Trends in Female Education in Low- and Middle-Income Countries: Coherence across Data Sources Trends in Female Education in Low- and Middle-Income Countries: Coherence across Data Sources Kristen Jeffers, Albert Esteve Abstract: Educational expansion and the closing of gender gaps in education are key objectives in national and international policy agendas. Monitoring progress towards these goals requires comparable data across countries and over time. The availability of international census and survey microdata allows for cross- national comparisons of education participation and completion. However, we lack systematic analyses of how trends vary across data sources and of the extent to which these data sources offer a consistent account of progress in education. In this paper, we examine coherence in estimates of educational attainment among women aged 25 to 29 in 75 countries across the three main repositories of international population microdata: IPUMS International, the Demographic and Health Surveys (DHS) and the Multiple Indicator Cluster Surveys (MICS). Coherence analysis of 535 census and survey observations from 1960 to 2017 shows high levels of consistency overall but also identifi es observations misaligned with trends. Results provide practical information to the research community about the validity of comparative investigations using three important data sources for demographic studies. The data also serve as benchmarks for assessing the quality of education information obtained in data sources not included in our analysis and the trend alignment of future estimates. Keywords: Educational expansion · Census microdata · DHS · MICS · Data coherence 1 Introduction Education is widely recognised as a vehicle for individual agency, social change and economic growth. It is also considered an important policy instrument for development in many low- and middle-income countries (LMICs), where enrolment in primary and secondary schooling is not universal and gender disparities in education remain. As a result, national and international policy agendas include as Comparative Population Studies Vol. 47 (2022): 349-388 (Date of release: 08.08.2022) Federal Institute for Population Research 2022 URL: www.comparativepopulationstudies.de DOI: https://doi.org/10.12765/CPoS-2022-14 URN: urn:nbn:de:bib-cpos-2022-14en6 • Kristen Jeffers, Albert Esteve350 key objectives expanded access to schooling and the closing of the gender gaps in education (see United Nations 2013; UNICEF 2019; UNESCO 2016, 2018). Monitoring progress towards these goals requires comparable data across countries and over time. The availability of international census and survey microdata allows for cross- national comparative assessments of education participation and completion. However, we lack systematic analyses of how trends in education vary across these data sources and of the extent to which these trends offer a consistent account of progress in education in recent decades. Do the multiple rounds of census and survey microdata available for LMICs provide a coherent picture of trends in educational expansion? Are there systematic differences across data sources? Which specifi c censuses or surveys stand apart from the trend presented by other census and survey observations in the country? To answer these questions, we pooled individual-level data from 535 censuses and surveys from 75 low- and middle-income countries to monitor trends in educational attainment and examine coherence across data sources. The data cover 1960 to 2017 and come from three sources: population censuses from the IPUMS database, the Demographic and Health Surveys (DHS) and the Multiple Indicator Cluster Surveys (MICS).1 All censuses available from IPUMS as well as all DHS and MICS surveys collect information on school attendance and educational attainment. We examine trends in educational attainment among females aged 25 to 29. While gender gaps in education have reversed in most high- and middle-income countries, women trail men in school enrolment and completion in many developing countries, particularly in sub-Saharan Africa and South Asia (Ilie/Rose 2016; Kebede et al. 2019). Accordingly, international policy agendas include specifi c targets to address gender disparities in education. We fi nd strong coherence across data sources overall with heterogeneity by data source and region. Our results provide practical information to the research community about the validity of comparative investigations using the data sources we examine. Our analysis focuses on young women, but sensitivity analyses examining trends in male educational attainment yield similar results and conclusions. 2 Background The contribution of basic education to economic growth is widely recognised (Schultz 1961; Becker 1994; Barro/Lee 1994; Cohen/Soto 2007). Its importance is not, however, confi ned to economic progress. Education has intrinsic positive effects for people and societies. Education imparts knowledge and skills that provide access to better paying jobs, alleviating poverty and increasing the availability of cultural and economic resources for households (Becker 1994). Highly educated 1 To access international census data from IPUMS, visit international.ipums.org. DHS data are available for download from the DHS Program, dhsprogram.com, and IPUMS-DHS, idhsdata. org. MICS data are available for download at mics.unicef.org. Trends in Female Education in Low- and Middle-Income Countries • 351 individuals tend to have better health and live longer lives (Hannum/Buchmann 2005; Kc/Lentzner 2010; Lutz et al. 2014; Masquelier/Garbero 2016). The children of highly educated parents are also more likely to have better health and cognitive outcomes (Caldwell 1981; Cochrane et al. 1982; Lutz et al. 2014; Bicego/Boerma 1993; Rosenzweig/Wolpin 1994; Mellington/Cameron 1999; Wang 2003; Schady 2011; Abuya et al. 2011; Vikram et al. 2012; Grépin/Bharadwaj 2015). Due to the link between female education and demographic outcomes, educational expansion has the potential to signifi cantly infl uence population growth (Lutz/Skirbekk 2014). Based on this evidence, policymakers and development practitioners view access to and completion of education as key policy instruments for development in low- and middle-income countries. The United Nations 2030 Agenda for Sustainable Development includes educational expansion and gender parity in schooling as primary goals, challenging LMICs to ensure universal and equal access at all levels of education. Many of these countries trail high-income countries in quantitative and qualitative progress in education. Enrolment in primary and secondary education is far from universal and college graduates represent less than fi ve percent of the population in much of sub-Saharan Africa (UNESCO Institute for Statistics 2019). Moreover, the expansion of primary, secondary and tertiary education has not been gender-neutral (Grant/Behrman 2010; Dorius/Firebaugh 2010; Dorius 2013; UNESCO 2019). During the Millennium Development Goals (MDGs) era, countries with better performance monitoring made better progress towards targets (Jacob 2017). To track progress towards new education targets, accurate measures of school participation and completion are fundamental. Policymakers and researchers worldwide have traditionally relied on offi cial reports and statistics from international agencies and some scholarly databases (Barro/Lee 2013, for example) to obtain macro level indicators on educational attainment. These statistics were compiled from censuses and surveys, resources for which microdata were rarely available to researchers. Only recently have researchers gained access to census and survey microdata from LMICs upon which to build new and large comparative analysis. These microdata come from three main sources: the IPUMS International database for population censuses and household surveys (Minnesota Population Center 2019), the Demographic Health Surveys Program (ICF 2004-2017) and the Multiple Indicators Cluster Surveys (UNICEF 2020). Among these sources, censuses have served as the underlying source for global databases on educational attainment like those maintained by the United Nations. They also provide the benchmark observations for modelled estimates and projections of educational attainment produced and used by the scholarly community (Barro/Lee 2013; Bauer et al. 2012; Cohen/Soto 2007; Kc et al. 2010; Lutz et al. 2014; De la Fuente/Doménech 2012; Jordá/Alonso 2017). For many countries, especially LMICs, population censuses represent the only available source for educational data. Compared to surveys, censuses offer some advantages – most importantly universal coverage of the population – which contribute to reliable and consistent trends over time. However, in the best of cases, censuses are conducted every ten years. In low-resource and/or confl ict settings, more than ten years may • Kristen Jeffers, Albert Esteve352 elapse between censuses. Furthermore, access to census microdata has been limited until recent years. The IPUMS International project has simplifi ed access to census microdata by collaborating with National Statistical Offi ces to compile, integrate and harmonise representative samples of census microdata from around the world. At the time of writing, harmonised microdata from more than 500 censuses and surveys and more than 100 countries were available to researchers free of charge through IPUMS International. Nearly every country in South and Central America disseminates microdata through IPUMS. Twenty-seven of 54 African countries