037 037 This paper is a product of the Poverty Global Practice Group and the Social Protection and Labor Global Practice Group. It is part of a larger effort by the World Bank to provide open access to its research and make a contribution to development policy discussions around the world. The authors may be contacted at Jan.Bauer@uni-hohenheim.de. The Poverty & Equity Global Practice Working Paper Series disseminates the findings of work in progress to encourage the exchange of ideas about development issues. An objective of the series is to get the findings out quickly, even if the presentations are less than fully polished. The papers carry the names of the authors and should be cited accordingly. The findings, interpretations, and conclusions expressed in this paper are entirely those of the authors. They do not necessarily represent the views of the International Bank for Reconstruction and Development/World Bank and its affiliated organizations, or those of the Executive Directors of the World Bank or the governments they represent. ‒ Poverty & Equity Global Practice Knowledge Management & Learning Team This paper is co-published with the World Bank Policy Research Working Papers. Subjective Well-Being across the Lifespan in Europe and Central Asia1 JAN MICHAEL BAUER1, VICTORIA LEVIN2, ANA MARIA MUNOZ BOUDET2, PENG NIE1, AND ALFONSO SOUSA-POZA1,3 1 Institute for Health Care & Public Management, University of Hohenheim, Germany 2 World Bank, Washington D.C., USA 3 IZA, Germany Keywords: Age; subjective well-being; Eastern Europe; Central Asia, life satisfaction JEL codes: C23, D1, I31, J1 1 We thank the staff at the various institutions that provided us with the data needed for this report, including the Integrated Values Survey, the Life in Transition Survey, and the Russia Longitudinal Monitoring Survey. Subjective Well-Being across the Lifespan in Europe and Central Asia 1. Introduction A large body of literature on the relation between age and subjective well-being has emerged in the past few years. This new field of research has been strongly influenced by the seminal paper of Blanchflower and Oswald (2008), which documents a U-shaped relation for well-being over the lifespan (with a minimum at around middle age) in most of the approximately 74 countries included in the analysis. Yet some controversy still remains over this U-shaped relation with as many papers refuting it (e.g., Frijters and Beatton, 2012; Kassenboehmer and Haisken-DeNew, 2012; Baetschmann, 2014) as supporting it (e.g., Clark, 2007; Van Landegham, 2008, 2012; Blanchflower and Oswald, 2011; Weiss et al., 2012).2 Most notably, Fischer (2009), using an OECD sub-sample of the World Values Survey (WVS), concludes that an S-shape with a cubic functional form in which well-being decreases in old age is a more accurate description of the well-being pattern across the lifecycle. Much of the ensuing literature also focuses almost exclusively on large-scale panel data in high-income countries; for example, the British Household Panel, the German Socioeconomic Panel, and the Household Income Labour Dynamics in Australia. Deaton (2008) expands this focus to include developing countries, arguing that the well- being/age relation may be mediated by culture. In his analysis of cross-sectional data from the 2006 Gallup World Poll, he reveals that the age profiles of self-reported life satisfaction differ significantly from country to country. Although this analysis does not control for covariates, his use of World Gallup Poll data facilitates cross-country comparisons and encompasses nationally representative samples of individuals from over 130 countries. Based on his observation of general life satisfaction among individuals from different age groups, Deaton (2008) argues that the U-shaped relation is present only in rich English-speaking countries. Similar results are obtained by a subsequent analysis of the World Gallup Poll; the age pattern of multiple well-being measures shows substantial differences between high-income English- speaking countries; countries in Sub-Saharan Africa; countries in Latin America and the Caribbean; and countries from the former Soviet Union and Eastern Europe. The U-shape is only present in the high-income English-speaking group, while well-being seems to decline with age in the last group (Steptoe et al., 2015). Our study examines the relation between age and subjective well-being in a region that to date has received relatively scarce attention in the literature: the Europe and Central Asia (ECA) 2 See Lopez Ulloa et al. (2013) for a recent literature review. 2 region. 3 This region warrants investigation not only because of the diverse cultures it encompasses but also because of the significant institutional and economic transformations many of its countries have undergone in the past few decades. Whereas the region’s cultural diversity permits us to test the universality of the U-shape hypothesis across different country contexts, the transformations experienced by most of these countries in the 1990s enable us to capture variations across cohorts whose economic and social environments differed radically at different points in the lifecycle. The finding of a U-shape across a diverse set of ECA countries and across cohorts within these countries would thus support the view that this U- shaped relation is fairly universal and resilient to contextual changes. With few exceptions (Blanchflower and Oswald, 2008; Deaton, 2008; Steptoe et al., 2015), studies of subjective well-being in the ECA region have not focused on an in-depth analysis of the interdependence between this variable and age. Rather, well-being regressions in most studies use age only as a potential covariate that is controlled when examining the association between well-being and such aspects as income (Frijters et al., 2006), the transition process (Hayo, 2007), and/or the correlations among different subjective welfare measures (Cojocaru and Diagne 2015). In contrast to Deaton (2008) and Steptoe et al. (2015), these ECA studies consistently find a U-shaped pattern across the region, although they differ in both the way that well-being is defined4 and the precise shape and location of the minimum. Country-specific evidence from the WVS also supports the U-shape for most countries in the ECA region (Namazie and Sanfey, 2001; Blanchflower and Oswald, 2008).5 For the Russian Federation, cross-sectional analyses using the Russia Longitudinal Monitoring Survey (RLMS) also find life satisfaction to be U-shaped in age, with a minimum in well-being between 45 and 55 (Graham and Pettianto, 2002; Ravallion and Lokshin, 2002; Graham et al., 2004). However, by exploiting the panel aspect of the RLMS (data from 1994 to 2010) and accounting for individual fixed effects, Massin and Kopp (2014) estimate a non- linear specification of age that places the lowest levels of well-being at age 30–39. They also find that the level of well-being for the oldest age group (70 and above) is statistically similar to that for people aged 20 years and younger. Our contribution to the research is twofold: first, by performing a systematic assessment of the relation between subjective well-being and age across the entire ECA region, we greatly 3 The ECA region is defined by the World Bank as covering 30 countries, ranging from Kazakhstan in the east to Poland in the west. The complete list of countries can be found at: http://www.worldbank.org/en/region/eca/brief/country-program-snapshots. 4 Minima vary from 37 (Hayo and Seifert, 2003) to 44 (Hayo, 2007) to 46.5 (48.2) for men (women) (Blanchflower and Oswald, 2008). 5 Specifically, the U-shaped relation is found for Azerbaijan, Belarus, Bosnia and Herzegovina, Bulgaria, Croatia, Czech Republic, Estonia, Kyrgyz Republic, Latvia, Lithuania, FYR Macedonia, Poland, Romania, Serbia, Slovakia, Turkey and Russian Federation. 