| HOME | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|---|
| ||||||||||||||||||||||||||||||||
TOPIC 1. STRUCTURAL PATTERNS OF HEALTH INEQUALITIES |
1 Survey Research Center and Department of Sociology
2 School of Public Health and Survey Research Center, University of Michigan, Ann Arbor.
3 Lyndon B. Johnson School of Public Affairs, University of Texas at Austin.
Address correspondence to James S. House, PhD, Institute for Social Research, 426 Thompson St., Box 1248, Ann Arbor, MI 48106. E-mail: jimhouse{at}umich.edu
Abstract
Objectives. This article overviews previously published and ongoing research from the Americans' Changing Lives (ACL) Study, a longitudinal study of a nationally representative sample of 3,617 adults aged 25 years and older when first interviewed in 1986, focusing on socioeconomic disparities in the way health changes with age during middle and later life, especially in terms of compression of morbidity/functional limitations.
Methods. A variety of descriptive and multivariate regression and growth curve analyses are done on the ACL sample, now surveyed over four waves spanning 15.5 years between 1986 and 2001/2002 with continuing mortality ascertainment via the National Death Index, death certificate searches, and informant reports.
Results. Both cross-sectional and longitudinal analyses indicate that socioeconomic disparities in health are small in early adulthood, increase through middle and early old age, and then lessen again in later old age. In other terms, compression of morbidity/functional limitations into the later stages of the life course is realized to a much greater degree among the better educated compared with the less educated. Cross-sectional evidence suggests that this reflects differential exposure to or experience of a wide range of psychosocial, environmental, and biomedical risk factors for health (and perhaps their differential impact at different ages and life stages), as well as variations in biological robustness and frailty and also perhaps in the strength of social welfare supports for health at different life stages. Longitudinal analyses reveal several new insights: (a) The flow of causality is much greater from socioeconomic position to health than vice versa; (b) education plays a greater role relative to income in the onset of functional limitations, whereas income has much stronger effects on their progression or course; and (c) educational disparities in the onset and hence of compression of functional limitations over the life course have increased strikingly in later middle and early old age (ages 55–84 years) since 1986.
Discussion. The results indicate that understanding and alleviating social disparities in health are both theoretically and methodologically quintessential problems of life course analysis and research.
A paper (House, Kessler, Herzog, Mero, Kinney, & Breslow, 1992
) presented at a conference on Aging, Health Behaviors, and Health Outcomes at the Pennsylvania State University in 1989 and subsequently published in an edited volume from that conference (Schaie, Blazer, & House, 1992) provided the first integrative summary of ideas and initial data that were then emerging from the Americans' Changing Lives (ACL) Study, a nationally representative sample of adults aged 25 years and older, first interviewed in 1986 and reinterviewed in 1989, 1994, and 2001/2002, as described more fully below. The ACL Study was designed to address one of the central dilemmas of research on aging and health in the late 1970s and early 1980s: Was the increasing life expectancy of the population of the United States and other developed countries foreshadowing a dystopian scenario of "longer life but worsening health" (Verbrugge, 1984) in which people are increasingly chronically ill and functionally limited and disabled, and thus increasing consumers of expensive medical and long-term care (see also Gruenberg, 1977; Manton, 1982; Schneider & Brody, 1983), or was increasing life expectancy adding "life to years" as well as "years to life" (Hauser, 1953) because increased understanding of psychosocial as well as biomedical risk factors for health were then or soon would be postponing or "compressing" the onset of serious morbidity and attendant functional limitations and disability into increasingly later years of a human life span that was relatively finite and subject to at most slight change relative to the large and continuing increases in life expectancy of the 19th and 20th centuries (Fries, 1980; Fries & Crapo, 1981).
At that time, the empirical validity of Fries's "compression of morbidity" hypothesis was hotly contested (e.g., Gerontological Perspecta, 1987), but evidence since the late 1980s has increasingly indicated that functional limitations and disability are declining in the older portions of the U.S. population (e.g., Crimmins & Saito, 2001; Manton & Stallard, 1991). Other evidence suggests that interventions to improve health risk factor profiles in middle and older age (and presumably also earlier in the life course) can postpone or compress the onset of health-compromising morbidity (Fries, 2001). The ACL Study was designed to understand the role of a broad range of psychosocial and behavioral factors—ranging from health behaviors to chronic and acute stress, social relationships and supports, productive activities, and personality dispositions—in enabling and predicting what the National Institute on Aging RFA that funded the ACL Study termed the "maintenance of health and effective functioning in middle and later life."
