The reciprocal relationship between material factors and health in the life course: evidence from SHARE and ELSA

Hoffmann, Rasmus; Kröger, Hannes; Pakpahan, Eduwin Article — Published Version The reciprocal relationship between material factors and health in the life course: evidence from SHARE and ELSA European Journal of Ageing Provided in Cooperation with: German Institute for Economic Research (DIW Berlin) Suggested Citation: Hoffmann, Rasmus; Kröger, Hannes; Pakpahan, Eduwin (2018) : The reciprocal relationship between material factors and health in the life course: evidence from SHARE and ELSA, European Journal of Ageing, ISSN 1613-9380, Springer, Heidelberg, Vol. 15, Iss. 4, pp. 379-391, https://doi.org/10.1007/s10433-018-0458-3 This Version is available at: https://hdl.handle.net/10419/200193 Standard-Nutzungsbedingungen: Die Dokumente auf EconStor dürfen zu eigenen wissenschaftlichen Zwecken und zum Privatgebrauch gespeichert und kopiert werden. Sie dürfen die Dokumente nicht für öffentliche oder kommerzielle Zwecke vervielfältigen, öffentlich ausstellen, öffentlich zugänglich machen, vertreiben oder anderweitig nutzen. Sofern die Verfasser die Dokumente unter Open-Content-Lizenzen (insbesondere CC-Lizenzen) zur Verfügung gestellt haben sollten, gelten abweichend von diesen Nutzungsbedingungen die in der dort genannten Lizenz gewährten Nutzungsrechte. Terms of use: Documents in EconStor may be saved and copied for your personal and scholarly purposes. You are not to copy documents for public or commercial purposes, to exhibit the documents publicly, to make them publicly available on the internet, or to distribute or otherwise use the documents in public. If the documents have been made available under an Open Content Licence (especially Creative Commons Licences), you may exercise further usage rights as specified in the indicated licence. http://creativecommons.org/licenses/by/4.0/ Vol.:(0123456789) 1 3 European Journal of Ageing (2018) 15:379–391 https://doi.org/10.1007/s10433-018-0458-3 ORIGINAL INVESTIGATION The reciprocal relationship betweenmaterial factors andhealth inthelife course: evidence fromSHARE andELSA RasmusHoffmann1,2 · HannesKröger1,3· EduwinPakpahan1 Published online: 19 February 2018 © The Author(s) 2018. This article is an open access publication Abstract The widely established health differences between people with greater economic resources and those with fewer resources can be attributed to both social causation (material factors affecting health) and health selection (health affecting material wealth). Each of these pathways may have different intensities at different ages, because the sensitivity of health to a lack of material wealth and the degree to which health can influence economic resources may change. We study the relative importance, in terms of explanatory power, of social causation and health selection, comparing the transitions from childhood to adulthood and from adulthood to old age. We use retrospective survey data from ten European countries from the Survey of Health, Ageing and Retirement in Europe (SHARELIFE, n=18,734) and the English Longitudinal Study of Ageing (ELSA, n=6117), and structural equations models in a cross-lagged panel design. Material wealth and health depend on their prior status, wealth more so than health. In the transition from childhood to adulthood, social causation and health selection are equally important: the standardized coefficients for men in SHARE are 0.07 and 0.06, respectively, i.e. one standard deviation increase in material wealth in childhood is associated with a 0.07 standard deviation increase in adult health. In the transition from adulthood to old age, social causation is more important than health selection (0.52 vs. 0.01), across gender and data sets. Both pathways contribute to the creation of health inequalities—however, their relative importance changes with age, which is important for understanding how health inequalities develop and how policies can address them. Keywords Material wealth· Health inequality· Life course· Structural equation model Introduction Health inequalities between social groups have been found in all periods and countries for which data are available. Morbidity and mortality are systematically higher among people with lower socio-economic status (SES), measured, for example, by education, occupational status, wealth, or income. Health inequalities usually amount to between 5 and 10years’ difference in life expectancy and between 10 and 20years’ difference in disability-free life expectancy (Mackenbach 2006), and they rate high on the political agenda (Elo 2009). Health inequalities in old age are mostly smaller than in adulthood, which has been attributed to a large extent to mortality selection (Crimmins etal. 2009; Hoffmann 2011a), but inequalities between income groups, in particular, are still substantial (Hoffmann 2011b; Huisman etal. 2003). For example, the difference in life expectancy at age 65 in Germany between the lowest and highest pension income quintile is 3.6years (Shkolnikov etal. 2008). Since the proportion of older people in societies is rapidly increasing, health inequalities among older people are of increasing importance for research and policy. Also increasing is the awareness that health inequalities in old age can be best explained by a life course approach (Mayer 2009), because they are the outcomes of the interaction of two life-long Responsible editor: D.J.H. Deeg. Electronic supplementary material The online version of this article (https ://doi.org/10.1007/s1043 3-018-0458-3) contains supplementary material, which is available to authorized users. * Rasmus Hoffmann [email protected] 1 European University Institute, Fiesole, Italy 2 Max Planck Institute forDemographic Research, Rostock, Germany 3 Socio-Economic Panel Study (SOEP), Deutsches Institut für Wirtschaftsforschung e.V. (DIW Berlin), Mohrenstraße 58, 10117Berlin, Germany 380 European Journal of Ageing (2018) 15:379–391 1 3 processes: first, the development and changes in a person’s SES, a process with critical periods and transitions, with path dependencies and accumulation; second, health trajectories, determined by fixed individual characteristics, social influences, behaviour, and institutional settings, also with critical periods, path dependencies, and accumulations of health problems or recovery. The social factors that contribute to health differences can be grouped into material living conditions (e.g. income or wealth), psychosocial conditions (e.g. social participation, job demand control), lifestyle (e.g. tobacco, alcohol, obesity, physical activity, diet), and access to essential services (e.g. education, health care) (Graham 2009). Especially for the association between material factors and health, there is widespread debate as to whether they reflect an effect of material factors on health (social causation) or the effect of health on material wealth (health selection) (Galama and van Kippersluis 2010; Martikainen etal. 2009). More specifically, it has been argued that the relative importance of the two pathways changes over the life course (Smith 2003). This may be because the sensitivity of health to a lack of material wealth and the degree to which health can influence economic resources depend on specific circumstances that change with age. We study the question as to whether pathways from material wealth to health have more explanatory power than pathways from health to material wealth, and whether the relative importance of the two reciprocal effects changes over the life course. The causal direction between SES and health is part of a long-running discussion with several important normative and political implications regarding the reduction in health inequalities, which raises complex methodological questions concerning empirical analysis in a longitudinal life course framework. The life course perspective is crucial for understanding the interrelated processes of SES and health, and how observed health inequalities in old age develop. In the remainder of the introduction, we will elaborate on the open question of the relative importance of social causation and health selection, and why it may change with age. The reciprocal relationship betweenmaterial wealth andhealth Mechanisms that create health inequalities are manifold and have been discussed extensively in the literature (Case and Deaton 2005; Galama and van Kippersluis 2010; Hoffmann 2008). A thorough investigation reveals mechanisms where SES influences health and those where health affects SES. The first model assumes that material wealth influences health through, for example, the affordability of health care, environmental hazards, consumption, and the psychological burden of being poor. The second model involves health influencing material wealth, both via the career benefits associated with good health, and the financial costs incurred during illness. A third model to explain health inequalities is that (unknown) background factors influence both SES and health (indirect selection) (Goldman 2001). These factors may be genetic