Education and Health Over the Life Course
The question of whether education differences in health increase or decrease with age has been examined for more than three decades. Initial research was motivated by two competing hypotheses: the cumulative (dis)advantage hypothesis and the age-as-leveler hypothesis. The cumulative (dis)advantage hypothesis states that education structures the distribution of determinants of health, such as living and working conditions, exposure to stress, social support, and health behaviors, which translates into increasing physical health differences between higher- and lower-educated people over the life course (Ross and Wu
1996). The
age-as-leveler hypothesis states that education differences in health increase only until early old age but decline thereafter (House et al.
1994). This hypothesis stresses the importance of social policy and selection effects. Social policy arguments concentrate on institutional interventions, such as Medicare (providing more equal access to health care) or Social Security (alleviating economic inequality among older adults) (Dannefer
2003). The selection argument attributes the decline of education health differences in older age to selective mortality and selective participation in surveys (Kitagawa and Hauser
1973; Wilkinson
1986).
Empirical tests of these hypotheses have focused on aggregate-level patterns rather than directly testing the individual-level mechanisms proposed by these hypotheses. Education gaps that widened with age have been regarded as support for the cumulative (dis)advantage hypothesis, whereas education gaps that narrowed with age have been regarded as support for the age-as-leveler hypothesis.
Pioneering research on education inequality in health trajectories has produced mixed findings and fueled an intense debate. In the course of this debate, theoretical views and empirical tools have been refined. Methodologically, the most important conclusion is that cross-sectional or short-term longitudinal data are not well suited to examine health trajectories of education groups. Longitudinal data are necessary to account for selection effects and to disentangle age effects from cohort effects because education differences in health trajectories were found to change across cohorts (Lynch
2003; Noymer
2001).
The finding of narrowing health gaps between education groups in older age is no longer considered to contradict the cumulative (dis)advantage hypothesis, as long as this trend results from selective mortality. Instead, higher rates of mortality among the lower-educated are seen as an outcome of processes described by the cumulative (dis)advantage perspective (Dupre
2007; Ferraro et al.
2009; Lynch
2003; Rohwer
2016; Willson et al.
2007). Thus, processes of cumulative (dis)advantage may suppress age-related increases in health inequality.
Current empirical tests of the cumulative (dis)advantage hypothesis assess whether education differences increase before later-life stages in which differential mortality may narrow the gaps. Most of the evidence comes from the United States and is consistent with aggregate-level health trajectories implied by the cumulative (dis)advantage hypothesis. Moreover, U.S. studies have found that this pattern is more pronounced among women and has intensified across cohorts (Brown et al.
2016; Liu and Hummer
2008; Ross and Mirowsky
2010). Evidence from the West German context of the present study, instead, points to a stronger age and cohort increase of education differences in self-rated health among men. Among women, education gaps in health do not widen with age, and this does not change across cohorts (Leopold and Leopold
2018).
Taken together, the advances from previous studies suggest that research on education differences in physical health trajectories should (1) use longitudinal data to account for selection effects and to separate age effects from cohort effects, and (2) examine gender differences given that conclusions about education differences may differ between men and women. In contrast to variation in estimation methods, variation across cohorts, and variation by gender, however, no systematic study has explored how variation in health measures affects conclusions about education differences in trajectories of physical health.
Does the Health Measure Matter?
Most studies have examined change in self-rated health, a measure of general physical health (Brown et al.
2016; Chen et al.
2010; Goesling
2007; Leopold
2016; Leopold and Leopold
2018; Lynch
2003; Mirowsky and Ross
2008; Sacker et al.
2005; Torres et al.
2016; van Kippersluis et al.
2009,
2010; Willson et al.
2007). The main reasons for the dominant role of this measure in the literature are (1) data availability, because self-assessments of health are often included in the core questionnaires of long-running panel studies; (2) validity, because self-rated health correlates with current and future health problems and mortality (Idler and Benyamini
1997; Mossey and Shapiro
1982); and (3) life course coverage, because self-rated health captures health differences across the entire lifespan. This contrasts with measures of specific conditions, symptoms, and functional limitations that become prevalent only at advanced ages (Willson et al.
2007).
Despite these benefits, recent research has raised doubts about the measure of self-rated health. The most important concern is a key assumption underlying studies on inequality: namely, that respondents in different social groups assess their overall health status in similar ways and on the basis of similar criteria, independent of their education level, age, cohort, and gender. An increasing number of studies have suggested that this assumption might be unwarranted (Au and Johnston
2014; Burgard and Chen
2014; Zajacova and Woo
2016).
The first problem concerns education differences in the validity of self-rated health. Studies have shown that for lower-educated persons, this measure is less predictive of mortality (Dowd and Zajacova
2007), and its association with physical health problems as assessed by biomarkers is weaker than for higher-educated persons (Dowd and Zajacova
2010). These findings suggest that education differences in physical health may not be adequately captured by a measure of self-rated health (Dowd and Zajacova
2010; d’Uva et al.
2008; Molina
2016).
Second, increasing evidence suggests that the validity of self-rated health depends on age. As people get older, their frame of reference for assessing their health status changes, and individuals tend to overestimate their health despite an increasing number of physical health problems (Krause and Jay
1994; Peersman et al.
2012). Accordingly, the power of poor self-rated health to predict chronic diseases and mortality weakens substantially with age (Helweg-Larsen et al.