3 expand on the findings from more geographically limited studies. We accomplish this expansion by using three large-scale surveys: the Integrated Values Survey (IVS), the Life in Transition Survey (LiTS), and the Russia Longitudinal Monitoring Survey (RLMS). Indeed, our analysis may be the first in-depth study of the relation between age and well-being using a large-scale panel data set from a non-Western country (i.e. the Russian Federation). Second, by adding indicators for birth decades into our regressions, we are able to disentangle some of the age-specific correlation with subjective well-being from the variation attributable to cohort effects. Our results generally support the presence of a U-shaped relation for most of the lifecycle, even when the panel data accounts reasonably well for the cohort effects. We also, however, find evidence of another turning point later in life (during the 60s or 70s), after which well-being in ECA begins to decline again, giving rise to an S-shape across the entire lifespan. The remainder of the paper proceeds as follows: Section 2 describes our three data sets and our different methodological approaches to determining the shape of subjective well-being over the lifespan in the ECA region. Section 3 reports our empirical results, together with the conclusions drawn about the (non-)universality of the U-shape. Section 4 rounds out the paper by summarizing the contributions our methodological approach makes to the research stream. 2. Data and methods 2.1 Data The empirical results of this study are based on three different data sets: the Integrated Values Survey (IVS), the Life in Transition Survey (LiTS), and the Russia Longitudinal Monitoring Survey (RLMS). The IVS data set merges the four-wave European Values Study (EVS, 1981– 2009) with the five-wave World Values Survey (WVS, 1981–2008). The EVS, begun in 1981 and repeated every 9 years, is a large-scale, cross-national, longitudinal survey that provides insights into the ideas, beliefs, preferences, attitudes, values, and opinions of citizens all over Europe. Nationally representative of the population 18 years and older, in its first wave, it sampled 1,000 individuals per country, and then for its fourth wave, administered in 2008, expanded to a multistage random sampling of 1,500 interviews per country. This last wave covers 25 of the 30 ECA countries: Albania, Armenia, Azerbaijan, Belarus, Bosnia and Herzegovina, Bulgaria, Croatia, Czech Republic, Estonia, Georgia, Hungary, Kosovo Republic, Latvia, Lithuania, FYR Macedonia, Moldova, Montenegro, Poland, Romania, Russian Federation, Serbia, Slovak Republic, Slovenia, Turkey and Ukraine. The WVS, administered in six waves from 1981 to 2014, is a global research project that investigates the changes in individual’s beliefs and values and how these changes influence social and political development throughout the world. It samples a minimum of 1,000 interviewees aged 18–85 in 23 ECA countries: Albania, Armenia, Azerbaijan, Belarus, Bosnia 4 and Herzegovina, Bulgaria, Croatia, Czech Republic, Estonia, Georgia, Hungary, Kyrgyz Republic, Latvia, FYR Macedonia, Moldova, Poland, Romania, Russian Federation, Serbia, Slovak Republic, Slovenia, Turkey and Ukraine. The resulting compilation, the IVS, assesses individual life satisfaction with the following question: All things considered, how satisfied are you with your life as a whole these days? Responses are measured on a scale from 1 (dissatisfied) to 10 (satisfied). The LiTS, a joint project of the European Bank for Reconstruction and Development (EBRD) and the World Bank, examines the impact of economic and political transition on individuals’ lives (and well-being), as well as their future hopes for Central Eastern Europe, the Baltic States, and South Eastern Europe. Its two waves, conducted in 2006 and 2010, covered approximately 58,000 individuals in 29 ECA countries (all except Kosovo). In the LiTS, individual life satisfaction is measured as follows: To what extent do you agree with the following statements? All things considered, I am satisfied with my life now. Responses on this question are assessed on a scale from 1 (strongly disagree) to 5 (strongly agree). The RLMS, a nationally representative survey of households in the Russian Federation, is designed to evaluate the impact of Russian reforms on welfare and health. Although the survey has collected data on about 10,000 individuals aged 14 and above 21 times since 1992, our analysis is based on 16 waves during the 1995–2013 period. The RLMS evaluates individual life satisfaction as follows: To what extent are you satisfied with your life in general at the present time? Responses are assessed on a 5-point scale from 1 (not at all satisfied) to 5 (fully satisfied). Summary statistics for subjective well-being and respondent socio-demographics from all the above data sources are given in Appendix 1. 2.2 Estimation approaches 2.2.1 Ordinary least squares (OLS): IVS For the IVS data, we adopt the standard regression approach applied in the majority of studies on subjective well-being (e.g. Blanchflower and Oswald, 2008). Because well-being is measured on an ordered 5-point or 10-point scale, a latent variable analysis might seem the most appropriate method. However, because the bias introduced by a standard ordinary least squares estimation (OLS) is fairly small (Ferrer-i-Carbonell and Frijters, 2004), we rely mostly on OLS. In general, the model is as follows: , 1 where denotes the life satisfaction of individual i. is included either as a linear or multinomial function of age (i.e. quadratic and cubic terms) or as a set of 5-year age group 5 indicators. We adopt the 5-year clustering for two reasons: (i) to balance the trade-off between cell size and within-group heterogeneity in well-being and (ii) to facilitate the distinction between age groups, cohort effects, and time.6 This specification, adopted in prior studies, also helps to avoid co-linearity problems between age, survey year, and birth year (the cohort effect).7 As in Blanchflower and Oswald (2008), cohort effects are modelled as birth decades. represents the characteristics of individual i (see Appendix 2 for a detailed description of the covariates included for each data source). Adding in self-reported health not only extends Blanchflower and Oswald’s (2008) model but is likely to reduce any bias in our estimates of age’s relation with well-being. C designates country indicators, Y denotes year dummies, and denotes the individual error term. Heteroskedasticity-adjusted standard errors are estimated for coefficients in all specifications. 2.2.2 Synthetic panel estimation: LiTS Following Deaton (1985) and Russell and Fraas (2005), we construct a synthetic (or pseudo) panel for the LiTS data that follows groups8 of individuals over the 2006–2010 inter-wave period and uses their group-specific means for inference. To ensure equal sample means for each group, these groups (or cells) must be created based on time-invariant characteristics, which in our case are birth-year decades, gender, and country. Because the estimation requires a sufficiently large cell size, we exclude the earliest and latest birth-year decade and drop all countries not covered by both LiTS waves.9 For this analysis, we create 672 cells based on 28 countries, 6 birth decades, 2 sexes, and 2 survey waves (2006 and 2010). For each cell, we estimate the mean of the covariates, including age, marital status, employment, education, self-reported health, and relative income. By assuming the existence of a fixed group that remains constant over both periods, we can apply a fixed-group effects estimator, a feasible step if the sample size in each cell is relatively large. To address any heteroskedasticity produced by unequal cell sizes, we also weight each cell by its sample size (Russell and Fraas, 2005).10 The model used is as follows: 6 We assume that the well-being of a 50-year-old does not differ systematically from that of a 51-year-old and that a certain category size is needed to capture possible differences in well-being with age. In addition, smaller age groups would result in small cell sizes, especially at the extremes of the age distribution. 7 For a critical discussion on the approach and identification, see Baetschmann (2014). 8 Such groups are sometimes referred to as “cohorts,” but we avoid this term because this paper also examines birth cohorts. 