endowment, family background, or individual characteristics (genetic or acquired), such as height, personality, or preferences in behaviour and lifestyle. There is disagreement concerning the relative importance of social causation versus health selection, due not only to different underlying ideas of the relation between social structure and presumably stable individual characteristics, but also to different research designs and methods, as well as to divergent concepts of causality. A central proposition of the health selection hypothesis is that social mobility is partly determined by health. While there are indications for a certain level of health-related social mobility at labour market entry (Smith 1999), the relationship between health and social mobility is fairly weak (Kröger 2015). Moreover, the chronology of social mobility at younger ages and increasing health problems at higher ages seems to contradict the proposition of the health selection hypothesis. Nevertheless, reverse causality from health to SES can bias the coefficients of conventional statistical models, and the direction of causality from SES to health should not be taken for granted. Few epidemiological studies have examined the possibility of health selection (e.g. Chandola etal. 2003), and many authors believe that health selection is of little importance (e.g. Manor etal. 2003). The assessment of its relative importance also depends on the exact pathway under study. While there is agreement that education influences health (Gathmann etal. 2015; Lleras-Muney 2005), some scholars think that the influence of material resources on health is low, and that the influence of health on material status is the strongest overall causality in the relationship between SES and health (Galama and van Kippersluis 2010). A recent systematic literature review of the relative importance of social causation versus health selection evaluated 34 out of 2952 reviewed studies from the past 20years, qualitatively and in a quantitative, statistical meta-analysis (Kröger etal. 2015), concluding that there is no preference for one of the two directions. Twelve studies supported social causation, and 10 supported health selection—the other studies supported both mechanisms equally. This study measures material wealth at three life course stages and covers large parts of the life course from childhood to old age, in order to estimate the relative explanatory power of social causation and health selection. The term material wealth is used in a broad sense to denote material living conditions in general, including income. We perform similar analyses with two comparable European data sets in order to explore whether a similar pattern of results can be found in two independent data sources. Our analysis does not allow us to study these differences 381European Journal of Ageing (2018) 15:379–391 1 3 in sufficient detail to interpret them as differences between countries or welfare systems. We use a long-term life course perspective and look at two broad age ranges, because the development of material wealth is accumulative (Dannefer 2003), and because specific mechanisms governing the interaction between material wealth and health may be relevant at specific stages of the life course. We are thus able to assess existing explanations for social causation and health selection at different stages of the life course. In particular, it has been suggested that health selection is relatively strong at labour market entry, when health influences occupation and income (Smith 1999), and in older working ages when many health problems start to become more prevalent (Oksanen and Virtanen 2012; Smith 2003). Unlike many existing studies, we do not aim to identify ‘local’ causal effects between a specific aspect of SES and a specific measure of health in a specific subgroup of the population. Instead, we use a long-term life course approach, use broad indicators of material wealth and health, and model their mutual influence simultaneously. Methods Data We use the third wave (SHARELIFE, version 5.0.0) of the Survey of Health Aging and Retirement in Europe (SHARE) (Börsch-Supan 2016; Börsch-Supan etal. 2013), which covers the whole life course of respondents retrospectively. The second data source is the third wave of the English Longitudinal Study of Ageing (ELSA) (Marmot etal. 2017; Steptoe etal. 2012). In both surveys, persons aged 50 and older were asked retrospectively about changes in their material wealth and health since childhood. The data are representative for the