The ACL research grew out of and was initially framed in terms of the rapidly growing body of theory and evidence deriving from the then and still active tradition of "stress and adaptation/coping" research (e.g., Cohen, Kessler, & Gordon, 1995; House, 1981; Levine & Scotch, 1970). Thus, the ACL survey tried to assess a variety of indicators of health and effective functioning and a broad range of psychosocial and behavioral factors that existing theory and research suggested were or should be predictive of the degree to which individuals were able to maintain health and effective functioning as they aged through middle and later life. The goal was that ACL be a long-term longitudinal study in some sense testing out Fries's compression of morbidity hypothesis, though the initial program project grant included funding for only two waves of data collection spaced 2.5 years apart in 1986 and 1989.
REDISCOVERING HEALTH DISPARITIES OVER THE LIFE COURSE
Over the same period of the late 1970s and early 1980s, the United States and other developed countries rediscovered the existence of large, persisting, and perhaps even increasing socioeconomic as well as racial–ethnic disparities in health (Black, Morris, Smith, & Townsend, 1982; House, 2002; House & Williams, 2000; Marmot, Kogevinas, & Elston, 1987; Pappas, Queen, Hadden, & Fisher, 1993). The ACL researchers had their own epiphany in this regard when, examining factors that influenced the cross-sectional relationship between age and health in the 1986 ACL baseline survey, they discovered that socioeconomic factors, especially education and income, were by far the strongest determinants of the degree to which health deteriorated or was maintained with increasing age. Further, socioeconomic disparities were not constant over the adult life course, but were small or nonexistent in early adulthood and later old age and increasingly large during the period between early adulthood and early old age (House, Kessler, Herzog, Mero, Kinney, & Breslow, 1990; House et al., 1992; House, Lepkowski, Kinney, Mero, Kessler, & Herzog, 1994)—a pattern of life course variation that has generally been consistently observed in other nationally representative data before and since (e.g., Beckett, 2000; Herd, in press; Kitagawa & Hauser, 1973; Newacheck, Butler, Harper, Piontkowski, & Franks, 1980).
Reframed in terms of the "compression of morbidity" debate, the evidence from ACL, replicated in the National Health Interview Survey (NHIS; see House et al., 1990), suggested that socioeconomically advantaged portions of the U.S. population were increasingly experiencing compression of morbidity and functional limitations into the later years of their lives, whereas the least advantaged manifested virtually linear declines in health and functional status over their entire adult life course. This is illustrated in Figure 1 (from House et al., 1994) which shows the proportion of people reporting no functional limitations (adjusted for race and sex) in the 1986 ACL baseline survey (described more fully below) across six age groups (25–34, 35–44, ..., 75+ years) and three levels of education (0–11, 12–15, and 16+ years). The measure of functional limitations ranged from being confined to a bed or chair through being able to "walk several blocks or climb several flights of stairs" with "little or no difficulty" to being able to "do heavy work around the house like shoveling snow or washing walls" with "little or no difficulty." Those reporting no functional limitations were able to do all of these things with little or no difficulty.
|
Disparities by income on the same measure of functional limitations were similar in form and as large as or larger than those seen for education in Figure 1 (House et al., 1990, 1992). In the cross section, education and income almost always produced similar results, and so we have focused on education in our cross-sectional work (House et al., 1994) where its temporal and hence putative causal priority with respect to health in middle and later life is much less ambiguous than is the case for income. However, when we turn to longer-term longitudinal work below, we will see emerging evidence and theory that education and income may play different roles with respect to health and the way health changes as people age over the adult life course. We have also essentially replicated the results in Figure 1 (and for income) using other indicators of health including both chronic conditions and self-rated limitations in both the 1986 ACL cross-section and the NHIS for 1985 (House et al., 1990).
Our analyses and the similar findings of others for other measures of health (e.g., mortality) and different samples/populations suggest that the apparent social stratification of the "compression" or postponement of morbidity/functional limitations (and also mortality) over the life course is an empirically robust phenomenon, at least in cross-sectional data. They also support the focus of our ACL research and now broader Public Health Service and National Institutes of Health (U.S. Department of Health and Human Services, 2002; Varmus, 1999) health research and policy on two major and related objectives, currently articulated as the two primary overarching goals of the Public Health Service's policy-defining Healthy People 2010 (U.S. Department of Health and Human Services, 2002): understanding and increasing health and active life expectancy or the "compression of morbidity" and functional limitations; and understanding and reducing health disparities by socioeconomic position (SEP) and race/ethnicity.
Beyond the obvious concern of social equity and justice, the major reason why understanding and reducing socioeconomic and racial/ethnic disparities are and should be priorities for health research and policy is that reducing such health disparities constitutes the major and most promising opportunity for improving overall population health and compression of morbidity and functional limitations in the United States (as well as many other nations); and hence arresting the decline in the relative population health standing of the United States in comparison with a growing number of developed and even some less developed countries, despite our spending far more than any nation on health care and health research. This is because the most socioeconomically advantaged portions of our (and other) populations are achieving increasingly optimal levels of compression of morbidity and functional limitations, leaving less and less room for further improvements, although these improvements are still ongoing as will be seen below. In contrast, as seen in Figure 1, lower socioeconomic groups have enormous room for improvements in health status and compression of morbidity/functional limitations, especially in the United States.