9 The gender differences in life expectancy are notably substantial in some ECA countries. In the Russian Federation, for instance, women live on average more than 10 years longer than men, which must be taken into account when looking at the joint distribution. However, the gender split analysis of the RLMS data shows no major differences in well-being/age patterns between the sexes. 10 This weighted method has some limitations related to data availability. On the one hand, if the sample size within each group does not exceed thousands of observations, the measurement error can be large (Devereux, 2007). On the other, the groups must be sufficiently homogeneous for the assumption of fixed group effects to remain justifiable, meaning that clustering groups over different countries would introduce another bias. We thus treat the results of this approach with caution but compare them with our pooled cross-sectional analysis estimates to check robustness. 6 , 1, ⋯ , 348; 1, 2, 2 where and represent the average life satisfaction and average time-variant characteristics, respectively, of individuals in group g at time t. denotes group-specific time- invariant unobservable characteristics, and is the disturbance error term (see Russell and Fraas, 2005). 2.2.3 Fixed-effects panel estimation: RLMS For the RLMS, we explore the U-shape in the age/well-being nexus using a fixed effects approach expressed as follows: , 3 where and indicates the life satisfaction and age, respectively, of individual i at time t. is a vector of individual i’s time-variant characteristics, represents unobservable time- invariant individual effects, and is the disturbance error term. 3. Results Although Blanchflower and Oswald (2008) provide comprehensive evidence of a robust U- shaped pattern for most subsamples of the IVS data set, Fischer (2009) proposes an S-shaped relation with decreasing well-being in higher age as a better fit. We thus test different age specifications to determine the most adequate model for the ECA region. First, we compare the age coefficients, in different forms, between a set of ECA countries11 and Western European countries.12 Table 1 presents the results of a comparison between the two IVS subsamples from 1981 until 2009 for individuals aged 18 to 108, while Figure 1 illustrates the corresponding average marginal effects for regressions (1) to (6). In line with the U-shape hypothesis, the squared terms are significant for both samples. The average individual in an ECA country reaches the nadir of well-being at age 48.8 (column 1), more than 10 years later than the average Western European, whose well-being starts increasing after age 35.4 (column 2). Estimations (3) and (4) in Table 1 add a cubic term for age, which allows for a more flexible relation with well-being. The coefficient on the cubic term is significant for both ECA and Western Europe, implying an S-shape with decreasing well-being in old age. For ECA countries, controlling for other covariates, well-being decreases with age to a local minimum 11 Albania, Armenia, Azerbaijan, Belarus, Bosnia and Herzegovina, Bulgaria, Croatia, Czech Republic, Estonia, FYR Macedonia, Georgia, Hungary, Kazakhstan, Kyrgyz Republic, Latvia, Lithuania, Moldova, Montenegro, Poland, Romania, Russian Federation, Serbia, Slovak Republic, Slovenia, Tajikistan, Turkey, Ukraine, and Uzbekistan. 12 France, United Kingdom, Austria, Belgium, Luxembourg, the Netherlands, Switzerland, West Germany, and East Germany. 7 at 42.8 and recovers to the average value reported by youth by the early 80s (see Figure 1, middle-left panel). Thereafter, well-being declines for the oldest old, even though the small sample size leads to larger standard errors, which makes precise interpretation difficult. Although the coefficient on the cubic term of age for Western Europe is also significant and negative, the maximum occurs much later (93 versus 82 in the ECA) and the level of well-being reached at the maximum is higher than that observed for youth (Figure 1, middle-right panel). The final two columns in Table 1 impose no specific functional form on age but rather use age group clusters. Compared to the omitted reference age group of 18–22, well-being in ECA countries is lower for all other age groups, a pattern evident in column (5). Figure 1 demonstrates that well-being in ECA decreases until the late 40s and then recovers slightly over the next 40 years (bottom-left panel). It does not, however, reach levels comparable to those experienced by youth until the early 80s and never significantly exceeds those levels. The largest negative coefficients occur in the 43–47 age group, echoing the results from regression (3). For the Western European comparison group (column 6), the age dummies show only a slight decrease in well-being from the levels attained in youth, with a minimum for those in their 30s. Thereafter, well-being seems to remain somewhat constant, with the coefficients losing significance until the 60s, which implies that well-being resembles that of the 18–22 reference age group. The point estimates then become positive and significant in the 70s and increase up to the second-oldest age group (83–87). Figure 1 clearly shows that well-being in the ECA countries tends to be generally lower and exhibits a different lifecycle pattern from the Western European comparison group. Whereas in Western Europe the highest level of well-being occurs among the oldest old across all specifications, in ECA countries higher levels among the old appear only in the squared specification. Subsequently, therefore, we employ a more flexible specification, which reveals a decrease in well-being at a very high age. Comparing (3) and (5) in Figure 1, for example, shows that the dummy mimics the cubic S-shape slightly better than the U-shape, which is in line with previous studies by Deaton (2008), Fischer (2009), and Gwozdz and Sousa-Poza (2010). 8 Figure 1. Average marginal effects of subjective well-being over the lifespan: IVS data (1981–2009) ECA Western Europe (1) squared (2) squared 9 9 Linear Prediction 8 8 7 7 6 6 5 5 23 28 33 38 43 48 53 58 63 68 73 78 83 88 93 23 28 33 38 43 48 53 58 63 68 73 78 83 88 93 (3) cubic (4) cubic 9 9 Linear Prediction 8 8 7 7 6 6 5 5 23 28 33 38 43 48 53 58 63 68 73 78 83 88 93 23 28 33 38 43 48 53 58 63 68 73 78 83 88 93 (5) non-parametric (6) non-parametric 9 9 Linear Prediction 8 8 7 7 6 6 5 5 23 28 33 38 43 48 53 58 63 68 73 78 83 88 93 23 28 33 38 43 48 53 58 63 68 73 78 83 88 93 Age Age Note: The panel numbers correspond to the regressions in Table 1 9 Table 1. OLS estimates for ECA countries and Western Europe: IVS data (1981–2009) (1) (2) (3) (4) (5) (6) ECA Western Europe ECA Western Europe ECA Western Europe Age -0.0478*** -0.0294*** -0.1367*** -0.1140*** (0.005) (0.004) (0.022) (0.024) Age squared 0.0005*** 0.0004*** 0.0024*** 0.0022*** (0.000) (0.000) (0.000) (0.000) Age cubic -0.00001*** -0.00001*** (0.000) (0.000) Age 23-27 -0.2740*** -0.1158* (0.074) (0.059) Age 28-32 -0.4131*** -0.1848** (0.066) (0.059) Age 33-37 -0.4818*** -0.2055* (0.068) (0.101) Age 38-42 -0.5272*** -0.2374 (0.094) (0.128) Age 43-47 -0.5741*** -0.2021 (0.085) (0.126) Age 48-52 -0.5096*** -0.1749 (0.081) (0.109) Age 53-57 -0.4985*** -0.0666 (0.072) (0.120) Age 58-62 -0.4050*** 0.1273 (0.070) (0.152) Age 63-67 -0.3085*** 0.3423* (0.074) (0.163) Age 68-72 -0.2791*** 0.3759** (0.086) (0.155) Age 73-77 -0.1847* 0.4839* (0.098) (0.206) Age 78-82 -0.2115* 0.6625** (0.110) (0.191) Age 83-87 -0.1106 0.7177* (0.154) (0.324) Age 88-108 -0.3627 0.5997** (0.359) (0.253) Minimum 48.8 35.4 42.9 36.6 Maximum 82.4 93.8 N 77527 17389 77527 17389 77527 17389 Adj. R2 0.246 0.241 0.246 0.242 0.246 0.242 Note: The dependent variable is self-rated individual life satisfaction (10-point scale). Age ranges from 18 to 108, with age 18–22 as the reference in (5) and (6). Controls are gender, employment status (full-time employed, part-time employed, self-employed, with not employed as the reference), marital status (married/living together, divorced/separated, widowed, with single/never married as the reference), education (medium level education and high level education, with low level education as the reference), income quintiles, self-reported health (5-point scale), country dummies, and year dummies. Robust standard errors clustered at the country level in parentheses; * p < 0.1, ** p < 0.05, *** p < 0.01. 