Life Expectancy by Education, Income and Occupation in Germany: Estimations Using the Longitudinal Survival Method Life Expectancy by Education, Income and Occupation in Germany: Estimations Using the Longitudinal Survival Method* Marc Luy, Christian Wegner-Siegmundt, Angela Wiedemann, Jeroen Spijker Abstract: Reliable estimates for differences in life expectancy (LE) by socio-eco- nomic position (SEP), that can be assessed in an international context and are com- prehensive in terms of considering different SEP dimensions, are missing for the German population so far. The aim of the present study is to fi ll this gap by providing estimates for differences in LE by education, household income, work status and vocational class. The lack of national mortality data by SEP required an innovative methodological approach to estimate LE from survey data with a mortality follow- up. The main strengths of the method are the low demand on the data, its simple applicability and the estimation of a set of age-specifi c probabilities of dying. We employed the method to the German Life Expectancy Survey and estimated period life tables for 45 male and 32 female SEP subpopulations. The results show strik- ing differences in LE across all analysed SEP indicators. Among men, LE at age 40 ranges by more than fi ve years between the lowest and highest household income quartiles, more than six years between individuals with low and high education, around ten years across the work status groups, and almost 15 years across the vocational classes. The proportion of those who reach the classic pension age of 65 years also varies considerably, as does the remaining LE at this age. The corre- sponding differences among women are smaller, yet still notable. The results yield an interesting fi nding for the ongoing discussion about the various consequences of an increased pension age. Moreover, they provide policy-makers, doctors, re- searchers and public health workers with insights into Germany’s most disadvan- taged SEP subpopulations and the potential extent of their disadvantages in terms of longevity and mortality. Keywords: Life expectancy · Mortality · Socioeconomic position · Longitudinal survival method · Germany Comparative Population Studies Vol. 40, 4 (2015): 399-436 (Date of release: 14.12.2015) © Federal Institute for Population Research 2015 URL: www.comparativepopulationstudies.de DOI: 10.12765/CPoS-2015-16en URN: urn:nbn:de:bib-cpos-2015-16en7 * This article contains supplementary material in the form of an online Appendix: DOI 10.12765/ CPoS-2015-17en, URL: http://www.comparativepopulationstudies.de/index.php/CPoS/article/ view/203/218. http://www.comparativepopulationstudies.de/index.php/CPoS/article/view/203/218 http://www.comparativepopulationstudies.de • Marc Luy, Christian Wegner-Siegmundt, Angela Wiedemann, Jeroen Spijker400 1 Introduction Differences in lifetime by socioeconomic position (SEP) are some of the most stud- ied phenomena in international mortality research. The social gradient in longevity – i.e., the higher the SEP, the lower the mortality – has been shown for nearly every industrialised society (e.g. Kunst 1997; Mackenbach et al. 1997; Spijker 2004). How- ever, estimates for Germany are missing in most international comparative studies (see for instance the recent compilation of life expectancy by educational attainment for the European Union and Norway by EUROSTAT in Corsini 2010). This is due to the fact that offi cial German population statistics do not provide data for mortality by SEP. Therefore, researchers who are interested in social differences in mortality in Germany must consult alternative data sources. Most commonly used are data from health insurances, pension registers and surveys with a mortality follow-up. Based on such data, the mortality gradient by SEP has also been confi rmed for the German population for each of the three central dimensions of SEP, i.e. education, income and occupation (an extensive overview can be found in Lampert/Kroll 2014). Due to the specifi c data situation in Germany, most studies analysing this topic provide results in terms of relative risks for specifi c age segments. For instance, Klein et al. (2001) found that among the 25-64 years old participants of the Augsburg MONICA study, the risk of dying decreases by 7.6 percent with every year of educa- tion. Such numbers are diffi cult to assess, however. This diffi culty holds especially true for estimates on young and middle adult ages where the general level of mor- tality is low, and thus even small absolute differences in mortality can easily cause large relative differences. For most users of data on mortality differentials – such as public health workers and policy-makers – estimates for life expectancy (LE) would be more informative. Unfortunately, studies containing such estimates for the Ger- man population are rare. Most of them analyse differences in LE by income (e.g. Doblhammer et al. 2008; Kibele et al. 2013; Lampert/Kroll 2006; Lampert et al. 2007; Luy 2006; Perna et al. 2010; Shkolnikov et al. 2008; von Gaudecker 2006; von Gaud- ecker/Scholz 2007), but we found only six which present differences by education level (Doblhammer et al. 2008; Klein 1996, 1999; Luy 2006; Perna et al. 2010; Unger/ Schulze 2013), and even less include estimates for LE by occupation status defi ned by main work status groups (Doblhammer et al. 2008; Luy 2006) or the basic differ- entiation between blue and white collar occupations (Kibele et al. 2013). What makes the thin information on differences in LE by SEP in Germany even more limited is the fact that the existing studies are very heterogeneous with regard to the data and methods used as well as the defi nitions of the analysed subpopu- lations. We illustrate this briefl y for the six studies providing estimates for differ- ences in LE by education level. Four of them (Doblhammer et al. 2008; Klein 1996, 1999; Unger/Schulze 2013) are based on data of the German Socio-Economic Panel (SOEP), whose mortality follow-up has been demonstrated to be not representative for the mortality of the overall German population (Schnell/Trappmann 2006). The study conducted by Luy (2006) is based on data of the German Life Expectancy Survey (LES), and the study of Perna et al. (2010) on data of the Augsburg MONICA/ KORA study. The longevity indicators used in these publications range from LE at Life Expectancy by Education, Income and Occupation in Germany • 401 birth (Perna et al. 2010) to LE at age 16 (Klein 1996), at age 40 (Klein 1999; Unger/ Schulze 2013), at age 45 (Luy 2006) and at age 50 (Doblhammer et al. 2008). How- ever, only Klein (1996) and Unger and Schulze (2013) derive these fi gures from the classical period life table approach. Klein (1999) presents estimates for education- specifi c LE only for a further subdivision of the education groups by income and marital status. Perna et al. (2010) translate an estimated hazard function for the study sample on the basis of odds ratios into education-specifi c survival functions. Doblhammer et al. (2008) estimate how variations in the education level change the expected remaining life years of a reference person defi ned by specifi c characteris- tics with regard to marital status, work status, income, satisfaction with health and number of persons in the household. Luy (2006) performs cohort projections for individuals born between 1934 and 1952. Another aspect which further complicates comparisons is that the six studies an- alyse different education levels, but none of them classifi es the subgroups accord- ing to the internationally recognised ISCED scale: Klein (1996) presents estimates for the graduation levels of tertiary and below tertiary, Klein (1999) for minimum secondary and maximum primary, Luy (2006) for tertiary, secondary and primary, Doblhammer et al. (2008) for tertiary versus primary, Perna et al. (2010) for lower secondary and tertiary, and Unger and Schulze (2013) for tertiary, secondary and having no graduation (without presenting data for the primary level). As a conse- quence of these different approaches, the estimated variation in LE by education level varies considerably, ranging from 2.3 years (Perna et al. 2010) to 7.7 years (Unger/Schulze 2013) among men and from 1.9 years (Luy 2006) to 6.4 years (Dobl- hammer et al. 2008) among women. Finally, all studies base their education-specifi c LE estimates on the assumption of age-invariant relative mortality risks, which were derived – with different methodological approaches – from survival experiences in young and mid-adult ages. This is questionable, because mortality differences by SEP have been shown to vary across ages and are usually higher in young and mid- adulthood than in older age groups (e.g. Zajacova et al. 2009). The main point of this brief overview of existing estimates for differences in LE by SEP is that reliable estimates with regard to the extent of the differentials, that can be assessed in an international context, and that consider different SEP dimen- sions are missing for the German population. The aim of the present study is to fi ll this gap by providing estimates for differences in period LE by the three central SEP dimensions – education, income and occupation – which (i) refer to the same age segment, (ii) are derived from the same data, (iii) are based on a method which takes age-specifi c mortality patterns into account, and (iv) are easily accessible and ap- plicable for further research. The lack of national mortality data by SEP in Germany requires the development of a methodological approach to estimate LE from survey data with a mortality follow-up. This method – to which we refer as the “Longitudi- nal Survival Method” (LSM) – is introduced in section 2 with its theoretical and for- mal derivation (section 2.1) and its implementation with the used LES data (section 2.2). The estimates for LE at ages 40 and 65 and the survival probabilities between these ages by education, income, and occupation are presented in section 3. In section 4 we discuss the specifi c characteristics of the LSM in comparison to other • Marc Luy, Christian Wegner-Siegmundt, Angela Wiedemann, Jeroen Spijker402 approaches for estimating LE from longitudinal survey data (section 4.1) and sum- marise the most important results of this study in an international context (section 4.2). The paper ends with the fi nal conclusions. 2 The Longitudinal Survival Method (LSM) 2.1 Theoretical and formal derivation The development of the LSM was inspired by the techniques of indirect mortality estimation, which are used for estimating LE in many developing countries. For most of these populations, detailed data on deaths and the populations at risk are still unavailable or of too bad quality to be useable (Luy 2010a). To produce de- mographic estimates despite this lack of data, specifi c estimation techniques – the so-called “indirect methods” – have been developed. These are based on particular survey questions which are included in censuses or special survey projects (for an up-to-date overview of these methods see Moultrie et al. 2013). The dominating indirect technique for the estimation of adult mortality is the so-called “orphanhood method”. Its basis is the information whether the survey respondents’ mothers and fathers are still alive at the time of interview. The proportion of respondents with mother and father alive is then transformed into a period survival probability from age 25 to age 25 plus the age of respondents. Because the underlying mortality levels and patterns are widely unknown in developing countries, the transforma- tion is based on theoretical population models (see Hill et al. 1983). With the help of model life tables, the best fi tting survival function is used as estimated life table for a specifi c period before the time of the survey. This reference period – i.e. the calendar year to which the life table is assumed to apply – is derived from the age of the respondents and the level of their parents’ mortality. Alternative and similarly functioning indirect techniques for estimating adult mortality are based on informa- tion about the survival of respondents’ siblings and spouses (more details can be found in the above references as well as in Hill et al. 2005; Timæus 1991). The functionality of these indirect techniques and thus the quality of the result- ing estimates depend predominantly on the fi t between the assumed theoretical mortality models and the actual mortality patterns. This has been demonstrated for the