and same-sex couples, our results suggest only small differences. the proportion of remarriages out of total marriages was slightly higher in different-sex marriages, comprising 28.56% of different-sex marriages in 2020, compared to 25.09% of same-sex marriages in 2020. for same-sex remarriages we note that, given same-sex marriage was illegal until late 2017, the majority, but not necessarily all, of those remarrying would have likely been in a heterosexual marriage for their first marriage. arguably the most interesting finding was that there was a substantial sex difference in same-sex remarriage compared to different-sex remarriages. our results find that in same-sex marriage, women represented a much higher proportion compared to men (table 3). this is supported by data in table 4 showing that the proportion of women remarrying after divorce is greater than men. in contrast, previous research has suggested that, in the broader australian population, men remarry more than women (de vaus, 2004; skew, evans, & gray, 2009). although this is reflected in the different-sex results of this study, with men accounting for a slightly higher proportion than women, this paper highlights that in same-sex remarriages, the opposite occurs. there could be two possible explanations for this. first, similar to our discussion in section 4.1, lesbian women are more inclined to get married than gay men for a range of reasons including attitudes towards marriage and having children (chamie & mirkin, 2011; dempsey, 2015; farr & goldberg, 2019; gates et al., 2007; gates & ost, 2004). secondly, for those remarrying, most, although not necessarily all, would have likely been in a different-sex marriage previously. given that women tend to come-out as queer later than men because they experience the coming-out process differently (grov et al., 2006), they are less likely to realise they are queer before first marriage (coleman, 1985). therefore, they may be more likely to initially marry a different-sex partner, and later marry a same-sex partner. more so than men, lesbian women are expected to conform to ideals of heteronormative marriage, especially if they have children (allen & goldberg, 2019). because of their sexual minority status, these women experience prejudice and stigma that comes with being a lesbian, on top of general stigma from divorce (allen & goldberg, 2019). for lesbian women, the nuances of motherhood, such as the normalisation of lesbian lives through pregnancy and childbirth, australian population studies 6 (1) 2022 heley & hewitt 11 amplify societal pressure that divorced lesbians experience (allen & goldberg, 2019; kolk & andersson, 2020). the link between parenthood and marriage, and the particular stigma that queer women face, could be why women choose to remarry more than men. differences between same-sex and different-sex remarriage patterns show how life courses are gendered and intersectional (carpenter, 2010), with sex trends playing out differently in marriage according to sexuality. female same-sex remarriages remain largely under researched, with the literature predominantly featuring gay and bisexual men (higgins, 2002; pearcey, 2005). this study fills a gap by highlighting differences in marriage trends between different-sex and same-sex couples in australia. this requires more research. unsurprisingly, remarriages for both different-sex and same-sex relationships occurred at older ages. same-sex couples also tended to get remarried at older ages than different-sex couples. 4.3 limitations categories in this study were defined by the abs, meaning some measures were limited. marriages where one or both partners neither identified as male nor female were not included within the category of same-sex marriages, although they were included in total marriage counts (abs, 2020b). by excluding marriages with partners who do not identify within the male-female gender binary, gender diverse populations are not adequately represented, thereby risking erasure and confounding data categories (lyons et al. 2020). this approach also does not take into account people who identify as bisexual. 5. conclusions we have described early trends in same-sex marriage and remarriage in the first few years since the legalisation of same-sex marriage in late 2017. while age at marriage was higher in same-sex marriages compared to different-sex marriages, further monitoring will be required to see if these trends persist or change in subsequent same-sex marriage cohorts. this study has also highlighted a gap in the literature: women who were previously in different-sex marriages but are now in samesex marriages have remained largely under-researched in previous studies. a difference in remarriage trends exists between same-sex and different-sex marriages, particularly for women. this paper has highlighted the need to view life courses and marriage trends through an intersectional lens, accounting for the unique experiences of sexually diverse women. key messages • same-sex marriages comprised around 5.49% (2018) to 3.69% (2020) of all marriages. • overall, remarriage rates for same-sex and different-sex couples were similar. • same-sex marriages and remarriages tended to be older than different-sex marriages and remarriages, although these age differences reduced between 2018 and 2020. • for females, same-sex remarriages comprised around 20% of all marriages, about double that for males. 12 heley & hewitt australian population studies 6 (1) 2022 references abs (2019). marriages and divorces, australia (3310.0) [data set]. https://www.abs.gov.au/statistics/people/people-and-communities/marriages-and-divorcesaustralia/2018. abs (2020a). marriages and divorces, australia [data set]. https://www.abs.gov.au/statistics/people/people-and-communities/marriages-and-divorcesaustralia/latest-release abs (2020b). marriages and divorces, australia, methodology. https://www.abs.gov.au/methodologies/marriages-and-divorces-australia-methodology/2019. abs (2021a). marriages and divorces, australia [data set]. https://www.abs.gov.au/statistics/people/people-and-communities/marriages-and-divorcesaustralia/latest-release#media-releases abs (2021b). marriages and divorces, australia, methodology. https://www.abs.gov.au/methodologies/marriages-and-divorces-australia-methodology/2020. allen, k. r., & goldberg, a. e. (2019). lesbian women disrupting gendered, heteronormative discourses of motherhood, marriage, and divorce. journal of lesbian studies, 24(1), 1–13. https://doi.org/10.1080/10894160.2019.1615356 australian institute of family studies (2022a). lgbtiqa+ glossary of common terms [fact sheet]. https://aifs.gov.au/cfca/sites/default/files/publication-documents/2202_rs_lgbtiqa_glossary_of_common_terms.pdf. australian institute of family studies (2022b). marriages in australia [fact sheet]. https://aifs.gov.au/facts-and-figures/marriages-australia. baiocco, r., argalia, m., & laghi, f. (2014). the desire to marry and attitudes toward same-sex family legalization in a sample of italian lesbians and gay men. journal of family issues, 35(2), 181–200. https://doi.org/10.1177/0192513x12464872 carpenter, l. m. (2010). gendered sexuality over the life course: a conceptual framework. sociological perspectives, 53(2), 155–178. https://doi.org/10.1525/sop.2010.53.2.155 chamie, j., & mirkin, b. (2011). same-sex marriage: a