population 50+ and their spouses living in households in the respective European countries (SHARE) and England (ELSA). From SHARE, we limit our analysis to ten countries (Austria, Belgium, Denmark, France, Germany, Italy, Netherlands, Spain, Sweden, and Switzerland), because for three SHARE countries information on wages was not comparable over the life course (Poland, Czech Republic) or contained too many missing values (Greece). We study persons aged 55–90 at the time of the interview in 2008/2009 (SHARE) and 2006/2007 (ELSA). Samples sizes are 18,734 (SHARE) and 6117 (ELSA). The average response rate across countries in SHARE Wave 1 is about 60% (ranging from about 40 to 80%). Details of participating countries are provided online (http://www.share -proje ct.org/data-docum entat ion/sampl e.html). SHARE added a refresher sample in Wave 2 to compensate for the loss of representativity due to follow-up attrition. The response rate in ELSA is 73%, and there was a refreshment sample in Wave 3. Details can be found in Steptoe etal. (Steptoe etal. 2012) for a description of the sample and the variables see Table1. Measures In the operationalization of our concepts, we divide the life course into three periods: childhood (age 0–15), adulthood (age 30–50), and old age (age 55–90). For childhood, we use indicators available in SHARE and ELSA that refer to childhood in general or to individuals at age ten; for adulthood, we use retrospective information and calculate averages for the age range 30–50; old age is represented by prospective information at the time of the interview, which ranges from age 55 to 90. The starting age of the oldest group is set at age 55, because we use the third wave of SHARE that was representative of the population aged 50+ in its first wave 6years earlier. We use heterogeneous measurement between age groups; it is the only way to combine retrospective and prospective survey data and allows us to measure health and material wealth at very different ages. We use two indicators for material wealth in childhood that are rough indicators for the general standard of living, but easy for interviewees to remember: the number of rooms per person and a summary index of features of the household (cold water, hot water, toilet, bath, heating). For adulthood, we use the two indicators homeownership and estimates of average monthly wages between age 30 and 50, corrected for purchasing power and inflation by purchasing power parities (PPP) relative to the level of Germany in the year 2006 (Weiss 2012). Respondents specified the beginning and the end of episodes in which they were home owners or rented, and we used the mode of this variable. Likewise, respondents specified job spells and reported their wages, which we averaged over the adult age range, taking into account the lengths of the spells. These reports also take into account periods of unemployment and promotions or wage changes within the same job. Wages represent monetary wealth, while home ownership provides information about the general stability of material circumstances, although it does not necessarily indicate the standard of living. In higher age groups, we measure material wealth with the net-equivalent household income at the time of the interview. Alternatively, we measure it with household net wealth per capita (property, cars, company shares, and liquid funds, minus debts). We analyse income and wealth separately in old age to establish whether they produce different results. This could be due not only to the different effects on health suggested in the literature (Avendano and Glymour 2008), but also to their different degree of responsiveness to health; wealth is even more 382 European Journal of Ageing (2018) 15:379–391 1 3 Table 1 Samples description of SHARE and ELSA (variables, categories, distributions) Latent construct Variable Category SHARE ELSA (England) N=18,734 (%) N=6117 (%) Country West Austria 945 5.0 Belgium 2584 13.8 France 2223 11.9 Germany 1762 9.4 Netherlands 2069 11.0 Switzerland 1157 6.2 South Italy 2292 12.2 Spain 2035 10.9 North Denmark 1806 9.6 Sweden 1861 9.9 Age in Wave 3 (SHARE 2008/2009, ELSA 2006/2007) Mean 68.3 68.7 SD 8.9 9.0 Min 55 55 Max 90 90 Gender Male 8598 45.9 2730 44.6 Female 10,136 54.1 3387 55.4 C-MW (childhood material wealth) Number of facilities Mean 2.0 3.0 SD 1.8 1.4 Min 0.0 0.0 Max 5.0 5.0 Missing 129 0.7 271 4.4 Rooms per capita Mean 0.8 0.6 SD 0.4 0.2 Min 0 0.1 Max 10 3.8 Missing 326 1.7 291 4.8 C-H (childhood health) Self-rated health Poor 457 2.4 207 3.4 Fair 1235 6.6 508 8.3 Good 