orphanhood method with data from industrialised countries where the transfor- mation of information about the survival of the respondents’ parents into a period life table can be based on empirical data of trends in age-specifi c mortality (Festy 1995; Luy 2012). The availability of detailed data on mortality trends in developed countries has been used by Luy (2012) to develop a “Modifi ed Orphanhood Method” (MOM) for application in these populations. The MOM has already been deployed by Luy et al. (2011) to estimate LE by education and occupation in Italy – where offi cial population statistics are also limited with respect to socially differentiated mortality – and by Wiedemann (2012) to estimate LE of immigrants in Germany. The LSM adopts the concept of indirect mortality estimation to transform longi- tudinal survival data into a period life table. In contrast to these techniques, the ba- Life Expectancy by Education, Income and Occupation in Germany • 403 sic information of the LSM is not the (indirectly surveyed) survival of respondents’ relatives but the (directly recorded) survival of the respondents themselves. The basic idea of the LSM can be loosely described as follows: when we have a survey with a mortality follow-up, then we can “ask” the respondents themselves whether they are still alive at the time of the follow-up. Thus, the proportion of respondents who survived until the mortality follow-up can be used for an almost identical trans- formation into period survivorship as it is done in indirect estimation approaches with the proportions of respondents with relatives alive. The basis of the LSM is the proportion of individuals in age x at the time of the fi rst survey who survived the time z until the mortality follow-up, denoted by )zx ,x(SR ˆ , with the bars indicating the average ages and average observation times of all x-year old survey respondents at the time of the fi rst interview. The hat on RŜ indicates that the survival rate is derived directly from the survey data, whereas survival rates without hat – being introduced below – are derived from data for the entire population. The subscript R marks the reference to the survey respon- dents. )zx ,x(SR ˆ can be calculated by dividing the number of survivors at the time of the mortality follow-up – derived from the difference between the number of respondents at age x at the time of the fi rst survey, R(x), and the number of those who died D(x) by R(x), thus In order to maximise our use of the available information for the transformation of these longitudinal survivorships into period survival probabilities, each individual i is used with the precise age x at the time of the fi rst survey and the precise time z until the mortality follow-up (calendar dates transferred into decimals). An ex- pected longitudinal survival )zx ,x(SL  is estimated for each age from the sum of the individual survival rates from age x to x+z derived from the corresponding survival probabilities p x of cohort life tables for the total population divided by R(x). The expected longitudinal survivals are reconstructed for each survey respondent’s precise age x and time z until the mortality follow-up by interpolation from the avail- able cohort life tables (which usually only include survival probabilities for exact ages x). In formal terms, the estimated expected longitudinal survival for the survey respondents aged x at the time of the fi rst interview until the mortality follow-up results from In the next step, the corresponding period survivals )zx ,x(SP  for the refer- ence period t are derived from the period survival probabilities l x , which are recon- . R(x) D(x)-R(x) )zx,x(SRˆ (1) . R(x) p p )zx,x(S R(x) 1i x zx L (2) • Marc Luy, Christian Wegner-Siegmundt, Angela Wiedemann, Jeroen Spijker404 structed – equivalently to the previous step – by interpolation from the available period life tables, thus An appropriate reference period t is the mid-year between the times of the fi rst survey and the mortality follow-up. Alternatively, the reference period can also be derived – if available – from the dates of death of the deceased survey respondents. The longitudinal survival rates of the survey respondents )zx ,x(SR ˆ can be trans- formed into period survival rates from the exact age x to x+z from with w( zx  , x + z)t being a weighting factor for converting the period survival rate from the age span between x and zx  to the age span from age x to x+z (for exact ages x and survival times z). These weighting factors can be derived from the pe- riod life table for the reference period t. The resulting survivorship functions – one for each age x – can be smoothed and extended to the highest ages with the Brass model life table system (Brass 1975), setting the model parameter β = 1.0 and using the reference life table as standard. From the resulting smoothed survival curves, the age-specifi c probabilities of dying can be derived and weighted averages can be calculated for each age x by taking R(x) into account. This results in a specifi c age pattern of probabilities of dying q(x) which constitute the fi nal basis for the life table construction. The LSM can be used to estimate life tables for any sub-population of the un- derlying survey. The only necessary assumption is that the relationship between cohort and period survival prevalent in the entire population, which is used for the transformation in formula (4), applies equivalently to each subpopulation. 2.2 Application to the data of the German Life Expectancy Survey (LES) Mortality data from surveys with mortality follow-ups are rarely large and complete enough to apply an estimation method such as the LSM without specifi c adjust- ments. For our analysis, we used data from the western sample of the German Life Expectancy Survey (LES). The LES is a panel that consists of two waves of interviews, restricted to individuals with German citizenship, and is based on the National Health Survey. The fi rst wave was carried out between 1984 and 1986 and included a representative random sample of the total West German population. In 1998, the German Federal Institute for Population Research (Bundesinstitut für Bevölkerungsforschung, BiB) carried out a follow-up survey among the individuals . R(x) l l )zx,x(S t R(x) 1i x zx tP (3) (4), )zx,x(S )zx,x(S )zx,zx(w)zx,x(Ŝ l l L tP tR tx zx Life Expectancy by Education, Income and Occupation in Germany • 405 interviewed in the 1984/86 National Health Survey. In this second survey, the initial questionnaires were slightly modifi ed – e.g. purely medical details were removed and replaced by questions on general living conditions and family situations – and the number of respondents was restricted to those born between 1914 and 1952 (more details can be found in Gärtner 2001). The LES contains demographic indi- cators as well as information about economic and social status, social networks, health behaviours, life attitudes and a variety of health indicators. The western German LES sample includes a total of 4,139 women and 4,335 men. Of those, 304 women (7.3 percent) and 653 men (15.1 percent) died between the two survey waves. For 1,047 women (25.3 percent) and 903 men (20.8 percent) the survival status by 1998 is unknown. Tests of the quality of the LES mortality data revealed that the refl ected survival of the LES sample between 1984 and 1998 is representative for the western German population, regardless of whether individu- als with unknown survival status in 1998 are included or not (Luy/Di Giulio 2005; Salzmann/Bohk 2008). As suggested by Luy and Di Giulio (2005) we excluded the individuals with unknown survival status from the analysis. A sensitivity analysis on the basis of the average health statuses of individuals with known and unknown survival statuses at the time of the mortality follow-up indicates that neither the reduced LES sample as a whole nor the small SEP subgroups (which are introduced in more detail below) are health-biased. The differences in average health status between individuals with known and unknown survival statuses are minor in most subpopulations and in no case statistically signifi cant (see Appendix, Tables A1 and A2). Thus, we can assume that the mortality of the SEP subpopulations is also re- fl ected properly in the reduced LES sample. In the following, we illustrate how we applied the LSM to the LES data. The fi rst step is the calculation of the observed longitudinal survival of the survey respon- dents. For instance, the LES sample includes 71 women who were 65 years old at the time of the fi rst survey, their average age being 65.4 years, thus x = 65.4. The fi rst wave interviews of these women took place in the year 1985.3 (on average), and the second interviews of those who survived until wave 2 were conducted in the year 1998.4 (on average). Thus, the observed survival time until the mortality follow-up was 13.1 years, hence z = 13.1. During this time, 18 of the women died and 53 survived. Applying these numbers to formula (1) indicates that In total, we obtain 70 )zx ,x(SR ˆ estimates for women and men and the single ages 35 to 69, covering the survival rates from age 35.5 to 48.6 until age 69.6 to 82.9 among women, and from age 35.5 to 48.6 until 69.4 to 82.6 among men. For each of these estimates, the expected longitudinal survival is determined by formula (2) on the basis of the German cohort life tables published by the German Statistical Offi ce (Statistisches Bundesamt 2006). As described in section 2.1, the calculation incorporates the observed survival times of each individual of the LES sample with the precise ages at the beginning and at the time of the mortality follow-up, assum- 0.7465. 71 53 78.5)(65.4,ˆ RS • Marc Luy, Christian Wegner-Siegmundt, Angela Wiedemann, Jeroen Spijker406 ing for the deceased individuals the average time of the second interview as time of mortality follow-up. When mortality estimation is conducted with survey data such as the LES, the sample sizes can in some cases become so small that not every single age group contains deceased individuals. To overcome this problem, we estimated the sur- vival rates from moving averages of 11 age groups, i.e. estimating the survival rate for age 40 from the average of the ages 35 to 45, the survival rate for age 41 from the average of the ages 36 to 46 and so on. Figure 1 shows the resulting )zx ,x(SR ˆ values for women and men and every single age (grey arrows) in comparison to the corresponding expected longitudinal survivals derived from the German cohort life tables (bold black lines). The next steps of the transformation of the proportions of surviving survey re- spondents into period survivorship estimates are performed as described in the previous section, by applying formulas (3) and (4). We chose the year 1992 as our reference period, as it lies between the average points in time at which the surveys were conducted. The corresponding period survival probabilities were taken from the offi cial period life table 1991/93 for western Germany, published by the Ger- man Statistical Offi ce. This life table ends with the open age interval of 90+ years. Fig. 1: Observed and expected longitudinal survival rates for the LES sample by single ages (11-year moving averages) Source: own calculations with LES data (a) Men (b) Women 0.5 0.6 0.7 0.8 0.9 1.0 30 40 50 60 70 80 Age Expected Observed Longitudinal survival rate Longitudinal survival rate 0.5 0.6 0.7 0.8 0.9 1.0 30 40 50 60 70 80 Age Observed Expected Life Expectancy by Education, Income and Occupation in Germany • 407 We extended the life table to age 110 by applying the Kannisto model, which has proven to be the most accurate method for extrapolating mortality at highest ages (Thatcher et al. 1998). After smoothing the single survival functions with the Brass logit life table model, we averaged the resulting probabilities of dying for each sin- gle age. We used weighted averages, giving higher weights to the probabilities of dying from survival functions being based on observations from the survey data, and giving lower weights to those that resulted from extrapolation of survival func- tions derived from observations of younger ages. This was done by weighting the probabilities with the number of survivors of the corresponding survival function. Figure 2 shows the estimated probabilities of dying, gained from the LES data, for all women and men with the LSM in comparison to the corresponding values of the offi cial German life table 1991/93 for ages 40 to 90. The fi gure illustrates the functionality of the