disseminate data through IPUMS. Twenty-three of 60 low- and middle-income countries in Asia, the Middle East and the Pacifi c disseminate through IPUMS. Prior to the publication of international census microdata through IPUMS international, cross-national research on the expansion and implications of education in LMICs mostly relied on surveys. Compared to censuses, surveys provide more timely data and cover a broader range of topics in greater depth. Data from two global survey programs are most frequently used to monitor development progress in LMICs: the Demographic and Health Surveys (DHS) and the Multiple Indicator Cluster Surveys (MICS). The Demographic and Health Surveys are nationally representative household surveys that collect data on a number of topics related to population, health and nutrition (ICF 2019). DHS surveys are conducted about every 5 years. Since 1984, more than 400 DHS surveys have been completed in over 90 countries in Africa, Latin America, Asia, Oceania and Eastern Europe. Data on literacy, school attendance and educational attainment are collected for all household members in every DHS survey. DHS surveys primarily target women in reproductive years and they have been the main source for large cross-national studies in trends in women’s education and their infl uence on family transitions and health (e.g. Castro Martín 1995; Lloyd 2005; Grant/Behrman 2010). A complement to DHS, the UNICEF MICS Surveys provide data on the wellbeing of women and children in LMICs. More than 300 surveys have been fi elded in 116 countries covering the period 1993 to the present (UNICEF 2020). Most MICS surveys are also nationally representative and collect education information similar to that collected in DHS surveys. DHS and MICS are important sources for measuring progress towards the Sustainable Development Goals (SDGs). Likewise, the data are used extensively by social scientists and policy and health researchers to study a wide range of topics related to child health and wellbeing and human development (e.g. Pace et al. 2019; Kang et al. 2018; Jeong et al. 2018). DHS and MICS surveys use similar multi-stage cluster sampling strategies. Primary sampling units (PSU) typically correspond to census enumeration areas and are stratifi ed by administrative regions and urban and rural areas before selection. A complete household listing is conducted in each selected PSU and households are selected for the survey by equal probability systematic sampling (ICF International 2012). This type of probability sampling relies on updated and reliable sampling frames. Censuses are considered the most suitable sampling frames for DHS and MICS surveys. For this reason, we expect strong coherence between census-based estimates and survey-based estimates. When a census has not been conducted recently, alternative sampling frames such as electoral rosters are used. Trends in Female Education in Low- and Middle-Income Countries • 353 Used together, IPUMS, DHS and MICS constitute a vast repository of individual- level data for cross-national research on population dynamics, development and health in LMICs. Despite the nearly limitless potential of these data, cross-national studies combining census, DHS and MICS microdata are still rare because microdata for a critical mass of countries only became available in the last decade. This paper – a fi rst attempt in this direction – examines the feasibility of combining these sources to examine trends in and implications of educational expansion. While the topical coverage and policy purposes of these data collection instruments differ, censuses and DHS and MICS surveys share basic features that make them suitable for large comparative studies. The collection of information on basic educational attainment is one of these common features. In theory, comparing educational attainment across countries and data sources should be straightforward. Most education systems are organised around three distinct levels: primary, secondary and tertiary. Systems within regions and among countries with shared colonial histories are usually quite similar. Moreover, the International Standard Classifi cation of Education (ISCED) provides clear guidelines for measuring educational attainment in censuses, surveys and population registers. In practice, accurate comparisons across countries are diffi cult because the defi nition of educational levels varies signifi cantly across countries. Scholars considering education at the global level have grappled with comparability across data sources. In their work to produce population projections disaggregated by level of education, researchers at the Wittgenstein Centre for Demography and Global Human Capital have relied on microdata from censuses, DHS, MICS and other household surveys to construct global datasets on educational attainment (Bauer et al. 2012; Goujon et al. 2016; Lutz et al. 2014, 2018; Speringer et al. 2019). Bauer et al. (2012) and Speringer and colleagues (2019) document the comparability issues and harmonisation challenges encountered in the construction of the base-year datasets used across these education and population projections, including discrepancies between national education systems and ISCED categories, changes over time in national education systems, inconsistencies across data sources in the treatment of complete versus incomplete levels, age-heaping in survey data and a lack of detail at the lowest and highest levels of education in developed and developing countries, respectively. Results of validation exercises confi rm that different data sources lead to different educational compositions and that census or register data are generally the most reliable. The few examples provided of discrepancies between DHS surveys and censuses and intra-country inconsistencies across DHS surveys support the need for further investigation of coherence across these important sources of population data. This paper seeks to fi ll this gap by providing a systematic evaluation of trends in educational attainment across the three primary sources of demographic data in low- and middle-income countries: population censuses, Demographic and Health Surveys (DHS) and UNICEF Multiple Cluster Indicator Surveys (MICS). We aim to identify coherence and inconsistencies in the empirical measurement of education within and across data sources and to contribute to a better understanding of the validity of a critical independent variable in development research. • Kristen Jeffers, Albert Esteve354 3 Data and Methodology We use individual-level data from 210 IPUMS census samples,2 219 DHS surveys and 106 MICS surveys to assess coherence in the measurement of education across data sources in 75 low- and middle-income countries. For a robust picture of trends over time, we include in the analysis all countries with four or more censuses and/or surveys across two or more data sources. The data were collected during the period 1960 to 2017 and cover more than half of the 138 countries designated as low or middle income by the World Bank. A plurality of countries studied are represented in all three data sources: IPUMS, DHS and MICS (Table 1). IPUMS census samples vary in size but typically include 10 percent of the population. Sample sizes for DHS and MICS surveys typically cover 20,000 to 50,000 households. See Table A1 in the appendix for analysis sample sizes. DHS and MICS data used in the analysis were collected in household questionnaires and come from corresponding household-member microdata fi les. DHS and MICS surveys collect basic demographic information on all usual residents and visitors in surveyed households via the household questionnaire. From these census and survey samples, we select women aged 25 to 29. In many low- and middle-income countries, women trail men in access to and completion of secondary education. Development agendas focus on confronting barriers to girls’ schooling and require accurate information on female educational attainment. Women in this age range represent the population that recently completed formal education, more accurately refl ecting the educational context in the country at the time of the survey compared to older cohorts. Coherence analysis based on men yield similar results. Most IPUMS census samples and DHS and MICS surveys include retrospective questions on years of schooling and highest level of education attended and/or completed. Our analysis focuses on the population completing secondary and higher education. Existing literature suggests secondary education is an important threshold for individual outcomes and development (Caldwell/McDonald 1982; Ainsworth et al. 1996; Desai/Alva 1998; Abuya et al. 2012; Makoka/Masibo 2015). 