| HOME | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|---|
| ||||||||||||||||||||||||||||||||
RESEARCH ARTICLE |
a Department of Health Studies, University of Chicago, Illinois
Kathleen A. Cagney, Department of Health Studies, University of Chicago, 5841 South Maryland Ave., MC 2007, Chicago, IL 60637 E-mail: kcagney{at}health.bsd.uchicago.edu.
Decision Editor: Toni C. Antonucci, PhD
 |
Abstract |
|---|
 TOP Abstract MethodsResultsDiscussionReferences |
|---|
STUDIES of elders consistently find a strong inverse association between years of formal schooling and cognitive functioning. The association has been seen regardless of whether the outcome is the score on a cognitive function screener such as the Mini-Mental State Exam, a threshold score used to define cognitive impairment, a decline in cognitive function, or the clinical diagnosis of dementia (Brayne and Calloway 1990
; Evans et al. 1993; Leibovici, Ritchie, Ledesert, and Touchon 1996; Ott et al. 1995; Scherr et al. 1988; Stern et al. 1994; White et al. 1994). Such relationships are seen in diverse cultural and geographic environments (Johnson et al. 1997; Yu et al. 1989).
The interpretation of the education–cognitive-function association is complicated by ambiguity as to what education represents. A substantial research literature has demonstrated a strong association between socioeconomic status (SES) and a broad range of health outcomes among the elderly (Feinstein 1993; House, Kessler, and Herzog 1990; Lantz et al. 1998; Liao, McGee, Kaufman, Cao, and Cooper 1999). Those with lower SES are more likely to experience limitations in functioning and increased morbidity and mortality. The majority of these studies use education as their measure of SES. Thus the question arises of whether the evidence of an association between education and cognitive functioning among elderly adults should be interpreted as primarily contributing to this broad pattern of differences in health attributable to an underlying SES gradient. An association between SES and cognitive function among elderly adults could, for example, be due to risk of chronic and infectious diseases throughout the life course, quality of health care, occupational or environmental exposures, or differences in health practices and lifestyle behaviors (Kubzansky, Berkman, Glass, and Seeman 1998).
By contrast, some hypotheses concerning the education–cognitive function association relate specifically to the process and consequences of education itself, not its correlation with economic status. These hypotheses fall into three categories. The first category is often referred to as the brain reserve capacity theory. Although there is variation in definition, the general concept is that a rich environment, such as being in school during formative years, promotes the development of brain reserve capacity through increased overall brain size, regional brain size, or greater dendritic branching (Katzman 1993; Satz 1993; Schmand, Smit, Geerlings, and Lindeboom 1997; Unverzagt, Hui, Farlow, Hall, and Hendrie 1998). Brain reserve capacity in turn delays the manifestation of neuropathology that occurs because of aging-related or pathologic processes. A related explanation is that educational attainment is a consequence of, and thus a marker for, genetically mediated cerebral capacity, which might be more directly measured by, for example, an intelligence assessment early in life (Pedersen, Reynolds, and Gatz 1996). A second type of explanation is that mental stimulation throughout the life course preserves cognitive function. Education would lead to occupations more likely to include mental stimulation (Schooler, Mulatu, and Oates 1999) and perhaps more contact with persons who have greater formal education, including a spouse. A third category of explanation is that the association is primarily artifactual, either because of better test-taking ability of those with more formal schooling or because of cultural bias in cognitive assessments (Schmand, Lindeboom, Hooijer, and Jonker 1995).
A few previous studies bear on the question of whether education has a uniquely strong association with cognitive functioning among elderly adults—one owing primarily to noneconomic mechanisms—or whether it serves fundamentally as an SES indicator. One approach to disentangling the effects of education from economic status is to select a study population with variation in either educational attainment or economic status but without variation in the other. Several studies have examined the education–cognitive-function (or dementia) association in rare populations with minimal SES variation, these being the Amish, Jesuit priests, or Roman Catholic nuns (Chibnall and Eastwood 1998; Johnson et al. 1997; Snowdon, Ostwald, and Kane 1989). Although the educational range varied markedly—with Amish having at most 8 years of school and Jesuits having as many as 23 years of postsecondary education—these studies found evidence of an effect for education within these special populations (in which education had little or no economic consequences).