method, as the estimates derived from the LES with the LSM are very close to the values from the offi cial German life table for the entire population. The corresponding LE estimates for the LES sample are also close to the offi cial German life table, being slightly lower among men (34.5 vs. 35.2 years) and slightly higher among women (41.1 vs. 40.7 years). We performed the illustrated LSM estimations for each examined SEP subpopu- lation. We used information from the LES about education, household income and Fig. 2: Estimated probabilities of dying for ages 40 to 90 with the LSM in comparison to the corresponding values of the offi cial German life table 1991/93 0.001 0.01 0.1 1 35 40 45 50 55 60 65 70 75 80 85 90 95 Age Life table 1991/93 LSM estimate Men Women Probability of dying (log. scale) Source: Statistical Offi ce of Germany, own calculations with LES data • Marc Luy, Christian Wegner-Siegmundt, Angela Wiedemann, Jeroen Spijker408 occupation at the time of fi rst interview in 1984/86 as indicators for individuals’ SEP. Educational attainment was grouped according to the ISCED-97 scale (United Nations Educational Scientifi c and Cultural Organization 1996) into the three lev- els of “low” (ISCED 0-2), “medium” (ISCED 3-4) and “high” (ISCED 5-6). The ISCED combines school and vocational training degrees and is used in most international studies. The monthly net household income was separated into quartiles. The fi rst quartile indicates the lowest income group (less than 895 €) while the fourth quartile indicates the group with the highest income (1,917 € and more). The current – or, in the case of retirement or unemployment, the last – posi- tion on the labour market is defi ned on the basis of the work status as well as the vocational class. The work status includes four main groups: manual workers, em- ployees, public servants and self-employed workers. Each main group was further subdivided into two or three specifi c work status subgroups, such as unskilled and skilled workers or employees in simple, qualifi ed and highly qualifi ed tasks. The vocational classes are recorded in the LES according to the German Classifi cation of Professions KldB-92 (Statistisches Bundesamt 1992). The KldB-92 provides a hie- rarchical order of all professions with respect to the industrial segment with the six main classes (i) husbandry, forestry and horticulture, (ii) mining and mineral work- ers, (iii) production jobs, (iv) technical occupations, (v) service sector, and (vi) other work force. Each of these main classes includes up to three sublevels of specifi c professions. Because of the case numbers it was only possible to analyse the main classes and the fi rst sublevel. Nonetheless, the main and fi rst level subclasses allow us to isolate specifi c professions with substantial health risks, such as miners and mineral workers, structural and civil engineers or metalworkers. In contrast to the more common International Standard Classifi cation of Occu- pation (International Labour Offi ce 2012), the KldB-92 does not defi ne the skill level and specifi cation of an individual’s profession. Thus, the KldB-92 does not present differentiations between chief or senior positions on the one hand and workers or clerks on the other in a specifi c occupational group. For instance, the jobs in the health sector, as a subclass of the service sector, contain physicians as well as nurs- es. As is to be expected, the number of different occupations depends on gender. The employment rate of women – particularly for the cohorts contained in the LES sample – is notably lower than for males, leading to less variations and smaller case numbers in the female occupation groups. Descriptive statistics for each of the analysed subpopulations can be found in the Appendix (Tables A3 and A4), including the number of respondents R, the number of deaths D, the average age at the time of fi rst survey x, the average survival time until the mortality follow-up z, and the entire subsamples’ survival rate )zx ,x(SR ˆ . We restricted the analysis to subpopulations with at least fi ve ob- served deaths and estimated 95 percent confi dence intervals for LE with the meth- od proposed by Chiang (1984). The required death numbers for each single age were derived from applying the estimated survival functions to the number of indi- viduals of the LES sample belonging to the particular subpopulation. The complete life tables for each subpopulation with the corresponding confi dence intervals are available in csv-format as online supplement. Life Expectancy by Education, Income and Occupation in Germany • 409 3 Results Table 1 presents an overview of the results for each analysed SEP subpopulation of men, including estimates for the remaining LE at age 40 (e 40 ), the probability to survive from age 40 to 65 ( 25 p 40 ), and the remaining LE at age 65 (e 65 ). All estimates refer to the period 1991/93. Cases in which the 95 percent confi dence interval for e 40 respective e 65 does not include the corresponding value for the total male LES population are marked with an asterisk. Figure 3 provides an additional visual sum- mary for selected subpopulations by showing the differences in e 40 (black bars) and e 65 (white bars) to the values of the total male population (Figures 3c and 3d for the work statuses and vocational classes include all main groups and the subgroups with the lowest and highest values of e 40 and e 65 ). The results confi rm the existence of striking differences in LE among men across all the four analysed SEP indicators (education, household net income, work status and vocational class). For the subpopulations with the highest and lowest LE, the differences to the total population are statistically signifi cant despite the low case numbers in the LES sample (i.e. the 95 percent confi dence intervals for these sub- populations do not include the LE value for the total male population). The LE at age 40 by education level varies by 6.3 years, being 32.0 years for low, 34.0 years for medium and 38.3 years for highly educated men. The probability of survival until age 65 differs by 0.145 between men with a high ( 25 p 40 = 0.870) and men with a low level of education ( 25 p 40 = 0.725), i.e. the proportion of 40-year-old individuals who survive until age 65 is 14.5 percentage points higher for highly educated men than for their counterparts with a low education. At age 65, remaining LE for men still shows signifi cant differences, with a gap of 3.7 years between a high (e 65 = 16.5) and a low education (e 65 = 12.8). Similar – but slightly smaller – differences can be seen with regard to the net household income quartiles. The LE gap between men belonging to the fi rst (e 40 = 31.1, e 65 = 12.8) and fourth quartile (e 40 = 36.8, e 65 = 15.6) is 5.7 years at age 40 and 2.8 years at age 65. The probabilities to survive from age 40 to age 65 are 0.838 for men in the highest and 0.697 for men in the lowest household income quartile. Figures 3a and 3b reveal that the social gradient in mortality does not only affect the individuals in the lowest and the highest SEP, but applies to all SEP groups: the higher the SEP, the higher the LE. The social gradient in mortality also becomes apparent when work status is used as indicator for SEP. Among the main work status groups, manual workers exhibit the lowest LE (e 40 = 32.4, e 65 = 13.2), followed by public servants (e 40 = 35.6, e 65 = 14.8), employees (e 40 = 36.1, e 65 = 15.2) and self-employed workers (e 40 = 36.6, e 65 = 15.3). Thus, compared to manual workers, the surplus in LE for public servants, employees and self-employed workers is 3.2, 3.7 and 4.2 years at age 40, and 1.6, 2.0 and 2.1 years at age 65. The corresponding differences in the probability to sur- vive from age 40 to age 65 are 0.101, 0.088 and 0.080, respectively. The differences in LE become larger when we consider the work status subgroups. At age 40, the lowest LE can be found for employees in simple tasks (e 40 = 30.1) and the highest LE for self-employed farmers (e 40 = 39.6), yielding an e 40 -difference of 9.5 years. The • Marc Luy, Christian Wegner-Siegmundt, Angela Wiedemann, Jeroen Spijker410 Tab. 1: Life expectancy at ages 40 (e 40 ) and 65 (e 65 ) and the survival probability between ages 40 and 65 ( 25 p 40 ) by different indicators for SEP, Men SEP indicator / subpopulation e 40 25 p 40 e 65 All men 34.5 0.789 14.2 Education according to ISCED-971 Low (ISCED 1-2) 32.0* 0.725 12.8* Medium (ISCED 3-4) 34.0 0.775 14.0 High (ISCED 5-6) 38.3* 0.870 16.5* Household net income 1st quartile (below € 895) 31.1* 0.697 12.8* 2nd quartile (€ 895–1,406) 34.8 0.798 14.3 3rd quartile (€ 1,406–1,917) 35.5 0.812 14.9 4th quartile (€ 1,917 and more) 36.8* 0.838 15.6* Work status Manual workers (all) 32.4* 0.737 13.2* Unskilled or semi-skilled workers 30.4* 0.678 12.4* Skilled workers 34.2 0.787 13.8 Master craftsmen, overseers, foremen 32.7 0.740 14.0 Employees (all) 36.1* 0.825 15.2* Simple tasks 30.1* 0.659 12.5 Qualifi ed tasks 35.4 0.808 14.6 Highly qualifi ed tasks 38.3* 0.869 16.6* Public servants (all) 35.6 0.817 14.8 Simple or medium service 33.4 0.756 13.8 Higher or senior service 37.8* 0.867 15.9* Self-employed workers (all) 36.6* 0.838 15.3 Entrepreneurs 35.0 0.800 14.7 Freelancers 37.4 0.848 15.8 Farmers 39.6* 0.902 16.8* Vocational classes according to KldB-922 Jobs in (animal) husbandry, forestry, horticulture 33.1 0.749 14.2 Miners and mineral workers 26.2 0.541 11.5 Production jobs (all) 33.3 0.761 13.5 Metal production and metal working 34.9 0.808 14.3 Metal constr., mech. engineering and similar 31.9 0.705 13.7 Electrical engineering 33.8 0.767 13.0 Foodstuffs sector 33.7 0.762 13.7 Structural and civil engineering 30.9 0.693 13.0 Wood and plastics processing 39.6* 0.913 16.3 Painting, varnishing and similar occupations 38.2 0.864 16.6 Unskilled workers without specifi cation 32.3 0.747 11.5 Life Expectancy by Education, Income and Occupation in Germany • 411 probabilities of surviving from age 40 to age 65 are 0.659 and 0.902 respectively, i.e. a difference of 24.3 percentage points. At age 65, the gap in remaining LE is 4.4 years between self-employed farmers with the highest LE (e 65 = 16.8) and unskilled or semi-skilled manual workers with the lowest LE (e 65 = 12.4). The largest differences in mortality and longevity can be found between the vocational classes. Miners and mineral workers are the main class with the high- est mortality. Their LE at age 40 is 26.2 years, and only slightly more than half of them survive until age 65 ( 25 p 40 = 0.541). In contrast, the technical occupations are the main vocational class exhibiting the lowest mortality, with LE at age 40 being 39.4 years and the probability of surviving until age 65 being almost 90 percent ( 25 p 40 = 0.895). In between are the service sector (e 40 = 35.0, 25 p 40 = 0.800), pro- duction jobs (e 40 = 33.3, 25 p 40 = 0.761), jobs in (animal) husbandry, forestry and horticulture (e 40 = 33.1, 25 p 40 = 0.749) and the residual category “other workforce” (e 40 = 32.9, 25 p 40 = 0.744). With regard to LE at age 65, we fi nd the same order of subgroups, with miners and mineral workers exhibiting the lowest (e 65 = 11.5), and the technical occupations the highest level of LE (e 65 = 16.8). SEP indicator / subpopulation e 40 25 p 40 e 65 Technical occupations (all) 39.4* 0.895 16.8* Engineering, chemistry, physics, maths 38.8* 0.871 17.1* Technicians, technical specialists 39.5* 0.900 16.7* Service sector (all) 35.0 0.800 14.5 Goods und service marketing 34.4 0.786 14.3 Transport industry 31.9* 0.725 13.0 Organisation, administration, clerical jobs 35.1 0.801 14.8 Public and private security sector 34.2 0.788 13.7 Health sector 38.8 0.864 17.0 Social service and education 40.6* 0.909 18.0* Other service occupations 38.2 0.874 15.9 Other work force 32.9 0.744 14.1 Tab. 1: Continuation * = statistical signifi cant deviation from total LE, p < 0.05 1 ISCED-97 = International Standard Classifi cation of Education 2 KldB-92 = German Classifi cation of Professions Note: Estimates for the small subgroups of work statuses and vocational classes have to be interpreted with caution. Some of them are based on very small case numbers (see Appendix, Table A3), and the shapes of the estimated period life tables might differ from the longitudinal survival patterns of the corresponding LES subsamples. Source: own calculations with LES data • Marc Luy, Christian Wegner-Siegmundt, Angela Wiedemann, Jeroen Spijker412 Fig. 3: Deviation of life expectancy at ages 40 and 65 of