new social phenomenon. population and development review, 37(3), 529–551. https://doi.org/10.1111/j.1728-4457.2011.00433.x cohler, b. j. (2005). life course social science perspectives on the glbt family. journal of glbt family studies, 1(1), 69-96. https://doi.org/10.1300/j461v01n01_06 coleman, e. (1985). bisexual women in marriages. journal of homosexuality, 11(1-2), 87-100. https://doi.org/10.1300/j082v11n01_08 de vaus, d. (2004). diversity and change in australian families: statistical profiles. australian institute of family studies. https://aifs.gov.au/publications/diversity-and-change-australian-families. digregorio, n. (2016). same-sex marriage policies and lesbian family life. sexuality research & social policy, 13(1), 58-72. https://doi.org/10.1007/s13178-015-0211-z ducharme, j., & kollar, m. (2012). does the “marriage benefit” extend to same-sex union? evidence from a sample of married lesbian couples in massachusetts. journal of homosexuality, 59(4), 580-591. https://doi.org/10.1080/00918369.2012.665689 elder, g. h. (1994). time, human agency, and social change: perspectives on the life course. social psychology quarterly, 57(1), 4-15. https://doi.org/10.2307/2786971 england, p., & mcclintock, e. a. (2009). the gendered double standard of ageing in us marriage markets. population & development review, 35(4), 797-816. https://doi.org/10.1111/j.17284457.2009.00309.x farr, r. h., & goldberg, a. e. (2019). same-sex relationship dissolution and divorce: how will children be affected? in a. e. goldberg & a. p. romero (eds.), lgbtq divorce and relationship dissolution: https://www.abs.gov.au/statistics/people/people-and-communities/marriages-and-divorces-australia/2018 https://www.abs.gov.au/statistics/people/people-and-communities/marriages-and-divorces-australia/2018 https://www.abs.gov.au/statistics/people/people-and-communities/marriages-and-divorces-australia/latest-release https://www.abs.gov.au/statistics/people/people-and-communities/marriages-and-divorces-australia/latest-release https://www.abs.gov.au/methodologies/marriages-and-divorces-australia-methodology/2019 https://www.abs.gov.au/statistics/people/people-and-communities/marriages-and-divorces-australia/latest-release#media-releases https://www.abs.gov.au/statistics/people/people-and-communities/marriages-and-divorces-australia/latest-release#media-releases https://www.abs.gov.au/methodologies/marriages-and-divorces-australia-methodology/2020 https://doi.org/10.1080/10894160.2019.1615356 https://aifs.gov.au/cfca/sites/default/files/publication-documents/22-02_rs_lgbtiqa_glossary_of_common_terms.pdf https://aifs.gov.au/cfca/sites/default/files/publication-documents/22-02_rs_lgbtiqa_glossary_of_common_terms.pdf https://aifs.gov.au/facts-and-figures/marriages-australia https://doi.org/10.1177/0192513x12464872 https://doi.org/10.1525/sop.2010.53.2.155 https://doi.org/10.1111/j.1728-4457.2011.00433.x https://doi.org/10.1300/j461v01n01_06 https://doi.org/10.1300/j082v11n01_08 https://aifs.gov.au/publications/diversity-and-change-australian-families https://doi.org/10.1007/s13178-015-0211-z https://doi.org/10.1080/00918369.2012.665689 https://doi.org/10.2307/2786971 https://doi.org/10.1111/j.1728-4457.2009.00309.x https://doi.org/10.1111/j.1728-4457.2009.00309.x australian population studies 6 (1) 2022 heley & hewitt 13 psychological and legal perspectives and implications for practice. oxford university press. https://doi.org/10.1093/med-psych/9780190635176.001.0001. gates, g. j., & ost, j. (2004). gay and lesbian atlas. urban institute press. gates, g. j., badgett, m. v. l., macomber, j. e., & chambers, k. (2007). adoption and foster care by gay and lesbian parents in the united states. the williams institute and urban institute. https://www.urban.org/research/publication/adoption-and-foster-care-lesbian-and-gay-parentsunited-states greenwich, a., & robinson, s. (2018). yes yes yes: australia's journey to marriage equality. sydney, newsouth publishing. grov, c., bimbi, d. s., nanín, j. e., & parsons, j. t. (2006). race, ethnicity, gender, and generational factors associated with the coming-out process among gay, lesbian, and bisexual individuals. the journal of sex research, 43(2), 115–121. https://doi.org/10.1080/00224490609552306 hewitt, b., & churchill, b. (2020). convergence and difference: marriage and family life from a crosscultural perspective. in w. k. halford & f. van de vijver (eds.), cross-cultural family research and practice (pp. 57-102). academic press. https://doi.org/10.1016/b978-0-12-815493-9.00003-x. higgins, d. j. (2002). gay men from heterosexual marriages: attitudes, behaviours, childhood experiences, and reasons for marriage. journal of homosexuality, 42(4), 15-34. https://doi.org/10.1300/j082v42n04_02 hord, l. c. r. (2020). specificity without identity: articulating post-gender sexuality through the “nonbinary lesbian”. sexualities, december 2020. https://doi.org/10.1177/1363460720981564 hughes, j. (2000). repartnering after divorce: marginal mates and unwedded women. family matters, 55, 16–21. kolk, m., & andersson, g. (2020). two decades of same-sex marriage in sweden: a demographic account of developments in marriage, childbearing, and divorce. demography, 57, 147–169. https://doi.org/10.1007/s13524-019-00847-6 lyons, a., anderson, j., rasmussen, m. l., & gray, e. (2020). toward making sexual and gender diverse populations count in australia. australian population studies, 4(2), 14-29. https://doi.org/10.37970/aps.v4i2.69 mcallister, i., & snagovsky, f. (2018). explaining voting in the 2017 australian same-sex marriage plebiscite. australian journal of political science, 53(4), 409-427. https://doi.org/10.1080/10361146.2018.1504877 meslay, g. (2019). five years of same-sex marriage in france: differences between male and female couples. population, 74(4), 465–482. https://www.ined.fr/fichier/s_rubrique/178/e_popu_1904_0499.en.pdf pearcey, m. (2005). gay and bisexual married men’s attitudes and experiences: homophobia, reasons for marriage, and self-identity. journal of glbt family studies, 1(4), 21-42. https://doi.org/10.1300/j461v01n04_02 philpot, s. p., ellard, j., duncan, d., dowsett, g. w., bavinton, b. r., down, i., keen, p., hammoud, m. a., & prestage, g. (2016). gay and bisexual men’s interest in marriage: an australian perspective. culture, health & sexuality, 18(12), 1347–1362. https://doi.org/10.1080/13691058.2016.1184314 reczek, c. (2020). sexual and gender minority families: a 2010 to 2020 decade in review. journal of marriage & family, 82(1), 300-325. https://doi.org/10.1111/jomf.12607 skew, a., evans, a., & gray, e. (2009). repartnering in the united kingdom and australia. journal of comparative family studies, 40(4), 563-585. https://doi.org/10.3138/jcfs.40.4.563 willett, g. (2010). howard and the homos. social movement studies, 9(2), 187-199. https://doi.org/10.1080/14742831003603323 wilson, t., temple, j., lyons, a., & shalley, f. (2020). what is the size of australia’s sexual minority population? bmc research notes, 13(1), 1–6. https://doi.org/10.1186/s13104-020-05383-w. https://doi.org/10.1093/med-psych/9780190635176.001.0001 https://www.urban.org/research/publication/adoption-and-foster-care-lesbian-and-gay-parents-united-states https://www.urban.org/research/publication/adoption-and-foster-care-lesbian-and-gay-parents-united-states https://doi.org/10.1080/00224490609552306 