A more frequent approach to separating the effects of education and economic status is to concurrently assess education and a different SES component to determine whether they contribute independently to cognitive functioning, cognitive impairment, or dementia (Dartigues et al. 1992; Evans et al. 1997; Scherr et al. 1988; Stern et al. 1994). The two other SES indicators likely to be ascertained in population-based studies are occupation and income. Both are problematic in research concerned with elders. Retirement introduces a measurement problem for occupation: Should one determine a "principal" occupation or the most recent occupation? The ordering of occupations into a scale is complex, and commonly used scales may not suit and rank women's occupations as appropriately as they do men's. For married women, the husband's occupation may be a better SES indicator than her own occupation. Many elderly women have had no lifetime employment outside the home. Also, occupation later in life may itself reflect ill health and diminished functioning. The most direct measure of economic status previously available in studies of cognition among elderly adults is household income. However, great disparities in income before retirement may not be apparent in income after retirement. Particularly for elderly adults, income is a less complete assessment of economic status than wealth is, because it does not reflect the value of accumulated assets such as home and other property, savings, retirement accounts, and investments (Berkman and Macintyre 1997; Smith 1997). To our knowledge, no previous study has examined the association between wealth and cognitive functioning among the elderly.
Using Wave 1 of the Asset and Health Dynamics Among the Oldest Old (AHEAD) study, a nationally representative survey first conducted in 1993 with Americans aged 70 or older, we examined the effect of economic status and educational attainment on cognitive function in later life. Specifically, we hypothesized that education, wealth, and household income would each be associated with all measures of cognitive function, such that higher levels of each would predict better cognitive status. Because we believe wealth to be a better indicator of lifetime economic status than income, we also expected it to be more strongly associated with cognitive function than income. Further, we hypothesized that these associations would not vary markedly by race and Latino ethnicity. Finally, we hypothesized that the education–cognition association would be independent of economic status, regardless of whether a better or poorer measure of economic status was used, and that economic status would predict cognitive function independent of educational attainment.
|
Methods |
|---|
|
TOPAbstractMethodsResultsDiscussionReferences |
|---|
, we report unweighted data. The weights in AHEAD take into account sample selection probabilities (i.e., adjust for oversampling of selected subgroups), nonresponse, and a poststratification adjustment factor, and thus allow the researcher to make population estimates of prevalence (Soldo, Hurd, Rodgers, and Wallace 1997). Because we do not make prevalence estimates, and because we conducted stratified analyses by race/ethnicity, we chose to use unweighted data. In addition, we did so because applying weights to these small groups would, by nature of the weight calculation, increase the standard error of the estimate—this inefficient estimator would lead to wider confidence intervals and less powerful hypothesis tests. A full description of the weighting procedures used in AHEAD can be found in Heeringa 1995.
Measures
Dependent variables.
The dependent variables in this analysis are the three measures that constitute cognitive status in the AHEAD survey—memory; working memory; and knowledge, language, and orientation. The development of these measures, their reliability and validity, and their relationship to other measures of cognitive function can be found in the work of Herzog and Wallace 1997. The first component, memory, consists of two questions that assess immediate free recall and delayed free recall. These ask respondents to recall 10 short, concrete high-frequency nouns. The delayed free-recall test is administered 5 min after the immediate free-recall test. This domain totals 20 points. The second component, working memory, is based on one question that asks respondents to begin at 100 and subtract by increments of 7 for five trials. One point is given for each correct subtraction for a total of five points. The final component, knowledge, language, and orientation, asks respondents to recall the day and the date, count backward from 20, name the president and the vice president, and name two familiar objects (i.e., cactus and scissors). This domain totals 10 points. Drawing from Herzog and Wallace 1997, we assigned respondents who refused at the inception of the task a score of zero and those who refused during the task the score they had obtained up until their refusal. To compare education, income, and wealth effects across measures, we standardized the 20-, 10-, and 5-point scores (M = 0, standard deviation = 1). Thus, a 1-unit change in the transformed score is equal to one standard deviation for each of the three scales.