subpopulations defi ned by specifi c SEP indicators from the total population, Men (a) Education (b) Household net income Difference to total LE Difference to total LE -4.0 -3.0 -2.0 -1.0 0.0 1.0 2.0 3.0 4.0 Low Medium High Education level Age 40 Age 65 -4.0 -3.0 -2.0 -1.0 0.0 1.0 2.0 3.0 4.0 1 2 3 4 Income quartile Age 40 Age 65 (c) Work status (d) Vocational class Difference to total LE Difference to total LE -9.0 -7.0 -5.0 -3.0 -1.0 1.0 3.0 5.0 7.0 9.0 E m p l. si m p le t as ks M an u al w o rk e rs ( al l) P u b lic s e rv an ts ( al l) E m p lo y e e s (a ll) S e lf -e m p lo y e d ( al l) S e lf -e m p l. fa rm e rs Age 40 Age 65 -9.0 -7.0 -5.0 -3.0 -1.0 1.0 3.0 5.0 7.0 9.0 M in e rs U n sk ill e d w o rk e rs O th e r w o rk f o rc e H u sb an d ry P ro d u ct io n jo b s S e rv ic e s e ct o r T e ch n ic al o cc . S o ci al s e rv ic e Age 40 Age 65 Source: own calculations with LES data Life Expectancy by Education, Income and Occupation in Germany • 413 The differences become even greater when we examine the subclasses. Miners and mineral workers remain the subgroup with the lowest LE at age 40. This voca- tional subclass also has the lowest LE at age 65, together with unskilled workers in production jobs, with a remaining LE of 11.5 years. However, compared to the miners and mineral workers, the unskilled workers in production jobs have a lower mortality at younger ages, with e 40 being 32.3 years and the probability of surviv- ing from age 40 to 65 being almost 75 percent ( 25 p 40 = 0.747). The lowest mortality level of all vocational subclasses can be found among men working in the social service and education sector, including mostly school teachers, but also scientists and social workers (e 40 = 40.6, e 65 = 18.0). Their advantage in LE compared to the vocational subclasses with the highest mortality levels is 14.4 years at age 40 and 6.5 years at age 65. Members of the social service and education sector have a 0.909 probability of surviving from age 40 to 65. The corresponding results for women are summarised in Table 2 and Figure 4. Compared to men, the differences in LE and survival probabilities between ages 40 and 65 are smaller and in fewer cases statistically signifi cantly different from the to- tal female LES population. To facilitate the comparison of the LE differentials among women and men, we use an identical axis scaling in Figures 3 and 4. With regard to education and household net income, we fi nd the same SEP gradient for women as for men. The difference in LE between women with a high (e 40 = 43.0, e 65 = 20.3) and a low education (e 40 = 40.7, e 65 = 18.3) is 2.3 years at age 40 and 2.0 years at age 65. Differently from men, we fi nd a larger LE span among the household net income quartiles than across the three education groups. The differences in LE at ages 40 and 65 between women belonging to the highest (e 40 = 43.4, e 65 = 20.1) and lowest household income quartiles (e 40 = 39.2, e 65 = 17.5) are 4.2 and 2.6 years, respectively. The corresponding spans in the probabilities to survive from age 40 to age 65 are 0.020 across the education groups and 0.063 across the household income quartiles. More interesting differences between women and men can be found with re- gard to the work statuses. Across the main work status groups, manual workers also show the lowest LE at age 40 among women (e 40 = 40.2), but the group of employees exhibits only a small advantage of 0.5 years (e 40 = 40.7). The highest LE at age 40, among the main work status groups, can be found for female pub- lic servants (e 40 = 46.7) with 6.5 years more than manual workers. Self-employed women (e 40 = 43.6) are placed between employees and public servants, with a 3.4- year advantage to the manual workers. The same order holds for LE at age 65. Compared to the manual workers (e 65 = 17.8), the surplus of employees (e 65 = 18.3), self-employed women (e 65 = 20.2) and public servants (e 65 = 22.9) is 0.5, 2.4 and 5.1 years, respectively. The highest probability of surviving from age 40 to age 65 can be found among public servants, with 25 p 40 being 0.960. The lowest survival probability is prevalent among the manual workers ( 25 p 40 = 0.900). Employees and self-employed women lie in between with 25 p 40 being 0.902 and 0.944, respectively. Across the work status subgroups, we fi nd the highest mortality among employees in highly qualifi ed tasks (e 40 = 38.3, e 65 = 16.5, 25 p 40 = 0.866) and the lowest mortal- ity among self-employed entrepreneurs (e 40 = 49.4, e 65 = 24.8, 25 p 40 = 0.986). The • Marc Luy, Christian Wegner-Siegmundt, Angela Wiedemann, Jeroen Spijker414 Tab. 2: Life expectancy at ages 40 (e 40 ) and 65 (e 65 ) and the survival probability between ages 40 and 65 ( 25 p 40 ) by different indicators for SEP, Women SEP indicator / subpopulation e 40 25 p 40 e 65 All women 41.1 0.911 18.5 Education according to ISCED-971 Low (ISCED 1-2) 40.7 0.903 18.3 Medium (ISCED 3-4) 41.3 0.915 18.6 High (ISCED 5-6) 43.0 0.923 20.3 Household net income 1st quartile (below € 895) 39.2* 0.877 17.5* 2nd quartile (€ 895–1,406) 41.0 0.910 18.5 3rd quartile (€ 1,406–1,917) 43.1* 0.929 20.2* 4th quartile (€ 1,917 and more) 43.4* 0.940 20.1* Work status Manual workers (all) 40.2 0.900 17.8 Unskilled or semi-skilled workers 39.9 0.894 17.7 Skilled workers 42.2 0.939 18.6 Employees (all) 40.7 0.902 18.3 Simple tasks 39.8 0.887 17.8 Qualifi ed tasks 42.2 0.923 19.3 Highly qualifi ed tasks 38.3 0.866 16.5 Public servants (all) 46.7* 0.960 22.9* Simple or medium service 40.2 0.864 19.1 Self-employed workers (all) 43.6* 0.944 20.2* Entrepreneurs 49.4* 0.986 24.8* Farmers 39.1 0.867 17.8 Vocational classes according to KldB-922 Production jobs (all) 40.7 0.893 18.7 Textile sector 41.5 0.901 19.4 Foodstuffs sector 41.6 0.898 19.6 Unskilled workers without specifi cation 37.9 0.842 17.3 Technical occupations (all) 42.6 0.917 19.8 Service sector (all) 41.4 0.916 18.7 Goods und service marketing 41.9 0.924 18.9 Transport industry 41.0 0.914 17.9 Organisation, administration, clerical jobs 41.6 0.917 18.8 Health sector 39.3 0.875 17.7 Social service and education 45.1* 0.939 22.1* Other service occupations 40.5 0.904 17.9 Other work force 38.0 0.875 15.5 * = statistical signifi cant deviation from total LE, p < 0.05 1 ISCED-97 = International Standard Classifi cation of Education 2 KldB-92 = German Classifi cation of Professions Note: Estimates for the small subgroups of work statuses and vocational classes have to be interpreted with caution. Some of them are based on very small case numbers (see Appendix, Table A4), and the shapes of the estimated period life tables might differ from the longitudinal survival patterns of the corresponding LES subsamples. Source: own calculations with LES data Life Expectancy by Education, Income and Occupation in Germany • 415 Fig. 4: Deviation of life expectancy at ages 40 and 65 of subpopulations defi ned by specifi c SEP indicators from the total population, Women (a) Education Difference to total LE -4.0 -3.0 -2.0 -1.0 0.0 1.0 2.0 3.0 4.0 Low Medium High Education level Age 40 Age 65 -4.0 -3.0 -2.0 -1.0 0.0 1.0 2.0 3.0 4.0 1 2 3 4 Income quartile Age 40 Age 65 (c) Work status (d) Vocational class -9.0 -7.0 -5.0 -3.0 -1.0 1.0 3.0 5.0 7.0 9.0 E m p l. h ig h ly q u al . M an u al w o rk e rs ( al l) E m p lo y e e s (a ll) S e lf -e m p lo y e d ( al l) P u b lic s e rv an ts ( al l) S e lf -e m p l. e n tr e p r. Age 40 Age 65 -9.0 -7.0 -5.0 -3.0 -1.0 1.0 3.0 5.0 7.0 9.0 U n sk ill e d w o rk e rs O th e r w o rk f o rc e P ro d u ct io n jo b s S e rv ic e s e ct o r T e ch n ic al o cc . S o ci al s e rv ic e Age 40 Age 65 Source: own calculations with LES data Difference to total LE (b) Household net income Difference to total LEDifference to total LE • Marc Luy, Christian Wegner-Siegmundt, Angela Wiedemann, Jeroen Spijker416 corresponding differences are 11.1 years in LE at age 40, 8.3 years in LE at age 65, and 0.120 in survival probability between ages 40 and 65. Across th e main vocational classes with suffi cient case numbers for analysis, we fi nd the same order as among men. The technical occupations have the highest LE (e 40 = 42.6, e 65 = 19.8), the residual category “other workforce” has the lowest (e 40 = 38.0, e 65 = 15.5). The LE gaps are 4.6 years at age 40 and 4.3 years at age 65. The corresponding survival probabilities from age 40 to age 65 are 0.917 and 0.875, respectively. In between these, we fi nd the production jobs (e 40 = 40.7, e 65 = 18.7, 25 p 40 = 0.893) and the service sector (e 40 = 41.4, e 65 = 18.7, 25 p 40 = 0.916). As for male workers, the vocational subclass with the lowest mortality is the social service and education sector. The surplus in LE compared to the vocational subclasses with the highest mortality levels is 7.2 years in e 40 (unskilled workers in production jobs and other work force), 6.6 years in e 65 (other work force), and 0.097 in the probability of surviving from age 40 to 65 (unskilled workers in production jobs). Figure 5 depicts the overall range of survival differences by showing the sur- vival functions for the subgroups of men and women with the lowest and highest mortality levels, together with the corresponding total populations. Among men, these are the vocational subclasses “miners and mineral workers” and “social ser- vice and education”, among women the vocational subclass “unskilled workers in production jobs” and the work status subgroup “self-employed entrepreneurs”. The graphs illustrate the substantial survival differences between the displayed popula- Fig. 5: Survival functions for the male and female subpopulations with lowest and highest life expectancy in comparison to the male and female total populations Source: own calculations with LES data (a) Men (b) Women Survivors in % Survivors in % 0.0 10.0 20.0 30.0 40.0 50.0 60.0 70.0 80.0 90.0 100.0 40 50 60 70 80 90 100 Age Social service All men Miners 0.0 10.0 20.0 30.0 40.0 50.0 60.0 70.0 80.0 90.0 100.0 40 50 60 70 80 90 100 Age Self-empl. entrepr. All women Unskilled workers Life Expectancy by Education, Income and Occupation in Germany • 417 tion subgroups. For instance, age 65 is reached by 90.9 percent of the 40-year-old men working in the social service and education sector, but only by 54.1 percent of miners and mineral workers, a difference of 36.8 percentage points (see the cross- ings of the survival functions with the vertical dotted line in Fig. 3a). Among the dis- played female subpopulations, the corresponding percentages are 98.6 for self-em- ployed entrepreneurs and 84.2 for unskilled workers in production jobs, resulting in a difference of 14.4 percentage points. Also, the vertical dotted lines indicating the median of the survival functions reveal the extent of the survival differences. The age to which 50 percent of the initial 40-year-old individuals survive is 82.5 among the men working in the social service and education sector, but only 66.6 among miners and mineral workers. This leads to a difference in the median survival age of 15.9 years. For women, the corresponding ages are 90.6 for self-employed entre- preneurs and 80.6 for unskilled workers in production jobs, yielding a difference in the median survival age of 10.0 years. 