Education and Physical Health Trajectories in Later Life: A Comparative Study

Research has consistently shown that health gaps between higher- and lower-educated people increase over the life course (Dupre 2007; Kim 2008; Kim and Durden 2007; Leopold 2016; Lynch 2003; Mirowsky and Ross 2008; Willson et al. 2007). This evidence supports the cumulative (dis)advantage hypothesis, which states that education is associated with health-related advantages and disadvantages that accumulate with age, enforcing a steady increase in health disparities (Ross and Wu 1996).

In recent years, tests of the cumulative (dis)advantage hypothesis have been refined by greater attention to the social conditions in which individual health trajectories unfold. In an influential line of research demonstrating the importance of sociohistorical context, cohort studies have indicated a “rising importance” (Mirowsky and Ross 2008) of the cumulative (dis)advantage process (Delaruelle et al. 2015; Goesling 2007; Kim 2008; Lynch 2003). According to these studies, newer cohorts are increasingly exposed to conditions that intensify the cumulative advantages and disadvantages that education brings, including rising inequality in economic returns to education, exposure to environmental stressors, health knowledge, and health behaviors (Goesling 2007).

Although these cohort studies have shown that health inequality strongly depends on the context in which individual life courses unfold, cross-national comparative studies of health trajectories remain scarce (Stolz et al. 2017). Longitudinal evidence on how education shapes health across the life course is largely limited to the U.S. context. This limitation is important given that other country contexts introduce variation in key institutional factors that target inequalities in health. Among these factors are education systems that are more or less inclusive or stratifying, employment policies that are more or less protective in terms of unemployment and exposure to adverse working conditions, public support after negative life events that is more or less generous, and social policy interventions that are more or less effective in targeting risky health behaviors (Beckfield et al. 2015).

Although comparative studies have examined health inequality in the United States and other developed societies (Avendano et al. 2009, 2010), these assessments were based on cross-sectional data that preclude the separation of age and cohort effects—a key requirement in the study of health trajectories (Lynch 2003). Recent research has added further single-country studies that offer longitudinal assessments of the cumulative (dis)advantage hypothesis (Chen et al. 2010; Leopold 2016), but this evidence remains limited in comparative perspective because the studies varied strongly in terms of sampling frames, observation periods, scope of panel data, measures of health, and estimation methods.

In view of these limitations, I designed the present study to provide a comparative longitudinal investigation of differences in health trajectories between education groups. I analyzed data from the United States, the United Kingdom, the Netherlands, and Sweden—four countries that offer sharp contrasts with regard to the social conditions shaping individual health trajectories. Among developed societies, the United States and Sweden can be placed at opposite poles in terms of institutional factors that promote or prevent disparities in health-related resources, resulting in larger or smaller inequalities in terms of social security, access to health care, working conditions, educational opportunity, economic means, risky health behaviors, and exposure to stress after adverse life events. The United Kingdom and the Netherlands can be placed between these two poles, allowing for a more nuanced comparative assessment.

The analysis was based on panel data from the Health and Retirement Study (HRS), the English Longitudinal Study of Ageing (ELSA), and the Survey of Health, Ageing and Retirement in Europe (SHARE). These surveys are ideally suited for longitudinal comparative research because the data are highly similar in terms of sampling frames, observation periods, cohort range, and measures of education and health. Given that the samples of all three surveys include only middle-aged and older people, my analysis focuses on health trajectories in later life.

The cumulative (dis)advantage hypothesis describes a process by which initial advantages or disadvantages associated with a structural position translate into a systematic divergence in life course resources, opportunities, and risks (Dannefer 1987; Ferraro and Shippee 2009; Ferraro et al. 2009; O’Rand 1996). Applied to education and health, the cumulative (dis)advantage hypothesis posits that education structures advantages and disadvantages in key determinants of health, producing widening health gaps between education groups as people age (Lynch 2003; Mirowsky and Ross 2008; Ross and Wu 1996).

Education reproduces and magnifies early advantages and disadvantages of social background and strongly determines income, occupational status, and wealth in later life (Kerckhoff 1995; Spring 1976). Depending on social background, children grow up in stable or unstable families, attend better or worse schools, earn more or less, reach higher or lower occupational positions, and experience more or less of the allostatic load of stress associated with economic hardship, unemployment, and other negative life events (McEwen 1998). Moreover, those who attain higher education increase their capacity of processing information and their sense of personal control (Mirowsky and Ross 2007)—skills that contribute to acquiring and maintaining a healthy lifestyle.