10 3.1 Impact of cohort effects To disentangle the age-specific correlation with well-being from that of the cohort effect, we add indicators for birth decades to the above regressions. The results for the IVS data and regressions with individual controls are given in Table 2, which reveals that not all functional forms of age remain significant when we control for cohort effects. For instance, although the U-shape remains robust for the ECA countries in column (1), the squared specification for Western Europe (column 2) fails to reach significance for the linear term even though it remains jointly significant. This latter could imply that well-being increases throughout the lifespan. The cohort effects themselves take varying signs but mostly fail to reach significance. Moreover, a within-comparison of the cohort dummies reveals only small differences, with only those born in 1990 and after showing significantly higher well-being in ECA countries. The second pair of columns, regressions (3) and (4), illustrate the S-shape of well-being for ECA and Western European countries. Comparing these specifications (see Tables 1 and 2) reveals that the inclusion of cohort effects has only a small impact on the age coefficients: as minima and maxima move toward younger age, well-being levels become lower in older age, a pattern observable for both ECA countries and Western Europe. The non-functional form of age mostly mimics the cubic term (Figure 2, bottom panels): the age coefficients for ECA lose their significance after the late 50s, but those for Western Europe show only a small decrease from 18–22 until the early 40s. In line with the functional form, the point estimate for well- being appears to decline for Western Europeans in the oldest age group, even though it is not significantly different from that for the young reference group. Interestingly, the cohort dummies for the ECA region (column 5) are also not significant, except for the 1930 birth cohort, which experienced the deprivation related to World War II at a very young age. Most cohort dummies for Western Europe, in contrast, are significant and negative, pointing to the relatively higher well-being of those in the oldest cohort (born before 1919).13 The marginal effects in Figure 2 show quite constant well-being over time, with slight downturns in middle and old age. To confirm the importance of cohort effects and their interplay with the age specification, we apply the same method used for the IVS data to the LiTS data set (age range of 18–99), although we cannot make direct comparisons between the ECA and Western Europe because the LiTS only covers ECA countries in more than one wave. The age pattern revealed by the LiTS results (Figure 3) is similar to that from the IVS with some noteworthy exceptions: First, the LiTS results for the cubic age specification provide little support for a second turning point in well-being, as maxima are estimated to be beyond normal life expectancy, and thus suggest 13 It should be noted, however, that the reference cohort (born before 1919) is a very select sample, not only because of their survival but also because their early years coincide with turbulent times (e.g., the 1917 October Revolution and WWI). 11 a U-shape. Second, adding cohort dummies to the regressions has little effect on the age coefficients and, therefore, minima and maxima remain fairly unchanged. Figure 2. Average marginal effects of subjective well-being over the lifespan with cohort dummies: IVS data (1981–2009) ECA Western Europe (1) squared (2) squared 9 9 Linear Prediction 8 8 7 7 6 6 5 5 23 28 33 38 43 48 53 58 63 68 73 78 83 88 93 23 28 33 38 43 48 53 58 63 68 73 78 83 88 93 (3) cubic (4) cubic 9 9 Linear Prediction 8 8 7 7 6 6 5 5 23 28 33 38 43 48 53 58 63 68 73 78 83 88 93 23 28 33 38 43 48 53 58 63 68 73 78 83 88 93 (5) non-parametric (6) non-parametric 9 9 Linear Prediction 8 8 7 7 6 6 5 5 23 28 33 38 43 48 53 58 63 68 73 78 83 88 93 23 28 33 38 43 48 53 58 63 68 73 78 83 88 93 Age Age Note: The panel numbers correspond to the regressions in Table 2 12 Table 2. Well-being over the lifespan in ECA countries and Western Europe with cohort dummies: IVS data (1981–2009) (1) (2) (3) (4) (5) (6) ECA Western Europe ECA Western Europe ECA Western Europe Age -0.0413*** -0.0143 -0.1641*** -0.1094*** (0.009) (0.011) (0.027) (0.021) Age squared 0.0004*** 0.0002** 0.0032*** 0.0023*** (0.000) (0.000) (0.001) (0.001) Age cubic -0.00002*** -0.00001*** (0.000) (0.000) Age 23-27 -0.2572*** -0.1204* (0.077) (0.059) Age 28-32 -0.3884*** -0.1453*** (0.084) (0.036) Age 33-37 -0.4387*** -0.1616* (0.094) (0.085) Age 38-42 -0.4219*** -0.2211* (0.140) (0.097) Age 43-47 -0.4410*** -0.1955 (0.133) (0.115) Age 48-52 -0.3163** -0.1898 (0.139) (0.161) Age 53-57 -0.2807** -0.0928 (0.135) (0.166) Age 58-62 -0.1555 0.0375 (0.149) (0.239) Age 63-67 -0.0565 0.2371 (0.150) (0.230) Age 68-72 -0.0462 0.1333 (0.165) (0.289) Age 73-77 0.0354 0.1932 (0.168) (0.263) Age 78-82 -0.0820 0.3045 (0.168) (0.316) Age 83-87 -0.0167 0.3564 (0.185) (0.339) Age 88-108 -0.3394 -0.0005 (0.381) (0.448) Born 1920-29 0.1519 -0.1624 -0.1199 -0.2128 -0.1340 -0.3534** (0.158) (0.131) (0.156) (0.145) (0.153) (0.104) Born 1930-39 0.1258 -0.1557 -0.2635 -0.2632* -0.2956* -0.4217** (0.171) (0.120) (0.169) (0.121) (0.162) (0.146) Born 1940-49 0.1634 -0.2329 -0.2231 -0.3112 -0.2697 -0.6026** (0.175) (0.171) (0.174) (0.167) (0.162) (0.187) Born 1950-59 0.1275 -0.3656 -0.1604 -0.3650 -0.2130 -0.6670** (0.189) (0.208) (0.189) (0.223) (0.179) (0.276) Born 1960-69 0.1523 -0.3405 -0.0440 -0.2737 -0.0978 -0.6782* (0.200) (0.223) (0.200) (0.258) (0.186) (0.318) Born 1970-79 0.2165 -0.2955 0.0208 -0.2218 -0.0181 -0.7127* (0.207) (0.246) (0.208) (0.294) (0.204) (0.362) Born 1980-89 0.2319 -0.1342 -0.0196 -0.1471 -0.0483 -0.6181 (0.209) (0.263) (0.214) (0.290) (0.220) (0.418) Born 1990-99 0.5546** -0.3161 0.1843 -0.4497 0.2565 -0.8493* (0.244) (0.381) (0.255) (0.377) (0.259) (0.400) Minimum 46.6 28.8 38.3 34.0 Maximum 76.0 85.1 N 77527 17389 77527 17389 77527 17389 Adj. R2 0.246 0.242 0.246 0.242 0.246 0.242 Note: The dependent variable is self-rated individual life satisfaction (10-point scale). Age ranges from 18 to 108, with age 18–22 as the reference in (5) and (6). For the cohort dummies, the reference group is those born between 1881 and 1919. The controls are gender, employment status (full-time employed, part-time employed, self-employed, with not employed as the reference), marital status (married/living together, divorced/separated, widowed, with single/never married as the reference), education (medium level education and high level education, with low level education as the reference), income quintiles, self-reported health (5-point scale), country dummies, and year dummies. Robust standard errors clustered at the country level in parentheses; * p < 0.1, ** p < 0.05, *** p < 0.01. 13 Figure 3. Average marginal effects of subjective well-being over the lifespan: LiTS data (2006–2010) ECA ECA with cohort dummies (1) squared (2) squared 4 4 Linear Prediction 3.5 3.5 3 3 2.5 2.5 2 2 23 28 33 38 43 48 53 58 63 68 73 78 83 88 93 23 28 33 38 43 48 53 58 63 68 73 78 83 88 93 (3) cubic (4) cubic 4 4 Linear Prediction 3.5 3.5 3 3 2.5 2.5 2 2 23 28 33 38 43 48 53 58 63 68 73 78 83 88 93 23 28 33 38 43 48 53 58 63 68 73 78 83 88 93 (5) non-parametric (6) non-parametric 4 4 Linear Prediction 3.5 3.5 3 3 2.5 2.5 2 2 23 28 33 38 43 48 53 58 63 68 73 78 83 88 93 23 28 33 38 43 48 53 58 63 68 73 78 83 88 93 Age Age 3.2 Well-being in a synthetic panel using LiTS To account for the time-constant effects shared by certain population groups, we construct a synthetic panel from the ECA countries in the LiTS data, whose results support the findings from the cross-sectional analysis (Table 3). Above all, once fixed-group effects are accounted for, the estimates find an overall negative linear relation between age and well-being. In addition, the squared specification is highly significant, with a minimum around 60, while the cubic term supports the U-shape until age 80, after which well-being decreases. The estimates become less precise, however, at older ages, when the standard errors are quite large (see Appendix 3 for the corresponding marginal plots). We are therefore unable to reject either the U-shape or a second turning point. 