4 Summary and discussion 4.1 Method The estimation of LE for specifi c subpopulations and the differentials between them is a common problem for demographers because offi cial population statistics usu- ally do not include the required detailed data on deaths and the population at risk. A few countries – for instance Austria, Belgium, the Netherlands, Switzerland, the United Kingdom and the Scandinavian countries – link their death registries with their census data, population registers or other registers. This enables research- ers to construct life tables for specifi c subpopulations, which can be identifi ed on the basis of the variables covered in the registries or censuses. For most popula- tions, however, such data do not exist, or the information of interest is not included in the registers. Therefore, longitudinal survey data with registration of deceased participants or mortality follow-ups must be used to derive the desired estimates. The case numbers of these data sources are in most cases too small to derive age- specifi c death rates, even for the larger subpopulations, what prohibits the use of classic life table techniques. Some scholars have employed different approaches to estimate LE on the basis of longitudinal survey data, including proportional haz- ards models (Li et al. 2014; Reuser et al. 2008; Reuser et al. 2009; Reuser et al. 2011), Bayesian Markov chain Monte Carlo methods (Lynch/Brown 2005), multi-state Markov models (Majer et al. 2011; Matthews et al. 2009), hidden Markov models (Van Den Hout et al. 2009) and the so-called “population attributable fraction” PAF (Preston/Stokes 2011). These approaches combine sophisticated statistical meth- ods with more classic (life table) techniques, resulting in particular statistical prop- erties and data requirements. In this paper, we developed an alternative but comparatively simple demographic approach for deriving life tables from survey data with a mortality follow-up. We refer to the method as the “Longitudinal Survival Method” (LSM) because it is based • Marc Luy, Christian Wegner-Siegmundt, Angela Wiedemann, Jeroen Spijker418 on longitudinal survival experiences of survey respondents, which are then trans- formed into a period life table. The transformation procedure of the LSM is basically a direct adoption of the transformation formula of the indirect Modifi ed Orphan- hood Method developed by Luy (2012). The differences are (i) that the observed life spans of respondents between the time of the survey and the time of the mor- tality follow-up can be used with their exact ages, times and lengths, (ii) that the transformation of the survival rates to a default age is not necessary, and (iii) that the reference period for the life table must not be estimated to a time some years before the survey, but can be determined directly to a time between the survey and the mortality follow-up. The applicability of the LSM is not restricted to the LES data used in this study. The method can be applied to all surveys with a mortality follow-up, such as SOEP, MONICA/KORA or SHARE. The LSM requires only three kinds of information for each age group x of the survey population (with x including single ages from x to x+1 or, if necessary, broader age intervals e.g. from x to x+5) and one assumption: • The observed longitudinal survival of the survey respondents )zx ,x(SR ˆ , i.e. the proportion of individuals aged x at the time of the survey who sur- vived until age x+z at the time of mortality follow-up, • the expected longitudinal survival of these survey respondents from age x to zx  , derived from cohort life tables, • the corresponding period survival for the same age interval x to zx  , de- rived from the reference period life table, and, • the assumption that the relationship between cohort and period survival prevalent in the entire population applies equivalently to each subpopula- tion. The specifi c strengths of the LSM include the low demand on data (e.g. no infor- mation on the date of deaths and the age at which they occur is necessary, no spe- cifi c statistical distributions of deaths need to be assumed), its simple applicability, and the estimation of a set of age-specifi c probabilities of dying. Consequently, the LSM is not based on the assumption of constant relative risks, but yields age-spe- cifi c mortality patterns for every subpopulation. The different age patterns in the es- timates become apparent from the varying orders of the subpopulations’ mortality levels according to the used indicators e 40 , 25 p 40 and e 65 , e.g. across the vocational subclasses among men. Another strength of the LSM is that it includes a strong lon- gitudinal component, which makes the estimated differences in period LE between subpopulations less susceptible to distortions caused by period tempo effects (see Luy 2010b). Moreover, when a constancy of the subpopulations’ age patterns of mortality and relative differences to the total population can be assumed, the LSM can also be used to produce estimates for other periods by varying the reference life table for the transformation. This can be particularly valuable when the life table for a specifi c subpopulation produced with the LSM is to be employed for estimat- ing Healthy LE with the basic health information stemming from data for a different period. Finally, the LSM can be used to estimate life tables for any subpopulation Life Expectancy by Education, Income and Occupation in Germany • 419 that can be identifi ed in the underlying data, even when the case numbers are ex- tremely low. 4.2 Results Lifetime inequalities by SEP are an important topic not only for mortality research- ers, but also for policy-makers, physicians, public health workers and the general population. Several German books have been dedicated to this subject, with (trans- lated) titles such as “If you are poor, you have to die earlier” (Oppolzer 1986), “Social inequality before death” (Ritz 1992; Spree 1981) or “Must poor people die earlier?” (Helmert et al. 2000). Mortality differences by occupation already sparked inter- est in the 19th and early 20th centuries (Prinzing 1931; Westergaard 1882). However, these publications only included relative risks for specifi c subgroups defi ned by occupation or work status, as also the most recent studies on this topic do (Helmert 2000; Linke 1990). As already described in the introduction, the situation is simi- lar with regard to the other indicators of SEP, and no comprehensive overview of the variation of LE by SEP in the German population exists. The central aim of this paper was therefore to bridge this gap of knowledge by producing corresponding estimates for the SEP dimensions of education, income and occupation. We found clear evidence for social disparities in mortality for all SEP dimensions that we con- sidered. In the following, we summarise the estimated ranges of longevity among the analysed SEP indicators, i.e. the differences between the subgroups with the highest and lowest mortality levels. According to our estimations for the period 1991/93, based on data of the Ger- man LES, LE at age 40 for men ranges by 6.3 years between individuals with a low and a high education, 5.7 years between the lowest and highest household income quartiles, 4.2 years across the main work statuses, and 13.2 years across the main vocational classes. The maximum difference across all analysed male SEP subpop- ulations is 14.4 years between miners and men working in the social service sector. The proportion of those who reach the classic pension age of 65 years also varies considerably. The spans in percentage points are 14.5 with regard to the education level, 14.1 with regard to household income, 10.1 with regard to the main work sta- tus, and 35.4 with regard to the main vocational class. The difference between the subpopulations with the highest and lowest survival probabilities until age 65 is 36.8 percentage points, referring again to miners and men working in the social service sector. Signifi cant differences are even prevalent in the remaining LE at this classic pension age. The ranges in LE at age 65 are 3.7 years for education, 2.8 years for household income, 2.1 years for the main work status, and 5.3 years for the main vocational class. The maximum range in LE at age 65 across all SEP subgroups is 6.5 years between miners and unskilled workers in production jobs on the one hand, and the social service sector on the other. The corresponding differences among women are smaller, but nonetheless no- table. At age 40, LE varies 2.3 years by education level, 4.2 years by household income, 6.5 years by main work status, and 4.6 years by main vocational class. The maximum difference across all analysed female SEP subpopulations is 11.5 years, • Marc Luy, Christian Wegner-Siegmundt, Angela Wiedemann, Jeroen Spijker420 between self-employed entrepreneurs and unskilled workers in production jobs. Due to the generally lower mortality of women in middle ages, the differences in survival until age 65 are less distinct than among men. The proportions of 40-year- old women reaching age 65 vary by 2.0 percentage points by education, 6.3 by household income, 6.0 by main work status, and 4.2 by main vocational class. The largest difference between the two mentioned subgroups with minimum and maxi- mum mortality levels is 14.4 percentage points. As another consequence of the relatively low female mortality between ages 40 and 65, the variations in LE at age 65 do not differ considerably from those at age 40. With regard to education, LE at age 65 varies by 2.0 years, 2.6 years by household income, 5.1 years by main work status, and 4.3 years by main vocational class. The maximum difference in LE at age 65 across all analysed female SEP subpopulations is 9.3 years between the self- employed entrepreneurs and the main vocational class “other work force”. We are confi dent that these results are reliable at least for the main SEP sub- groups for four reasons: • the LES data is representative for the total population of western Germany, • the mortality follow-up of the LES is representative for the total population of western Germany (and our sensitivity analysis suggests that this is likely to also hold true for the LES subpopulations), • subgroup-specifi c age patterns of mortality are taken into account, and • our estimates for LE differentials by education level are similar to corre- sponding data for the neighbouring countries. The last point is an especially strong indicator of the results’ validity. Calcula- tions for Austria on the basis of Austrian census data with one-year mortality follow- ups published by Statistics Austria provide differences in LE (at age 35) between the same ISCED levels of high and low education of 6.1 years among men and 3.3 years among women, averaged for the years 1981-2001 (Klotz/Asamer 2014). Cor- responding estimates for the German-speaking part of Switzerland on the basis of the Swiss National Cohort (a linkage of the 1990 census with death certifi cate data registered between 1990 and 1997) reveal a difference in LE at age 40 of 6.3 years among men and 3.3 years among women (Spoerri et al. 2006). Thus, our estimates for the German population provide an almost identical education-specifi c difference in LE among men and a gap approximately one year smaller among women. Similar LE differentials for education among men in the period between 1980 and 2000 can be found, for instance, in the Czech Republic (Shkolnikov et al. 2006), Finland (Shkolnikov et al. 2006; Valkonen et al. 1993; Valkonen et al. 1997) and among the white US population (Crimmins/Saito 2001; Meara et al. 2008). Slightly smaller differences (of about fi ve years) are reported for Belgium (Deboosere et al. 2009), Denmark (Brønnum-Hansen/Baadsgaard 2012) and Lithuania (Kalediene/ Petrauskiene 2000). Considerably larger differentials are prevalent in Russia and Es- tonia, with a LE difference at young adulthood of more than ten years (Shkolnikov et al. 2006). In general, the same regional differences can be found among the female populations. According to our estimates, however, German women are closer to Life Expectancy by Education, Income and Occupation in Germany • 421 the populations with smaller education-specifi c differences in LE, such as Denmark and Lithuania, whereas the female Belgian population is – according to the estima- tions of Deboosere et al. (2009) – closer to the populations with medium-sized dif- ferentials (Austria, Czech Republic, Finland, the white US population, Switzerland). Also among women, Russia and Estonia exhibit signifi cantly larger inequalities in LE by education level. Our results for income and occupation are in line with fi ndings for other populations as well, for instance with regard to the order of professions’ or work status groups’ LE levels and the generally larger differentials among men (see e.g. Brønnum-Hansen 2000; Burström et al. 2005; Hattersley 1999; Valkonen et al. 1993). Nonetheless, the estimates for subclasses have to be interpreted with cau- tion. Some of them are based on very small case numbers. Although the estimated LE values refl ect the observed average survival time of the LES subsamples, it can happen that the shape of the estimated period life table differs from the longitudinal survival pattern. This holds true, for instance, for the male workers in wood and plastics processing and in painting, varnishing and similar occupations. Our results for the German population imply that the variations in survival and longevity are much larger with respect to occupation than with respect to educa- tion and household income. The spans are more than double among men and ap- proximately tripled among women. This is due to the fact that the subpopulations defi ned by specifi c occupations are more selected than those defi ned by education and income. When assessing