2 Certain microdata samples disseminated by IPUMS come from large-scale household surveys rather than censuses. See appendix for more information. Tab. 1: Number of analysis countries by data source and world region Africa Asia Europe Latin Americ /Caribbean Total DHS, MICS, IPUMS 17 5 1 3 26 IPUMS, DHS 7 8 – 7 22 IPUMS, MICS 2 4 1 9 16 DHS, MICS 8 2 – 1 11 Total 34 19 2 20 75 Source: own design Trends in Female Education in Low- and Middle-Income Countries • 355 We generate a dichotomous educational attainment variable that distinguishes women who have completed secondary education or higher from those who have not. For census samples, we recode the IPUMS EDATTAIN variable. DHS survey data were accessed directly from the DHS Program and from IPUMS DHS. For DHS surveys, we use the educational attainment summary variable constructed by the DHS Program during data processing (hv109 or EDSUMM in IPUMS). The MICS datasets do not include a summary educational attainment variable. We construct a summary measure of educational attainment for MICS samples according to each country’s educational system using variables indicating the highest education level attended and highest grade completed at level. We calculate the percentage of women aged 25 to 29 who have completed secondary education for each census or survey, yielding 535 country-year observations in total. The mean number of census and survey observations among countries examined is 7.7. We use polynomial regression to fi t trends over time in educational attainment within each country. More specifi cally, we have specifi ed a second-degree polynomial (quadratic) regression model to account for the curvilinear nature of the relationship between female educational attainment and time. Previous studies of global educational expansion during the time period of interest identifi ed a sigmoidal pattern in most countries, with initially slow growth, a phase of rapid expansion and then a slow approach towards universal participation (Barakat/Durham 2014). We explored both quadratic and cubic specifi cations, but found the quadratic model to provide the most parsimonious fi t to the data for the developing countries included in our analysis, many of which have not yet reached the fi nal phase of educational expansion. In the model specifi ed above, P is the proportion of women aged 25-29 completing secondary or higher education. Each observation of P is weighted by sample size. By doing this, we give more importance to census observation over DHS and MICS. Census microdata samples are usually drawn from complete-count census databases. Results derived from census samples are therefore extremely close to those derived from complete-count databases. We use the parameters estimated by the model to generate predicted values of P. We then compare observed and predicted values, calculating and summarising absolute differences between the two and classifying observations based on absolute deviance from the predicted trend. We disregarded the use of confi dence intervals as samples vary greatly in size and large samples usually yield very narrow confi dence intervals such that even a tiny deviation from that value represents an outlier. On the contrary, small sample sizes generate wider confi dence intervals and relatively large deviations may fall within the interval. ln 1 • Kristen Jeffers, Albert Esteve356 4 Results 4.1 Observed trends in educational attainment Figure 1 presents, by country and world region, estimates of the percent of women aged 25 to 29 completing upper secondary education or higher for the 75 countries included in our analysis (see also Table A1 in the appendix). Each line represents a country, showing change over time based on multiple data sources. The colour and shape of the symbol marking each point estimate indicates the data source: Census, DHS or MICS. This fi gure corroborates trends well-known to researchers and policymakers. Over the last several decades, there has been a general upward trend in all world regions in the proportion of young women completing at least secondary education in LMICs. The percentage of women aged 25 to 29 completing secondary education or higher has increased between the earliest census or survey observation and the most recent in nearly every country examined. However, levels of secondary or higher completion vary widely across countries and regions. Educational attainment is lowest in Africa. Less than 10 percent of women aged 25 to 29 had completed secondary education according to the most recent census or survey in Burkina Faso, Burundi, Benin, Central African Republic, Chad, Ethiopia, Mali, Mozambique, Niger, Senegal and Togo. Levels of secondary completion are much higher in Latin America. Since 2010, in two-thirds of the Latin American countries studied, more than 50 percent of women complete secondary or higher. Levels of completion were highest in Cuba (2014) and Trinidad and Tobago (2011), where more than 80 percent of young women complete secondary or higher, and lowest in Haiti (2016) and Guatemala (2014), where less than 25 percent of young women complete secondary or higher. Recent levels of secondary completion vary within Asia, where the most heterogeneity is observed. Female educational attainment remains low in several of the Southern and Southeast Asian countries studied according to recent surveys. Less than 30 percent of women aged 25 to 29 had completed secondary or higher in Cambodia (2014), Bangladesh (2014), Laos (2017), India (2015) and Pakistan (2012). In contrast, educational attainment is high in Central Asia and Eastern Europe: more than 95 percent of women aged 25 to 29 complete secondary education or higher in Armenia, Belarus, Kazakhstan, Kyrgyz Republic and Ukraine according to recent surveys. The rate of increase in secondary completion in LMICs during the last several decades also varies across countries and regions, as refl ected in the variation in the slope of country trend lines in Figure 1. Change has been slowest in Africa, where the median annual increase in the percentage of women aged 25 to 29 completing secondary or higher among the countries examined is 0.43 percentage points. By comparison, the median annual increase in the percentage of women aged 25 to 29 completing secondary or higher is 0.66 percentage points among Asian countries examined is 1.1 percentage points among Latin American countries studied. The dominant trend of increasing female educational attainment over time and the relative levels of educational attainment we observe in recent censuses and surveys refl ect what we understand about access to education across and Trends in Female Education in Low- and Middle-Income Countries • 357 F ig . 1: P e rc e n t o f w o m e n a g e d 2 5 -2 9 c o m p le ti n g s e co n d a ry e d u ca ti o n o r h ig h e r b y y e a r, c o u n tr y, d a ta s o u rc e a n d w o rl d r e g io n S o u rc e : o w n d e si g n • Kristen Jeffers, Albert Esteve358 within world regions. A closer look at country-specifi c patterns, however, reveals inconsistencies within and across data sources, even among observations that are only a few years apart. We do not expect that educational expansion is linear in all countries. This expectation is refl ected in the data. In some countries, secondary or higher completion has expanded more rapidly in recent decades and we see this trend refl ect in a J-shaped curve for many countries. In other countries, educational attainment increased rapidly during the MDG era but has slowed in recent years as countries approach universal secondary participation. This trend is refl ected in an inverse J-shaped curve. We do, however, expect that in most countries educational attainment among young women will always be non-decreasing. As clearly visible in Figure 1, this is not the case. In 49 of 75 countries examined, at least one point estimate of secondary completion among women aged 25 to 29 is lower than the previous point estimate. In 11 countries, we observe a decline of 10 percentage points or more between adjacent point estimates. 4.2 Generating new trends: country examples To address incoherent trends over time, we calculate adjusted estimates of female educational attainment. Providing adjusted estimates poses some challenges. The application of inferential models to correct biases due to poor data quality and to fi ll in missing values are commonplace in international research databases (Bauer et al. 2012; Barro/Lee 2013; Cohen/Soto 2007; Kc et al. 2010; De la Fuente/Doménech 2012; Jordá/Alonso 2017) and not only for data on education but also for databases measuring mortality, fertility and various economic indicators. The variety of methods used to carry out these estimations makes it diffi cult to pool or compare data across sources. A parsimonious yet effi cient approach is to infer trends from as many observations as possible within a country. Thanks to the availability of international census and survey microdata, we can now perform such an analysis. In our analysis, we use weighted regression to infer trends in educational expansion based on observations from the same country. Figure 2 presents observed point estimates of educational attainment among young females for nine illustrative countries. For each of these countries, at least 8 observations from censuses, DHS surveys and/or MICS surveys are available. All nine countries experienced an increase in the proportion of females aged 25 to 29 completing secondary education or higher between the earliest and most recent census or survey observations, but no country experienced a monotonic increase during the observed period. In Benin, for example, 1.8 percent of females aged 25 to 29 had completed secondary or higher according to the 1992 census but only 0.3 percent of females aged 25 to 29 had completed secondary or higher according to the 1996 DHS survey. In Nepal, 27.5 percent of females aged 25 to 29 had completed secondary or higher according to the 2011 census, but only 15.9 percent in 2014 according to that year’s MICS survey. To identify which and to what extent observations diverge from overall country- level trends, we specify polynomial regression models for each country that estimate the best-fi tting trend line according to the survey and census observations, weighted Trends in Female Education in Low- and Middle-Income Countries • 359 Fig. 2: Percent of women aged 25-29 completing secondary education or higher in nine countries Source: own design • Kristen Jeffers, Albert Esteve360 by sample size. As the dependent variable, we use the logit of the proportion of women aged 25 to 29 with secondary or more education (ln(P/1-P)) to account for non-linearity in educational expansion. The grey regression line in each country graph depicts the fi tted relationship between time and the percent of females aged 25 to 29 completing secondary or higher. By design, regression lines favour census observations, which are based on sample sizes much larger than those provided by DHS and MICS surveys. We use regression parameters to produce predicted values of the percentage of females aged 25 to 29 completing secondary education or higher that align with expected trends (results in Appendix 1). 