Important aspects of the process of cumulative (dis)advantage are education differences in the timing and duration of exposure to stress, hardship, and unhealthy lifestyles as well as the occurrence and consequences of negative life events. Smokers and nonsmokers, for example, are almost equally healthy in their 20s. In middle and later stages of the life course, differences in functional limitations and chronic conditions that are attributable to smoking gradually unfold. The same applies to other health-related factors that are structured along educational lines. Because it takes years until adverse working conditions, economic hardship, and exposure to stress take their toll on health (Shuey and Willson 2014), the cumulative (dis)advantage hypothesis posits that education gaps widen steadily with age.

Figures 2 and 3 show model-based predictions for initial health levels and for change in health among the higher-educated (solid lines) and the lower-educated (dashed lines). The left upper panel shows the results for Sweden, the right upper panel shows the results for the Netherlands, the left bottom panel shows the results for the United Kingdom, and the right bottom panel shows the results for the United States. To account for cohort effects, I fixed the variable for cohort at six values of age at initial observation, corresponding to the birth years of 1939, 1942, 1945, 1948, 1951, and 1954.³ The y-axis indicates the predicted average number of chronic conditions (Fig. 2) and the predicted average number of functional limitations (Fig. 3). In both figures, higher values indicate worse health. The scales are identical across countries, and a grey dashed reference line is included to ease the comparison of effect sizes.

The x-axis shows each cohort’s age at the beginning and at the end of the observation period. The youngest cohort, for example, was observed from age 50 until approximately age 60. An important advantage of the graphical age-vector analysis is that it shows not only education gaps in age trajectories of health but also cohort differences in these trajectories as well as the associated levels of health (Mirowsky and Kim 2007). Cohort effects are visible at the age overlaps across cohorts. If cohort effects are small, the cohort-specific curves connect; if cohort effects are larger, the pattern appears increasingly ragged.

The findings illustrated in Figs. 2 and 3, and the associated Tables 3 and 4, provide answers to the three hypotheses guiding the present study. First, with regard to average levels of health, I found the expected cross-national gradient using both measures of health. In the United States, the number of chronic conditions and functional limitations was higher than in the United Kingdom and much higher than in the Netherlands and Sweden. This cross-national gradient was most pronounced for functional limitations: it applied to higher- and lower-educated people in all cohorts and across the age range covered. A comparison between the United States and Sweden illustrates the magnitude of cross-national differences: the dashed curves pertaining to lower-educated Swedes remained below all corresponding curves for the United States across the range of ages and cohorts. Thus, lower-educated Swedes were in better health than both lower-educated and higher-educated Americans. Health levels of higher-educated Americans were comparable to those of lower-educated people in the United Kingdom and the Netherlands. The estimated number of chronic conditions and functional limitations in lower-educated Americans vastly exceeded those for all other groups.

Second, with regard to the magnitude of health gaps between education groups, the results show the smallest gaps in Sweden, followed by the Netherlands, the United Kingdom, and the United States. This cross-national gradient of health inequality in later life was most pronounced for functional limitations (Fig. 3) but also visible for chronic conditions (Fig. 2). Beyond this general pattern, the age-vector graphs indicate cross-national differences in terms of cohort effects. In the United States, the United Kingdom, and the Netherlands, most of the curves show a ragged cohort pattern, whereas the curves tend to connect in Sweden. Cohort effects were most pronounced among lower-educated Americans. In this group, the number of chronic conditions and functional limitations grew rapidly across cohorts observed at the same ages. As a result, the magnitude of health gaps increased across cohorts. In the United Kingdom, cohort effects did not differ appreciably between both education groups for either measure of health, resulting in health gaps that remained stable across cohorts. In the Netherlands, higher-educated people showed health improvements across cohorts, whereas lower-educated people showed health declines across cohorts. These diverging trends applied to both measures of health, resulting in a cross-cohort trend of growing health gaps between education groups.

Third, with regard to cumulative (dis)advantage over the later life course, Figs. 2 and 3 indicate the extent to which health gaps between education groups increased with age. Health gaps at initial observation were largest in the United States and smallest in Sweden, with the United Kingdom and the Netherlands positioned in between. These cross-national differences were more pronounced in functional limitations than in chronic conditions and applied for most birth cohorts. For example, among people born around 1950,⁴ education differences in the number of functional limitations at the initial observation at age 53 amounted to only 0.4 in Sweden. These differences were twice as large in the Netherlands, 2.5 times larger in the United Kingdom, and 3.5 times larger in the United States (see Table 4). As this cohort moved through their 50s, the gaps increased to 0.65 in Sweden, 0.91 in the Netherlands, 1.1 in the United Kingdom, and 1.7 in the United States. However, in Sweden, the Netherlands, and the United States, this divergent pattern in functional limitations emerged only in more recent cohorts and within the age range between the early 50s and the late 60s, after which the observation windows for younger cohorts were censored. In the United Kingdom, education differences in the number of functional limitations increased in each cohort and across the entire age range 50–76. For the other health measure—number of chronic conditions—results showed an age-related increase in education gaps for each country and each cohort (see Table 3), consistent with the pattern postulated by the cumulative (dis)advantages hypothesis.