14 Table 3. Well-being over the lifespan in a synthetic panel with group-fixed effects: LiTS data (2006–2010) (1) (2) (3) Age -0.0141*** -0.0783*** -0.1813*** (0.005) (0.016) (0.051) Age² 0.0007*** 0.0029*** (0.000) (0.001) Age³ -0.00001** (0.000) Minimum 59.2 51.2 Maximum 82.2 Adj. R2 0.255 0.285 0.293 Number of groups 672 672 672 Note: Individuals are grouped by birth-year decades, country, gender, and survey year. The dependent variable is group average life satisfaction (5-point scale). Age ranges from 18 to 99. The individual controls are group averages of marital status (married versus unmarried as the reference), self-reported health (5-point scale), education (medium level education and high level education, with low level education as the reference), income quintiles, and employment (employed versus not employed as the reference). Robust standard errors in parentheses; * p < 0.1, ** p < 0.05, *** p < 0.01. 3.3 Well-being in a panel data set for the Russian Federation The comprehensive RLMS panel data for the Russian Federation permit us to follow the same individuals over time and include individual fixed effects (Clark and Oswald, 2006; Clark, 2007; Gwozdz and Sousa-Poza, 2010). Because the first four RLMS waves are problematic and Wave 5 lacks important socio-demographic controls, for this analysis, we rely on RLMS rounds VI (1995) to XXI (2013). We first compare the results of the pooled OLS to those for the model with fixed effects (see Table 4) using specifications similar to the cross-sectional analysis above.14 By comparing age as a linear, quadratic, and cubic term, as well as a dummy specification (age range of 14–104), we are able to quantify the differences in the results produced by the two methods and thereby show the robustness of the pattern in our other regressions. Although the linear specification yields a negative correlation between age and subjective well-being (column 1) that loses significance when fixed effects are included (column 2), when age is treated as linear, the cohort effects are significantly different from those of the comparison birth cohort (born 1919 and earlier). Interestingly, the coefficients on cohort dummies exhibit a U-shape, with the lowest well- being observed among those born in the 1960s, who are middle aged (37–52) during the survey period. This latter provides evidence of the cohort dummies picking up the residual variation left by the linear (as opposed to quadratic) age specification. The quadratic and cubic results for the pooled OLS and FE specifications are fairly similar: the minima of the U-shape differ slightly, with the fixed effects regression indicating a minimum 5 years later than the pooled OLS. Here, the cohort effects mostly remain insignificant, although the cubic specification yields a maximum for the pooled OLS that is 10 years later in life than in 14 As we do not include survey weights into the cross-sectional regressions, we followed the recommendation of Heeringa (1997) and re-estimated the models including dummies for rural/PTG/urban and 176 areas to account for the non-random sampling procedure: the results remain robust and do not substantially differ from Table 4. 15 the fixed effects regression. Moreover, the cohort coefficients, although significant, show no particular pattern. A similar pattern does emerge, however, for the age dummies, although the pooled OLS exhibits a more pronounced shape. Separate regressions by gender, on the other hand, yield no major pattern differences even though men tend to have higher overall levels of well- being. The corresponding average marginal effects for the non-functional forms in Table 4 and the gender split analysis are plotted in Figure 4. Figure 4. Average marginal effects of a methodical and gender comparison of well-being over the lifespan: RLMS data (1995–2013) (1) POLS (2) FE full sample full sample 4.5 4.5 4 4 Linear Prediction 3.5 3.5 3 3 2.5 2.5 2 2 23 28 33 38 43 48 53 58 63 68 73 78 83 88 93 23 28 33 38 43 48 53 58 63 68 73 78 83 88 93 (3) FE (4) FE men women 4.5 4.5 4 4 Linear Prediction 3.5 3.5 3 3 2.5 2.5 2 2 23 28 33 38 43 48 53 58 63 68 73 78 83 88 93 23 28 33 38 43 48 53 58 63 68 73 78 83 88 93 Age Age 16 Table 4. Comparison of well-being over the lifespan in the pooled OLS versus fixed-effects regressions: RLMS data (1995–2013) (1) (2) (3) (4) (5) (6) (7) (8) POLS FE POLS FE POLS FE POLS FE Age -0.0051*** -0.0068 -0.0598*** -0.0442*** -0.1586*** -0.1351*** (0.002) (0.008) (0.003) (0.008) (0.006) (0.010) Age squared 0.0006*** 0.0004*** 0.0028*** 0.0025*** (0.000) (0.000) (0.000) (0.000) Age cubic -0.00002*** -0.00001*** (0.000) (0.000) Age 23-27 -0.3316*** -0.1460*** (0.018) (0.020) Age 28-32 -0.3929*** -0.1265*** (0.022) (0.027) Age 33-37 -0.4405*** -0.1341*** (0.027) (0.033) Age 38-42 -0.4444*** -0.0738* (0.032) (0.040) Age 43-47 -0.4363*** 0.0005 (0.038) (0.047) Age 48-52 -0.4124*** 0.0652 (0.044) (0.054) Age 53-57 -0.3120*** 0.1970*** (0.050) (0.060) Age 58-62 -0.2060*** 0.3405*** (0.055) (0.068) Age 63-67 -0.1424** 0.4113*** (0.062) (0.075) Age 68-72 -0.0310 0.5069*** (0.068) (0.083) Age 73-77 0.0458 0.5871*** (0.075) (0.090) Age 78-82 0.1310 0.6647*** (0.082) (0.099) Age 83-87 0.2576*** 0.7894*** (0.091) (0.110) Age 88-104 0.1696 0.7036*** (0.108) (0.135) Born 1920-29 -0.2613*** 0.0508 -0.2168*** -0.1073** (0.049) (0.052) (0.052) (0.051) Born 1930-39 -0.4766*** 0.0450 -0.3233*** -0.1538*** (0.055) (0.059) (0.060) (0.057) Born 1940-49 -0.6815*** -0.0029 -0.3663*** -0.1686*** (0.066) (0.071) (0.072) (0.065) Born 1950-59 -0.8258*** -0.0781 -0.3556*** -0.1340* (0.077) (0.083) (0.083) (0.073) Born 1960-69 -0.8946*** -0.1688* -0.3480*** -0.0973 (0.090) (0.095) (0.094) (0.082) Born 1970-79 -0.8218*** -0.2069* -0.3480*** -0.0141 (0.104) (0.107) (0.106) (0.092) Born 1980-89 -0.6666*** -0.1982* -0.3986*** 0.0597 (0.117) (0.119) (0.118) (0.101) Born 1990-99 -0.4030*** -0.0635 -0.3728*** 0.2093* (0.129) (0.130) (0.129) (0.108) Minimum 51.5 56.1 42.2 42.9 Maximum 82.6 74.4 N 162808 162808 162808 162808 162808 162808 162808 162808 Adj. R2 0.197 0.085 0.204 0.087 0.208 0.090 0.204 0.087 Note: The dependent variable is self-rated individual life satisfaction (5-point scale). Age ranges from 14 to 104, with age 14–22 as the reference in (5) and (6). For cohort dummies, the reference group is those born between 1894 and 1919. The controls are gender, employment status (employed versus not employment as the reference) marital status (married/ living together, divorced/separated, widowed, with single/never married as the reference), education (medium level education and high level education, with low level education as the reference), income quintiles, self-reported health (5-point scale), and year dummies. Robust standard errors clustered at the individual level are in parentheses.*p < 0.1, **p < 0.05, ***p < 0.01. 17 3.4 Country-specific differences in ECA countries To test whether the results described above hold within the diverse ECA sample, we conclude by conducting country-specific analysis using the LiTS data, which covers most relevant countries in two waves. The results, shown in Table 5, produce a heterogeneous picture, with 17 countries exhibiting an S-shaped relation between age and well-being. For all other countries, the cubic term is insignificant, but the squared-term specification yields significant coefficients, suggesting a U-shape relation; a heterogeneity also evident from the variation in minima (32.9 to 51.3) and maxima (67.4 to 101.8). Table 5. Country-specific differences of life satisfaction in age minima and maxima: LiTS data (2006-2010) Country Minimum Maximum Observations Countries with age minimum and maximum (S shape) Albania 37.4 81.3 1973 Bosnia and Herzegovina 42.2 70.7 1988 Czech Republic 39.5 79.0 1977 Hungary 38.5 84.6 2028 Moldova 40.5 92.0 1943 Montenegro 37.2 77.6 1790 Poland 39.6 84.1 2557 Romania 39.6 80.8 2030 Serbia 41.8 93.2 2420 Slovak Republic 39.7 79.4 1975 Turkey 36.2 73.2 1970 Ukraine 44.3 92.9 2369 Azerbaijan 40.1 78.2 1978 Estonia 40.1 95.2 1977 Kyrgyz Republic 32.9 67.4 1964 Latvia 41.8 91.7 1981 Lithuania 44.0 101.8 1968 Countries with age minimum (U shape) Belarus 43.8 1587 Bulgaria 48.1 1981 Croatia 44.8 1878 FYR Macedonia 38.3 1967 Slovenia 49.2 1917 Armenia 37.8 1899 Georgia 51.3 1978 Kazakhstan 45.7 1938 Russian Federation 49.8 2477 Tajikistan 40.4 1968 Uzbekistan 45.0 2443 Note: The dependent variable is self-rated individual life satisfaction (5-point scale). Age is specified using linear, squared, and cubic terms. The controls are gender, employment status (employed versus not employed as the reference), marital status (married, not married as the reference), education (medium level education and high level education, with low level education as the reference), income quintiles, self-reported health (5-point scale), and year dummies. 