4.3 Adjusted trends Figure 3 plots predicted values of educational attainment among young females for all countries studied. Adjusted data show a monotonically increasing pattern of educational expansion in nearly every country. The few exceptions might refl ect actual population trends or lack of observations, in particular from censuses. To assess coherence in the measurement of female educational attainment across censuses, DHS surveys and MICS surveys, we produced Figure 4 and Table 2. Figure 4 plots observed and predicted values of the percentage of women aged 25 to 29 completing secondary education or higher for all countries studied. Table 2 summarises the absolute difference between observed and predicted values by data source, region and time period. Overall, coherence across data sources is strong: the mean absolute difference between observed and predicted values is 2.2 percentage points. Of 535 census and survey observations, two-thirds deviate from expected trends by less than two percentage points. Only 11 percent of observations deviate from expected trends by fi ve or more percentage points. The difference between observed and predicted values is 10 or more in 19 of 535 observations (labelled in Fig. 4). As expected, census-based estimates deviate least from expected trends. At the global level, predicted values for DHS-based estimates deviate from observed values by 2.7 percentage points on average compared to 3.2 percentage points for MICS-based estimates. MICS and DHS are similar in terms of the number of observations diverging from predicted estimates by 0 to 2 percentage points, but the plurality of observations with large deviations – 10 percentage points or more – are derived from MICS data. In relative terms, the percent of observations deviating more than 10 percentage points from predicted values for MICS is more than twice (8.5 percent) the number than for DHS (3.2 percent). In terms of regional differences, Africa, Asia and Eastern Europe show the highest levels of conformity between observed and expected patterns and across data sources. In Africa, nearly 80 percent of observed point estimates deviate from predicted values by less than 2 percentage points. The share of African observations that diverge 10 or more percentage points is 2 percent or less for all data sources. Compared to MICS-based observations in other regions, MICS surveys in Africa are more in line with expected country trends. In Asia, 70 percent of observations fall within the minimum deviation category (0 to 2 percentage points). MICS surveys in Asia show higher levels of deviation than DHS surveys Trends in Female Education in Low- and Middle-Income Countries • 361 F ig . 3: P re d ic te d p e rc e n t o f w o m e n a g e d 2 5 -2 9 c o m p le ti n g s e co n d a ry e d u ca ti o n o r h ig h e r b y y e a r, c o u n tr y, d a ta so u rc e a n d w o rl d r e g io n S o u rc e : o w n d e si g n • Kristen Jeffers, Albert Esteve362 in the region. Finally, coherence across data sources in Latin America are lowest than in other regions. More than half of all survey and census observations from the region deviate from predicted values by 2 or more percentage points. Results for Latin America are infl uenced by the predominance of census samples for the region compared to surveys. There are three times as many census samples as DHS surveys and four times as many census samples than MICS surveys available for Latin America. The relative quantity of large census samples means the model demands higher precision from both census and survey observations for most Latin American countries. Accordingly, survey observations perform particularly poorly Fig. 4: Percent of women aged 25 to 29 completing secondary or higher, observed versus predicted values Source: own design Trends in Female Education in Low- and Middle-Income Countries • 363 in the region. Thirty percent of DHS observations and more than 40 percent of MICS observations deviate from predicted values by 5 percentage points or more. Our results do not reveal signifi cant differences in the coherence of observations from the year 2000 or earlier compared to observations after 2000. We also considered the temporal proximity of surveys to censuses. There is no discernible relationship between the number of years elapsed since the previous census and the magnitude of the difference between observed and predicted values for survey observations (Pearson correlation coeffi cient = -0.06). Tab. 2: Distribution of samples based on the gap between the observed and predicted percentages of women aged 25-29 with secondary or more Gap between observed and predicted values 0-1.9 2-4.9 5-9.9 10+ N All Countries 67.9 20.7 7.9 3.6 535 Census (IPUMS) 76.2 19.5 2.9 1.4 210 DHS 61.6 24.2 11.0 3.2 219 MICS 64.2 16.0 11.3 8.5 106 Africa 79.1 14.8 4.9 1.2 244 Census (IPUMS) 91.7 6.9 0.0 1.4 72 DHS 71.9 19.8 7.4 1 121 MICS 78.4 13.7 5.9 2.0 51 Asia & Eastern Europe 69.9 17.8 8.2 4.1 146 Census (IPUMS) 87.3 10.9 1.8 0.0 55 DHS 56.1 28.1 12.3 3.5 57 MICS 64.7 11.8 11.8 11.8 34 Latin America 46.9 33.8 12.4 6.9 145 Census (IPUMS) 55.4 36.1 6.0 2.4 83 DHS 39.0 31.7 19.5 9.8 41 MICS 28.6 28.6 23.8 19.0 21 2000 or earlier 68.3 23.8 5.0 2.9 240 Census (IPUMS) 75.4 20.8 2.3 1.5 130 DHS 54.9 32.9 7.3 4.9 82 MICS 75 10.7 10.7 3.6 28 2001 or later 67.5 18.3 10.2 4.1 295 Census (IPUMS) 77.5 17.5 3.8 1.3 80 DHS 65.7 19 13.1 2.2 137 MICS 60.3 17.9 11.5 10.3 78 Samples are classifi ed by data source, region, and time period. Source: own design • Kristen Jeffers, Albert Esteve364 5 Conclusion and Discussion Improved access to census and survey microdata for LMICs has opened new opportunities for large-scale comparative investigations based on multiple sources of data. Used together, these data sources provide more comprehensive temporal and geographic coverage than any single source used in isolation. Yet this type of comparative work is still rare in socio-demographic analysis. Researchers tend to work with a single data source, shying away from the challenges of harmonisation and coherence across sources. In this paper, we challenge perceived limitations of cross-national research to examine the measurement of educational expansion across data sources. Literature from a variety of disciplines documents the role of education in personal and social development. Many studies considering this relationship use data from a single country; those studies that do offer a cross- national perspective typically use only one of the three data sources utilised in this paper: censuses, DHS or MICS. As a result, our understanding of the links between education and individual, household and societal outcomes has been informed by studies using different datasets. Whether these datasets yield similar results was previously unknown. To address this gap, we carried out a simple coherence exercise to measure the consistency of trends in female educational expansion across data sources. We focus on women aged 25 to 29 completing secondary education because gender gaps persist in post-primary education in many LMICs. Analyses of men and other categories of educational attainment produced similar results. We pooled nearly 20 million individual-level observations from 535 censuses and surveys and 75 countries from IPUMS, DHS and MICS. To maximise comparability across data sources, we use a dichotomous measure of education to identify females completing upper secondary education or higher. These data confi rm previously observed patterns of educational expansion: Access to education is increasing in all regions, but levels of educational attainment among young women remain low in the majority of LMICs. Less than half of women aged 25 to 29 had completed secondary education or higher in the most recent survey or census available in 51 of 75 countries studied. Our coherence analysis, however, shows that the general upward trend in educational attainment among young females conceals erratic trajectories in the majority of countries studied. Our model smooths these trajectories. To assess the accuracy of each point estimate, we compared observed estimates to predicted values. Overall, coherence across data sources is high. Two- thirds of observed point estimates differ from predicted point estimates by two or fewer percentage points. Still, thirty percent of observed point estimates diverge from predicted values by two or more percentage points. Censuses perform better than DHS and MICS. Coherence across data sources is strongest in Africa. Coherence across data sources is lowest in Latin America, where nearly 20 percent of observed point estimates deviate from predicted values by 5 or more percentage points. Discrepancies across data sources may originate during survey design, data collection and/or data processing. While DHS and MICS use similar