https://doi.org/10.1300/j082v42n04_02 https://doi.org/10.1177/1363460720981564 https://doi.org/10.1007/s13524-019-00847-6 https://doi.org/10.37970/aps.v4i2.69 https://doi.org/10.1080/10361146.2018.1504877 https://www.ined.fr/fichier/s_rubrique/178/e_popu_1904_0499.en.pdf https://doi.org/10.1300/j461v01n04_02 https://doi.org/10.1080/13691058.2016.1184314 https://doi.org/10.1111/jomf.12607 https://doi.org/10.3138/jcfs.40.4.563 https://doi.org/10.1080/14742831003603323 https://doi.org/10.1186/s13104-020-05383-w 14 heley & hewitt australian population studies 6 (1) 2022 wilson, t., shalley, f., & perales, f. (2020). the geography of australia’s marriage law postal survey outcome. area, 52(1), 164-175. https://doi.org/10.1111/area.12558 https://doi.org/10.1111/area.12558 australian population studies 2022 | volume 6 | issue 2 | pages 14–26 © wilson, zou, and sigler 2022. published under the creative commons attribution-noncommercial licence 3.0 australia (cc by-nc 3.0 au). journal website: www.australianpopulationstudies.org were there really 1 million unoccupied dwellings in australia on census night 2021? tom wilson* university of melbourne frank zou the university of queensland thomas sigler the university of queensland *corresponding author. email: wilson.t1@unimelb.edu.au. address: melbourne school of population and global health, the university of melbourne, 207 bouverie st, melbourne, vic 3010, australia paper received 2 september 2022; accepted 13 december 2022; published 19 december 2022 abstract background the 2021 census in australia revealed that just over 1 million dwellings were ‘unoccupied’ on census night. this finding was widely reported and may have given the impression of a large number of vacant dwellings ready for households to move into, potentially offering a solution to homelessness and those struggling to find suitable or affordable accommodation. aims the aim of the paper is to investigate whether there really were 1 million unoccupied dwellings in australia in 2021, to shed some conceptual and empirical light on exactly what is meant by an ‘occupied’ and an ‘unoccupied’ dwelling, and also try to understand why dwellings were unoccupied. data and methods we used a variety of census, population, and dwelling data to estimate the number of private dwellings disaggregated by occupancy on both a de facto basis (whether people were present in dwelling on census night or not) and on a usual residence basis (whether people are usually resident in a dwelling or not). a comparison with the situation at the time of the 2016 census is made. results the results show that there were indeed about 1 million dwellings unoccupied on a usual residence basis in australia in 2021. but they were not the exact same 1 million unoccupied on census night, and not all of these dwellings were available to households to live in. there was a substantial increase in the number of dwellings unoccupied by usual residents between 2016 and 2021; we suggest some possible reasons for this, including covid-related effects. conclusions greater clarity and more detail are needed in census dwelling data. in addition, it would be useful if there were detailed annual official statistics on dwellings and households to better inform housing policy and research. key words unoccupied dwellings; census of population and housing; de facto; de jure; australia. http://www.australianpopulationstudies.org/ mailto:wilson.t1@unimelb.edu.au australian population studies 6 (2) 2022 wilson et al. 15 1. introduction shortly after the release of 2021 census data in mid-2022, several media outlets focused on the “1,043,776 unoccupied dwellings” in australia recorded by the census (abs 2022a). for example, news.com.au announced that “almost one-in-10 australian homes were ‘vacant’ on census night” (smith 2022), while abc news declared “one million homes were unoccupied on census night. how that could help people struggling to find housing” (lohberger 2022). the daily mail australia ran with headline “alarming statistic reveals the depths of australia’s housing crisis as more than one million homes remain empty while renters struggle to find a place to live” (heaton 2022). these headlines might give the impression that the 2021 census counted 1 million perfectly habitable dwellings sitting permanently empty, or at least unused for most of the time. this sits uncomfortably in the context of homelessness (o’donnell 2020), overcrowded housing (dockery et al. 2022), people living in precarious circumstances, and those struggling to find affordable accommodation to live in (gurran et al. 2021). is the interpretation of an unoccupied dwelling as a vacant property ready for a household to move into correct? the abs count of unoccupied dwellings strictly refers to private dwellings where no-one was present on census night (abs 2021a). however, many of these ‘unoccupied’ dwellings have people usually living in them; it is just that all members of the household were absent on census night – perhaps on holiday, visiting relatives, or away for a few days for work. at the same time, the census counted many ‘occupied’ dwellings where there were no usual residents. the people in them on census night were just there temporarily – in holiday rentals or second homes, for example. in the past, the abs recorded reasons for private dwellings not having anyone present on census night. according to the 1986 census, 35% of dwellings were empty due to the usual residents being temporarily absent, 27% were for sale, for rent, newly-built, under repair, or scheduled for demolition, 25% were holiday homes, and 14% were for other/not stated reasons (abs 1988 p. 175). so about 6 in 10 ‘unoccupied’ dwellings were only unoccupied for a short period because the residents were away, or because of the normal operations of the housing market and turnover of housing stock. in later censuses, the reasons for dwellings being unoccupied on census night have not been provided. the classification of populations by usual residence (de jure) or location at a particular point in time (de facto) is a fundamental concept of demography, and discussed early on in introductory demography textbooks (e.g. rowland 2003). it is rarely applied to dwellings – yet it an essential concept when considering the use of dwellings by people. census dwelling data on occupied and unoccupied dwellings is published by the abs on a de facto basis – which means that dwellings are classified exactly as they were on census night, and census night only. this differs from how dwellings are occupied on a usual residence basis. the abs classifies usual residence as a place where someone is living for at least 6 months (abs 2021b). we accept that in the messiness of the real world the concept of usual residence is problematic for some regularly mobile people (charles-edwards 2021), but addressing this issue is beyond the scope of the paper, and for simplicity we assume that usual residence, at least approximately, applies to everyone. the academic literature contains many contributions on temporary movements of population (from the perspective of people) and temporary populations (from the perspective of places) (e.g., bell and 16 wilson et al. australian population studies 6 (2) 2022 ward 2000; panczak et al. 2020). yet the demography of dwellings, and the connections between populations and dwellings remains very undeveloped in the literature, a sentiment expressed by myers (1990) over 30 years ago, which still largely applies today. good conceptual frameworks and high-quality data on the demography of dwellings are essential for understanding the usage of housing, determining housing need, developing well-informed housing policy, and monitoring the impacts of policy change. in this paper, we hope to shed some conceptual and empirical light on the issue of unoccupied dwellings in australia. the aim of the paper is to provide estimates of the number of private dwellings in australia by occupancy on census night and by occupancy on a usual residence basis. the estimates were calculated for the time of the 2021 census, and also the 2016 census for comparison. by doing so, we will be able to answer the question ‘were there really 1 million unoccupied dwellings in australia on census night 2021?’ the paper also attempts to answer the question why the number of dwellings unoccupied on a usual residence basis increased so much between 2016 and 2021. our study focuses only on private dwellings, and therefore excludes non-private dwellings such as boarding schools, staff quarters, student halls of residence, prisons, and nursing homes. following this introduction, we outline a simple dwelling classification framework for describing dwelling occupancy on both a usual residence and a de facto basis. the method for calculating dwelling estimates for this classification is described, along with the census and other input data used. section 3 presents the dwelling estimates by occupancy status for census night 2021 and 2016, while section 4 suggests some reasons for the growth in the number of dwellings unoccupied on a usual residence basis between 2016 and 2021. we conclude by recommending more comprehensive official statistics on dwellings and households in australia. 2. data and methods 2.1 dwelling classification framework a simple four-way classification of private dwellings by occupancy on both a de facto (census night) basis and a usual residence (de jure) basis is shown in table 1. the abs census classification of private dwellings as either occupied or unoccupied depends on whether there are people present in those dwellings on census night or not, and are represented by the totals in the bottom row of the table. the totals in the right-hand column of the table classify dwellings according to whether people are usually resident in dwellings or not, which is probably how census counts of occupied and unoccupied dwellings are interpreted by many. cell a in the table comprises the majority of dwellings, those with usual residents where people (generally, members of the household) were present on census night. cell b refers to dwellings with usual residents but without anyone present on census night, and includes situations such as: • the whole household being away on holiday or visiting relatives, • single person households temporarily away for work, and • individuals from single person households in living-apart-together relationships spending a couple of nights at their partner’s address. australian population studies 6 (2) 2022 wilson et al. 17 table 1: private dwellings by census night and usual residence occupancy dwelling occupancy on census night occupied unoccupied total dwelling occupancy on a usual residence basis occupied a. dwellings with usual residents, and person(s) present on census night† b. dwellings with usual residents, but without anyone present on census night all private dwellings with usual residents unoccupied c. dwellings without usual residents, but person(s) present on census night† d. dwellings without usual residents, and without anyone present on census night all private dwellings without usual residents total all private dwellings with people present on census night† all private dwellings without anyone present on census night† all private dwellings† source: authors. note: † italicised text denotes where dwelling numbers are directly available from the census. cell c represents dwellings without usual residents but where people were present on census night, such as: • holiday rentals, and • second homes. cell d refers to dwellings which were unoccupied on census night and without any usual residents currently living there. these include: • properties for let or for sale where one household has departed and a new household has not yet moved in • newly built dwellings without residents yet • properties where the resident(s) recently died (deceased estates) • properties undergoing renovation or repairs • properties slated for demolition where the previous residents have moved out • holiday rentals with no-one present on census night • second homes with no-one present on census night, and • empty investment properties. these dwellings can essentially be classified into two broad groups: (i) those intended for usual residents but without them at a particular point in time (the first four dot points), and (ii) those not available to usual residents (the last four dot points). 2.2 dwelling estimation method private dwelling numbers for the bottom row of table 1 can be obtained directly from available census data. the number of dwellings with usual residents with one or more persons present on census night (cell a), and the number of dwellings with no usual residents but with people temporarily present on census night (cell c), are also available. the values of cells b and d remain to be estimated. due to data availability, we calculate the value of cell b and then find the value of cell 18 wilson et al. australian population studies 6 (2) 2022 d as the column residual, i.e. the total number of private dwellings without anyone present on census night minus the value of cell b. the number of wholly-absent households – dwellings usually occupied but empty on census night because all residents were temporarily away – was calculated by taking the number of people temporarily away from home from these households and dividing by average household size. the number of people in these households cannot be determined directly, but can be estimated indirectly from census data, erp rebasing data, and some assumptions about non-private dwelling residents. the calculation is: people temporarily away from home from wholly-absent households = people counted away from their home address in australia (1) – non-private dwelling residents counted away from their non-private dwelling in australia (2) + australian residents temporarily overseas (3) – non-private dwelling residents temporarily overseas (4) – private dwelling residents away from households where at least 1 person was present (5). the subtotal from data inputs 1-4 provides an estimate of the number of private dwelling residents temporarily away from home. the last item (5) in the equation is a count of people temporarily away from dwellings which were not unoccupied on census night. the difference between them is an estimate of the number of people away from wholly-absent households. each of the five data inputs to the equation is described in turn. the first of the data items, the number of people present in dwellings in australia on census night who are visiting and away from home, is directly measured by the census. visitors should be included in every household’s census return, and can be identified from the