4777 25.5 1252 20.5 Very good 5814 31.0 2070 33.8 Excellent 6265 33.4 2021 33.0 Missing 186 1.0 59 1.0 Missed school Yes 2166 11.6 1341 22.1 No 16,441 87.8 4728 77.3 Missing 127 0.7 48 0.8 Hospitalized Yes 1150 6.2 664 10.9 No 17,489 93.4 5407 88.4 Missing 95 0.5 46 0.8 A-MW (adult material wealth) Age 30–50 Owner of house or apartment Yes 13,726 74.1 5054 83.5 No 4792 25.6 996 16.3 Missing 216 1.2 67 1.1 Average wages (in SHARE corrected for purchasing power and inflation by purchasing power parities (PPP) relative to German € in 2006) Mean 1301 1994 SD 868 1673 Min 8 83 Max 6126 12,826 Missing 9714 51.9 2182 35.7 383European Journal of Ageing (2018) 15:379–391 1 3 stable than (pension) income because it has been acquired over the whole life course. Results using wealth are in the Online Resources. Health in childhood is measured retrospectively by three indicators: self-assessed health in five categories, whether school was missed because of health for 1month or more, and whether 1month or more was spent in hospital. At ages 30–50, our health measure for SHARE is based on two indicators reflecting how many years (as a share of the years between age 30 and 50) individuals reported being either in bad health in general, or suffering from an acute or chronic illness. For ELSA, only the first of these indicators was Table 1 (continued) Latent construct Variable Category SHARE ELSA (England) N=18,734 (%) N=6117 (%) A-H (adult health) Age 30–50 Percentage of years of non-illness Mean 97.3 86.3 Min 0 0 Max 100 100 Percentage of years of non-poor health Mean 97.5 NA Min 0 NA Max 100 NA O-MW (old age material wealth) Age 55–90 Household income in € (SHARE) and £ (ELSA) Mean 35,290 15,396 SD 54,677 11,832 Min 0 0 Max 755,089 224,203 Missing 2132 11.4 102 1.7 Household wealth in € (SHARE) and £ (ELSA) Mean 161,356 66,842 SD 222,142 162,377 Min −784,644 −81,495 Max 7153,102 3,631,500 Missing 618 3.3 103 1.7 O-H (old age health) Age 55–90 Self-rated health Poor 2336 12.5 32 0.5 Fair 5116 27.3 199 3.3 Good 6895 36.8 1127 18.4 Very good 2886 15.4 2088 34.1 Excellent 1424 7.6 1181 19.3 Missing 77 0.4 1490 24.4 Grip strength Mean 33.6 29.6 SD 12.2 11.4 Min 1 0.0 Max 85 70 Missing 1576 8.4 1502 24.6 Alternative measures Lung function (spirometer) Mean 3.6 3.8 SD 1.7 1.4 Min 0.3 0.4 Max 10.0 9.5 Missing 3612 19.3 1772 29.0 Number of limitations in activities of daily living (ADL) Mean 0.2 0.3 SD 0.7 0.9 Min 0.0 0.0 Max 6.0 6.0 Missing 2585 13.8 160 2.6 To measure limitations in ADL, respondents are asked whether they have any difficulty with (1) dressing, including putting on shoes and socks, (2) walking across a room, (3) bathing or showering, (4) eating, such as cutting up your food, (5) getting in or out of bed, and (6) using the toilet, including getting up or down. They are asked to exclude any difficulties that they expect to last less than 3months 384 European Journal of Ageing (2018) 15:379–391 1 3 available, used as a manifest variable. In old age, health is measured with the indicators current self-rated health (SRH) and grip strength. SRH is considered a good health measure and predictor for mortality. It measures health not only as the absence of disease, but comprehensively (Idler and Benyamini 1997). Grip strength is an objective measure and has been shown to be related to income, and even more so to wealth (Mohd Hairi etal. 2010). In sensitivity analyses, we use an index of six limitations in activities in daily living (ADL) and a measurement of lung function (a spirometer measurement of how much air respondents can exhale, which has been shown to be related to general health) (Sabia etal. 2010). Table1 shows that the distributions of SRH differ strongly between SHARE and ELSA. Since our models exploit the covariation of variables within one data set, this does not bias our results. We do not include further control variables because our aim is to estimate the total effects between material wealth and health, and we expect other variables either to mediate these effects (e.g. health behaviour) or to affect health through income (e.g. occupation). Analysis We chose a model-based approach to study the interplay between material factors and health across the life course. The advantage of a model-based approach, compared to design-based approaches such as quasi-experiments, is the potential for simultaneously modelling two related processes (social causation and health selection) in which the outcome of one process is the predictor of the other. We estimate the parameters of a structural equation model (Bollen 