Making headway toward these goals however, requires that we move beyond our and others' early empirical work in two ways. First, we need a better theoretical understanding of how and why education and income (and perhaps other variables of social stratification such as occupation and wealth) have such strong and persisting relationships to health. Although other dimensions of stratification, such as occupation and wealth, are appropriately receiving increased attention in our and others' work, education and income have received priority in our and others' work at least in the United States, for several reasons: (a) Education (at least in terms of years) and income are each relatively easy to measure in a simple metric across the total range of the population in terms of both age and social stratification; (b) evidence is still limited that variables such as occupation or wealth add appreciably to our ability to understand the social stratification of aging and health across the full range of the population and adult life course, net of education and income; and (c) both education and income are amenable to change via the efforts of individuals and planned and unplanned social interventions and change in both the private and the public sectors.
Second, we need to move empirically from collecting and analyzing mainly cross-sectional and short-term longitudinal data to developing and analyzing long-term prospective studies of multiple indicators of health and functioning in representative national populations. We have been engaged in both of these processes for over 15 years in our ACL Study, which represents only one small component part of the burgeoning fields of research on our and the Public Health Service/National Institutes of Health priorities of understanding and reducing socioeconomic disparities in health while increasing the compression of morbidity/functional limitations. However, our work remains unique and distinctive at this point in following a nationally representative cohort of adults aged 25 years and over for a long period of time, now 15.5 years on average for our ACL respondents. The remainder of this article reviews the current state of our and more general theoretical thinking on these issues, the initial testing of these ideas in our own cross-sectional and longitudinal data, and the more recent work we have done and are doing in prospective data over 7.5–15.5 years of follow-up.
DEVELOPING THEORY AND SOME CONFIRMING DATA ON THE SOCIAL STRATIFICATION OF AGING AND HEALTH
Still-developing bodies of theory and data indicate why SEP manifests such a strong, presumably causal, relationship with health and the way health changes with age. We increasingly understand that SEP is a powerful master status or "fundamental cause" (Link & Phelan, 1995) that shapes people's exposure to and experience of almost all risk factors for health—past, present, and future. Figure 2 (House, 2002) graphically depicts how indicators of SEP such as education and income are products of macrosocioeconomic conditions and policy as well as ascribed statuses or positions in terms of race/ethnicity, gender, and age. Education and income in turn shape people's access to and utilization of medical care and insurance, but also their health behaviors, social relationships and support, chronic and acute stress, psychological dispositions, and social roles and productive activities, all of which have been shown to predict morbidity and/or mortality in prospective studies (House, 2002; House & Williams, 1996). Education and income also affect people's exposure to physical, chemical, biological and environmental hazards—ranging from infectious disease agents to toxic chemical and physical conditions at home and work—and to social environmental hazards such as lack of safety and security at work and especially at home (Davey-Smith, Neaton, & Wentworth, 1998; Diez Roux, 1998; Krieger, Rowley, Herman, Avery, & Phillips, 1993). These explanatory factors in turn affect a wide range of health outcomes directly (e.g., accidents) or via psychoneuroendocrine pathways not shown in Figure 2 (e.g., blood pressure, immune response) through which environmental psychological factors "get under the skin." There is a rich literature on these psychoneuroendocrine processes that is beyond the scope of this article (Ader, Felton, & Cohen, 1991; McEwen & Wingfield, 2000; Taylor, Repetti, & Seemen, 1997).
|
). Figures 3 and 4 show that the same is true among ACL adults of working age (where socioeconomic disparities are greatest) across a broad range of psychological dispositions (hostility and self-efficacy), social relationships and supports, negative affective states, and life event stresses. We do not have in the early waves of ACL good measures of access to and use of medical care and insurance and environmental hazards, but evidence noted above indicates that these are also both differentially distributed by education and income and consequential predictors of health and hence further potential explanations of socioeconomic health disparities. Many forms of chronic stress—financial, marital, and familial—also are more prevalent at lower levels of education and income and predictive of adverse health outcomes (e.g., Cohen et al., 1995; Lantz, House, Mero, & Williams, in press; Turner, 1995).