2003; Idler and Cartwright
2018; Lindeboom and van Doorslaer
2004; Schnittker
2005; Van Doorslaer and Gerdtham
2003; Zajacova and Woo
2016).
Third, self-ratings of health vary across cohorts. As demonstrated in a recent study, the predictive power of self-rated health for mortality increases among more recently born (Schnittker and Bacak
2014). The explanation proposed by this study is that because of education expansion, improvements in health knowledge, and medical progress, younger cohorts are better informed about their physical health status and thus provide more accurate self-assessments.
Finally, some doubts have been raised regarding the validity of self-rated health as a measure of gender differences in health. This evidence, however, is not consistent. One recent study found no gender differences in terms of criteria that individuals use to report on their self-rated health status Zajacova et al.
2017). Other studies found that women take milder symptoms and health complaints into account, whereas men focus on more severe or even life-threatening conditions as well as on health behaviors (Benyamini et al.
2003; Grol-Prokopczyk et al.
2011; Peersman et al.
2012). These findings indicate that self-rated health measures may not adequately capture gender differences in physical health.
Taken together, the evidence from these studies questions whether self-rated health accurately measures differences in terms of education, age, cohort, and gender. Given that all of these four factors represent key analytical constructs in studies of health inequality over the life course, this raises doubts about the conclusions of previous research. In studies on the cumulative (dis)advantage hypothesis, however, these issues are rarely considered because self-rated health remains the primary outcome measure.
A few studies have complemented or replaced self-rated health with alternative measures, although these measures have also relied on respondents’ self-reports (House et al.
2005; Hu et al.
2016; Kim
2008; Kim and Durden
2007; Leopold
2016; Torres et al.
2016; van Kippersluis et al.
2009). Among the measures used are different indices of self-reported physical impairments or functional limitations. These measures are based on a battery of questions about limitations in daily activities, such as walking several blocks, climbing several flights of stairs, lifting something heavy, or picking up a coin from a table. Studies on the intergroup validity of these measures, however, have suggested that concerns about self-rated health may apply similarly to self-reported measures that are based on specific health problems (Burgard and Chen
2014; Molina
2016; Ziebarth
2010).
A potential solution to these problems is to complement subjective measures by objective measures of health. Although no study has used objective measures in tests of the cumulative (dis)advantage hypothesis, the literature suggests that they may constitute a suitable alternative to self-reported measures of health given that they are not affected by systematic group differences in reporting behavior (Burgard and Chen
2014; Ploubidis and Grungy
2011).
In particular, grip strength (as assessed by a dynamometer) is considered a valid indicator of general physical health status that captures changes in health across adulthood (Peterson et al.
2016). Because of its potential as a low-cost and noninvasive objective measure of overall health in population studies, the properties and validity of grip strength have been studied extensively over the past decade. Meta-studies have shown that similar to self-rated health and other self-reported health measures, grip strength predicts all-cause and cause-specific mortality, and correlates with current and future disability (Bohannon
2001; de Lima et al.
2017; Rijk et al.
2016; Wu et al.
2017). Moreover, grip strength was found to be strongly related to aging, declining from mid-adulthood onward (Steiber
2016; Vianna et al.
2007).
The main factors that underlie the decline of grip strength with age are reductions in muscle strength and skeletal muscle mass (Abe et al.
2014). This decline reflects biological processes that lead to a number of physical conditions, diseases, and causes of death. First, declining muscle strength is caused by a decrease in serum levels of testosterone and adrenal androgens, increased action of inflammatory mediators, and the onset of anabolic resistance (Montalcini et al.
2013; Schlüssel et al.
2008). Changes in these factors are often caused by declines in physical activity as a result of chronic and acute diseases, injuries, and (to a smaller degree) stress and depression (Syddall et al.
2003). Moreover, low grip strength is a strong predictor of falls, physical disability, and frailty, all of which lead to increased mortality risk, especially in old age.
Second, grip strength is an indicator of skeletal muscle mass, which is responsible for the disposal of blood glucose. The amount of skeletal muscle mass indicates the ability to respond to insulin, which in turn is a major predictor of metabolic diseases and mortality from metabolic diseases, such as type 2 diabetes. Moreover, sarcopenia (degenerative loss of skeletal muscle mass) is associated with chronic low-grade inflammation, which occurs in a number of common chronic diseases that lead to premature mortality (Metter et al.
1997; Peterson et al.
2016).
Although grip strength is per definition not affected by reporting heterogeneity, biological differences between men and women have to be taken into account. Men have more muscle mass than women because of higher plasma concentrations of the major anabolic hormones (testosterone, GH, and IGF-1) but also because of more physical activity on the job and during leisure time (Montalcini et al.
2013). Moreover, men (but not women) with higher body weight and higher levels of body fat—common predictors for health impairment—have higher grip strength (Gale et al.
2007). Although gender differences in the predictive validity of grip strength as a general health measure are not fully understood, the most recent and comprehensive meta-study has concluded that there are no substantial gender differences in the relationship between grip strength and all-cause mortality as well as mortality from cardiovascular diseases, stroke, coronary heart disease, and cancer (Wu et al.
2017).
Taken together, the literature on health measures suggests that grip strength may accurately capture group differences arising from the unequal distribution of health-related resources and exposures to health risks over the life course. These benefits render grip strength an interesting alternative to self-rated health or other self-reported health measures to study the core association between education and health, as well as how it changes with age, across cohorts, and for men and women.