1989; Pakpahan etal. 2015) that includes social causation and health selection in different stages of the life course. Our model is represented by a cross-lagged panel design (Fig.1). We model material wealth and health at three different ages as latent variables with measurement models, except for adult health in ELSA (which only offers one observed variable) and material wealth in old age, where we explicitly compare the results for the observed variables income and wealth. The parameters are estimated using mean and variance-adjusted weighted least squares (WLSMV) (Finney and DiStefano 2006). We present standardized coefficients in a uniform value range of −1 to 1, making them comparable across paths and models. Our model estimates the correlation between material wealth and health in childhood that can be jointly influenced by common unobserved background factors, e.g. genetic factors or unobserved characteristics of the family. Consequently, we address the common background factors mentioned above to the extent that such factors create a correlation between health and wealth in childhood. The path parameters can be divided in two groups: first, the autoregressive parameters showing the effect of wealth at t1 on wealth at t2 (and the same for health); second, the cross-lagged parameters showing how wealth at t1 influences health at t2 (social causation) or health at t1 influences wealth at t2 (health selection). In the SHARE analysis, we use country dummies to control for unobserved national differences. All models are calculated separately for men and women, and age at interview, in 5-year categories reflecting the birth cohort, is also controlled for. Data preparation is performed in Stata 14.1 and analyses in Mplus 7.4 (Muthen and Muthen 2015). Results Results from the structural equation models are shown in Fig.1 (as a graphical illustration of the model and overview with results for SHARE and ELSA), in Table2 (all coefficients, standard errors, and goodness-of-fit measures), and in Fig.2 (only results that are relevant for our main question, the comparison between social causation and health selection). Goodness-of-fit statistics for the measurement models are provided in Online Table1. As expected, the correlation between childhood material wealth (C-MW) and childhood health (C-H) is positive. However, the coefficient is only statistically significant in SHARE. All but one of the autoregressive coefficients for MW and health in both phases of the life course are statistically significant, and they range from 0.01, which means almost no path dependency from adult to old age health (among men in ELSA), to 0.61 in the same group, which means that one standard deviation increase in A-MW is associated with a 0.61 standard deviation increase in O-MW. The effects from C-H to A-MW (health selection) are somewhat stronger than from C-MW to A-H (social causation), but the total cross-lagged effects from childhood to adulthood suggest similar explanatory power for social causation and health selection among men. This is also displayed in Fig.1. Among women in SHARE, health selection seems to be somewhat stronger than social causation, with 0.22 versus 0.08. Altogether, out of eight coefficients that concern reciprocal effects in the first phase, seven are statistically significant, ranging between 0.06 (health selection for men in SHARE) and 0.22 (health selection for women in SHARE), classifiable as relatively small effects that do not show systematic differences between data sets and gender. Summarizing the transition from childhood to adulthood, both reciprocal effects are of equally small importance, with a tendency towards stronger health selection among women. For the transition to old age, our model shows that all coefficients for social causation are positive and statistically significant, and range between 0.34 (women in ELSA) and 0.77 (women in SHARE), showing a strong predictive power of adult MW on health in old age. The coefficients for health 385European Journal of Ageing (2018) 15:379–391 1 3 selection are much smaller than for social causation, around zero, and none of them is statistically significant. The goodness-of-fit indicators show variable results: the Chi-square is always highly significant and the root-meansquare error of approximation (RMSEA) shows good fit for SHARE and acceptable fit for ELSA, the latter with a maximum value of 0.062 (CI 90% 0.057–0.067). The comparative fit index (CFI) and the Tucker–Lewis index are borderline, with ranges between 0.782 and 0.882 and between 0.610 and 