|
|
can account for a modest portion of these differences (House et al., 1992, 1994; Lantz et al., 1998, 2001). However, taken together, a full set of risk factors may account for or explain how and why SEP is so strongly associated with health and the way health changes with age, doing so more or less fully depending on the number of risk factors considered in any single analysis. To this point, we have examined this possibility most fully in our 1986 ACL cross-sectional data, where, as shown in Figure 5, adjusting for income and a total of 11 risk factors (including health behaviors, several indicators of chronic and life event stress, several measures of social relationships and supports, and an index of self-efficacy) accounted for one half to two thirds of the differences we observed by education in Figure 1. We find similar results with respect to income disparities in health and the way health changes with age and clearly are missing in these analyses adequate measures of other major explanatory variables such as health insurance, health care utilization, some key psychological dispositions (e.g., anger/hostility), and social and physical–chemical–biological environmental exposures. Others have shown similar results with different sets of risk factors and different indicators of SEP and health (Lynch, Kaplan, Cohen, Tuomilehto, & Salonen, 1996; Marmot et al., 1991), suggesting that a more comprehensive set of risk factors might fully explain these educational and income disparities in health and the way health changes with age.
|
FROM CROSS-SECTIONAL (AND SHORT-TERM LONGITUDINAL) ANALYSIS TO LONGER-TERM PROSPECTIVE STUDIES
The "Age, Period, Cohort" Problem
Age differences in cross-sectional research may represent patterns of aging over the life course, but they may also be the product of historical or period effects that produce differences between birth cohorts that manifest themselves as age differences in the cross-section. For example, members of the age group (55–64 years) in which we observe the greatest socioeconomic disparities in health in our ACL in 1986 were born between 1922 and 1931 and thus grew into adulthood primarily during the period of the Great Depression and World War II, the period of the greatest and most extensive socioeconomic deprivation in American society during the 20th century. Thus, the large socioeconomic disparities in health that we observe in this 10-year birth cohort and also those adjacent to it may reflect to some degree, and perhaps entirely, the residues of the lower socioeconomic strata in those cohorts having lived some or all of their formative years between birth and early adulthood under conditions of relatively extreme socioeconomic deprivation, which current theories and research on the life course and health suggest may be important determinants of later adult health (Kuh & Ben-Shlomo, 1999, 2003). Methodologically, disentangling aging and cohort effects requires long-term longitudinal data.
Further, the current state and the trajectory over time of both individuals and population cohorts are constantly subject to alteration by the forces of planned and unplanned social change. Thus, just as cross-sectional age differences are confounded with cohort effects, changes with age within individuals or cohorts over time may be confounded with historical or period effects. Methodologically, this suggests the need for cohort-sequential designs, where sequential birth cohorts are followed longitudinally over time.
ACL As a Long-Term Cohort Longitudinal Study
In order to help disentangle aging from cohort effects as well as address other questions noted below, the ACL Study was designed as a cohort longitudinal study. Our original funding allowed a two-wave longitudinal study, with the waves spaced 2.5 years apart. Results from those two waves shifted our focus, however, from just studying the impact of various psychosocial factors on change in health over relatively short intervals of time toward documenting and understanding socioeconomic disparities in health over the full adult life course, making it clear that a much longer-term longitudinal study was needed. Further, the intervals of 2.5 years between our two initial waves were insufficient to adequately observe and test whether consequential changes in physical health (i.e., development of major chronic diseases and functional limitations and ultimately mortality) within individuals over time were consistent with cross-sectional patterns depicted above in Figures 1 and 5, though initial analyses using the first two waves of ACL in 1986 and 1989 suggested that this was the case (House et al., 1994). Thus, we have sought and obtained funding on a wave-by-wave basis to realize ACL as a long-term cohort longitudinal (or prospective) study with longer intervals of 5–7.5 years between waves.
The result is to this point a four-wave cohort longitudinal study as described in Figure 6, with continuous mortality tracking via the National Death Index and other methods yielding over 99% mortality ascertainment, with over 97% of deaths confirmed via death certificates. This design is now proving exceedingly useful and informative in demonstrating that the age differences in health, and socioeconomic disparities therein, observed in the 1986 ACL baseline data are a function primarily of socioeconomic disparities in individual aging over the life course, though the explanations for this remain to be explored in our and other longitudinal data.
|
A More Differentiated View of Health and Socioeconomic Position, and the Reciprocal Relations Between Them
The realization of ACL as a long-term cohort longitudinal study, and its planned extension into a cohort-sequential design, not only allows us to deal with the problems of distinguishing between and among age (or aging), period, and cohort effects, but also to address other important issues. Extending a cross-sectional survey into a longitudinal or prospective one allows for the study of entirely new indicators and processes of health. A cross-sectional study is, in epidemiological parlance, merely a study of the prevalence of health problems and their differential distributions in the population. A longitudinal study allows for the study of incidence and change in health problems as well as observation of the ultimate health outcome of mortality. It is not necessarily the case that any given indicator of SEP will relate in the same way to the incidence of given health outcomes as it does to its course or its prevalence or that it will relate in the same way to different health outcomes (e.g., morbidity vs functional limitations vs mortality, not to mention specific causes of all of these) in either the cross-section or longitudinally. For example, although education and income have behaved very similarly in relation to multiple health indicators in the cross-section, this is less the case when our analysis is extended longitudinally.