question ‘where does the person usually live?’ by the answer ‘elsewhere in australia’ (abs 2020). second, because we wished to calculate the number of private dwelling residents only, we then subtracted the estimated number of non-private dwelling residents temporarily away. this information is not available from the census, and an assumption was made that the ratio of people counted at home to those counted away amongst private dwelling residents also applies to the portion of the non-private dwelling resident population who can potentially travel (which excludes those with travel limitations, such as people in prisons, detention centres, and nursing homes). third, we added the number of australian residents temporarily overseas on census night. the number temporarily overseas is estimated by the abs for the purposes of rebasing the erp from overseas arrivals and departures statistics (abs 2022b). fourth, we subtracted the estimated number of non-private dwelling residents temporarily overseas. we simply assumed the ratio of all australian residents temporarily overseas to those counted at home applied to the portion of the non-private dwelling resident population with the ability to travel. this is a relatively small number. then, to obtain the required estimate of the number of people temporarily away from home from wholly-absent households, we subtracted those private dwelling residents temporarily away from dwellings where at least one person was present on census night. these people return to households = all private dwelling residents temporarily away from home australian population studies 6 (2) 2022 wilson et al. 19 which were counted as occupied on census night, and are therefore not from wholly-absent households. this count is available from the census, and is derived from information requested towards the end of the household census form. the person completing the form is asked to provide limited information on ‘each person who usually lives in this dwelling but was away on [census night]’ (abs 2020). finally, the number of persons temporarily away from wholly-absent households was divided by average household size to give the estimated number of dwellings where people were usually resident but temporarily away on census night. average household size was calculated from the census variable on the number of persons usually resident in a dwelling. obviously, there are several assumptions inherent in this dwelling estimate, and it should be regarded only as an approximate figure. we assume that: the number of people listed as visitors to dwellings (whose usual address is ‘elsewhere in australia’) is accurately recorded by the census; that our estimates of the non-private dwelling population temporarily away in australia and overseas are reasonable; that the count of persons listed at the end of the census form as temporarily away is reliable (noting there is only room for a maximum of 3 persons to be listed); and that overall average household size is a reasonable approximation for the average size of wholly-absent households. none of these assumptions holds perfectly. 2.3 data to calculate the number of dwellings usually occupied but empty on census night the following 2016 and 2021 census data was obtained from the abs. the variable usual address indicator census night (uaicp) gives counts of people counted ‘at home’ or at another address ‘elsewhere in australia’. the dwelling variable count of persons temporarily absent from household (cpad) lists the number of dwellings where 1, 2, or 3 usual residents were temporarily absent. a cross-tabulation of persons by type of non-private dwelling (npdd) by uaicp allows an estimate to be made of non-private dwelling residents counted ‘at home’ who have the ability to travel. excluded were those in establishments such as prisons, detention centres, and nursing homes. dwelling type (dwtd) lists the number of private dwellings occupied and unoccupied on census night. a cross-tabulation of dwtd by uaicp enables the calculation of the ratio of people in private dwellings counted away to those counted at home. the same census table provides the denominator for the ratio of australian residents temporarily overseas to those at home. the dwelling variable number of persons usually resident in dwelling (nprd) lists the number of dwellings where 1, 2, 3, 4, 5, 6, 7 or 8+ people are usually resident. this permits the calculation of average household size. the variable household composition (hhcd) lists the number of dwellings with only visitors present on census night. census variables are described in the census dictionary (abs 2021c). in addition to census data, estimates of the number of australians temporarily overseas on census nights 2016 and 2021 were supplied directly by the abs. for section 4 of the paper, we also made use of abs private dwelling estimates (abs 2022c) and estimated resident populations (abs 2022d). we also obtained data on the number of short-term rental properties from airdna, the main provider of data and analytics for the short-term rental industry (airdna, no date). the database contains booking and listing information that identifies the number of dwellings either available or 20 wilson et al. australian population studies 6 (2) 2022 reserved for rental in august 2016 and august 2021 (corresponding with the census timing). the data are primarily derived from bookings on digital platforms such as airbnb and vrbo (vacation rentals by owner), and while the coverage of short-term rentals is thought to be high, it is not 100%. 3. dwelling estimates by occupancy status table 2 presents the estimated distribution of private dwellings in australia by occupancy on census night 2021 and by usual residence. the widely reported 1.04 million dwellings unoccupied on census night includes an estimated 186,821 (18%) where the usual residents were only temporarily absent. a further 856,955 dwellings (82%) were unoccupied on census night but did not have any usual residents. the number of dwellings unoccupied on a usual residence basis included these 856,955 dwellings plus an additional 160,883 which contained only visitors on census night. in total, this means there were just over 1 million dwellings unoccupied on a usual residence basis at the time of the 2021 census (row total of 1,017,838). although the total number is similar, it differs in occupancy composition from the total number of dwellings unoccupied on census night, although there is substantial overlap. table 3 shows the equivalent figures in 2016 for comparison. the total number of dwellings unoccupied on census night 2016 was very similar at just over 1 million, though this represents a slightly higher share of the total number of private dwellings at the time (10.5% compared to 9.6% in 2021). the much larger estimated number of dwellings where the usual residents were only temporarily absent, 435,224, reflects the population mobility of pre-covid times, and includes wholly-absent households temporarily away from home both in australia and overseas. the large drop in the number of these table 2: estimated number of private dwellings by census night and usual residence occupancy, australia, 10th august 2021 dwelling occupancy on census night occupied unoccupied total dwelling occupancy on a usual residence basis occupied 9,647,543† 186,821 9,834,364 unoccupied 