sampling Trends in Female Education in Low- and Middle-Income Countries • 365 strategies and typically rely on recent censuses as sampling frames, practices vary across countries to adapt to national circumstances. In cases where surveys are fi elded soon after the most recent national census, an updated census-based sampling frame may not yet be available, producing discrepancies between data sources even when collected during the same year. For example, in Nepal, the 2001 DHS survey uses the 1991 census as its sampling frame and 2011 DHS survey uses the 2001 census (though there was a census fi elded in 2011). This may explain gaps between DHS- and census-based estimates for these years in Nepal (see Fig. 2). Non-response and other non-sampling errors occur in both censuses and surveys, but there may be systematic differences by data source. For example, refusal to participate may contribute more to non-response for surveys than for censuses, which are obligatory or well-promoted in many countries. Likewise, depending on the cultural context, the reporting of educational attainment among female household members might vary depending on the characteristics of interviewers and perceived use of collected information, which is likely to differ by data source. As indicated in previous studies (Bauer et al. 2012; Speringer et al. 2019), discrepancies related to data processing are particularly likely with education data due to varied treatment of complete versus incomplete levels of education. In countries where secondary completion rates are still low, the distinction between secondary attendance and completion may not be reported or recorded consistently across sources. Our results identify the countries for which further investigation of these sources of discrepancies may be required for analyses that combine census and survey data. Results presented here have other practical applications. Adjusted estimates of the percent of women aged 25 to 29 completing at least secondary education can be used in a variety of demographic and economic investigations. The data also serve as benchmarks for assessing the accuracy of estimates reported in international databases, the trend alignment of future DHS, MICS and census-based estimates and the quality of education information obtained in other sources not included in our analysis such as household surveys with small sample sizes. In future research, we will evaluate consistency across data sources based on other dimensions captured in censuses and household surveys such as demographic composition of the population, family structure and living arrangements. When data are scarce, as is the case for many LMICs, pooling data resources can be a useful strategy. Still, in the interest of reliable empirical evidence for policymaking, we should not take the coherence of data within and across sources for granted. Our analysis provides empirical validity for most analyses that combine census, DHS and MICS and identifi es survey and census samples that should be used with caution. References Abuya, Benta A.; Ciera, James; Kimani-Murage, Elizabeth 2012: Effect of mother’s education on child’s nutritional status in the slums of Nairobi. In: BMC Pediatrics 12,1: 80. https://doi.org/10.1186/1471-2431-12-80 • Kristen Jeffers, Albert Esteve366 Abuya, Benta A. et al. 2011: Infl uence of maternal education on child immunization and stunting in Kenya. In: Maternal and Child Health Journal 15,8: 1389-1399. https://doi.org/10.1007/s10995-010-0670-z Ainsworth, Martha; Beegle, Kathleen; Nyamete, Andrew 1996: The impact of women's schooling on fertility and contraceptive use: A study of fourteen sub-Saharan African countries. In: The World Bank Economic Review 10,1: 85-122. https://doi.org/10.1093/wber/10.1.85 Barakat, Bilal F.; Durham, Rachel, E. 2014: Future education trends. World Population and Human Capital in the 21st Century. Oxford: Oxford University Press: 397-433. Barro, Robert J.; Lee, Jong-Wha 1994: Sources of economic growth. In: Carnegie- Rochester Conference Series on Public Policy 40: 1-46. https://doi.org/10.1016/0167-2231(94)90002-7 Barro, Robert J.; Lee, Jong-Wha 2013. A new data set of educational attainment in the world, 1950-2010. In: Journal of Development Economics 104: 184-198. https://doi.org/10.1016/j.jdeveco.2012.10.001 Bauer, Ramon et al. 2012: Populations for 171 countries by age, sex, and level of education around 2010: Harmonized estimates of the baseline data for the Wittgenstein Centre projections. Becker, Gary S. 1994: Human capital: A theoretical and empirical analysis, with special reference to education. University of Chicago press. Bicego, George T.; Boerma, J. Ties 1993: Maternal education and child survival: a comparative study of survey data from 17 countries. In: Social Science & Medicine 36,9: 1207-1227. https://doi.org/10.1016/0277-9536(93)90241-u Caldwell, John C. 1981: Maternal education as a factor in child mortality. In: World Health Forum 2,1: 75-8. Caldwell, John C.; McDonald, Peter 1982: Infl uence of maternal education on infant and child mortality: Levels and causes. In: Health Policy and Education 2,3-4: 251-267. https://doi.org/10.1016/0165-2281(82)90012-1 Castro Martin, Teresa 1995: Women's education and fertility: results from 26 Demographic and Health Surveys. In: Studies in Family Planning 26,4: 187-202. https://doi.org/10.2307/2137845 Cochrane, Susan H.; Leslie, Joanne; O'Hara, Donald J. 1982: Parental education and child health: Intracountry evidence. In: Health Policy and Education 2,3-4: 213-250. https://doi.org/10.1016/0165-2281(82)90011-X Cohen, Daniel; Soto, Marcelo 2007: Growth and human capital: good data, good results. In: Journal of Economic Growth 12,1: 51-76. https://doi.org/10.1007/s10887-007-9011-5 De la Fuente, Angel; Doménech, Rafael 2012: Educational Attainment in the OECD, 1960- 2010. BBVA Research Working Paper 12/20. BBVA (Banco Bilbao Vizcaya Argentaria) Research. De la Fuente, Angel; Domenech, Rafael 2013: Cross-country data on the quantity of schooling: a selective survey and some quality measures. BBVA working papers 13/27. Desai, Sonalde; Alva, Soumya 1998: Maternal education and child health: Is there a strong causal relationship? In: Demography 35,1: 71-81. https://doi.org/10.2307/3004028 Dorius, Shawn F. 2013: The rise and fall of worldwide education inequality from 1870 to 2010: Measurement and trends. In: Sociology of Education 86,2: 158-173. https://doi.org/10.1177/0038040712456558 Trends in Female Education in Low- and Middle-Income Countries • 367 Dorius, Shawn F.; Firebaugh, Glenn 2010: Trends in global gender inequality. In: Social Forces 88,5: 1941-1968. https://doi.org/10.1353/sof.2010.0040 Fukuda-Parr, Sakiko; Greenstein, Joshua; Stewart, David 2013: How should MDG success and failure be judged: Faster progress or achieving the targets? In: World Development 41: 19-30. https://doi.org/10.1016/j.worlddev.2012.06.014 Goujon, Anne et al. 2016: A harmonized dataset on global educational attainment between 1970 and 2060 – An analytical window into recent trends and future prospects in human capital development. In: Journal of Demographic Economics 82,3: 315-363. https://doi.org/10.1017/dem.2016.10 Grant, Monica J.; Behrman, Jere R. 2010: Gender gaps in educational attainment in less developed countries. In: Population and Development Review 36,1: 71-89. https://doi.org/10.1111/j.1728-4457.2010.00318.x Grépin, Karen A.; Bharadwaj, Prashant 2015: Maternal education and child mortality in Zimbabwe. In: Journal of Health Economics 44: 97-117. https://doi.org/10.1016/j.jhealeco.2015.08.003 Hannum, Emily; Buchmann, Claudia 2005: Global educational expansion and socio- economic development: An assessment of fi ndings from the social sciences. In: World Development 33,3: 333-354. https://doi.org/10.1016/j.worlddev.2004.10.001 ICF 2004-2017: Demographic and Health Surveys (various) [Datasets]. Funded by USAID. Rockville, Maryland: ICF [Distributor]. ICF 2019: "Survey Types." The DHS Program website. Funded by USAID [http://www. dhsprogram.com, 15.12.2019]. ICF International 2012: Demographic and Health Surveys Methodology. Sampling and Household Listing Manual. Calverton, Maryland. Ilie, Sonia; Rose, Pauline 2016: Is equal access to higher education in South Asia and sub-Saharan Africa achievable by 2030? In: Higher Education 72,4: 435-455. https://doi.org/10.1007/s10734-016-0039-3 Jacob, Arun 2017: Mind the gap: Analyzing the impact of data gap in Millennium Development Goals’ (MDGs) indicators on the progress toward MDGs. In: World Development 93: 260-278. https://doi.org/10.1016/j.worlddev.2016.12.016 Jeong, Joshua; Bhatia, Amiya; Fink, Günther 2018: Associations between birth registration and early child growth and development: evidence from 31 low-and middle-income countries. In: BMC public health 18,1: 673. https://doi.org/10.1186/s12889-018-5598-z Jordá, Vanesa; Alonso, José M. 2017: New estimates on educational attainment using a continuous approach (1970-2010). In: World Development 90: 281-293. https://doi.org/10.1016/j.worlddev.2016.10.005 Kang, Yunhee et al. 2018: Association between stunting and early childhood development among children aged 36-59 months in South Asia. In: Maternal & Child Nutrition 14: e12684. http://dx.doi.org/10.1111/mcn.12684 KC, Samir et al. 2010: Projection of populations by level of educational attainment, age, and sex for 120 countries for 2005-2050. In: Demographic Research 22,15: 383-472. https://dx.doi.org/10.4054/DemRes.2010.22.15 KC, Samir; Lentzner, Harold 2010: The effect of education on adult mortality and disability: a global perspective. In: Vienna Yearbook of Population Research 8: 201- 235. https://doi.org/10.1553/populationyearbook2010s201 • Kristen Jeffers, Albert Esteve368 Kebede, Endale; Goujon, Anne; Lutz, Wolfgang 2019: Stalls in Africa’s fertility decline partly result from disruptions in female education. In: Proceedings of the National Academy of Sciences 116,8: 2891-2896. https://doi.org/10.1073/pnas.1717288116 Lloyd, Cynthia B. 2005: Growing up global. The transition to adulthood in developing countries. Washington DC: National Academies. Lutz, Wolfgang; Skirbekk, Vegard 2014: How education drives demography and knowledge informs projection. In: Lutz, Wolfgang; Butz, William P.; KC, Samir (Eds.): World Population and Human Capital in the 21st Century. Oxford: Oxford University Press: 14-38. https://doi.org/10.1093/acprof:oso/9780198703167.003.0002 Lutz, Wolfgang; Butz, William P.; KC, Samir (Eds.) 2014: World Population and Human Capital in the 21st Century. Oxford: Oxford University Press. https://doi.org/10.1093/acprof:oso/9780198703167.001.0001 Lutz, Wolfgang et al. 2018: Demographic and human capital scenarios for the 21st century: 2018 assessment for 201 countries. Publications Offi ce of the European Union. https://doi.org/10.2760/835878 Makoka, Donald; Masibo, Peninah Kinya 2015: Is there a threshold level of maternal education suffi cient to reduce child undernutrition? Evidence from Malawi, Tanzania and Zimbabwe. In: BMC Pediatrics 15,1: 96. https://doi.org/10.1186/s12887-015-0406-8 Masquelier, Bruno; Garbero, Alessandra 2016: Educational differentials in adult mortality in low-and middle-income countries. In: Quetelet Journal 4,1: 7-28. https://doi.org/10.14428/rqj2016.04.01.01 Mellington, Nicole; Cameron, Lisa 1999: Female education and child mortality in Indonesia. In: Bulletin of Indonesian Economic Studies 35,3: 115-144. https://doi.org/10.1080/00074919912331337717 Minnesota Population Center 2019: Integrated Public Use Microdata Series, International: Version 7.2 [dataset]. Minneapolis, MN: IPUMS. https://doi.org/10.18128/D020.V7.2 OECD 2018: Education at a Glance 2018: OECD Indicators. Paris: OECD Publishing. https://doi.org/10.1787/eag-2018-en Pace, Garrett T.; Lee, Shawna J.; Grogan-Kaylor, Andrew 2019: Spanking and young children’s socioemotional development in low-and middle-income countries. In: Child Abuse & Neglect 88: 84-95. https://doi.org/10.1016/j.chiabu.2018.11.003 Rosenzweig, Mark R.; Wolpin, Kenneth I. 1994: Are there increasing returns to the intergenerational production of human capital? Maternal schooling and child intellectual achievement. In: The Journal of Human Resources 29,2: 670. Schady, Norbert 2011: Parents’ education, mothers’ vocabulary, and cognitive development in early childhood: Longitudinal evidence from Ecuador. In: American Journal of Public Health 101,12: 2299-2307. https://doi.org/10.2105/AJPH.2011.300253 Schultz, Theodore W. 1961: Investment in human capital. In: The American Economic Review 51,1: 1-17. Speringer, Markus et al. 2019: Global reconstruction of educational attainment, 1950 to 2015: Methodology and assessment. In: Vienna Institute of Demography Working Paper 02/2019. https://doi.org/10.1553/0x003cb434 UNESCO 2016: Creating sustainable futures for all; Global education monitoring report. Gender review. Paris: UNESCO. UNESCO 2018: Global Education Monitoring Report 2019: Migration, Displacement and Education – Building Bridges, not Walls. Paris: UNESCO. Trends in Female Education in Low- and Middle-Income Countries • 369 Kristen Jeffers. Universitat Autònoma de Barcelona. Bellaterra, Spain. E-mail: kristen.m.jeffers@gmail.com Albert Esteve (). Universitat Pompeu Fabra and Centre d’Estudis Demogràfi cs – CERCA. Barcelona, Spain. E-mail: albert.esteve@upf.edu URL: https://www.upf.edu/en/web/politiques/entry/-/-/196854/adscripcion/albert- esteve UNESCO Institute for Statistics (UIS) 2019: New Methodology Shows that 258 Million Children, Adolescents and Youth Are Out of School. Fact Sheet no. 56 [http://uis. unesco.org/sites/default /files/documents/new-methodology-shows-258-million- children-adolescents-and-youth-are-out-school.pdf, 10.10.2019]. UNESCO 2019: From access to empowerment: UNESCO strategy for gender equality in and through education 2019-2025. Paris: UNESCO. UNICEF 2019: Every Child Learns: UNICEF Education Strategy 2019-2030. New York: UNICEF. UNICEF 2020: Multiple Indicator Cluster Surveys (MICS) [http://mics.unicef.org/., 15.01.2020]. United Nations 2013: The Millenium Development Goals Report 2013. New York: United Nations. Vikram, Kriti; Vanneman, Reeve; Desai, Sonalde 2012: Linkages between maternal education and childhood immunization in India. In: Social Science & Medicine 75,2: 331-339. https://doi.org/10.1016/j.socscimed.2012.02.043 Wang, Limin 2003: Determinants of child mortality in LDCs: empirical fi ndings from demographic and health surveys. In: Health Policy 65,3: 277-299. https://doi.org/10.1016/s0168-8510(03)00039-3 Date of submission: 15.11.2021 Date of acceptance: 24.05.2022 • Kristen Jeffers, Albert Esteve370 Appendix Tab. A1: Percent of women aged 25-29 completing secondary education or higher (P) by country, year and data source: observed and predicted values Country/Year Data Source Observed Predicted Difference1 Sample P P size2 Argentina 1970 Census (IPUMS) 18.10 15.76 2.34 17162 1980 Census (IPUMS) 28.58 28.94 0.36 100336 1991 Census (IPUMS) 40.47 42.52 2.05 153473 2001 Census (IPUMS) 53.92 50.14 3.78 136778 2010 Census (IPUMS) 50.47 52.24 1.77 156901 2011 MICS 68.20 52.20 16.00 3317 Armenia 2000 DHS 95.39 93.24 2.15 849 2001 Census (IPUMS) 92.63 92.10 0.53 11471 2005 DHS 28.13 89.83 61.70 953 2010 DHS 95.02 93.79 1.23 943 2011 Census (IPUMS) 95.87 94.98 0.89 13888 2015 DHS 95.51 98.57 3.06 1111 Bangladesh 1991 Census (IPUMS) 4.89 4.89 0 471616 1993 DHS 7.08 5.75 1.33 2208 1996 DHS 9.56 7.22 2.34 2166 1999 DHS 12.28 8.89 3.39 2371 2001 Census (IPUMS) 10.03 10.10 0.07 602539 2004 DHS 12.55 12.06 0.49 2303 2006 MICS 18.06 13.44 4.62 12378 2007 DHS 14.49 14.14 0.35 2252 2011 DHS 17.67 17.01 0.66 3851 2011 Census (IPUMS) 16.96 17.01 0.05 362601 2012 MICS 17.27 17.74 0.47 10765 2014 DHS 19.57 19.16 0.41 3771 Belarus 1999 Census (IPUMS) 97.16 97.16 0 34310 2005 MICS 95.93 96.64 0.71 1252 2009 Census (IPUMS) 96.94 96.91 0.03 36923 2012 MICS 97.08 97.38 0.30 1358 Trends in Female Education in Low- and Middle-Income Countries • 371 Tab. A1: Continuation Country/Year Data Source Observed Predicted Difference1 Sample P P size2 Benin 1979 Census (IPUMS) 0.45 0.52 0.07 15000 1992 Census (IPUMS) 1.78 1.18 0.60 21115 1996 DHS 0.33 1.59 1.26 1092 2001 DHS 1.28 2.41 1.13 1314 2002 Census (IPUMS) 2.20 2.63 0.43 29520 2006 DHS 2.33 3.77 1.44 4079 2011 DHS 5.61 6.08 0.47 3417 2013 Census (IPUMS) 8.32 7.42 0.90 42883 2014 MICS 4.59 8.20 3.61 2859 Bolivia 1976 Census (IPUMS) 7.91 7.88 0.03 17702 1992 Census (IPUMS) 24.56 24.98 0.42 24420 1994 DHS 29.98 27.79 2.19 1496 1998 DHS 37.75 33.57 4.18 1834 2000 MICS 44.74 36.48 8.26 771 2001 Census (IPUMS) 37.91 37.92 0.01 31187 2003 DHS 39.42 40.77 1.35 2877 2008 DHS 46.69 47.52 0.83 2928 Botswana 1981 Census (IPUMS) 4.21 3.88 0.33 3696 1991 Census (IPUMS) 9.67 11.30 1.63 5470 2000 MICS 23.22 22.96 0.26 1188 2001 Census (IPUMS) 26.86 24.47 2.39 7424 2011 Census (IPUMS) 39.18 39.90 0.72 10408 Brazil 1970 Census (IPUMS) 7.25 8.43 1.18 175630 1980 Census (IPUMS) 17.83 14.89 2.94 242723 1991 DHS 4.87 26.12 21.25 1108 1991 Census (IPUMS) 27.25 26.12 1.13 362413 1996 DHS 28.84 32.73 3.89 2291 2000 Census (IPUMS) 34.21 38.57 4.36 407987 2010 Census (IPUMS) 56.35 54.27 2.08 416354 • Kristen Jeffers, Albert Esteve372 Tab. A1: Continuation Country/Year Data Source Observed Predicted Difference1 Sample P P size2 Burkina Faso 1985 Census (IPUMS) 0.54 0.53 0.01 33016 1993 DHS 1.25 1.28 0.03 1296 1996 Census (IPUMS) 1.62 1.67 0.05 38956 1998 DHS 1.41 1.95 0.54 1171 2003 DHS 2.23 2.71 0.48 2145 2006 MICS 1.81 3.16 1.35 1451 2006 Census (IPUMS) 3.40 3.16 0.24 56225 2010 DHS 2.01 3.68 1.67 3064 Burundi 2000 MICS 3.63 3.16 0.47 721 2005 MICS 1.96 2.38 0.42 1501 2010 DHS 2.94 2.51 0.43 1689 2016 DHS 4.02 4.12 0.10 3023 Cambodia 1998 Census (IPUMS) 2.48 2.46 0.02 45814 2000 DHS 2.98 2.96 0.02 2193 2004 Survey3 (IPUMS) 3.37 4.40 1.03 2990 2005 DHS 3.58 4.89 1.31 2320 2008 Survey3 (IPUMS) 6.96 6.80 0.16 62643 2010 DHS 8.60 8.56 0.04 3396 2013 Census (IPUMS) 12.29 12.21 0.08 6015 2014 DHS 12.28 13.77 1.49 3111 Cameroon 1976 Census (IPUMS) 0.69 0.66 0.03 27308 1987 Census (IPUMS) 2.43 2.60 0.17 34447 1991 DHS 4.14 3.99 0.15 754 1998 DHS 5.33 7.71 2.38 1033 2000 MICS 4.92 9.10 4.18 975 2004 DHS 6.70 12.29 5.59 1933 2005 Census (IPUMS) 14.31 13.17 1.14 72900 2006 MICS 8.45 14.08 5.63 1595 2011 DHS 11.18 18.94 7.76 2859 2014 MICS 16.02 22.00 5.98 1854 Trends in Female Education in Low- and Middle-Income Countries • 373 Tab. A1: Continuation Country/Year Data Source Observed Predicted Difference1 Sample P P size2 Central African Rep. 1994 DHS 1.72 1.77 0.05 1098 2000 MICS 2.39 2.32 0.07 3255 2006 MICS 3.16 3.34 0.18 2257 2010 MICS 4.62 4.51 0.11 2290 Chad 1996 DHS 0.17 0.18 0.01 1480 2000 MICS 0.62 0.56 0.06 1130 2004 DHS 1.34 1.33 0.01 1179 2010 MICS 2.88 2.99 0.11 3305 2014 DHS 3.76 3.69 0.07 3579 Colombia 1964 Census (IPUMS) 2.05 2.72 0.67 12319 1973 Census (IPUMS) 7.48 7.55 0.07 70229 1985 Census (IPUMS) 25.36 21.66 3.70 122574 1990 DHS 33.99 30.00 3.99 1786 1993 Census (IPUMS) 31.57 35.35 3.78 152653 1995 DHS 37.99 38.94 0.95 2045 2000 DHS 43.02 47.67 4.65 1919 2005 DHS 56.03 55.52 0.51 6363 2005 Census (IPUMS) 55.75 55.52 0.23 153211 2010 DHS 65.37 62.15 3.22 8165 2015 DHS 74.93 67.46 7.47 6322 Congo 2005 DHS 8.47 8.49 