Finally, longitudinal data give us a better (though by no means perfect) handle on understanding the causal priorities between SEP and health. Although most sociological and social epidemiological research on socioeconomic disparities on health including our own presumes, based on considerable evidence from longitudinal or prospective research studies, that the cross-sectional associations between SEP and health reflect a predominant causal flow from SEP to health, other bodies of theory and research, especially in economics (Grossman, 1972; Smith, 1999), suggest that considerable, and perhaps the predominant, causal flow should be and is from health to socioeconomic position. ACL data are now able to shed new light on this issue in terms of the potential reciprocal relations between income and health over the adult life span.
In sum, the extension of the ACL Study to 15 years of longitudinal follow-up has enabled us currently to begin to address a number of important issues on questions in understanding the nature and causal dynamics of socioeconomic disparities in health over the life course:
The remainder of this article will overview initial results regarding questions 1–4, especially 1 and 2, and what they imply as to future work on question 5.
CURRENT FINDINGS AND RESEARCH DIRECTIONS REGARDING SOCIOECONOMIC DISPARITIES IN HEALTH OVER THE LIFE COURSE
Aging Versus Cohort Differences (and Their Implications for the Changing Nature of Socioeconomic Disparities in Health)
With our 15-year longitudinal ACL data, we are able to construct a synthetic representation of aging over the entire life course by utilizing the experience over the 15 years between 1986 and 2001/2002 of each of four 15-year age cohorts defined as of 1986, that is, persons aged 25–39, 40–54, 55–69, and 70–84 years in 1986. Splicing together the experience of these four cohorts over the 15 years from the first to the fourth waves of ACL creates a synthetic life course cohort that approximates the way health changes within individuals as they age from age 25 to 99. The result is Figure 7, which consists of four sets of lines for each of three levels of education, plotting, within levels of education and for each of the four age groups in 1986, the mean proportion reporting no functional limitations at the four waves in 1986, 1989, 1994, and 2001/2002. These are computed using wave-specific sample weights for each respondent, which keep the sample at each wave representative of the surviving members of the U.S. population in 1986, and are plotted at the mean age for that group at each wave of ACL (e.g., ages 32, 34.5, 39.5, and 47 years for the 25–39 age group, etc.). Thus, for example, among surviving persons aged 25–39 years with higher education (16+ years) in 1986, approximately 98%, 100%, 96%, and 94% report having no functional limitations in 1986, 1989, 1994, and 2001/2002, respectively; whereas among persons aged 25–39 years but with low education (0–11 years) in 1986, only about 93%, 92%, 83%, and 82% are free of functional limitations in those same years—an educational disparity that grows from about 5 to 13 percentage points as this age cohort ages between 1986 and 2001/2002.
|
In general, these transitional points between age groups are quite contiguous in Figure 7, with a few notable exceptions, suggesting a general pattern of socioeconomic disparities in the way health changes with age similar to that inferred from the cross-section in 1986, but also with some notable evidence of cohort change. To better illustrate this, Figure 8 superimposes on Figure 7 the observed cross-sectional results previously seen in Figure 1. For the most highly educated group, the 1986 cross-sectional and 1986/2001/2002 longitudinal results are strikingly corresponding, with one major exception: The health of highly educated persons who were 55–69 years old in 1986 is markedly better (by almost 25 percentage points) in 2001/2002 when they are 70–84 years than was the health of 70–84 year olds in 1986. No such improvements are evident in these age ranges for those with medium (12–15 years) and lower (0–11 years) levels of education. In other words, higher-educated persons (16+ years) manifest a rapidly increasing compression of morbidity, such that in 2001/2002, almost 85% of them report being free of any functional limitations until they are on average 77 years old, which was the average life expectancy for the U.S. population in 1999 estimated by the National Center for Health Statistics (2002). In contrast, in this same age cohort, only 60% and 37% of persons with medium (12–15 years) or low (0–11 years) education, respectively, report no functional limitation, levels that are no better or even worse than for persons of the same age and education in 1986.
|
We also see in Figures 7 and 8 evidence for the highest-educated persons of much smaller improvements in health among those aged 40–54 in 1986 and 55–69 in 2001/2002 compared with those who were 55–69 years in 1986, but no improvement for those 25–39 years in 1986 and 40–54 years in 2001/2002 (compared with those aged 40–54 years in 1986). This is considerably due to the higher-educated group increasingly being at or near the absolute ceiling on this measure of health up through age 69. Among the medium- and low-educated groups, we also observe in Figure 8 a generally striking similarity between the cross-sectional age differences plotted in Figure 1 and the longitudinal changes with persons over 15 years plotted in Figure 7. However, among persons aged 55–69 years in 2001/2002 (or 40–54 years in1986), we also see evidence of health improvements at lower educational levels as large as or larger than those among the higher educated in this age cohort. This suggests that less-educated persons in middle and early old age may have achieved or experienced some improvements in health behaviors, health care, or general economic well-being similar to those among the higher educated. In addition, the less educated may have benefited from social welfare (e.g., earned income tax credits, food stamps, housing assistance) and health care (e.g., Medicaid) policies targeted toward middle and especially lower socioeconomic persons.