these differences from an analytical point of view, it is also important to note that our estimates are based on empirical life tables – thus refl ecting the actual mortality of the survey sample – and not on theoretical ones. The former can include a 100 percent survival rate for some ages when no case of death is observed, while a theoretical life table never features a zero percent prob- ability of dying. Among the subpopulations analysed in our study, this is the case among female self-employed entrepreneurs between ages 40 and 56, for instance (see Fig. 5b). Nonetheless, the specifi cs of the resulting survival curve are not as unrealistic as they might seem at fi rst glance. The survivorship function tells us that almost all (98.6 percent) female self-employed entrepreneurs reach age 65, and that approximately one in ten (10.8 percent) reaches age 100. Although this survival curve might look exceptional, these fi gures are not totally impossible. Another fact to keep in mind is that the presented estimates refer to the early 1990s, and are thus not fully refl ective of current developments. However, changes in mortality – especially the relative differences between subpopulations – do not occur very suddenly. This is confi rmed by the trends of LE by education level in Austria from 1981 to 2011, where only minor variations took place over time, with the tendency of a slight increase among men and a small decrease among women (Klotz/Asamer 2014). Unger and Schulze (2013) draw a similar picture for the Ger- man population in their analysis of education-specifi c differentials in LE, which did not change between 1989 and 2009 for either sex. Thus, it is likely that the currently prevalent SEP-specifi c differences in LE among German women and men are close to those presented in this paper. Finally, it is important to note that this paper was neither intended to analyse the causal mechanisms behind the SEP-specifi c differences in LE, nor did we aim • Marc Luy, Christian Wegner-Siegmundt, Angela Wiedemann, Jeroen Spijker422 at contributing to the question of which dimension of SEP affects the mortality dif- ferentials among the others. Obviously, the different dimensions of SEP are inter- related, but they still have independent effects on mortality (Geyer/Peter 2000). The network of interrelations and causal factors is very complex (see e.g. Schneider 2008; Spijker 2004) and it is possible that the so-called “status syndrome” is only driven by the factors education, income and occupation to a minor extent, but is instead rooted in an overarching mechanism determined by socially-produced hier- archical differences in individuals’ autonomy and opportunities for social participa- tion (Marmot 2004). All these questions are still unanswered, and more research is necessary to provide decisive knowledge for developing appropriate strategies for reducing the existing inequalities in LE. 5 Conclusions The LSM provides improved estimates for differentials in LE of subgroups on the basis of survey data with a mortality follow-up. The particular strength of the LSM is the incorporation of age-specifi c mortality patterns for each subpopulation. The only necessary assumption is that the relationship between cohort and period sur- vival prevalent in the entire population applies equivalently to each subpopulation. Application of the LSM can be extended to estimate cohort life tables as well as multi-state life tables, and it enables researchers to produce estimates even for subpopulations with extremely low case numbers. However, the quality of the data must always be taken into account. If the mortality follow-up is not representative of the population’s mortality, no valid estimates can be derived with the LSM (or any other method). Thus, our assessment is in line with Charafeddine et al. (2014), who concluded in their comparative analysis of Healthy LE estimated with Belgium’s census and survey data including a mortality follow-up that the latter is a useful and valid data source in the absence of population-wide data when it is of good quality. This paper illustrates the variation of LE in Germany by SEP. No information about causality can be extracted from these results, and it has to be kept in mind that the estimates are derived in a cross-sectional setting which does not necessar- ily refl ect the factual longitudinal survival. Nonetheless, the results of this study are useful for all researchers, policy-makers and other parties interested in the extent of differences in lifetime by SEP across the German population. In particular, the presented fi gures provide an interesting subject of discussion for the ongoing de- bate about the implementation of an increasing pension age, which is considered to be necessary by many scholars in order to reduce the consequences of the de- mographic change (for a special focus on the German population see e.g. Birg 2001; Schimany 2003; Schmid et al. 2000). Moreover, the results presented in this study provide doctors, scientists and public health workers with insights into the most disadvantaged SEP subpopulations and the potential extent of their disadvantages in both longevity and mortality. Life Expectancy by Education, Income and Occupation in Germany • 423 Acknowledgements This research was supported by the European Research Council, within the Euro- pean Community’s Seventh Framework Programme (FP7/2007–2013), ERC Grant Agreement No. 262663 (HEMOX) and ERC Grant Agreement No. 323947 (Re-Age- ing). We thank Karla Gärtner (German Federal Institute for Population Research, BiB) for providing us with the LES codes for the specifi c occupations which are not in- cluded in the publically available data set. Author contributions ML developed the idea of the study as well as the method and wrote most parts of the text. CWS helped in refi ning the method, performed the analyses and contrib- uted to the text. AW contributed to this study by collecting literature on mortality differences by SEP in Germany and giving inputs to the text. JS performed a sys- tematic review of the existing methods for estimating mortality from longitudinal survey data and commented on all parts of the paper. References Birg, Herwig 2001: Die demographische Zeitenwende. Der Bevölkerungsrückgang in Deutschland und Europa. München: Beck. Brass, William 1975: Methods for estimating fertility and mortality from limited and de- fective data. Chapel Hill: University of North Carolina. Brønnum-Hansen, Henrik 2000: Socioeconomic differences in health expectan- cy in Denmark. In: Scandinavian Journal of Public Health 28,3: 194-199 [doi: 10.1177/14034948000280030801]. Brønnum-Hansen, Henrik; Baadsgaard, Mikkel 2012: Widening social inequality in life expectancy in Denmark. A register-based study on social composition and mortality trends for the Danish population. In: BMC Public Health 12,1: 994 [doi: 10.1186/1471- 2458-12-994]. Burström, Kristina; Johannesson, Magnus; Diderichsen, Finn 2005: Increasing socio- economic inequalities in life expectancy and QALYs in Sweden 1980-1997. In: Health Economics 14,8: 831-850 [doi: 10.1002/hec.977]. Charafeddine, Rana et al. 2014: Using mortality follow-up of surveys to estimate so- cial inequalities in healthy life years. In: Population Health Metrics 12,1: 13 [doi: 10.1186/1478-7954-12-13]. Chiang, Chin Long 1984: The life table and its applications. Malabar, Florida: Krieger. Corsini, Veronica 2010: Highly educated men and women likely to live longer. Life ex- pectancy by educational attainment. Statistics in focus 24/2010. Brussels: Eurostat. Crimmins, Eileen M.; Saito, Yasuhiko 2001: Trends in healthy life expectancy in the Unit- ed States, 1970-1990: gender, racial, and educational differences. In: Social Science & Medicine 52,11: 1629-1641 [doi: 10.1016/S0277-9536(00)00273-2]. • Marc Luy, Christian Wegner-Siegmundt, Angela Wiedemann, Jeroen Spijker424 Deboosere, Patrick; Gadeyne, Sylvie; Van Oyen, Herman 2009: The 1991-2004 evolution in life expectancy by educational level in Belgium based on linked census and popu- lation register data. In: European Journal of Population 25,2: 175-196 [doi: 10.1007/ s10680-008-9167-5]. Doblhammer, Gabriele; Muth, Elena; Kruse, Anne 2008: Lebenserwartung in Deutsch- land: Trends, Prognose, Risikofaktoren und der Einfl uss ausgewählter Medizininno- vationen. Abschlussbericht für den Verband Forschender Arzneimittelhersteller e.V. Rostock: Rostocker Zentrum zur Erforschung des Demografi schen Wandels. Festy, Patrick 1995: Adult mortality and proportions orphaned in Austria in 1991. In: Population: An English Selection 7: 232-238. Gärtner, Karla 2001: Lebensstile und ihr Einfl uss auf Gesundheit und Lebenserwartung. Der Lebenserwartungssurvey des BiB. Materialien zur Bevölkerungswissenschaft 102a. Wiesbaden: Bundesinstitut für Bevölkerungsforschung. Geyer, Siegfried; Peter, Richard 2000: Income, occupational position, qualifi cation and health inequalities – competing risks? (Comparing indicators of social status). In: Jour- nal of Epidemiology and Community Health 54,4: 299-305 [doi: 10.1136/jech.54.4.299]. Hattersley, Lin 1999: Trends in life expectancy by social class – an update. In: Health Statistics Quarterly 2, Summer: 16-24. Helmert, Uwe 2000: Der Einfl uß von Beruf und Familienstand auf die Frühsterblichkeit von männlichen Krankenversicherten. Eine Längsschnittanalyse von Routinedaten der Gmünder Ersatzkasse von 1989 bis 1997. In: Helmert, Uwe et al. (Eds.): Müssen Arme früher sterben? Soziale Ungleichheit und Gesundheit in Deutschland. Weinheim, Mün- chen: Juventa: 243-268. Helmert, Uwe et al. 2000: Müssen Arme früher sterben? Soziale Ungleichheit und Ge- sundheit in Deutschland. Weinheim, München: Juventa. Hill, Kenneth; Choi, Yoonjoung; Timæus, Ian M. 2005: Unconventional approaches to mortality estimation. In: Demographic Research 13,12: 281-300 [doi: 10.4054/Dem- Res.2005.13.12]. Hill, Kenneth; Zlotnik, Hania; Trussell, James 1983: Manual X. Indirect techniques for demographic estimation. Population Studies 81. New York: United Nations. International Labour Offi ce 2012: International Standard Classifi cation of Occupations ISCO-8. Vol. 1: Structure, group defi nitions and corresponding tables. Geneva: Inter- national Labour Offi ce. Kalediene, Ramune; Petrauskiene, Jadvyga 2000: Inequalities in life expectancy in Lithu- ania by level of education. In: Scandinavian Journal of Public Health 28,1: 4-9 [doi: 10.1177/140349480002800103]. Kibele, Eva U. B.; Jasilionis, Domantas; Shkolnikov, Vladimir M. 2013: Widening socio- economic differences in mortality among men aged 65 years and older in Germany. In: Journal of Epidemiology and Community Health 67,5: 453-457 [doi: 10.1136/jech- 2012-201761]. Klein, Thomas 1996: Mortalität in Deutschland – Aktuelle Entwicklungen und soziale Un- terschiede. In: Zapf, Wolfgang; Schupp, Jürgen; Habich, Roland (Eds.): Lebenslagen im Wandel: Sozialberichterstattung im Längsschnitt. Frankfurt/New York: Campus: 366-377. Klein, Thomas 1999: Soziale Determinanten der aktiven Lebenserwartung. In: Zeitschrift für Soziologie 28,6: 448-464. Life Expectancy by Education, Income and Occupation in Germany • 425 Klein, Thomas; Schneider, Sven; Löwel, Hannelore 2001: Bildung und Mortalität. Die Be- deutung gesundheitsrelevanter Aspekte des Lebensstils. In: Zeitschrift für Soziologie 30,5: 385-400. Klotz, Johannes; Asamer, Eva-Maria 2014: Bildungsspezifi sche Sterbetafeln 2006/2007 sowie 2011/2012. In: Statistische Nachrichten 2014,3: 209-214. Kunst, Anton 1997: Cross-national comparisons of socio-economic differences in mor- tality. Rotterdam: Department of Public Health, Erasmus University. Lampert, Thomas; Kroll, Lars Eric 2014: Soziale Unterschiede in der Mortalität und Le- benserwartung. In: GBE Kompakt 5,2: 1-13. Lampert, Thomas; Kroll, Lars Eric; Dunkelberg, Annalena 2007: Soziale Ungleichheit der Lebenserwartung in Deutschland. In: Aus Politik und Zeitgeschichte 42: 11-18. Lampert, Thomas; Kroll, Lars Erik 2006: Einkommensdifferenzen in der Gesundheit und Lebenserwartung – Quer-und Längsschnittbefunde des Sozio-oekonomischen Panels (SOEP). In: Das Gesundheitswesen 68,4: 219-230 [doi: 10.1055/s-2006-926638]. Li, Kuanrong; Husing, Anika; Kaaks, Rudolf 2014: Lifestyle risk factors and residual life expectancy at age 40: a German cohort study. In: BMC Medicine 12,1: 59 [doi: 10.1186/1741-7015-12-59]. Linke, Wilfried 1990: Differentielle Sterblichkeit nach Berufen – Eine Auswertung der Beschäftigtenstatistiken 1984 und 1985. In: Zeitschrift für Bevölkerungswissenschaft 16,1: 29-51. Luy, Marc 2006: Differentielle