18 4. Conclusions This analysis of the relation between age and well-being in Europe and Central Asia (ECA) takes advantage of panel and cross-sectional data from three large surveys: the Integrated Values Survey (IVS), the Life in Transition Survey (LiTS), and the Russia Longitudinal Monitoring Survey (RLMS). By focusing on the ECA region, the analysis makes two important contributions. First, by expanding the current research focus on high-income countries, it is better able to assess the universality of the U-shaped relation identified for Western countries. Second, it is able to leverage the disparate impacts of recent political and economic transformation in most ECA countries on the well-being of different cohorts to address the methodological problem of distinguishing between age and cohort effects. As Frijters and Beatton (2012) point out, the ideal would be to ‘follow representative individuals throughout their whole life, starting at birth’ (p. 529), but such data are not available. We overcome this limitation by examining well-being across the lifespan in a synthetic panel data set for ECA countries and a true panel data set for the Russian Federation.15 We hypothesize that if a U- shape is observable in societies that have witnessed dramatic economic and social changes in past decades then, once cohort effects are corrected for (albeit imperfectly), they are most probably not the explanation for any observed U-shaped relation. In the aggregate, our findings mostly support the existence of a U-shape in individual well-being in the ECA up until the 60s and 70s but point to an apparent decrease in well-being among the oldest old, even after we control for self-reported health. This pattern remains robust for most of our cross-sectional and panel analyses, although within the ECA, there is significant heterogeneity in the pattern of well-being across the lifespan. 15 In the absence of large-scale panels that follow the same individuals across their lifespan, the most promising way to address the cohort issue is most probably by using less subjective measures of well-being such as in Weiss et al.’s (2012) ape study. 19 References Baetschmann, G. (2014). ‘Heterogeneity in the relationship between happiness and age: Evidence from the German Socio-Economic Panel’, German Economic Review, 15(3), pp. 393–410. Blanchflower, D. G. and Oswald, A. J. (2008). ‘Is well-being U-shaped over the life cycle?’, Social Science & Medicine, 66(8), pp. 1733–1749. Blanchflower, D. G. and Oswald, A. J. (2011). ‘Antidepressants and age’, IZA Discussion Paper No. 5785. Clark, A. (2007), ‘Born to be mild? Cohort effects don’t (fully) explain why well-being is U- Shaped in age’, IZA Discussion Paper No. 3170. Clark, A. and Oswald, A. J. (2006), ‘The curved relationship between subjective well-being and age’, Paris-Jourdan Sciences Economiques Working Paper No. 29, Paris School of Economics. Cojocaru, A. and Diagne, M. F. (2015). ‘How reliable and consistent are subjective measures of welfare in Europe and Central Asia?’, Economics of Transition, 23(1), pp. 75–103. Deaton, A. (1985). ‘Panel data from time series of cross-sections’, Journal of Econometrics, 30, pp. 109–126. Deaton, A. (2008). ‘Income, health, and well-being around the world: Evidence from the Gallup World Poll’, Journal of Economic Perspectives, 22(2), pp. 53–72. Devereux, P. J. (2007). ‘Small-sample bias in synthetic cohort models of labour supply’, Journal of Applied Econometrics, 22(4), pp. 839–848. Ferrer-i-Carbonell, A. and Frijters, P. (2004). ‘How important is the methodology for the estimates of determinants of happiness?’ Economic Journal, 140, pp. 641–-659. Fischer, J. A. (2009). ‘Happiness and age cycles-return to start…? On the functional relationship between subjective well-being and age’, OECD Social, Employment and Migration Paper No. 99, OECD Publishing. Frijters, P. and Beatton, T. (2012). ‘The mystery of the U-shaped relationship between happiness and age’, Journal of Economic Behavior and Organization, 82(2–3), pp. 525– 542. Frijters, P., Geishecker, I., Hasken-DeNew J. P., and Shields, M. A. (2006). ‘Can the large swings in Russian life satisfaction be explained by ups and downs in real incomes?’ Scandinavian Journal of Economics, 108(3), pp. 433–458. Graham, C., Eggers, A., and Skhtankar, S. (2004). ‘Does happiness pay? An exploration based on panel data from Russia’, Journal of Economic Behaviour and Organization, 55, pp. 319–342. Graham, C. and Pettinato S. (2002). ‘Frustrated achievers: Winners, losers and subjective well- being in new market economies’, Journal of Development Studies, 38(4), pp. 100–140. 20 Gwozdz, W. and Sousa-Poza, A. (2010). ‘Ageing, health and life satisfaction of the oldest old: An analysis for Germany’, Social Indicators Research, 97(3), pp. 397–417. Hayo, B. (2007). ‘Happiness in transition: An empirical study on Eastern Europe’, Economic Systems, 31(2), pp. 204–221. Hayo, B. and Seifert, W. (2003). ‘Subjective economic well-being in Eastern Europe’, Journal of Economic Psychology, 24(3), pp. 329–348. Heeringa, S. G. (1997).’Russia Longitudinal Monitoring Survey (RLMS) Sample Attrition, Replenishment, and Weighting in Rounds V-VII’, available at: http://www.cpc.unc.edu/ projects/rlms-hse/project/samprep/samprep.pdf Kassenboehmer, S. C. and Haisken-DeNew, J. P. (2012). ‘Heresy or enlightenment? The well- being age U-shape effect is flat’, Economics Letters, 117(1), pp. 235–238. Lopez Ulloa, B. F., Moller, V., and Sousa-Poza, A. (2013). ‘How does subjective well-being evolve with age? A literature review’, Journal of Population Ageing, 6(3), pp. 227–246. Massin, S. and Kopp, P. (2014). ‘Is life satisfaction hump-shaped with alcohol consumption? Evidence from Russian panel data’, Addictive Behaviors, 39(4), pp. 803-810. Namazie, C. and Sanfey, P. (2001). ‘Happiness in transition: The case of Kyrgyzstan’, Review of Development Economics, 5(3), pp. 392–405. Ravallion, M. and Lokshin, M. (2002). ‘Self-rated economic welfare in Russia’, European Economic Review, 46(8), pp. 1452–1473. Russell, J. E. and Fraas, J. W. (2005). ‘An application of panel regression to pseudo panel data’, Multiple Linear Regression Viewpoints, 31(1), pp. 1–15. Steptoe, A., Deaton, A., and Stone, A. A. (2015). ‘Subjective wellbeing, health, and ageing’, Lancet, 385, pp. 640-648. Van Landeghem, B. (2008). ‘Human well-being over the life cycle: Longitudinal evidence from a 20-year panel’, LICOS Discussion Paper No. 213/2008, Catholic University, Leuven. Van Landeghem, B. (2012). ‘A test for the convexity of human well-being over the life cycle: Longitudinal evidence from a 20-year panel’, Journal of Economic Behavior and Organization, 81(2), pp. 571–582. Weiss, A., King, J. E., Inoue-Murayama, M., Matsuzawa, T., and Oswald, A. J. (2012). ‘Evidence for a midlife crisis in great apes consistent with the U-shape in human well- being’, Proceedings of the National Academy of Sciences, 109(49), pp. 19,949–19,952. 