The results (not shown) of comparable analyses by income are generally similar to those by education with two notable exceptions. First, the dramatic improvement in health of highly educated persons aged 70–84 years in 2001/2002 is evident only in very attenuated form for higher-income persons and is hardly evident at all at younger ages. However, the improvements in health among lower- and medium-income persons aged 40–54 and 55–69 years in 2001/2002 compared with those of the same age and education in 1986 are as great as or greater than those seen in Figures 7 and 8 for persons of lower and medium levels of education. Second, as a result of the differences just noted, the evidence of increasing socioeconomic disparities with increasing age is much more muted in the case of income than education, though still significant in growth curve analyses over the entire age range, as discussed below. These results may partially reflect the somewhat smaller and hence absolutely and relatively more advantaged nature of the higher-education versus the higher-income group. But it may also reflect the nature of the dependent health variable in these analyses, which mainly indexes the ability to prevent the onset of health problems, as opposed to the progression or course of health problems after their onset, with education perhaps being more predictive and determinative of the former (onset prevention) and income being more predictive and determinative of the latter (course and progression).
In sum, the now-15-year longitudinal follow-up on our nationally representative ACL sample indicates that there is a genuine social stratification by both education and income of the way health changes with age, with much greater compression of functional limitation achieved at higher levels of education and income and the greatest socioeconomic disparities in health at middle and early old ages. However, in other ways, the longitudinal results also suggest differences in how health and the way health changes with age are stratified by education and income. First, educational disparities in health and the way health changes with age appear to be increasing on average, especially in early to middle old age between ages 55 and 84 years (while possibly diminishing at somewhat younger ages). Income disparities have increased only very slightly at older ages, while diminishing somewhat more at younger ages. Second, education may be more important in preventing the onset of health problems, whereas income is more important in shaping the course of progression of health problems once they emerge. We turn now to a more direct test of this latter proposition.
Differential Roles of Education Versus Income in Onset Versus Progression of Health Problems
The availability of longitudinal data over a substantial period of follow-up allows the exploration of a range of health outcomes, including differentiating between the onset and course and progression of health problems. Our analyses of the ultimate and final health outcome, mortality, have given further indication of the potentially different roles of education and income in the onset versus progression of health problems. A number of analyses of mortality and health change through ACL waves 3 and preliminary analyses through ACL 4 have shown that when education and income are considered separately, income is a stronger predictor of mortality than education. When they are considered together in the same analyses, the effects of education are weakened, often to the point of no longer being statistically significant, while the effects of income remain strong (e.g., Lantz et al., 1998), suggesting that the effects of education on mortality are largely mediated through income. Since mortality follows upon a progression of deteriorating health status, these results are consistent with the emerging hypothesis that education plays a greater role in the onset of health problems, whereas income is more consequential for the progression and course of health problems (Lantz et al., 2001; Zimmer & House, 2003).
Zimmer and House (2003) confirmed this hypothesis in analyses completed on the first three waves of ACL before the wave 4 data were fully collected and ready for analysis. They found education to be more predictive than income of the onset of functional health problems by wave 3 (in 1994) among those free of such problems at wave 1 (in 1986). Conversely, income was more predictive than education of change in health by wave 3 (in 1994), including progression to mortality, among those with some degree of functional limitations at wave 1 (in 1986). Melzer, Izmirlian, Leveille, and Guralnik (2001) have produced congruent findings with respect to education's prospective prediction of disability and change therein.
Analyses now being readied for submission as a separate paper (Herd, Goesling & House, in preparation) have analyzed the relative predictive power of education and income (net of each other) as predictors of transitions (n = 8,388) in functional health status between successive waves of ACL (i.e., wave 1 to wave 2, wave 2 to wave 3, and wave 3 to wave 4, intervals of about 2.5, 5.0, and 7.5 years, respectively). The specific transitions are from "no" versus "some" (or any) functional limitation at the initial wave to dying or having the same, better, or worse level of functional limitations by the succeeding wave. Controlling for age, sex, race, and reports of self-rated health and number of chronic conditions reported at the initial wave (all of which predict transitions in expected ways), education has a stronger effect than income in predicting onset of limitations (transitions from "no" to "some" limitations versus remaining without limitations). However, education has almost no significant effect on any of the other transitions, all of which are predicted by income and involve the course or progression of a health problem either explicitly (in transitions from the initial state of "some limitations") or implicitly (in the case of death after reporting no limitations where presumably there was generally also onset of a health problem leading to death).