Sterblichkeit: die ungleiche Verteilung der Lebenserwar- tung in Deutschland. In: Sozialverband VdK. Bayern (Ed.): Soziale Verunsicherung ohne Ende? Das politische System setzt die Bürger auch weiter unter Druck. München: VdK-Dimetria: 61-82. Luy, Marc 2010a: A classifi cation of the nature of mortality data underlying the estimates for the 2004 and 2006 United Nations’ World Population Prospects. In: Comparati- ve Population Studies – Zeitschrift für Bevölkerungswissenschaft 35,2: 315-334 [doi: 10.4232/10.CPoS-2010-08en]. Luy, Marc 2010b: Tempo effects and their relevance in demographic analysis. In: Com- parative Population Studies – Zeitschrift für Bevölkerungswissenschaft 35,3: 415-446 [doi: 10.4232/10.CPoS-2010-11en]. Luy, Marc 2012: Estimating mortality differences in developed countries from survey in- formation on maternal and paternal orphanhood. In: Demography 49,2: 607-627 [doi: 10.1007/s13524-012-0101-4]. Luy, Marc; Di Giulio, Paola 2005: Der Einfl uss von Verhaltensweisen und Lebensstilen auf die Mortalitätsdifferenzen der Geschlechter. In: Gärtner, Karla; Grünheid, Evelyn; Luy, Marc (Eds.): Lebensstile, Lebensphasen, Lebensqualität. Interdisziplinäre Ana- lysen von Gesundheit und Sterblichkeit aus dem Lebenserwartungssurvey des BiB. Schriftenreihe des Bundesinstituts für Bevölkerungsforschung 36. Wiesbaden: VS Verlag für Sozialwissenschaften: 365-392. Luy, Marc; Di Giulio, Paola; Caselli, Graziella 2011: Differences in life expectancy by edu- cation and occupation in Italy, 1980-94: indirect estimates from maternal and paternal orphanhood. In: Population Studies 65,2: 137-155 [doi: 10.1080/00324728.2011.568192]. Lynch, Scott M.; Brown, J. Scott 2005: A new approach to estimating life tables with covariates and constructing interval estimates of life table quantities. In: Sociological Methodology 35,1: 177-225 [doi: 10.1111/j.0081-1750.2006.00168.x]. Mackenbach, Johan P. et al. 1997: Socioeconomic inequalities in morbidity and mortal- ity in western Europe. In: The Lancet 349,9066: 1655-1659. • Marc Luy, Christian Wegner-Siegmundt, Angela Wiedemann, Jeroen Spijker426 Majer, Istvan M. et al. 2011: Socioeconomic inequalities in life and health expectancies around offi cial retirement age in 10 Western-European countries. In: Journal of Epide- miology and Community Health 65,11: 972-979 [doi: 10.1136/jech.2010.111492]. Marmot, Michael 2004: The status syndrome. How social standing affects our health and longevity. London: Bloomsbury. Matthews, Fiona E. et al. 2009: Education differences in life expectancy with cognitive impairment. In: The Journals of Gerontology Series A: Biological Sciences and Medi- cal Sciences 64A,1: 125-131 [doi: 10.1093/gerona/gln003]. Meara, Ellen R.; Richards, Seth; Cutler, David M. 2008: The gap gets bigger: changes in mortality and life expectancy, by education, 1981-2000. In: Health Affairs 27,2: 350- 360 [doi: 10.1377/hlthaff.27.2.350]. Moultrie, Tom et al. 2013: Tools for demographic estimation. Paris: IUSSP. Oppolzer, Alfred 1986: Wenn du arm bist, musst du früh sterben. Hamburg: VSA. Perna, Laura et al. 2010: Socio-economic differences in life expectancy among persons with diabetes mellitus or myocardial infarction: results from the German MONICA/ KORA study. In: BMC Public Health 10: 135 [doi: 10.1186/1471-2458-10-135]. Preston, Samuel H.; Stokes, Andrew 2011: Contribution of obesity to international differ- ences in life expectancy. In: American Journal of Public Health 101,11: 2137-2143 [doi: 10.2105/AJPH.2011.300219]. Prinzing, Friedrich 1931: Handbuch der medizinischen Statistik. Zweiter Halbband: Die Sterbefälle. Jena: Gustav Fischer. Reuser, Mieke; Bonneux, Luc G.; Willekens, Frans J. 2009: Smoking kills, obesity disa- bles: a multistate approach of the US Health and Retirement Survey. In: Obesity 17,4: 783-789 [doi: 10.1038/oby.2008.640]. Reuser, Mieke; Bonneux, Luc; Willekens, Frans J. 2008: The burden of mortality of obe- sity at middle and old age is small. A life table analysis of the US Health and Re- tirement Survey. In: European Journal of Epidemiology 23,9: 601-607 [doi: 10.1007/ s10654-008-9269-8]. Reuser, Mieke; Willekens, Frans J.; Bonneux, Luc 2011: Higher education delays and shortens cognitive impairment. A multistate life table analysis of the US Health and Retirement Study. In: European Journal of Epidemiology 26,5: 395-403 [doi: 10.1007/ s10654-011-9553-x]. Ritz, Hans-Günther 1992: Soziale Ungleichheit vor Tod in der Bundesrepublik Deutsch- land. Gesundheit – Arbeit – Medizin 3. Bremerhaven: Wirtschaftsverlag NW, Verlag für neue Wissenschaft GmbH. Salzmann, Thomas; Bohk, Christina 2008: Überprüfung der im Rahmen des „Lebenser- wartungssurveys“ gemessenen Sterblichkeit auf Bevölkerungsrepräsentativität unter Berücksichtigung rechts- und intervallzensierter Ereignisse mit dem Konzept „Relati- ve Survival“. In: Zeitschrift für Bevölkerungswissenschaft 33,2: 121-152 [doi: 10.1007/ s12523-009-0008-3]. Schimany, Peter 2003: Die Alterung der Gesellschaft. Ursachen und Folgen des demo- graphischen Umbruchs. Frankfurt am Main: Campus. Schmid, Josef; Heigl, Andreas; Mai, Ralf 2000: Sozialprognose. Die Belastung der nach- wachsenden Generation. München: Olzog. Schneider, Sven 2008: Der Schichtgradient von Morbidität und Mortalität. Vorschlag für ein theoretisches Erklärungsmodell gesundheitlicher Ungleichheit. In: Österreichische Zeitschrift für Soziologie 33,1: 43-66 [doi: 10.1007/s11614-008-0003-2]. Life Expectancy by Education, Income and Occupation in Germany • 427 Schnell, Rainer; Trappmann, Mark 2006: Konsequenzen der Panelmortalität im SOEP für Schätzungen der Lebenserwartung. Arbeitspapier 2/2006. Konstanz: Universität Konstanz. Shkolnikov, Vladimir M. et al. 2006: The changing relation between education and life expectancy in central and eastern Europe in the 1990s. In: Journal of Epidemiology and Community Health 60,10: 875-881 [doi: 10.1136/jech.2005.044719]. Shkolnikov, Vladimir M. et al. 2008: Length of life and the pensions of fi ve million retired German men. In: European Journal of Public Health 18,3: 264-269 [doi: 10.1093/eur- pub/ckm102]. Spijker, Jeroen 2004: Socioeconomic determinants of regional mortality differences in Europe. Amsterdam: Dutch University Press. Spoerri, Adrian et al. 2006: Educational inequalities in life expectancy in the German speaking part of Switzerland between 1990 and 1997: Swiss National Cohort. In: Swiss Medical Weekly 136: 145-148 [doi: 10.5167/uzh-81153]. Spree, Reinhard 1981: Soziale Ungleichheit vor Krankheit und Tod. Zur Sozialgeschichte des Gesundheitsbereichs im Deutschen Kaiserreich. Göttingen: Vandenhoeck & Rup- recht. Statistisches Bundesamt 1992: Klassifi kation der Berufe 1992 (KldB 92). Gliederungs- struktur bis zur 4 Steller-Ebene. Wiesbaden: Statistisches Bundesamt. Statistisches Bundesamt 2006: Generationensterbetafeln für Deutschland. Modellrech- nungen für die Geburtsjahrgänge 1871-2004. Wiesbaden: Statistisches Bundesamt. Thatcher, A. Roger; Kannisto, Väinö; Vaupel, James W. 1998: The force of mortality at ages 80 to 120. Odense Monographs on Population Aging 5. Odense: Odense Univer- sity Press. Timæus, Ian M. 1991: Measurement of adult mortality in less developed countries: a comparative review. In: Population Index 57,4: 552-568. Unger, Rainer; Schulze, Alexander 2013: Können wir (alle) überhaupt länger arbeiten? Trends in der gesunden Lebenserwartung nach Sozialschicht in Deutschland. In: Com- parative Population Studies – Zeitschrift für Bevölkerungswissenschaft 38,3: 545-564 [doi: 10.4232/10.CPoS-2013-03de]. United Nations Educational Scientifi c and Cultural Organization 1996: International Standard Classifi cation of Education ISCED 1997. Montreal: UNESCO-UIS. Valkonen, Tapani et al. 1993: Socio-economic mortality differences in Finland 1981-90. Statistics Finland Population. Helsinki: Statistics Finland. Valkonen, Tapani; Sihvonen, Ari-Pekka; Lahelma, Eero 1997: Health expectancy by level of education in Finland. In: Social Science & Medicine 44,6: 801-808. Van Den Hout, Ardo; Jagger, Carol; Matthews, Fiona E. 2009: Estimating life expec- tancy in health and ill health by using a hidden Markov model. In: Journal of the Royal Statistical Society: Series C (Applied Statistics) 58,4: 449-465 [doi: 10.1111/j.1467- 9876.2008.00659.x]. von Gaudecker, Hans-Martin 2006: Differentielle Sterblichkeit in der GRV: Problem- aufriss und erste Berechnungen. In: Deutsche Rentenversicherung Bund (Ed.): For- schungsrelevante Daten der Rentenversicherung: Bericht vom zweiten Workshop des Forschungsdatenzentrums der Rentenversicherung (FDZ-RV) vom 27. bis 29. Juni 2005 in Würzburg. DRV-Schriften 55. Bad Homburg: Wdv Gesellschaft für Medien und Kommunikation: 242-252. • Marc Luy, Christian Wegner-Siegmundt, Angela Wiedemann, Jeroen Spijker428 von Gaudecker, Hans-Martin; Scholz, Rembrandt D. 2007: Differential mortality by life- time earnings in Germany. In: Demographic Research 17,4: 83-108 [doi: 10.4054/Dem- Res.2007.17.4]. Westergaard, Harald 1882: Die Lehre von der Mortalität und Morbilität. Anthropolo- gisch-statistische Untersuchungen. Jena: Gustav Fischer. Wiedemann, Angela 2012: Die Mortalität erwachsener Immigranten in Deutschland: Eine indirekte Schätzung der Sterblichkeitsunterschiede zwischen türkisch- und italie- nischstämmigen Einwanderern und der deutschen Bevölkerung. Diplomarbeit: Uni- versität Wien. Zajacova, Anna; Goldman, Noreen; Rodríguez, Germán 2009: Unobserved heterogene- ity can confound the effect of education on mortality. In: Mathematical Population Studies 16,2: 153-173 [doi: 10.1080/08898480902790528]. Date of submission: 05.05.2015 Date of acceptance: 07.09.2015 Dr. Marc Luy (), Dr. Christian Wegner-Siegmundt, Angela Wiedemann. Wittgenstein Centre for Demography and Global Human Capital (IIASA, OEAW/VID, WU), Vienna Institute of Demography of the Austrian Academy of Sciences. Vienna, Austria. E-mail: mail@marcluy.eu, christian.wegner@oeaw.ac.at, angela.wiedemann@oeaw.ac.at URL: http://www.marcluy.eu http://www.oeaw.ac.at/vid/staff/staff_christian_wegner.shtml http://www.oeaw.ac.at/vid/staff/staff_angela_wiedemann.shtml Dr. Jeroen Spijker. Centre d’Estudis Demogràfi cs. Bellaterra, Spain E-mail: jspijker@ced.uab.es URL: http://www.ced.uab.es/content/spijker-jeroen Life Expectancy by Education, Income and Occupation in Germany • 429 Appendix Tab. A1: Average health status at the time of fi rst survey of respondents with known and unknown survival status at the time of mortality follow-up (second survey) and unknown-known health status ratio of average health statuses for all analysed subsamples, Men SEP indicator / subpopulation Survival status Unknown- Known Unknown known ratio All men 2.67 2.66 0.99 Education according to ISCED-971 Low (ISCED 1-2) 2.94 2.95 1.00 Medium (ISCED 3-4) 2.70 2.69 1.00 High (ISCED 5-6) 2.46 2.38 0.97 Household net income 1st quartile (below € 895) 2.85 2.85 1.00 2nd quartile (€ 895–1,406) 2.73 2.68 0.98 3rd quartile (€ 1,406–1,917) 2.63 2.70 1.03 4th quartile (€ 1,917 and more) 2.48 2.41 0.97 Work status Manual workers (all) 2.85 2.79 0.98 Unskilled or semi-skilled workers 2.95 2.95 1.00 Skilled workers 2.82 2.72 0.96 Master craftsmen, overseers, foremen 2.67 2.42 0.91 Employees (all) 2.60 2.51 0.96 Simple tasks 3.12 2.90 0.93 Qualifi ed tasks 2.67 2.68 1.00 Highly qualifi ed tasks 2.47 2.30 0.93 Public servants (all) 2.50 2.56 1.02 Simple or medium service 2.60 2.39 0.92 Higher or senior service 2.40 2.61 1.09 Self-employed workers (all) 2.57 2.57 1.00 Entrepreneurs 2.66 2.67 1.01 Freelancers 2.45 2.17 0.89 Farmers 2.49 2.57 1.03 Vocational classes acc. to KldB-92 (Germany)2 Jobs in (animal) husbandry, forestry, horticulture 2.62 2.62 1.00 Miners and mineral workers 3.51 4.06 1.16 Production jobs (all) 2.80 2.76 0.98 Metal production and metal working 2.76 2.76 1.00 Metal constr., mech. engineering and similar 2.66 2.68 1.01 Electrical engineering 2.74 2.61 0.95 • Marc Luy, Christian Wegner-Siegmundt, Angela Wiedemann, Jeroen Spijker430 SEP indicator / subpopulation Survival status Unknown- Known Unknown known ratio Textile sector 3.28 2.71 0.83 Foodstuffs sector 2.70 2.53 0.94 Structural and civil engineering 2.96 2.40 0.81 Wood and plastics processing 2.86 2.94 1.03 Painting, varnishing and similar occupations 2.60 2.60 1.00 Unskilled workers without specifi cation 2.96 3.11 1.05 Technical occupations (all) 2.50 2.49 1.00 Engineering, chemistry, physics, maths 2.44 2.23 0.92 Technicians, technical specialists 2.55 2.56 1.00 Service sector (all) 2.61 2.62 1.00 Goods und service marketing 2.69 2.59 0.96 Transport industry 2.77 2.63 0.95 Organisation, administration, clerical jobs 2.57 2.54 0.99 Public and private security sector 2.53 2.40 0.95 Health sector 2.35 2.23 0.95 Social service and education 2.45 2.77 1.13 Other service occupations 2.60 2.71 1.04 Other work force 2.94 3.35 1.14 Tab. A1: Continuation Notes: Health status is measured by means of self-perceived health (“How is your health in general?” Very good=1 / good=2 / satisfactory=3 / not so good=4 / poor=5); Average health statuses are age standardized with the total male LES population in 5-year age groups as standard (known + unknown survival status); Subpopulations with case numbers of fi ve or less among individuals with known or unknown survival status are marked in italic; None of the differences in average health status between individuals with known and unknown survival status at the time of mortality follow-up is statistically signifi cant (95 percent confi dence level); Figures exclude LES participants below age 33 at the time of the fi rst survey. 