0.02 1399 2009 DHS 11.55 11.36 0.19 1237 2011 DHS 13.80 13.94 0.14 2101 2014 MICS 20.20 20.14 0.06 1926 Congo DR 2000 MICS 7.53 7.64 0.11 2188 2007 DHS 12.96 11.50 1.46 1782 2010 MICS 11.86 13.50 1.64 2299 2013 DHS 16.20 15.71 0.49 3590 • Kristen Jeffers, Albert Esteve374 Tab. A1: Continuation Country/Year Data Source Observed Predicted Difference1 Sample P P size2 Costa Rica 1963 Census (IPUMS) 8.34 9.65 1.31 2641 1973 Census (IPUMS) 14.16 15.60 1.44 6271 1984 Census (IPUMS) 30.36 24.00 6.36 10629 2000 Census (IPUMS) 31.57 37.75 6.18 15029 2011 MICS 43.35 46.54 3.19 985 2011 Census (IPUMS) 49.01 46.54 2.47 19652 Cote d'Ivoire 1994 DHS 0.86 1.37 0.51 1534 1998 DHS 5.46 2.40 3.06 569 2000 MICS 5.87 3.07 2.80 1926 2005 DHS 3.63 5.18 1.55 1067 2006 MICS 3.83 5.66 1.83 2268 2011 DHS 8.12 8.15 0.03 2123 2016 MICS 11.46 10.34 1.12 2325 Cuba 2002 Census (IPUMS) 59.00 58.91 0.09 43868 2006 MICS 60.62 69.74 9.12 1274 2010 MICS 83.03 78.59 4.44 1791 2014 MICS 84.15 85.28 1.13 1961 Dominican Republic 1960 Census (IPUMS) 2.37 2.55 0.18 7222 1970 Census (IPUMS) 4.88 6.01 1.13 9222 1981 Census (IPUMS) 17.92 13.53 4.39 17021 1991 DHS 38.93 24.62 14.31 1424 1996 DHS 35.32 31.44 3.88 1564 1999 DHS 37.75 35.78 1.97 224 2000 MICS 41.60 37.25 4.35 653 2002 Census (IPUMS) 34.35 40.21 5.86 35614 2002 DHS 39.78 40.21 0.43 3894 2007 DHS 46.35 47.54 1.19 4434 2010 Census (IPUMS) 53.86 51.79 2.07 38209 2013 DHS 59.06 55.85 3.21 1474 2014 MICS 63.22 57.15 6.07 5545 Trends in Female Education in Low- and Middle-Income Countries • 375 Tab. A1: Continuation Country/Year Data Source Observed Predicted Difference1 Sample P P size2 Egypt 1986 Census (IPUMS) 23.83 23.76 0.07 261427 1992 DHS 30.48 30.58 0.10 2419 1995 DHS 37.42 34.85 2.57 3293 1996 Census (IPUMS) 36.03 36.41 0.38 226212 2000 DHS 44.21 43.29 0.92 3424 2005 DHS 54.86 53.11 1.75 4658 2006 Census (IPUMS) 55.44 55.19 0.25 323772 2008 DHS 59.54 59.39 0.15 4029 2014 DHS 66.28 71.79 5.51 5593 Eswatini 2000 MICS 25.62 25.94 0.32 882 2006 DHS 34.12 32.33 1.79 809 2010 MICS 32.68 34.63 1.95 897 2014 MICS 35.91 35.11 0.80 803 Ethiopia 1984 Census (IPUMS) 1.41 1.44 0.03 117671 1994 Census (IPUMS) 3.81 3.70 0.11 192189 2000 DHS 5.21 5.08 0.13 2907 2005 DHS 5.94 5.74 0.20 2791 2007 Census (IPUMS) 5.20 5.82 0.62 55486 2011 DHS 6.59 5.65 0.94 3490 2016 DHS 9.79 4.86 4.93 3185 Ghana 1984 Census (IPUMS) 5.60 5.58 0.02 54221 1993 DHS 4.43 8.72 4.29 903 1998 DHS 7.03 11.38 4.35 940 2000 Census (IPUMS) 12.92 12.69 0.23 78720 2003 DHS 12.09 14.99 2.90 1058 2006 MICS 16.81 17.73 0.92 1019 2008 DHS 20.18 19.83 0.35 1968 2010 Census (IPUMS) 22.06 22.17 0.11 110632 2011 MICS 24.68 23.44 1.24 1701 2014 DHS 30.98 27.64 3.34 1678 • Kristen Jeffers, Albert Esteve376 Tab. A1: Continuation Country/Year Data Source Observed Predicted Difference1 Sample P P size2 Guatemala 1964 Census (IPUMS) 0.63 0.88 0.25 7460 1973 Census (IPUMS) 3.50 2.64 0.86 10147 1981 Census (IPUMS) 6.83 5.65 1.18 11083 1994 Census (IPUMS) 11.94 12.69 0.75 29376 1995 DHS 12.30 13.22 0.92 1947 1998 DHS 12.49 14.72 2.23 1026 2002 Census (IPUMS) 15.70 16.35 0.65 41847 2014 DHS 24.25 17.61 6.64 4206 Guinea 1983 Census (IPUMS) 4.53 4.52 0.01 20630 1996 Census (IPUMS) 1.52 1.54 0.02 31379 1999 DHS 2.44 1.60 0.84 1383 2005 DHS 1.75 2.38 0.63 1332 2012 DHS 8.77 6.37 2.40 1677 2016 MICS 11.94 13.79 1.85 1964 Guyana 2000 MICS 34.65 34.16 0.49 814 2005 DHS 37.73 39.02 1.29 418 2006 MICS 38.59 40.41 1.82 769 2009 DHS 48.24 45.46 2.78 763 2014 MICS 56.12 56.66 0.54 976 Haiti 1971 Census (IPUMS) 0.76 0.74 0.02 16771 1982 Census (IPUMS) 3.17 3.56 0.39 6648 1994 DHS 5.37 11.10 5.73 902 2000 DHS 4.80 15.73 10.93 1589 2003 Census (IPUMS) 20.08 17.81 2.27 36205 2005 DHS 11.59 19.02 7.43 1769 2012 DHS 13.80 21.60 7.80 2403 2016 DHS 22.22 21.72 0.50 2256 Trends in Female Education in Low- and Middle-Income Countries • 377 Tab. A1: Continuation Country/Year Data Source Observed Predicted Difference1 Sample P P size2 Honduras 1961 Census (IPUMS) 3.21 2.44 0.77 700 1974 Census (IPUMS) 5.40 6.08 0.68 9166 1988 Census (IPUMS) 14.39 12.54 1.85 15661 2001 Census (IPUMS) 18.07 19.58 1.51 22690 2005 DHS 19.91 21.58 1.67 3566 2011 DHS 29.42 24.17 5.25 3981 India 1983 Survey4 (IPUMS) 8.72 7.86 0.86 24520 1987 Survey4 (IPUMS) 9.64 9.77 0.13 27059 1992 DHS 8.92 12.51 3.59 22344 1993 Survey4 (IPUMS) 13.01 13.10 0.09 23612 1998 DHS 20.38 16.24 4.14 22949 1999 Survey4 (IPUMS) 17.98 16.90 1.08 24926 2004 Survey4 (IPUMS) 22.54 20.30 2.24 24527 2005 DHS 15.39 20.99 5.60 23759 2009 Survey4 (IPUMS) 29.96 23.78 6.18 19333 2015 DHS 27.50 27.88 0.38 124239 Indonesia 1971 Census (IPUMS) 2.65 2.62 0.03 26274 1976 Survey5 (IPUMS) 4.86 4.74 0.12 9953 1980 Census (IPUMS) 7.28 7.25 0.03 283374 1985 Survey5 (IPUMS) 9.51 11.54 2.03 26639 1990 Census (IPUMS) 17.10 17.07 0.03 41494 1991 DHS 18.78 18.29 0.49 5832 1994 DHS 24.07 22.14 1.93 6739 1995 Survey5 (IPUMS) 27.03 23.47 3.56 31556 1997 DHS 28.12 26.15 1.97 6860 2000 MICS 31.59 30.15 1.44 1834 2000 Census (IPUMS) 30.13 30.15 0.02 951041 2002 DHS 32.27 32.76 0.49 6744 2005 Survey5 (IPUMS) 34.27 36.47 2.20 48676 2007 DHS 37.52 38.77 1.25 7956 2010 Census (IPUMS) 41.94 41.90 0.04 1059239 2012 DHS 45.21 43.74 1.47 7682 • Kristen Jeffers, Albert Esteve378 Tab. A1: Continuation Country/Year Data Source Observed Predicted Difference1 Sample P P size2 Iraq 1997 Census (IPUMS) 20.61 20.60 0.01 79448 2000 MICS 19.12 19.68 0.56 3759 2006 MICS 18.82 18.49 0.33 4452 2011 MICS 18.04 18.10 0.06 9025 Jamaica 1982 Census (IPUMS) 31.80 27.76 4.04 8181 1991 Census (IPUMS) 34.04 44.08 10.04 10401 2001 Census (IPUMS) 82.02 72.92 9.10 8461 2005 MICS 69.20 83.32 14.12 492 2011 MICS 72.77 93.53 20.76 736 Jordan 1990 DHS 45.51 46.88 1.37 2089 1997 DHS 57.83 54.09 3.74 1971 2004 Census (IPUMS) 57.12 57.18 0.06 21753 2007 DHS 54.49 57.27 2.78 3167 2009 DHS 57.72 56.92 0.80 2874 2012 DHS 56.84 55.77 1.07 3145 Kazakhstan 1995 DHS 87.13 86.84 0.29 592 1999 DHS 85.58 87.36 1.78 734 2006 MICS 91.83 90.05 1.78 1943 2010 MICS 90.39 92.17 1.78 2059 2015 MICS 95.10 94.81 0.29 2215 Kenya 1969 Census (IPUMS) 1.79 1.66 0.13 26558 1979 Census (IPUMS) 5.18 5.98 0.80 45921 1989 Census (IPUMS) 16.97 14.43 2.54 42454 1993 DHS 22.44 18.41 4.03 1338 1998 DHS 25.40 23.09 2.31 1412 1999 Census (IPUMS) 22.99 23.92 0.93 60253 2000 MICS 25.87 24.72 1.15 1853 2003 DHS 24.13 26.78 2.65 1539 2008 DHS 25.09 28.98 3.89 1497 2009 Census (IPUMS) 28.99 29.22 0.23 167262 2014 DHS 35.13 29.30 5.83 6334 Trends in Female Education in Low- and Middle-Income Countries • 379 Tab. A1: Continuation Country/Year Data Source Observed Predicted Difference1 Sample P P size2 Kyrgyz Republic 1997 DHS 88.07 91.01 2.94 598 1999 Census (IPUMS) 90.72 90.46 0.26 18992 2005 MICS 70.27 89.60 19.33 1019 2009 Census (IPUMS) 90.24 89.76 0.48 23637 2012 DHS 90.65 90.26 0.39 1331 2014 MICS 87.59 90.75 3.16 1191 Laos 2000 MICS 7.70 7.20 0.50 1488 2005 Census (IPUMS) 11.54 11.59 0.05 21626 2006 MICS 10.57 12.56 1.99 1157 2011 MICS 18.75 17.50 1.25 3884 2017 MICS 22.18 22.58 0.40 4260 Lesotho 1996 Census (IPUMS) 12.34 12.34 0 6976 2000 MICS 13.82 14.42 0.60 1102 2004 DHS 17.21 17.22 0.01 1456 2006 Census (IPUMS) 19.49 18.96 0.53 8026 2009 DHS 19.44 22.07 2.63 1789 2014 DHS 29.72 28.82 0.90 1572 Liberia 1974 Census (IPUMS) 2.83 2.83 0 6725 2007 DHS 11.24 11.10 0.14 1296 2008 Census (IPUMS) 12.18 12.20 0.02 15151 2013 DHS 20.04 20.01 0.03 1718 Malawi 1987 Census (IPUMS) 1.85 1.85 0 30849 1992 DHS 2.76 3.06 0.30 887 1998 Census (IPUMS) 5.09 5.10 0.01 39520 2000 DHS 6.52 5.92 0.60 2479 2004 DHS 8.62 7.69 0.93 2301 2006 MICS 6.07 8.63 2.56 5007 2008 Census (IPUMS) 9.87 9.58 0.29 57404 2010 DHS 10.94 10.52 0.42 4538 2013 MICS 9.97 11.87 1.90 4502 2016 DHS 13.84 13.11 0.73 4225 • Kristen Jeffers, Albert Esteve380 Tab. A1: Continuation Country/Year Data Source Observed Predicted Difference1 Sample P P size2 Mali 1987 Census (IPUMS) 1.98 1.89 0.09 30491 1995 DHS 0.27 0.90 0.63 1903 1998 Census (IPUMS) 0.84 0.89 0.05 35398 2001 DHS 2.10 1.04 1.06 2492 2006 DHS 1.26 1.86 0.60 2774 2009 MICS 4.83 3.21 1.62 4935 2009 Census (IPUMS) 3.42 3.21 0.21 54723 2012 DHS 3.32 6.34 3.02 2335 2015 MICS 9.24 13.88 4.64 3523 Mexico 1960 Census (IPUMS) 1.70 1.77 0.07 18860 1970 Census (IPUMS) 2.61 4.71 2.10 17035 1990 Census (IPUMS) 22.55 19.96 2.59 337284 1995 Census (IPUMS) 21.58 25.58 4.00 13721 2000 Census (IPUMS) 30.57 31.41 0.84 430071 2005 Census (IPUMS) 35.15 37.10 1.95 440673 2010 Census (IPUMS) 41.19 42.35 1.16 458103 2015 Survey6 (IPUMS) 49.08 46.94 2.14 440085 2015 MICS 47.80 46.94 0.86 2290 Mongolia 1989 Census (IPUMS) 40.60 40.81 0.21 8623 2000 MICS 73.75 64.35 9.40 1863 2000 Census (IPUMS) 63.52 64.35 0.83 10967 2005 MICS 60.68 68.14 7.46 1387 2010 MICS 67.31 68.19 0.88 1489 2013 MICS 68.10 66.46 1.64 2103 Morocco 1982 Census (IPUMS) 4.01 4.01 0 38493 1992 DHS 10.10 10.92 0.82 1562 1994 Census (IPUMS) 12.22 12.16 0.06 53354 2003 DHS 10.06 13.97 3.91 2852 2004 Census (IPUMS) 13.88 13.71 0.17 64611 Trends in Female Education in Low- and Middle-Income Countries • 381 Tab. A1: Continuation Country/Year Data Source Observed Predicted Difference1 Sample P P size2 Mozambique 1997 DHS 0.25 0.76 0.51 1887 1997 Census (IPUMS) 0.78 0.76 0.02 65059 2003 DHS 1.47 1.32 0.15 2542 2007 Census (IPUMS) 2.42 2.44 0.02 83801 2008 MICS 3.50 2.93 0.57 2882 2009 DHS 3.30 3.56 0.26 1076 2011 DHS 5.52 5.41 0.11 2424 Nepal 1996 DHS 5.69 5.64 0.05 1838 2001 DHS 7.72 12.49 4.77 1863 2001 Census (IPUMS) 12.66 12.49 0.17 82787 2006 DHS 12.27 20.78 8.51 1841 2011 DHS 21.47 27.05 5.58 2219 2011 Census (IPUMS) 27.45 27.05 0.40 142388 2014 MICS 15.86 28.69 12.83 2461 2016 DHS 34.24 28.73 5.51 2210 Nicaragua 1971 Census (IPUMS) 4.68 4.67 0.01 6485 1995 