The explanation of these differential effects of education and income remains to be determined. One possible explanation is that education is set early in the life course and therefore can more strongly affect onset of health problems that develop only gradually over time. Another is that education and income are different types of resources, with education affecting a pattern of opportunities and choices to maximize health-promoting and minimize health-damaging exposures and experiences. Income, on the other hand, provides resources that can be used to care for or adapt to a given health problem or restructure one's life situation to slow the progression of the problem, for example, by moving away from or avoiding occupational or residential experiences or exposures that might exacerbate it. This represents a major area for further research.
Potential Reciprocal Relation Between Income and Health
Long-term longitudinal data of the type now available in ACL also allow for asking fundamental questions as to whether observed associations between income and health reflect the causal impact of income on health, as has been assumed in most of the social and biomedical science literature on this topic including our own, or the causal impact of health on income, as is suggested in some other literatures, particularly those stemming from human capital theories in economics (Grossman, 1972; Smith, 1999), or both. We believe it is particularly important to ask these questions with respect to longer-term changes in consequential indicators of physical health as these, rather than more minor short-term perturbations in health, are the basis of current concerns with socioeconomic disparities in health.
The ACL data are well suited for approaching these issues from two different perspectives (Herd, House, Morenoff, & Goesling, in preparation). First is the standard prospective study approach of social epidemiology in which baseline characteristics of individuals are used to predict subsequent changes in a dependent variable (usually health or health change), adjusting for baseline levels of the dependent variable and other confounding variables. Surprisingly, this approach has seldom been applied to estimating the potential impact of health on income. We have used growth curve analysis to estimate models predicting individuals' trajectories in levels of functional limitations between 1986 and 2001/2002 from their income in 1986 and similarly predicting trajectories in their level of income from their 1986 levels of functional limitations. The findings reveal a significant adverse effect of income in 1986 on trajectories of health (as indexed by functional limitations) between 1986 and 2001/2002, but no significant effect of health in 1986 (again indexed by levels of functional limitations) on trajectories of income between 1986 and 2001/2002.
From another perspective, we can evaluate the effect of adverse health and financial events that occur between each pair of waves of ACL on change in health and income between the waves. At each wave, individuals are asked whether they have experienced a number of events including "involuntary loss of a job for reasons other than retirement," a "serious financial problem," and a "life threatening" and/or a "serious but not life threatening" "illness or injury." Table 1 shows the numbers of people reporting each of these types of events and combinations thereof (including none of the events) between each pair of adjacent waves of ACL.
|
Conversely, we have used ordinary least-squares regression to predict income at waves 2, 3, and 4 as a function of job loss and health events (but not financial problems in order to avoid confounding and maximize the chance of the health events having an impact), with the same controls. The signs of the health events variables were all negative, but none was significant. There are some significant negative effects of the level of health at a preceding wave on subsequent income, but these are generally smaller and less significant than the comparable effects of income (and education) on health shown in the probit analyses.
In sum, both the growth curve and "event shocks" analyses indicate that income (as well as education) has significant and substantial effects on changes in functional health over periods of 2.5–15.5 years, but functional health has only small and largely nonsignificant effects on subsequent income over similar time periods. To our knowledge, this is the only extant effort to compare the effects of income on health and health on income in the same data using a consequential indicator of health and considering these changes over a substantial period of time. The results suggest that although some of the association observed in ACL between income and health may (and must) reflect effects of health on income, the vast bulk of the association is a product of the strong and consistent impact of income on health over follow-up periods ranging now up to 15 years.
Conclusion
The extensions of the ACL Study to 15 years of follow-up have largely confirmed the initial understandings of the social stratification of aging and health over the adult life course derived from the initial cross-sectional and short-term longitudinal analyses from ACL and related studies and first overviewed at a Pennsylvania State University conference and in a resulting book about 15 years ago. Some of the most important of these understandings include the following:
) that structures people's exposure to and experience of almost all variables or risk factors that shape health (though the full longer-term longitudinal analysis to support this proposition remains to be done). However, the new analyses from our recently completed fourth wave of ACL, extending the follow-up to 15 years, also reveal new findings and raise new questions that are simply impossible to see in cross-sectional and short-term longitudinal data. First, the pattern of socioeconomic disparities in health and the compression of morbidity and functional limitations have been undergoing very rapid cohort change in just the 15 years between 1986 and 2001/2002. Most notably, educational disparities in the compression of functional limitations have increased dramatically in early to middle old age, owing to a massive improvement of the health of higher-educated persons aged 70–84 in 2001/2002 compared with those 70–84 in 1986, with no evidence of any such change in this same age cohort at lower educational levels. These results suggest that much, and perhaps almost all, of the improvement in the health and functional status at older ages that has been observed in the total population by multiple researchers since the late 1980s is a function of changes in the college-educated portions of that population, with a resulting fourfold increase in the disparities between this group and those of lower education.