21 Appendix 1: Descriptive statistics IVS LiTS RLMS Variable Mean Variable Mean Variable Mean Well-being 5.876 Well-being 3.101 Well-being 2.865 Scales of well-being 1-10 Scales of well-being 1-5 Scales of well-being 1-5 Age 43.878 Age 46.070 Age 45.093 Male 0.456 Male 0.399 Male 0.424 Employment Unemployed 0.119 Yes 0.455 Yes 0.545 Other 0.382 Part time 0.051 Full time 0.383 Self employed 0.064 Marital status Single 0.200 Married 0.574 Single 0.169 Married 0.641 Married/living together 0.602 Divorced 0.059 Divorced/separated 0.098 Widowed 0.100 Widowed 0.131 Education Low 0.358 Low 0.186 Low 0.407 Medium 0.427 Medium 0.626 Medium 0.380 High 0.215 High 0.188 High 0.212 Health 3.488 Health 3.389 Health 3.142 Observations 77,527 57,950 (person/year) 162,808 Note: Well-being in the IVS is measured from 1 (dissatisfied) to 10 (satisfied); in the LiTS and RLMS, it is measured from 1 = not at all satisfied to 5 = fully satisfied. Health is measured from 1 = very poor to 5 = very good. Income is not reported because there is no useful interpretation for the mean of income quintiles. 22 Appendix 2: Description of variables European Value Study and World Value Survey (IVS) Life satisfaction: The IVS assesses individual life satisfaction as follows: All things considered, how satisfied are you with your life as a whole these days? Responses are measured on a scale from 1 (dissatisfied) to 10 (satisfied). Employment: It classifies individual employment status into five types: non-employed (reference group), full-time employed, part-time employed, self-employed and other, which includes retired, housewife, student, and those who selected other in the survey. Marital status: The survey categorizes marital status into three states: single/never married (reference category), married/living together as married, and divorced/separated/widowed. Health: In our analysis, we evaluate health based on self-reported health status, a subjective health indicator assessed by the following: All in all, how would you describe your state of health these days? Responses are measured on a 5-point scale: 1 = very poor, 2 = poor, 3 = fair, 4 = good and 5 = very good. Income: The IVS data contain income deciles based on survey year and country. We reduce these to quintiles to facilitate income comparisons. Education: For manageability, we reduce the original IVS education levels to only three states: Low (omitted reference), containing inadequately completed elementary education, completed (compulsory) elementary education, incomplete secondary school: technical/vocational type/ (compulsory) elementary education, and basic vocational qualification incomplete secondary: university-preparatory type/secondary, intermediate general qualification; Medium, consisting of complete secondary school: technical/vocational type/secondary, intermediate vocational qualification and complete secondary: university-preparatory type/full secondary, maturity level certificate; and High, covering some university without degree/higher education lower-level tertiary certificate and university with degree/higher education – upper-level tertiary certificate. The Life in Transition Survey Life satisfaction: In the LiTS, individual life satisfaction is based on the following question: To what extent do you agree with the following statements? All things considered, I am satisfied with my life now, with answers assessed on a scale from 1 = strongly disagree to 5 = strongly agree. Household satisfaction: The wording of the item on household life satisfaction differs slightly in the 2006 and 2010 LiTS waves: My household lives better nowadays than around 1989 and my household lives better nowadays than around 4 years ago, respectively. Responses to both are assessed on a 5-point scale from 1 = strongly disagree to 5 = strongly agree. Employment status: Because precise data are unavailable, employment status is a binary variable that equals 1 if the respondent is currently employed and 0 otherwise. Marital status: Marital status is rescaled as a binary variable (1=married, 0=not married). 23 Health: As before, we assess self-reported health status using a subjective health measure based on the following question: How would you assess your health? evaluated on a 5-point scale from 1 = very bad to 5 = very good. This variable is very close to the IVS, in which only wording and category names differ slightly. Income: Because none of the income measures are sufficient for an international comparison, we proxy household expenses with spending, which differs slightly between the two waves. In 2006, spending included food, beverages, and tobacco, clothing and footwear, transport and communication (fixed-line phone, mobile phone, Internet) expenses, and recreation, entertainment, meals outside the home, etc. In 2010, it covered food, beverages and tobacco, utilities (electricity, water, gas, heating, fixed line phone) and transportation (public transportation, fuel for car). Based on these expenses, we calculate country-specific expense quintiles for every survey year. Education: We reduce the original six LiTS education levels to only three: low, medium and high, constructed as follows: For 2006, low includes no degree/education and compulsory school education; medium comprises secondary education and professional, vocational school/training, and high consists of higher professional degree (university, college) and post- graduate degree. For 2010, the medium category contains three different classifications: lower secondary, (upper) secondary and post-secondary non tertiary. Both the other categories are constructed as in 2006. Russia Longitudinal Monitoring Survey Life satisfaction: The RLMS evaluates individual life satisfaction as follows: To what extent are you satisfied with your life in general at the present time? , evaluated on a 5-point scale from 1 = not at all satisfied to 5 = fully satisfied. Employment: Using the RLMS data, the individual employment status is a binary measure equal to 1 if the respondent is currently employed, 0 otherwise. The reference group is non-employed. Marital status: We recode the 5-point scale for marital status (1 = never married, 2 = married, 3 = married but not living together, 4 = divorced and 5 = separated/ widowed) as a 3-point scale: 1= never married, 2 = married/ married but not living together and 3 = divorced/separated/widowed, with never married as the reference group. It should be noted, however, that no marital status information is available for Round V. Health: Self-rated health status is measured by the question, Tell me, please, how would you evaluate your health status?, assessed on a 5-point scale: 1 = very bad, 2 = bad, 3 = average, 4 = good and 5 = very good. Income: To enable comparison with other data sets, we recode nominal household income into household income quintiles. Education: We recode education levels on a 7-point scale: 0 = no education, 1 = obtaining a diploma of professional course (e.g. typing or accounting courses), 2 = obtaining a PTU, FZU or 24 FZO diploma without a secondary education, 3 = obtaining a PTU diploma with a secondary education, 4 = obtaining a technical education diploma, including medical, music, pedagogical and art school, 5 = obtaining a university degree and 6 = obtaining a graduate degree and/or residency. The reference category is further education. 25 Appendix 3: Additional Material Average marginal effects of well-being over the lifespan in the synthetic panel regressions using LiTS data (2006–2010) Squared Cubic (1) full sample (2) full sample 5 5 Linear Prediction 4.5 4.5 4 4 3.5 3.5 3 3 2.5 2.5 2 2 23 28 33 38 43 48 53 58 63 68 73 78 83 88 93 98 23 28 33 38 43 48 53 58 63 68 73 78 83 88 93 98 (3) men (4) men 5 5 4.5 4.5 Linear Prediction 4 4 3.5 3.5 3 3 2.5 2.5 2 2 23 28 33 38 43 48 53 58 63 68 73 78 83 88 93 98 23 28 33 38 43 48 53 58 63 68 73 78 83 88 93 98 (5) women (6) women 5 5 Linear Prediction 4.5 4.5 4 4 3.5 3.5 3 3 2.5 2 2.5 2 23 28 33 38 43 48 53 58 63 68 73 78 83 88 93 98 23 28 33 38 43 48 53 58 63 68 73 78 83 88 93 98 Age Age 26 Poverty & Equity Global Practice Working Papers (Since July 2014) The Poverty & Equity Global Practice Working Paper Series disseminates the findings of work in progress to encourage the exchange of ideas about development issues. An objective of the series is to get the findings out quickly, even if the presentations are less than fully polished. The papers carry the names of the authors and should be cited accordingly. The findings, interpretations, and conclusions expressed in this paper are entirely those of the authors. They do not necessarily represent the views of the International Bank for Reconstruction and Development/World Bank and its affiliated organizations, or those of the Executive Directors of the World Bank or the governments they represent. This series is co‐published with the World Bank Policy Research Working Papers (DECOS). It is part of a larger effort by the World Bank to provide open access to its research and contribute to development policy discussions around the world. For the latest paper, visit our GP’s intranet at http://POVERTY. 