Census (IPUMS) 19.29 19.64 0.35 17001 1998 DHS 25.82 22.29 3.53 2367 2001 DHS 25.41 25.01 0.40 2245 2005 Census (IPUMS) 28.55 28.68 0.13 21272 Niger 1992 DHS 0.45 0.49 0.04 1443 1998 DHS 0.79 0.69 0.10 1388 2000 MICS 0.97 0.76 0.21 996 2006 DHS 0.74 0.95 0.21 1847 2012 DHS 1.17 1.09 0.08 2432 Nigeria 1990 DHS 15.29 14.55 0.74 1836 1999 DHS 22.16 24.11 1.95 1637 2003 DHS 27.06 28.95 1.89 1471 2006 Survey7 (IPUMS) 31.73 32.68 0.95 3674 2007 MICS 31.87 33.93 2.06 5394 2007 Survey7 (IPUMS) 35.49 33.93 1.56 3679 • Kristen Jeffers, Albert Esteve382 Tab. A1: Continuation Country/Year Data Source Observed Predicted Difference1 Sample P P size2 2008 Survey7 (IPUMS) 34.19 35.17 0.98 4748 2008 DHS 34.67 35.17 0.50 6815 2009 Survey7 (IPUMS) 38.74 36.41 2.33 3385 2010 Survey7 (IPUMS) 42.51 37.64 4.87 3121 2011 MICS 44.54 38.87 5.67 6342 2013 DHS 36.15 41.27 5.12 7381 2016 MICS 45.24 44.75 0.49 6201 Pakistan 1973 Census (IPUMS) 2.10 2.87 0.77 49036 1981 Census (IPUMS) 3.34 3.06 0.28 340212 1990 DHS 4.92 4.37 0.55 1974 1998 Census (IPUMS) 7.15 7.60 0.45 469545 2006 DHS 24.72 15.88 8.84 29428 2012 DHS 27.87 29.24 1.37 3982 Palestine 1997 Census (IPUMS) 35.19 35.20 0.01 9101 2007 Census (IPUMS) 44.28 44.12 0.16 8145 2010 MICS 52.50 53.14 0.64 2743 2014 MICS 68.96 68.72 0.24 2065 Panama 1960 Census (IPUMS) 9.12 8.11 1.01 1926 1970 Census (IPUMS) 13.81 15.97 2.16 5335 1980 Census (IPUMS) 28.66 26.98 1.68 7740 1990 Census (IPUMS) 41.70 39.37 2.33 10055 2000 Census (IPUMS) 47.93 50.77 2.84 12097 2010 Census (IPUMS) 59.78 59.67 0.11 13446 2013 MICS 67.27 61.78 5.49 1497 Paraguay 1962 Census (IPUMS) 5.95 5.65 0.30 2896 1972 Census (IPUMS) 8.53 7.40 1.13 7511 1982 Census (IPUMS) 7.89 11.00 3.11 11577 1990 DHS 23.46 16.28 7.18 1054 1992 Census (IPUMS) 21.74 18.11 3.63 15747 2002 Census (IPUMS) 30.22 31.40 1.18 18117 2016 MICS 59.26 61.82 2.56 1372 Trends in Female Education in Low- and Middle-Income Countries • 383 Tab. A1: Continuation Country/Year Data Source Observed Predicted Difference1 Sample P P size2 Peru 1991 DHS 53.41 45.41 8.00 2848 1993 Census (IPUMS) 49.22 49.49 0.27 92731 1996 DHS 54.24 54.72 0.48 5309 2000 DHS 57.16 59.94 2.78 4644 2004 DHS 63.28 63.19 0.09 6546 2007 Census (IPUMS) 65.06 64.43 0.63 116384 2009 DHS 46.83 64.70 17.87 3792 2010 DHS 63.75 64.68 0.93 3576 2011 DHS 64.27 64.55 0.28 3502 2012 DHS 67.00 64.30 2.70 3650 Philippines 1990 Census (IPUMS) 51.82 52.56 0.74 249175 1993 DHS 58.95 54.64 4.31 2614 1995 Census (IPUMS) 58.12 56.28 1.84 286916 1998 DHS 63.88 59.11 4.77 2420 2000 Census (IPUMS) 59.56 61.22 1.66 292832 2003 DHS 67.38 64.67 2.71 2221 2008 DHS 72.56 70.98 1.58 2279 2010 Census (IPUMS) 73.94 73.60 0.34 375739 2013 DHS 74.74 77.52 2.78 2461 2017 DHS 76.59 82.51 5.92 4133 Rwanda 1992 DHS 1.56 1.71 0.15 1131 2000 DHS 4.14 3.47 0.67 1647 2000 MICS 2.88 3.47 0.59 739 2002 Census (IPUMS) 4.21 4.17 0.04 30290 2005 DHS 5.28 5.54 0.26 1820 2010 DHS 6.48 8.96 2.48 2567 2012 Census (IPUMS) 11.01 10.88 0.13 46952 2014 DHS 13.47 13.21 0.26 2352 Senegal 1988 Census (IPUMS) 1.85 1.87 0.02 28965 1992 DHS 3.09 2.52 0.57 1181 2000 MICS 3.31 4.13 0.82 2434 2002 Census (IPUMS) 4.87 4.58 0.29 38821 2005 DHS 3.06 5.26 2.20 2646 • Kristen Jeffers, Albert Esteve384 Tab. A1: Continuation Country/Year Data Source Observed Predicted Difference1 Sample P P size2 2010 DHS 5.59 6.37 0.78 2974 2012 DHS 5.30 6.77 1.47 1648 2014 DHS 5.99 7.15 1.16 1624 2015 DHS 7.98 7.33 0.65 1713 2016 DHS 8.00 7.49 0.51 1657 2017 DHS 9.06 7.65 1.41 3016 Sierra Leone 2000 MICS 1.41 0.82 0.59 980 2004 Census (IPUMS) 1.63 1.73 0.10 22401 2005 MICS 2.22 2.08 0.14 2070 2008 DHS 5.04 3.55 1.49 1811 2010 MICS 5.01 5.01 0 2679 2013 DHS 9.04 8.21 0.83 2891 2017 MICS 14.05 15.09 1.04 3103 South Africa 1996 Census (IPUMS) 32.41 33.31 0.90 158301 1998 DHS 36.84 35.61 1.23 1907 2001 Census (IPUMS) 41.74 39.19 2.55 166246 2007 Survey8 (IPUMS) 35.31 46.71 11.40 42443 2011 Census (IPUMS) 52.91 51.86 1.05 209390 Sudan 2000 MICS 25.35 24.83 0.52 6951 2008 Census (IPUMS) 10.95 11.05 0.10 212414 2010 MICS 25.62 13.55 12.07 3444 2014 MICS 27.07 30.63 3.56 3718 Tajikistan 2000 MICS 47.57 46.88 0.69 936 2005 MICS 43.98 45.02 1.04 1503 2012 DHS 48.61 47.67 0.94 1676 2017 DHS 53.00 53.33 0.33 1938 Tanzania 1988 Census (IPUMS) 3.36 3.36 0 98157 1992 DHS 3.85 4.09 0.24 1737 1996 DHS 5.65 5.07 0.58 1554 1999 DHS 6.59 6.03 0.56 818 2002 Census (IPUMS) 7.24 7.23 0.01 165233 2003 DHS 7.4 7.7 0.30 1365 Trends in Female Education in Low- and Middle-Income Countries • 385 Tab. A1: Continuation Country/Year Data Source Observed Predicted Difference1 Sample P P size2 2004 DHS 9.3 8.21 1.09 2009 2007 DHS 10.02 9.99 0.03 1697 2010 DHS 11.18 12.23 1.05 1774 2012 Census (IPUMS) 13.97 114.04 0.07 183634 2015 DHS 23.1 17.32 5.78 2307 Thailand 1970 Census (IPUMS) 2.83 2.97 0.14 27739 1980 Census (IPUMS) 8.92 8.12 0.80 15679 1990 Census (IPUMS) 20.67 18.62 2.05 24278 2000 Census (IPUMS) 30.70 34.73 4.03 27316 2012 MICS 63.07 55.96 7.11 3072 2015 MICS 65.36 60.65 4.71 3625 Togo 1960 Census (IPUMS) 0.99 0.49 0.50 805 1970 Census (IPUMS) 0.17 0.37 0.20 1181 1998 DHS 1.19 1.31 0.12 1751 2000 MICS 3.60 1.61 1.99 960 2006 MICS 3.36 3.27 0.09 1275 2010 MICS 4.85 5.60 0.75 1242 2010 Census (IPUMS) 5.73 5.60 0.13 26741 2013 DHS 5.71 8.60 2.89 1768 Trinidad and Tobago 1970 Census (IPUMS) 28.90 24.92 3.98 2115 1980 Census (IPUMS) 34.12 38.76 4.64 4336 1990 Census (IPUMS) 55.40 55.02 0.38 5262 2000 MICS 75.63 70.56 5.07 527 2000 Census (IPUMS) 72.84 70.56 2.28 4263 2006 MICS 82.56 78.31 4.25 666 2011 MICS 86.14 83.60 2.54 756 2011 Census (IPUMS) 82.07 83.60 1.53 5449 Türkiye 1985 Census (IPUMS) 13.64 13.66 0.02 98502 1990 Census (IPUMS) 16.46 16.40 0.06 118632 1993 DHS 17.32 18.52 1.20 1577 1998 DHS 21.70 23.01 1.31 1587 2000 Census (IPUMS) 25.16 25.20 0.04 146356 2003 DHS 31.63 28.96 2.67 2043 • Kristen Jeffers, Albert Esteve386 Tab. A1: Continuation Country/Year Data Source Observed Predicted Difference1 Sample P P size2 Uganda 1991 Census (IPUMS) 1.28 1.30 0.02 63199 1995 DHS 8.21 2.58 5.63 1462 2000 DHS 5.69 5.14 0.55 1494 2002 Census (IPUMS) 6.40 6.43 0.03 94313 2006 DHS 7.20 9.25 2.05 1649 2011 DHS 10.63 12.52 1.89 1792 2016 DHS 15.98 14.47 1.51 3392 Ukraine 2001 Census (IPUMS) 90.55 90.56 0.01 170044 2005 MICS 94.75 86.67 8.08 1358 2007 DHS 70.48 88.71 18.23 1087 2012 MICS 97.72 97.48 0.24 1985 Uruguay 1963 Census (IPUMS) 9.68 11.07 1.39 9449 1975 Census (IPUMS) 21.63 18.93 2.70 9735 1985 Census (IPUMS) 33.35 26.29 7.06 10917 1996 Census (IPUMS) 22.62 33.83 11.21 10898 2006 Survey9 (IPUMS) 41.85 39.18 2.67 8300 2011 Census (IPUMS) 43.76 41.13 2.63 11590 2012 MICS 42.03 41.45 0.58 478 Venezuela 1971 Census (IPUMS) 7.38 7.82 0.44 39993 1981 Census (IPUMS) 21.66 19.57 2.09 62069 1990 Census (IPUMS) 30.99 33.28 2.29 71554 2000 MICS 27.70 46.14 18.44 745 2001 Census (IPUMS) 47.81 47.11 0.70 96038 Vietnam 1989 Census (IPUMS) 15.75 15.79 0.04 125526 1997 DHS 22.58 17.58 5.00 1336 1999 Census (IPUMS) 18.59 18.50 0.09 102748 2000 MICS 21.05 19.05 2.00 1336 2002 DHS 19.65 20.33 0.68 1256 2005 DHS 20.75 22.77 2.02 992 2006 MICS 27.15 23.74 3.41 1245 2009 Census (IPUMS) 27.07 27.16 0.09 621601 Trends in Female Education in Low- and Middle-Income Countries • 387 Tab. A1: Continuation Country/Year Data Source Observed Predicted Difference1 Sample P P size2 2010 MICS 42.60 28.50 14.10 1853 2013 MICS 49.02 33.13 15.89 1423 Zambia 1990 Census (IPUMS) 10.24 9.99 0.25 29844 1992 DHS 7.31 9.77 2.46 1285 1996 DHS 3.98 9.92 5.94 1427 1999 MICS 10.51 10.60 0.09 1780 2000 Census (IPUMS) 11.22 10.95 0.27 37791 2001 DHS 10.08 11.36 1.28 1465 2007 DHS 13.96 15.65 1.69 1479 2010 Census (IPUMS) 19.77 19.42 0.35 54052 2013 DHS 20.01 24.80 4.79 3049 Zimbabwe 1994 DHS 4.07 8.50 4.43 1026 1999 DHS 38.01 9.81 28.20 1088 2005 DHS 5.61 10.66 5.05 1664 2009 MICS 10.57 10.68 0.11 2238 2010 DHS 7.85 10.62 2.77 1858 2012 Census (IPUMS) 10.58 10.41 0.17 30674 2014 MICS 9.96 10.10 0.14 2519 2015 DHS 11.04 9.91 1.13 1800 1 Absolute value of observed value less predicted value 2 Number of females aged 25 to 29 (unweighted) 3 Cambodia Intercensal Population Survey 4 National Sample Survey Organisation Socio-Economic Survey of India 5 Indonesia Intercensal Population Survey 6 Intercensal Survey 7 Nigeria: National Bureau of Statistics General Household Survey 8 South Africa Community Survey 9 Uruguay Extended National Survey of Homes 2006 Source: own design Published by Federal Institute for Population Research (BiB) D-65180 Wiesbaden / Germany Managing Publisher Dr. Nikola Sander 2022 Editor Prof. Frans Willekens Managing Editor Dr. Ralina Panova Dr. Katrin Schiefer Editorial Assistant Beatriz Feiler-Fuchs Wiebke Hamann Layout Beatriz Feiler-Fuchs E-mail: cpos@bib.bund.de Scientifi c Advisory Board Karsten Hank (Cologne) Ridhi Kashyap (Oxford) Michaela Kreyenfeld (Berlin) Natalie Nitsche (Rostock) Zsolt Spéder (Budapest) Alyson van Raalte (Rostock) Rainer Wehrhahn (Kiel) Comparative Population Studies www.comparativepopulationstudies.de ISSN: 1869-8980 (Print) – 1869-8999 (Internet) Board of Reviewers Bruno Arpino (Barcelona) Kieron Barclay (Rostock) Laura Bernardi (Lausanne) Gabriele Doblhammer (Rostock) Anette Eva Fasang (Berlin) Michael Feldhaus (Oldenburg) Tomas Frejka (Sanibel) Alexia Fürnkranz-Prskawetz (Vienna) Birgit Glorius (Chemnitz) Fanny Janssen (Groningen) Frank Kalter (Mannheim) Stefanie Kley (Hamburg) Bernhard Köppen (Koblenz) Anne-Kristin Kuhnt (Duisburg) Hill Kulu (St Andrews) Nadja Milewski (Wiesbaden) Roland Rau (Rostock) Thorsten Schneider (Leipzig) Tomas Sobotka (Vienna) Jeroen J. A. Spijker (Barcelona) Heike Trappe (Rostock) Helga de Valk (The Hague) Sergi Vidal (Barcelona) Michael Wagner (Cologne)