There is also some evidence of decline in educational disparities in functional limitations at earlier ages in the adult life course, resulting from improvements in the health of less-educated individuals in 2001/2002 relative to their age counterparts in 1986. This suggests that at these younger age ranges, where higher-educated persons have little opportunity for further improvement of health or compression of functional limitations, persons of lower education may be gradually catching up owing to improvements in general economic conditions, social welfare programs, and health behaviors, as well as access to and utilization of preventive as well as therapeutic health care. The monitoring and understanding of cohort changes in the social stratification of aging and health over the life course constitute a major continuing agenda for future research.
Similar decreases in income disparities in health at younger age ranges are evident, but there is little evidence that higher income has promoted the postponement or compression of functional limitations in the way that higher education has. This result and evidence that income is a much stronger predictor of mortality than education, along with other current research, suggest the emerging hypothesis that education plays a larger role in preventing the onset of health problems, whereas income plays a larger role in the course or progression of health problems once they exist—a hypothesis confirmed in analysis of the ability of education and income to predict the onset versus course or progression of functional limitations (up to and including death) over periods from 2.5 to 15.5 years in the ACL Study. The reasons or explanations for this largely remain to be elucidated.
In sum, our knowledge and understanding of socioeconomic disparities in health and the way health changes with age over the adult life course have expanded greatly over almost two decades of the ACL Study. Yet an enormous amount remains to be understood. One major issue is whether and how levels of or changes in the wide array of explanatory risk factors that are shaped by socioeconomic position can explain the changes with aging and across cohorts that we observe in the social stratification of health in our ACL cohort.
Another is how and why socioeconomic disparities in health are so small in both early adulthood and later old age—an issue that will require careful consideration and integration of psychosocial and biomedical factors and forces and may have important implications for social and health policy. For reasons of both space constraints and current gaps in understanding, the discussion here and more generally in this article neglects fully exploring this issue. Clearly, there must be some combination of psychosocial and biomedical factors that account for this (see Robert & House, 1994, for fuller review and discussion of these). One possibility is that either the socioeconomic disparities in exposure to and experience of explanatory risk factors and/or the impact of those factors is lower at younger and older ages and greatest during middle adulthood and early old age. We have presented some data to this effect (House et al., 1992) and incorporated this into the multivariate model from which Figure 5 is derived by including interactions between age and all risk factor variables. As discussed elsewhere (House et al., 1990, 1992, 1994), the ability of SEP and risk factors associated with it to affect health may also be buffered or moderated by either biological robustness (in early adulthood) or frailty (in later old age), which leave less opportunity for the influence of psychosocial factors at these ages. The impact of socioeconomic factors on health may be similarly moderated of buffered among persons of (or just past) school age and among the elderly by the existence of social welfare programs and/or institution (e.g., schools and the military at younger ages or Social Security and Medicare at older ages), which moderate the degree to which individual SEP may affect health. This remains a fertile area for future research, with existing studies serving mainly to indicate that selection via mortality is unlikely to be a major part of the explanation of the variation in socioeconomic disparities that we and others observe (Figure 5; Beckett, 2000; Herd, in press).
Finally, there is a clear need to continue to extend the long-term longitudinal follow-up in ACL and similar studies and to convert them into true cohort-sequential designs by beginning to observe the social stratification of health and the way health changes in younger cohorts as they enter adulthood and move through it into older age. The problem of understanding and alleviating socioeconomic disparities in health has truly become a quintessential problem of life course analysis and research.
Acknowledgments
This research was supported by grants to the first author from NIH/National Institute on Aging (PO1 AG0551 and RO1 AG018418) and from a Robert Wood Johnson Foundation Health Investigators Award (030987).
This work was originally presented at a conference on Health Inequalities Over the Life Course, Pennsylvania State University, State College, Pennsylvania, June 6 and 7, 2004.
References
This article has been cited by other articles:
 |
|
 |
|
  S. Schieman and G. Plickert Functional Limitations and Changes in Levels of Depression Among Older Adults: A Multiple-Hierarchy Stratification Perspective J. Gerontol. B. Psychol. Sci. Soc. Sci., January 1, 2007; 62(1): S36 - S42. [Abstract] [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||
| HOME | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|---|