160,883† 856,955 1,017,838 total 9,808,426† 1,043,776† 10,852,202† source: authors’ calculations based on abs census and population data described in section 2.3. note: † obtained directly from the census; other values are approximate estimates. table 3: estimated number of private dwellings by census night and usual residence occupancy, australia, 9th august 2016 dwelling occupancy on census night occupied unoccupied total dwelling occupancy on a usual residence basis occupied 8,713,202† 435,224 9,148,426 unoccupied 148,418† 604,655 753,073 total 8,861,620† 1,039,879† 9,901,499† source: authors’ calculations based on abs census and population data described in section 2.3. note: † obtained directly from the census; other values are approximate estimates. australian population studies 6 (2) 2022 wilson et al. 21 dwellings in 2021 to 186,821 is not surprising given the mobility restrictions imposed by state/territory governments in response to the covid pandemic and the federal government’s closure of the international border. another notable difference between 2016 and 2021 can be seen in the number of private dwellings unoccupied on both census night and on a usual residence basis. these dwelling numbers increased from an estimated 604,655 in 2016 to 856,955 in 2021 (an increase of 42%). by comparison, the dwelling stock overall increased by 9.6% over the five year period 2016 to 2021. the total number of dwellings unoccupied on a usual residence basis rose from about ¾ million in 2016 to just over a million in 2021 (an increase of 35%). if this total had increased in line with overall dwelling growth between 2016 and 2021, there would have been about 825,000 dwellings unoccupied on a usual residence basis in 2021. 4. why were there so many unoccupied dwellings? why were there about 1 million dwellings unoccupied on a usual residence basis in 2021? unfortunately, given the available data, it is difficult to go much beyond the disaggregation shown in table 2, which shows that out of an estimated 1,017,838 dwellings unoccupied on a usual residence basis, 160,883 (16%) were occupied with visitors on census night, while 856,955 (84%) were empty. a large proportion of the latter category of dwellings were probably empty on census night for the same set of reasons recorded in the 1986 census. these include housing market or turnover reasons (for sale, for let, deceased estates, under renovation, or newly-built), while others may have been empty second homes, unlet holiday rentals, or empty investment properties. however, it is possible to consider why the number of dwellings unoccupied on a usual residence basis increased so much from 2016. two potential proximate (and connected) reasons that we briefly explore here are: net additions to the dwelling stock outpacing household growth, and an increase in the number of holiday rental properties and second homes over the last few years. did dwelling construction outpace household growth? there are no annual household estimates produced by the abs. however, approximate household estimates can be easily calculated for 2016 and 2021. the mid-year estimated resident population in private dwellings for these two years can be calculated by multiplying the census share of the population usually resident in private dwellings by the total erp (to give the private household erp). dividing this by average household size results in the estimated number of households. the census share of people in private dwellings and average household size can then be interpolated for years between 2016 and 2021, and household estimates then created for these years. net dwelling growth was calculated from private dwelling estimates in the new abs publication estimated dwelling stock (abs 2022c). the results of these estimations are shown in figure 1. net dwelling growth exceeded household growth for each of the five years shown. this was especially the case in 2020-21 when household growth was subdued due to slow population growth from the international border closure and negative net overseas migration (abs 2022d). dwelling construction declined only moderately in 2019-20 and 2020-21, likely assisted by various housing market stimulus measures. these include the federal government’s homebuilder program, which provided households below a certain income threshold with a grant of $25,000 to spend on building a new house, and several state and territory 22 wilson et al. australian population studies 6 (2) 2022 government incentives (leishman et al. 2022). it seems plausible, therefore, that part of the explanation for the increase in the number of unoccupied dwellings on a usual residence basis lies in net dwelling growth outpacing household growth. however, in the absence of comprehensive annual official statistics on households and dwellings, we cannot be sure of the extent of this effect. figure 1: estimated net dwelling and household growth, australia, 2016-17 to 2020-21 source: authors’ calculations based on abs erp and dwelling data another possibility is that there was a large increase in the number of holiday rental properties and second homes between 2016 and 2021. was this the case? data from airdna reveals a much greater number of short-term rental properties in australia in august 2021 than in 2016. we focus on the ‘entire home’ category in the airdna data because it refers to a self-contained property with its own entrance, and closely resembles the abs definition of a private dwelling. we excluded the airdna count of individual rooms available for rent because it is not possible to determine if these are in dwellings which are unoccupied on a usual residence basis or not. it is important to emphasise that the data we present here only provides a partial picture of the trend in the number of short-term holiday rentals and only includes second homes if they are rented out for some of the year. figure 2 shows the number of short-term rentals of ‘entire homes’ in the month of august in 2016 and 2021. it shows both active and professional listings. an ‘active’ listing is defined as one which is available for booking (or reserved) for at least 1 day in a month, while a professional listing is a particular type of active listing available for almost all of a month. the total number of active shortterm rental listings in august 2021 (120,500) was about 2½ times the number in 2016, while the number of professional listings in 2021 (90,400) increased by a similar factor since 2016. thus, the growth in short-term rental properties between 2016 and 2021 could provide part of the answer to the question about the increase in the number of dwellings without usual residents. however, the situation is complicated by the possibility that many holiday homes, not previously listed on shortterm rental websites, are now let out for parts of the year. these dwellings would not have added to the stock of dwellings unoccupied on a usual residence basis. australian population studies 6 (2) 2022 wilson et al. 23 figure 2: number of short-term rental listings of entire homes, australia, at the time of the 2016 and 2021 censuses source: airdna data. notes: an ‘active’ listing is available (or reserved) for at least 1 day in a month; a ‘professional’ listing is available (or reserved) for all but a maximum of 5 days in a month. 