1 Estimating poverty in the absence of consumption data: the case of Liberia Dabalen, A. L., Graham, E., Himelein, K., Mungai, R., September 2014 2 Female labor participation in the Arab world: some evidence from panel data in Morocco Barry, A. G., Guennouni, J., Verme, P., September 2014 3 Should income inequality be reduced and who should benefit? redistributive preferences in Europe and Central Asia Cojocaru, A., Diagne, M. F., November 2014 4 Rent imputation for welfare measurement: a review of methodologies and empirical findings Balcazar Salazar, C. F., Ceriani, L., Olivieri, S., Ranzani, M., November 2014 5 Can agricultural households farm their way out of poverty? Oseni, G., McGee, K., Dabalen, A., November 2014 6 Durable goods and poverty measurement Amendola, N., Vecchi, G., November 2014 7 Inequality stagnation in Latin America in the aftermath of the global financial crisis Cord, L., Barriga Cabanillas, O., Lucchetti, L., Rodriguez‐Castelan, C., Sousa, L. D., Valderrama, D. December 2014 8 Born with a silver spoon: inequality in educational achievement across the world Balcazar Salazar, C. F., Narayan, A., Tiwari, S., January 2015 Updated on December 2015 by POV GP KL Team | 1 9 Long‐run effects of democracy on income inequality: evidence from repeated cross‐sections Balcazar Salazar,C. F., January 2015 10 Living on the edge: vulnerability to poverty and public transfers in Mexico Ortiz‐Juarez, E., Rodriguez‐Castelan, C., De La Fuente, A., January 2015 11 Moldova: a story of upward economic mobility Davalos, M. E., Meyer, M., January 2015 12 Broken gears: the value added of higher education on teachers' academic achievement Balcazar Salazar, C. F., Nopo, H., January 2015 13 Can we measure resilience? a proposed method and evidence from countries in the Sahel Alfani, F., Dabalen, A. L., Fisker, P., Molini, V., January 2015 14 Vulnerability to malnutrition in the West African Sahel Alfani, F., Dabalen, A. L., Fisker, P., Molini, V., January 2015 15 Economic mobility in Europe and Central Asia: exploring patterns and uncovering puzzles Cancho, C., Davalos, M. E., Demarchi, G., Meyer, M., Sanchez Paramo, C., January 2015 16 Managing risk with insurance and savings: experimental evidence for male and female farm managers in the Sahel Delavallade, C., Dizon, F., Hill, R., Petraud, J. P., el., January 2015 17 Gone with the storm: rainfall shocks and household well‐being in Guatemala Baez, J. E., Lucchetti, L., Genoni, M. E., Salazar, M., January 2015 18 Handling the weather: insurance, savings, and credit in West Africa De Nicola, F., February 2015 19 The distributional impact of fiscal policy in South Africa Inchauste Comboni, M. G., Lustig, N., Maboshe, M., Purfield, C., Woolard, I., March 2015 20 Interviewer effects in subjective survey questions: evidence from Timor‐Leste Himelein, K., March 2015 21 No condition is permanent: middle class in Nigeria in the last decade Corral Rodas, P. A., Molini, V., Oseni, G. O., March 2015 22 An evaluation of the 2014 subsidy reforms in Morocco and a simulation of further reforms Verme, P., El Massnaoui, K., March 2015 Updated on December 2015 by POV GP KL Team | 2 23 The quest for subsidy reforms in Libya Araar, A., Choueiri, N., Verme, P., March 2015 24 The (non‐) effect of violence on education: evidence from the "war on drugs" in Mexico Márquez‐Padilla, F., Pérez‐Arce, F., Rodriguez Castelan, C., April 2015 25 “Missing girls” in the south Caucasus countries: trends, possible causes, and policy options Das Gupta, M., April 2015 26 Measuring inequality from top to bottom Diaz Bazan, T. V., April 2015 27 Are we confusing poverty with preferences? Van Den Boom, B., Halsema, A., Molini, V., April 2015 28 Socioeconomic impact of the crisis in north Mali on displaced people (Available in French) Etang Ndip, A., Hoogeveen, J. G., Lendorfer, J., June 2015 29 Data deprivation: another deprivation to end Serajuddin, U., Uematsu, H., Wieser, C., Yoshida, N., Dabalen, A., April 2015 30 The local socioeconomic effects of gold mining: evidence from Ghana Chuhan-Pole, P., Dabalen, A., Kotsadam, A., Sanoh, A., Tolonen, A.K., April 2015 31 Inequality of outcomes and inequality of opportunity in Tanzania Belghith, N. B. H., Zeufack, A. G., May 2015 32 How unfair is the inequality of wage earnings in Russia? estimates from panel data Tiwari, S., Lara Ibarra, G., Narayan, A., June 2015 33 Fertility transition in Turkey—who is most at risk of deciding against child arrival? Greulich, A., Dasre, A., Inan, C., June 2015 34 The socioeconomic impacts of energy reform in Tunisia: a simulation approach Cuesta Leiva, J. A., El Lahga, A., Lara Ibarra, G., June 2015 35 Energy subsidies reform in Jordan: welfare implications of different scenarios Atamanov, A., Jellema, J. R., Serajuddin, U., June 2015 36 How costly are labor gender gaps? estimates for the Balkans and Turkey Cuberes, D., Teignier, M., June 2015 37 Subjective well‐being across the lifespan in Europe and Central Asia Bauer, J. M., Munoz Boudet, A. M., Levin, V., Nie, P., Sousa‐Poza, A., July 2015 Updated on December 2015 by POV GP KL Team | 3 38 Lower bounds on inequality of opportunity and measurement error Balcazar Salazar, C. F., July 2015 39 A decade of declining earnings inequality in the Russian Federation Posadas, J., Calvo, P. A., Lopez‐Calva, L.‐F., August 2015 40 Gender gap in pay in the Russian Federation: twenty years later, still a concern Atencio, A., Posadas, J., August 2015 41 Job opportunities along the rural‐urban gradation and female labor force participation in India Chatterjee, U., Rama, M. G., Murgai, R., September 2015 42 Multidimensional poverty in Ethiopia: changes in overlapping deprivations Yigezu, B., Ambel, A. A., Mehta, P. A., September 2015 43 Are public libraries improving quality of education? when the provision of public goods is not enough Rodriguez Lesmes, P. A., Valderrama Gonzalez, D., Trujillo, J. D., September 2015 44 Understanding poverty reduction in Sri Lanka: evidence from 2002 to 2012/13 Inchauste Comboni, M. G., Ceriani, L., Olivieri, S. D., October 2015 45 A global count of the extreme poor in 2012: data issues, methodology and initial results Ferreira, F.H.G., Chen, S., Dabalen, A. L., Dikhanov, Y. M., Hamadeh, N., Jolliffe, D. M., Narayan, A., Prydz, E. B., Revenga, A. L., Sangraula, P., Serajuddin, U., Yoshida, N., October 2015 46 Exploring the sources of downward bias in measuring inequality of opportunity Lara Ibarra, G., Martinez Cruz, A. L., October 2015 47 Women’s police stations and domestic violence: evidence from Brazil Perova, E., Reynolds, S., November 2015 48 From demographic dividend to demographic burden? regional trends of population aging in Russia Matytsin, M., Moorty, L. M., Richter, K., November 2015 49 Hub‐periphery development pattern and inclusive growth: case study of Guangdong province Luo, X., Zhu, N., December 2015 50 Unpacking the MPI: a decomposition approach of changes in multidimensional poverty headcounts Rodriguez Castelan, C., Trujillo, J. D., Pérez Pérez, J. E., Valderrama, D., December 2015 51 The poverty effects of market concentration Rodriguez Castelan, C., December 2015 52 Can a small social pension promote labor force participation? evidence from the Colombia Mayor program Pfutze, T., Rodriguez Castelan, C., December 2015 Updated on December 2015 by POV GP KL Team | 4 53 Why so gloomy? perceptions of economic mobility in Europe and Central Asia Davalos, M. E., Cancho, C. A., Sanchez, C., December 2015 54 Tenure security premium in informal housing markets: a spatial hedonic analysis Nakamura, S., December 2015 For the latest and sortable directory, available on the Poverty & Equity GP intranet site. http://POVERTY WWW.WORLDBANK.ORG/POVERTY Updated on December 2015 by POV GP KL Team | 5