Taken together, the findings for most of the cohorts and for both health measures showed a clear gradient across the four countries in the size of health gaps at initial and final observation. Gaps were largest in the United States, followed by the United Kingdom, the Netherlands, and Sweden. Differences between the United States and all three European countries were large, whereas differences between the European countries were smaller.

Although the general pattern of findings supported my expectations, it was important to examine a number of factors that could influence comparative conclusions about health inequality. I conducted four additional analyses to assess the role of (1) gender differences, (2) racial differences, (3) regional differences, and (4) selective attrition.

According to the cumulative (dis)advantage hypothesis, education reproduces and magnifies differences in health-related resources over the life course, producing growing education gaps in health from younger age to older age (Ross and Wu 1996). The present study focused on cross-national differences in this process, using comparative data on health trajectories in four countries. Sweden and the United States offer sharp contrasts with respect to inequality in health-related resources and risks over the life course (DiPrete 2002). The Netherlands and the UK can be placed between these poles, allowing for a nuanced test of comparative arguments.

First, across all ages and cohorts, physical health problems were more prevalent in older Americans among both the lower-educated and the higher-educated. Second, health gaps by education were largest in the United States, followed by the United Kingdom, the Netherlands, and Sweden. These differences in the magnitude of health differences between education groups indicate that the divergent pattern associated with the cumulative (dis)advantage hypothesis was most pronounced in the United States, less pronounced in the United Kingdom and the Netherlands, and least pronounced in Sweden. Although my data did not cover life course stages before the age of 50, my results for health differences between education groups at the start of the observation window suggest that these gaps opened up at younger ages.

Third, initial health gaps further widened across the observation window in all countries and in most of the cohorts studied within each country. These findings lend support to the cumulative (dis)advantage hypothesis for health trajectories in later life. As expected, the extent to which health inequality increased with age differed cross-nationally, indicating that divergence was smallest in Sweden, larger in the Netherlands and the United Kingdom, and largest in the United States.

In contrast to previous comparative studies of health inequality, these findings are based on a longitudinal design and comparable measures of education and health. Furthermore, unlike previous cross-sectional research, my findings are corroborated by attrition analyses that examined whether observed country differences in health inequality were biased by selective dropout. Additional sensitivity analyses showed that the cross-country gradient found in this study applied to men and women and was robust to racial and regional differences within the United States.

The longitudinal comparative data used in this study allowed me to account for cohort effects, an important requirement for the analysis of age trajectories in health (Lynch 2003). Although cohort effects should be interpreted with caution given the limited age overlaps between cohorts in my data, some notable patterns emerged from the analysis. In the United States, cohort effects were strong and consistent with the rising importance hypothesis, which posits that cumulative (dis)advantage has intensified in newer cohorts (Goesling 2007; Mirowsky and Ross 2008). This trend was mainly due to faster health declines among lower-educated Americans rather than slower declines among higher-educated Americans. I also found a pattern of rising importance for the Netherlands, where education groups increasingly drifted apart, resulting from accelerating health declines among the lower-educated combined with slowing health declines among the higher-educated. In the United Kingdom and Sweden, the data showed no indication of a rising importance of cumulative (dis)advantage in physical health trajectories.

The differences found across countries in the size of cohort effects are broadly in line with substantive arguments about the sociohistorical conditions surrounding successive birth cohorts. In the United States, the finding of increasing disparities in health by education has been attributed to a sharp and steady increase in economic inequality since the 1970s (Hayward et al. 2015; Mirowsky and Ross 2008). This rise of inequality was much larger in the United States than in other developed societies, including the United Kingdom and the Netherlands (Bambra and Beckfield 2012). In Sweden, inequality in health-related resources increased only slightly across the cohorts covered in the present study. These cohorts spent major stages of adult life during the golden age of the Swedish welfare state (Esping-Andersen 1999). Yet, although Sweden still ranks among the least unequal countries, income inequality has increased since the 1990s; the welfare state has been cut back; and social security, health care, and the labor market have become more liberal (Freeman et al. 2010). These changes suggest that health inequality could increase in newer cohorts, and this idea was supported in a recent study of cumulative (dis)advantage in Sweden (Leopold 2016).