0.805, respectively. However, we think that the lower values of CFI and TLI are acceptable, because the goal of the study is the comparison of the social causation and health selection pathways, and not a best possible model of all interrelationships of material factors and health throughout the life course. For a more direct comparison of the two reciprocal effects, Fig.2 shows only the related coefficients for SHARE and ELSA, by life course stages and gender. For men in the transition from childhood to adulthood (Phase 1), both pathways have the same explanatory power and they are both Fig. 1 Structural equation model for reciprocal relationships between material wealth and health over the life course, with standardized coefficients, factor loadings, and simultaneous correlations between health and material wealth, for SHARE (upper panel) and ELSA (lower panel). Notes (a) observed variables are shown as boxes and latent variables as ellipses; uni-directed arrows are path coefficients or factor loading from the measurement models, bi-directed arrows are correlations; (b) SHARE countries in the upper panel are Austria, Germany, Netherlands, France, Switzerland, Belgium, Sweden, Denmark, Spain, Italy; (c) C childhood (0–15), A adulthood (30–50), O old age (55–90), MW material wealth, M/F male/female; (d) bold numbers are statistically significant (p<0.05) 386 European Journal of Ageing (2018) 15:379–391 1 3 statistically significant (indicated by the box-plots), but the difference between them is not statistically significant (indicated by the p values in the graph). Health selection is unimportant during the transition to old age (Phase 2), but social causation is stronger than before and significantly stronger than health selection. Among women, we see similar results in the second phase, but in the first phase, health selection seems to be stronger than social causation in SHARE, which means a reversal of the relative importance of social causation and health selection during the life course among women. The findings are the same for SHARE and ELSA. The same results as in Fig.2, but with wealth as a measure for material wealth in old age, can be found in Online Fig.1. Using wealth, the main findings stay the same, except that social causation and health selection in Phase 1 are equally important for women. In other words, the only exception to the overall pattern that we found in the results based on income disappears when we use wealth. This highlights the fact that alternative measures influence the results, but the overall pattern is robust with different indicators for material wealth. The results are also similar between models where age at interview is controlled for and those where it is not. Table 2 Results from structural equation models on the relationship between material wealth and health over the life course (a) Standardized regression coefficients; SE standard errors, C childhood, A adulthood (30–50), O old age (55–90), MW material wealth; Phase 1=transition from childhood to adulthood; Phase 2=transition from adulthood to old age; (b) for interpretation of the coefficients, e.g. 0.5 means that one standard deviation change in the independent variable results in 0.5 standard deviation change in the dependent variable; (c) statistically significant coefficients (p<0.05) are printed in bold; (d) the total results for SHARE are weighted to account for unequal probability in the sampling process and to represent the different sizes of the population in the countries in Europe Parameter Male Female SHARE ELSA SHARE ELSA Correlation C-MW↔CHEALTH Coef. 0.04 0.04 0.06 0.07 SE 0.02 0.04 0.02 0.04 A-MW↔AHEALTH Coef. 0.07 0.11 0.04 0.37 SE 0.07 0.04 0.08 0.11 O-MW↔OHEALTH Coef. −0.04 −0.11 −0.04 −0.08 SE 0.03 0.07 0.07 0.13 Phase 1 Autoregression C-MW→A-MW Coef. 0.23 0.51 0.36 0.43 SE 0.03 0.04 0.04 0.04 CHEALTH→AHEALTH Coef. 0.15 0.09 0.16 0.43 SE 0.02 0.03 0.02 0.05 Causation C-MW→AHEALTH Coef. 0.07 0.09 0.08 0.07 SE 0.03 0.04 0.03 0.06 Selection CHEALTH→A-MW Coef. 0.06 0.14 0.22 0.09 SE 0.02 0.04 0.04 0.04 Phase 2 Autoregression A-MW→O-MW Coef. 0.38 0.61 0.19 0.52 SE 0.08 0.03 0.05 0.04 AHEALTH→OHEALTH Coef. 0.12 0.01 0.26 0.60 SE 0.02 0.03 0.04 0.08 Causation A-MW→OHEALTH Coef. 0.52 0.47 0.77 0.34 SE 0.08 0.05 0.08 0.08 Selection AHEALTH→O-MW Coef. 0.01 −0.02 0.00 −0.04 SE 0.02 0.02 0.02 0.06 Chi2 2105 458 1393 542 p Value 0 0 0 0 CFI 0.782 0.882 0.877 0.879 TLI 0.610 0.805 0.781 0.800 RMSEA 0.038 0.062 0.029 0.061 RMSEA 90% confidence interval: Lower 0.036 0.057 0.028 0.056 Upper 0.039 0.067 0.031 0.066