5. conclusions this paper has presented estimates of the number of private dwellings in australia cross-classified by occupancy on census night and occupancy by usual residence in 2021 (and also 2016). the answer to the question ‘were there really 1 million unoccupied dwellings in australia on census night 2021?’ is yes – and no. yes, there were about 1 million dwellings with no-one present in them on census night, and yes there were also about 1 million dwellings with no usual residents – though not the exact same 1 million. consequently, the characteristics of dwellings unoccupied on census night as revealed by census data, including geographical patterns, will not be the same as those unoccupied on a usual residence basis. but the answer to the question is also no in the sense that many of the 1 million dwellings unoccupied on a usual residence basis were not available for people to move into. many of these dwellings will have been only temporarily without usual residents (e.g. newly-built, or vacant for a couple of weeks following the sale of a property) or they were second homes, holiday rentals, or other dwellings which do not contain usual residents. the distribution of dwellings by occupancy status was substantially affected by covid-related restrictions and lockdowns, as a comparison of tables 2 and 3 makes clear. mobility was restricted and very few australians were temporarily overseas in 2021. not surprisingly, a greater proportion of dwellings with usual residents had people present in them on census night in 2021 (98%) than in 2016 (95%). the number of dwellings without usual residents but with people present on census night 2021 (cell c) was still reasonably high (and greater in number than in 2016), suggesting movement to holiday homes during lockdowns. in future research we aim to investigate the geographical pattern of dwelling occupancy by usual residence. we presented evidence which provides some support for the suggestion that the growth in dwellings unoccupied on a usual residence basis between 2016 and 2021 was due to net additions to the dwelling stock exceeding household growth, and growth in the number of short-term rental properties. however, the evidence is broad brush and not conclusive due to data limitations, and other influences may also have contributed to the growth in the number of dwellings without usual residents. much remains unknown. we were not able to examine trends in the number of second 24 wilson et al. australian population studies 6 (2) 2022 homes which are not rented out, and the extent to which there are properties are being left vacant over the long-term for other reasons. it is difficult to determine the number of dwellings only temporarily unoccupied on a usual residence basis due to them being newly-built, temporarily between tenants or owners, or under renovation. and there are commercial property datasets which may also provide some of the answers, but these are not in the public domain. the paper has also demonstrated the complexity of census dwelling data, along with some of its strengths and limitations. while census data on population is available on both a de facto (location on census night) and a de jure (place of usual residence) basis, census data on dwellings is presented on a de facto basis. some census variables, do in fact, include some information on persons normally resident in a dwelling but not present on census night, such as number of persons usually resident in dwelling (nprd), but most do not. it is essential for data users to be aware of the scope of each census dwelling and household variable. for example, the scope of census data on relationship in household (rlhp) is ‘persons present in the household on census night’ (abs 2021d). but actually, relationships recorded in this variable (e.g. lone parent, in a registered marriage, brother/sister) are only given for people usually resident in the dwelling and who were present there on census night (cell a in table 1). people present in the dwelling, but not usually resident, are classified as visitors in rlhp. household relationship data covers only about 90% of the census usually resident population, and in some local areas, much less. if a census dataset of dwellings on a usual residence basis was available, then it would provide a more comprehensive picture of dwelling occupancy in australia. but there is a case for further detail in dwelling and household data. it would be very beneficial to have regular, timely, accurate, and freely available statistics on australia’s dwelling stock and its occupancy. the recently-announced national housing supply and affordability council, which will focus on “increasing housing supply and improving housing affordability” (collins, 2022), would greatly benefit from such statistics. to support the work of the council and housing policy more generally, the federal government should fund a program of annual dwelling and household statistics in the abs. importantly, these statistics should distinguish between dwellings without usual residents (i.e. dwellings used as second homes, short-term rentals, or not occupied for other reasons) and those intended for usual residents (most dwellings). ideally, the data would also provide estimates of the number of dwellings temporarily unoccupied on a usual residence basis by broad reason for being without usual residents at the reference date of the data (newly-built, about to be demolished, etc.). if such comprehensive and regular dwelling data was available, then we would know whether dwelling occupancy was currently returning to ‘normal’ after covid restrictions and lockdowns. doubtless, the creation of detailed official statistics on dwellings and households would be a substantial, and challenging, undertaking. but they would provide a strong evidence base for housing policy, facilitate good-quality research to inform policy, and enable myths and assumptions made about the nation’s housing market and dwelling stock to be countered. key messages • it is useful to distinguish between the occupancy of dwellings on a de facto basis (whether people were present on census night) and on usual residence basis (whether people are usually resident there). australian population studies 6 (2) 2022 wilson et al. 25 • the 2021 census recorded just over 1 million dwellings unoccupied on census night. our approximate estimates suggest a similar number of dwellings unoccupied on a usual residence basis – but not the exact same 1 million. • the increase in dwellings unoccupied on a usual residence basis between 2016 and 2021 is probably due, in part, by net dwelling growth exceeding household growth during the covid pandemic and the increase in the number of short-term rental properties. • greater clarity around census dwelling data is needed. in addition, housing policy and research would benefit from a series of annual dwelling and household estimates from the abs. acknowledgments author tw was supported by the australian research council centre of excellence in population ageing research (project number ce1101029). authors fz and ts were supported by the australian research council discovery projects scheme (dp200100506). the authors are most grateful to the australian bureau of statistics for providing 2021 census data and estimates of residents temporarily overseas on census night. the reviewers and abs experts also provided very helpful comments and suggestions on an earlier draft of the paper. references abs (1988). census 86 – cross-classified characteristics of persons and dwellings. census of population and housing, 30 june 1986. catalogue no. 2498.0. canberra, abs. https://www.abs.gov.au/ausstats/abs@.nsf/detailspage/2102.01986?opendocument abs (2020). 