1 ISCED-97 = International Standard Classifi cation of Education 2 KldB-92 = German Classifi cation of Professions Source: own calculations with LES data Life Expectancy by Education, Income and Occupation in Germany • 431 Tab. A22: Average health status at the time of fi rst survey of respondents with known and unknown survival status at the time of mortality follow-up (second survey) and unknown-known health status ratio of average health statuses for all analysed subsamples, Women SEP indicator / subpopulation Survival status Unknown- Known Unknown known ratio All women 2.78 2.82 1.01 Education according to ISCED-971 Low (ISCED 1-2) 2.86 2.88 1.01 Medium (ISCED 3-4) 2.76 2.81 1.02 High (ISCED 5-6) 2.53 2.39 0.94 Household net income 1st quartile (below € 895) 2.91 2.86 0.98 2nd quartile (€ 895–1,406) 2.85 2.84 1.00 3rd quartile (€ 1,406–1,917) 2.78 2.92 1.05 4th quartile (€ 1,917 and more) 2.60 2.60 1.00 Work status Manual workers (all) 2.90 2.93 1.01 Unskilled or semi-skilled workers 2.91 2.92 1.00 Skilled workers 2.88 2.92 1.02 Master craftsmen, overseers, foremen 2.71 2.70 1.00 Employees (all) 2.71 2.72 1.00 Simple tasks 2.80 2.75 0.98 Qualifi ed tasks 2.63 2.66 1.01 Highly qualifi ed tasks 2.66 2.61 0.98 Public servants (all) 2.71 2.38 0.88 Simple or medium service 2.77 2.40 0.87 Higher or senior service 2.54 2.24 0.88 Self-employed workers (all) 2.81 2.83 1.01 Entrepreneurs 2.67 2.73 1.02 Freelancers 2.60 2.91 1.12 Farmers 3.20 3.05 0.96 Vocational classes acc. to KldB-92 (Germany)2 Jobs in (animal) husbandry, forestry, horticulture 2.79 3.05 1.10 Miners and mineral workers 3.00 2.00 0.67 Production jobs (all) 2.88 2.89 1.00 Metal production and metal working 3.10 3.00 0.97 Metal constr., mech. engineering and similar 3.02 3.36 1.11 Electrical engineering 2.19 3.80 1.74 • Marc Luy, Christian Wegner-Siegmundt, Angela Wiedemann, Jeroen Spijker432 SEP indicator / subpopulation Survival status Unknown- Known Unknown known ratio Textile sector 2.83 2.74 0.97 Foodstuffs sector 2.79 2.97 1.07 Structural and civil engineering --- 3.00 --- Wood and plastics processing 3.00 5.00 1.67 Painting, varnishing and similar occupations 3.68 --- --- Unskilled workers without specifi cation 2.87 2.95 1.03 Technical occupations (all) 2.63 1.86 0.71 Engineering, chemistry, physics, maths 2.35 --- --- Technicians, technical specialists 2.65 1.86 0.70 Service sector (all) 2.74 2.76 1.01 Goods und service marketing 2.81 2.81 1.00 Transport industry 2.76 2.72 0.98 Organisation, administration, clerical jobs 2.71 2.70 1.00 Public and private security sector 2.79 2.66 0.95 Health sector 2.58 2.57 1.00 Social service and education 2.66 2.52 0.95 Other service occupations 2.83 2.85 1.00 Other work force 2.74 2.68 0.98 Tab. A2: Continuation Notes: Health status is measured by means of self-perceived health (“How is your health in general?” Very good=1 / good=2 / satisfactory=3 / not so good=4 / poor=5); Average health statuses are age standardized with the total female LES population in 5-year age groups as standard (known + unknown survival status); Subpopulations with case num- bers of fi ve or less among individuals with known or unknown survival status are marked in italic; None of the differences in average health status between individuals with known and unknown survival status at the time of mortality follow-up is statistically signifi cant (95 percent confi dence level); Figures exclude LES participants below age 33 at the time of the fi rst survey. 1 ISCED-97 = International Standard Classifi cation of Education 2 KldB-92 = German Classifi cation of Professions Source: own calculations with LES data Life Expectancy by Education, Income and Occupation in Germany • 433 Tab. A3: Number of respondents R, number of deaths D, average age at the time of fi rst survey x, average survival time z until the mortality follow-up, and respondents’ survival rate for the total subsamples, Men SEP indicator / subpopulation R D x z ŜR All men 3,364 651 50.03 13.11 0.8065 Education according to ISCED-971 Low (ISCED 1-2) 448 123 52.51 13.09 0.7254 Medium (ISCED 3-4) 2,165 442 50.21 13.12 0.7958 High (ISCED 5-6) 748 85 48.01 13.08 0.8864 Household net income 1st quartile (below € 895) 747 232 52.86 13.12 0.6894 2nd quartile (€ 895–1,406) 1,091 196 49.63 13.13 0.8203 3rd quartile (€ 1,406–1,917) 744 115 48.52 13.06 0.8454 4th quartile (€ 1,917 and more) 692 89 49.16 13.10 0.8714 Work status Manual workers (all) 1,282 305 50.31 13.11 0.7621 Unskilled or semi-skilled workers 471 139 51.69 13.12 0.7049 Skilled workers 558 110 49.03 13.12 0.8029 Master craftsmen, overseers, foremen 253 56 50.60 13.06 0.7787 Employees (all) 1,121 168 49.34 13.10 0.8501 Simple tasks 104 30 51.40 13.06 0.7115 Qualifi ed tasks 459 70 48.60 13.15 0.8475 Highly qualifi ed tasks 558 68 49.57 13.07 0.8781 Public servants (all) 392 73 51.25 13.09 0.8138 Simple or medium service 218 46 50.40 13.13 0.7890 Higher or senior service 277 35 48.60 13.14 0.8736 Self-employed workers (all) 495 81 49.39 13.14 0.8364 Entrepreneurs 185 41 51.70 13.08 0.7784 Freelancers 109 17 50.22 13.02 0.8440 Farmers 93 14 51.37 13.17 0.8495 Vocational classes acc. to KldB-92 (Germany)2 Jobs in (animal) husbandry, forestry, horticulture 126 27 50.48 13.17 0.7857 Miners and mineral workers 26 11 55.35 13.22 0.5769 Production jobs (all) 996 216 49.86 13.11 0.7831 Metal production and metal working 73 14 49.21 13.09 0.8082 Metal constr., mech. engineering and similar 277 57 48.53 13.12 0.7942 Electrical engineering 84 16 46.40 13.06 0.8095 )zx ,x(SR ˆ • Marc Luy, Christian Wegner-Siegmundt, Angela Wiedemann, Jeroen Spijker434 SEP indicator / subpopulation R D x z ŜR Textile sector 14 3 54.36 13.07 0.7857 Foodstuffs sector 44 8 49.09 13.09 0.8182 Structural and civil engineering 126 33 50.76 13.13 0.7381 Wood and plastics processing 51 14 55.78 13.04 0.7255 Painting, varnishing and similar occupations 47 8 49.11 13.11 0.8298 Unskilled workers without specifi cation 61 21 53.13 13.13 0.6557 Technical occupations (all) 364 40 49.49 13.14 0.8901 Engineering, chemistry, physics, maths 146 16 50.16 13.17 0.8904 Technicians, technical specialists 218 24 49.05 13.12 0.8899 Service sector (all) 1,553 277 49.57 13.11 0.8216 Goods und service marketing 294 54 49.33 13.06 0.8163 Transport industry 258 63 50.09 13.13 0.7558 Organisation, administration, clerical jobs 563 100 50.16 13.13 0.8224 Public and private security sector 160 35 49.97 13.13 0.7813 Health sector 46 5 49.17 13.16 0.8913 Social service and education 159 12 46.61 13.09 0.9245 Other service occupations 39 6 50.51 12.97 0.8462 Other work force 71 15 50.28 12.94 0.7887 Tab. A3: Continuation Note: Figures exclude LES participants below age 33 at the time of fi rst survey. 1 ISCED-97 = International Standard Classifi cation of Education 2 KldB-92 = German Classifi cation of Professions Source: own calculations with LES data Life Expectancy by Education, Income and Occupation in Germany • 435 SEP indicator / subpopulation R D x z ŜR All women 3,044 304 50.05 13.11 0.9001 Education according to ISCED-971 Low (ISCED 1-2) 1,215 146 52.46 13.13 0.8798 Medium (ISCED 3-4) 1,630 150 48.78 13.10 0.9080 High (ISCED 5-6) 196 8 45.45 13.15 0.9592 Household net income 1st quartile (below € 895) 867 142 54.32 13.12 0.8362 2nd quartile (€ 895–1,406) 930 93 49.45 13.13 0.9000 3rd quartile (€ 1,406–1,917) 599 29 47.06 13.09 0.9516 4th quartile (€ 1,917 and more) 528 25 47.31 13.09 0.9527 Work status Manual workers (all) 842 96 50.27 13.13 0.8860 Unskilled or semi-skilled workers 726 83 50.39 13.12 0.8857 Skilled workers 108 13 49.18 13.18 0.8796 Master craftsmen, overseers, foremen 8 0 53.88 13.41 1.0000 Employees (all) 1,348 119 48.59 13.09 0.9117 Simple tasks 626 67 49.61 13.12 0.8930 Qualifi ed tasks 616 42 47.96 13.06 0.9318 Highly qualifi ed tasks 106 10 46.20 13.07 0.9057 Public servants (all) 259 24 51.31 13.13 0.9073 Simple or medium service 35 5 50.09 13.14 0.8571 Higher or senior service 87 0 42.40 13.14 1.0000 Self-employed workers (all) 122 5 44.61 13.14 0.9590 Entrepreneurs 83 6 50.73 13.14 0.9277 Freelancers 27 1 46.56 13.13 0.9630 Farmers 61 9 55.16 13.13 0.8525 Vocational classes acc. to KldB-92 (Germany)2 Jobs in (animal) husbandry, forestry, horticulture 60 4 50.77 13.23 0.9333 Miners and mineral workers 1 0 56.00 12.21 1.0000 Production jobs (all) 387 41 50.94 13.17 0.8941 Metal production and metal working 5 0 48.40 13.13 1.0000 Metal constr., mech. engineering and similar 7 1 48.29 13.15 0.8571 Electrical engineering 5 3 57.20 13.23 0.4000 Tab. A4: Number of respondents R, number of deaths D, average age at the time of fi rst survey x, average survival time z until the mortality follow-up, and respondents’ survival rate for the total subsamples, Women )zx ,x(SR ˆ • Marc Luy, Christian Wegner-Siegmundt, Angela Wiedemann, Jeroen Spijker436 SEP indicator / subpopulation R D x z ŜR Textile sector 136 14 51.72 13.11 0.8971 Foodstuffs sector 34 2 49.74 13.24 0.9412 Structural and civil engineering 0 0 --- --- --- Wood and plastics processing 1 0 62.00 13.19 1.0000 Painting, varnishing and similar occupations 3 0 46.00 13.49 1.0000 Unskilled workers without specifi cation 96 13 49.54 13.22 0.8646 Technical occupations (all) 56 5 49.88 12.92 0.9107 Engineering, chemistry, physics, maths 5 2 53.00 12.56 0.6000 Technicians, technical specialists 51 3 49.57 12.95 0.9412 Service sector (all) 1,879 157 48.21 13.11 0.9164 Goods und service marketing 390 32 48.17 13.14 0.9179 Transport industry 67 7 48.30 13.17 0.8955 Organisation, administration, clerical jobs 709 56 48.26 13.08 0.9210 Public and private security sector 14 2 49.93 13.14 0.8571 Health sector 145 13 46.63 13.04 0.9103 Social service and education 172 6 45.56 13.13 0.9651 Other service occupations 358 41 49.85 13.15 0.8855 Other work force 43 8 50.72 12.98 0.8140 Note: Figures exclude LES participants below age 33 at the time of the fi rst survey. 1 ISCED-97 = International Standard Classifi cation of Education 2 KldB-92 = German Classifi cation of Professions Source: own calculations with LES data Tab. A4: Continuation Published by / Herausgegeben von Prof. Dr. Norbert F. Schneider Federal Institute for Population Research D-65180 Wiesbaden / Germany Managing Editor / Verantwortlicher Redakteur Frank Swiaczny Assistant Managing Editor / Stellvertretende Redakteurin Katrin Schiefer Copy Editor (German) / Lektorat (deutsch) Dr. Evelyn Grünheid Layout / Satz Beatriz Feiler-Fuchs E-mail: cpos@bib.bund.de Scientifi c Advisory Board / Wissenschaftlicher Beirat Paul Gans (Mannheim) Johannes Huinink (Bremen) Michaela Kreyenfeld (Rostock) Marc Luy (Wien) Clara H. Mulder (Groningen) Notburga Ott (Bochum) Peter Preisendörfer (Mainz) Zsolt Spéder (Budapest) Comparative Population Studies www.comparativepopulationstudies.de ISSN: 1869-8980 (Print) – 1869-8999 (Internet) Board of Reviewers / Gutachterbeirat Martin Abraham (Erlangen) Laura Bernardi (Lausanne) Hansjörg Bucher (Bonn) Claudia Diehl (Konstanz) Andreas Diekmann (Zürich) Gabriele Doblhammer-Reiter (Rostock) Jürgen Dorbritz (Wiesbaden) Anette Eva Fasang (Berlin) E.-Jürgen Flöthmann (Bielefeld) Alexia Fürnkranz-Prskawetz (Wien) Beat Fux (Salzburg) Joshua Goldstein (Berkeley) Karsten Hank (Köln) Sonja Haug (Regensburg) Hill Kulu (Liverpool) Aart C. Liefbroer (Den Haag) Kurt Lüscher (Konstanz) Emma Lundholm (Umeå) Nadja Milewski (Rostock) Dimiter Philipov (Wien) Roland Rau (Rostock) Tomáš Sobotka (Wien) Jeroen Spijker (Barcelona) Olivier Thévenon (Paris) Helga de Valk (Brussel) Heike Trappe (Rostock) Michael Wagner (Köln) © Federal Institute for Population Research 2015 – All rights reserved