Cross-country differences in cohort effects constitute another important source of contextual variation that should be addressed in future research. The data I used were not ideal to study these differences in detail, given the limitations regarding cohort range, observation periods, and overlaps between age and cohort. Internationally comparative panel data that cover more birth cohorts and follow them for a longer period of time are needed to analyze change in cross-country differences in health inequality.

Perhaps the most striking comparative result I found concerns cross-national differences in older people’s health found across all education groups. Most notably, higher-educated Americans were in similar or worse health than lower-educated people in the United Kingdom, the Netherlands, and Sweden. Confidence in this finding is strengthened by specific analytical benefits of the present study, which compared education groups that belonged to similar birth cohorts, were observed in similar historical periods, reported on the same health measures, and were not affected by compositional change related to differential rates of attrition. Moreover, this finding is consistent with studies by Avendano et al. (2009, 2010) reporting substantial disadvantages of wealthy and higher-educated Americans compared with Europeans in terms of life expectancy, chronic conditions, and functional limitations. Similarly, Banks et al. (2006) found that Americans belonging to the top third of the income distribution were comparable with those in the bottom third of the income distribution in England with regard to self-reported hypertension, diabetes, and a wide range of biological measures. Similar to the current study, these results were not driven by compositional effects with respect to race or region.

The present study adds to these findings by showing that the health disadvantage of lower- and higher-educated Americans compared with their Swedish, Dutch, and British counterparts did not level off with age or across cohorts. Explanations that have been advanced for these patterns are societal factors that expose both lower-educated and higher-educated Americans to more adverse living conditions and higher levels of stress. These societal factors include the obesity epidemic as well as social policy regarding infrastructure, employment protection, housing, family, and health care (Avendano and Kawachi 2014). In view of the current study’s results, a more detailed examination of the mechanisms underlying the prevalence of health problems among higher-educated Americans appears worthwhile.

On a more general level, this study’s findings add to knowledge on cross-national differences in health inequality between education groups. Previous comparative research was based largely on cross-sectional data (Avendano et al. 2010; Jürges 2010) and could not provide insight into life course–specific and cohort-specific processes that produce health gaps between education groups. My results have shown that these processes vary substantially across developed countries. An objective for future research is to study the mechanisms that underlie these differences. Addressing such mechanisms empirically is beyond the scope of this study, but the results support the general idea that education differences in health are profoundly shaped by contextual factors.

Understanding the role of these factors would require extending the comparative life course perspective to cover earlier stages in which health differences between education groups emerge. In this regard, I discussed a number of institutional factors that are relevant at different stages of life and in different domains of life and have the potential to intensify or to inhibit the increase in health inequality across lives and cohorts. Most, but not all, of the institutional factors are in line with cross-country gradient found in the analyses (United States > United Kingdom > Netherlands > Sweden). For example, some of the factors discussed suggest that the Dutch education system is more conducive to the growth of inequality over the life course than education systems in Sweden, the United Kingdom, and the United States. My findings on relatively small health differences between education groups in the Netherlands suggest that if these factors operate in the expected direction, they are less influential than countervailing factors. Future research is needed to understand the role of particular contextual factors in cross-country differences in health inequality over the life course.

Because addressing all these explanations simultaneously is difficult, future research should also examine key factors in detail. For example, the institutional factors that I outlined suggest country differences in stress and economic hardship, which may harm health directly or indirectly by triggering unhealthy behaviors. From a comparative point of view, the adverse effects of critical life events, such as union dissolution or job loss, could be larger in liberal economies (such as the United States and the United Kingdom) and smaller in protective welfare states (such as Sweden and the Netherlands).

Another important question for research and policy concerns country differences in terms of selection into education based on social background and health in childhood. Numerous studies have shown that parental education as well as early-life health and economic hardship predict educational attainment and partly explain the relationship between education and adult health (Behrman et al. 2011; Duke and Macmillan 2016; Lynch and Hippel 2015; Ross and Mirowsky 2011). In the absence of comparative studies, the role of institutional factors in moderating health selection into education remains unexplored. Certain features of education systems, such as an early tracking age, appear likely to tighten the link between childhood health and educational attainment. Studies that examine these connections would also clarify the extent to which cross-national differences in health inequality are due to social background and health selection into education.

Comparative research along these lines will improve our understanding of how to target health inequality by policy measures. The results of the present study suggest that the potential for reduction is substantial.