2021 census topics and data release plan. https://www.abs.gov.au/statistics/research/2021census-topics-and-data-release-plan. accessed 2 august 2022 abs (2021a). dwelling type (dwtd). census of population and housing: census dictionary. https://www.abs.gov.au/census/guide-census-data/census-dictionary/2021/variablestopic/housing/dwelling-type-dwtd. accessed 11 august 2022. abs (2021b). place of usual residence (purp). census of population and housing: census dictionary. https://www.abs.gov.au/census/guide-census-data/census-dictionary/2021/variablestopic/location/place-usual-residence-purp. accessed 11 august 2022. abs (2021c). census of population and housing: census dictionary. https://www.abs.gov.au/census/guide-census-data/census-dictionary/2021. accessed 11 august 2022. abs (2022a). housing: census. information on housing type and housing costs. https://www.abs.gov.au/statistics/people/housing/housing-census/latest-release. accessed 2 august 2022. abs (2022b). methodology used in rebased population estimates, june 2021. https://www.abs.gov.au/statistics/detailed-methodology-information/informationpapers/methodology-used-rebased-population-estimates-june-2021. accessed 11 august 2022. abs (2022c). estimated dwelling stock. https://www.abs.gov.au/statistics/industry/building-andconstruction/estimated-dwelling-stock/latest-release. accessed 29 november 2022. abs (2022d). national, state and territory population. https://www.abs.gov.au/statistics/people/population/national-state-and-territorypopulation/dec-2021. accessed 11 august 2022. airdna (no date). dataset on short-term rentals. https://www.airdna.co/about. bell, m., & ward, g. (2000). comparing temporary mobility with permanent migration. tourism geographies, 2(1), 87–107. https://doi.org/10.1080/146166800363466 https://www.abs.gov.au/ausstats/abs@.nsf/detailspage/2102.01986?opendocument https://www.abs.gov.au/statistics/research/2021-census-topics-and-data-release-plan https://www.abs.gov.au/statistics/research/2021-census-topics-and-data-release-plan https://www.abs.gov.au/census/guide-census-data/census-dictionary/2021/variables-topic/housing/dwelling-type-dwtd https://www.abs.gov.au/census/guide-census-data/census-dictionary/2021/variables-topic/housing/dwelling-type-dwtd https://www.abs.gov.au/census/guide-census-data/census-dictionary/2021/variables-topic/location/place-usual-residence-purp https://www.abs.gov.au/census/guide-census-data/census-dictionary/2021/variables-topic/location/place-usual-residence-purp https://www.abs.gov.au/census/guide-census-data/census-dictionary/2021 https://www.abs.gov.au/statistics/people/housing/housing-census/latest-release https://www.abs.gov.au/statistics/detailed-methodology-information/information-papers/methodology-used-rebased-population-estimates-june-2021 https://www.abs.gov.au/statistics/detailed-methodology-information/information-papers/methodology-used-rebased-population-estimates-june-2021 https://www.abs.gov.au/statistics/industry/building-and-construction/estimated-dwelling-stock/latest-release https://www.abs.gov.au/statistics/industry/building-and-construction/estimated-dwelling-stock/latest-release https://www.abs.gov.au/statistics/people/population/national-state-and-territory-population/dec-2021 https://www.abs.gov.au/statistics/people/population/national-state-and-territory-population/dec-2021 https://www.airdna.co/about https://doi.org/10.1080/146166800363466 26 wilson et al. australian population studies 6 (2) 2022 charles-edwards, e. (2021). living in more than one place: capturing dual-local lives in the 2026 census. australian population studies, 5(1), 3-8. https://doi.org/10.37970/aps.v5i1.75 collins, j. (2022). national homelessness conference 2022. https://ministers.dss.gov.au/speeches/8756. accessed 26 august 2022. dockery, a. m., moskos. m., isherwood, l., & harris, m. (2022). how many in a crowd? assessing overcrowding measures in australian housing. ahuri final report no. 382, australian housing and urban research institute limited, melbourne. https://doi.org/10.18408/ahuri8123401. gurran, n., pill, m., & maalsen, s. (2021). hidden homes? uncovering sydney’s informal housing market. urban studies, 58(8), 1712–1731. https://doi.org/10.1177/0042098020915822 heaton, t. (2022). alarming statistic reveals the depths of australia's housing crisis as more than one million homes remain empty while renters struggle to find a place to live. daily mail australia. https://www.dailymail.co.uk/news/article-10960139/one-million-vacant-homes-australia-2021census.html. accessed 2 august 2022. leishman, c., aminpour, f., baker, e., beer, a., crowe, a., goodall, z., horton, e., jacobs, k., lester, l., torchia, s., maclennan, d., martin, c., nash, m., pawson, h., rowley, s., stone, w., & vifor, j. r. (2022). australia’s covid-19 pandemic housing policy responses. ahuri final report no. 376, australian housing and urban research institute limited, melbourne, https://doi.org/10.18408/ahuri3227801. lohberger, l. (2022). one million homes were unoccupied on census night. how that could help people struggling to find housing. https://www.abc.net.au/news/2022-06-29/census-finds-1-millionempty-houses-amid-affordability-crisis/101190794. accessed 2 august 2022 myers, d. (ed.) (1990). housing demography: linking demographic structure and housing markets. university of wisconsin press. o’donnell, c. (2020). estimating annual rates of homelessness. demographic research, 43(1), 1-34. https://doi.org/10.4054/demres.2020.43.1 panczak, r., charles-edwards, e., & corcoran, j. (2020). estimating temporary populations: a systematic review of the empirical literature. humanities and social sciences communications, 6, 87. https://doi.org/10.1057/s41599-020-0455-y rowland, d. t. (2003). demographic methods and concepts. oxford university press. smith, r. (2022). almost one-in-10 australian homes were ‘vacant’ on census night. https://www.news.com.au/finance/money/investing/almost-onein10-australian-homes-werevacant-on-census-night/news-story/e5da221268ad7331c2882a4feefd0db1. accessed 2 august 2022. https://doi.org/10.37970/aps.v5i1.75 https://ministers.dss.gov.au/speeches/8756 https://doi.org/10.18408/ahuri8123401 https://doi.org/10.1177/0042098020915822 https://www.dailymail.co.uk/news/article-10960139/one-million-vacant-homes-australia-2021-census.html https://www.dailymail.co.uk/news/article-10960139/one-million-vacant-homes-australia-2021-census.html https://doi.org/10.18408/ahuri3227801 https://www.abc.net.au/news/2022-06-29/census-finds-1-million-empty-houses-amid-affordability-crisis/101190794 https://www.abc.net.au/news/2022-06-29/census-finds-1-million-empty-houses-amid-affordability-crisis/101190794 https://doi.org/10.4054/demres.2020.43.1 https://doi.org/10.1057/s41599-020-0455-y https://www.news.com.au/finance/money/investing/almost-onein10-australian-homes-were-vacant-on-census-night/news-story/e5da221268ad7331c2882a4feefd0db1 https://www.news.com.au/finance/money/investing/almost-onein10-australian-homes-were-vacant-on-census-night/news-story/e5da221268ad7331c2882a4feefd0db1 