Explanatory variables.
The explanatory variables in this analysis are three—education, wealth, and household income. Education is measured in years completed. Years range from 0 to 17, grouped as 0 to 3, 4 to 7, 8 to 11, 12, 13 to 15, 16+. We treated education as a categorical variable to allow for nonlinear (e.g., threshold) effects. The wealth measure was based on the assets data and was the sum of net worth (items such as value of home owned, checking or savings accounts, individual retirement accounts, certificates of deposit, savings bonds, and shares of stocks or mutual funds), with reported debt subtracted. For comparability, we used the assets-less-debts summary variable of wealth constructed, with imputation, by the AHEAD research group. The income measure was reported current household income from the survey. The variable we used was the imputed measure provided by the AHEAD research group. Further information regarding wealth and household income imputation can be found in Smith 1997.
Control variables.
Age was included as a control variable in the multivariate analyses because evidence suggests that cognitive function declines sharply with increasing age (Evans et al. 1993). Gender was also included as a control variable. Introducing controls for health-status measures such as body mass index, hypertension, stroke, or current smoking did not alter the relationships among education, wealth, and cognitive performance so we present the results from the more parsimonious models.
Statistical Methods
Our analytic strategy included five components. First, we examined descriptive statistics, stratified by gender and race/ethnicity, for our explanatory and outcome variables. Second, we produced Spearman correlations for categories of education, wealth, and household income by race/ethnicity. Third, we plotted the standardized cognition scores for the three domains of cognitive function by education level—this was done for each race/ethnicity and gender group. Fourth, we conducted bivariate analyses for each cognitive function domain to assess the relationship between cognition score and the covariates of interest (i.e., education level, age, gender, household income, and wealth). Finally, we conducted multivariate analyses for each domain; age, gender, and education acted as the baseline model, then wealth and household income were introduced sequentially in two additional models.
All models used ordinary least squares (OLS) regression. Because the inclusion of spousal pairs from the same household violates statistical assumptions about the independence of observations, we used a cluster correction. This allows for the retention of all cases by taking into account the correlation among observations from the same household and thus allowed us to maximize sample size. The regression parameters reported in Table 3 Table 4 Table 5 Table 6 indicate that for a one unit change in the covariate, there would be a one standard deviation change in cognitive score.
|
|
|
|
|
Results |
|---|
|
TOPAbstractMethodsResultsDiscussionReferences |
|---|
|
Table 2 reports Spearman correlation coefficients for our explanatory variables by race/ethnic group. Although all are statistically significant, the correlations among education, income and wealth are only moderate, ranging from .26 to .48. All correlations are weaker for Blacks than for Whites or Latinos.
|
|
|
|
Income and net worth had a much smaller impact on cognitive function than education, with net worth having the greater predictive power. However, for all race/ethnicity groups, there was still a consistent pattern of higher cognitive function scores with increases in net worth. The effect of net worth on cognitive performance was quite similar in each domain across race/ethnicity. For example, those with net worth less than or equal to zero were, on average, 0.85 to 0.98 standard deviations lower in the working memory score than those with net worth at $200,000 or more.
Table 4 , Table 5 , and Table 6 present multivariate models for each cognitive function domain. In each table, three models are presented for each race/ethnic group—the first model shows the association of education with cognitive function, adjusted for age and gender, the second model introduces net worth into this baseline model, and the third model adds household income. Adjusting for age and gender, we found that the education effect remained extremely strong across all three domains. When net worth was introduced as a control variable, the education effect was modestly attenuated. The introduction of household income into the model had little impact on the relationship between education and cognitive function. Only for Whites did net worth have a consistent contribution after adjusting for education and income; this was observed most readily in the memory and working memory components. In general, the effect of net worth in the presence of education was greatly attenuated as compared with the bivariate models, especially for Blacks and Latinos. Household income did not have a consistent association with cognitive function for any group after controls for education and net worth were introduced. Across all three domains, the difference in cognitive function scores between the highest and lowest educational levels was greater for Blacks and Whites than for Latinos. For instance, in the knowledge, language, and orientation component, Blacks with fewer than 4 years of formal education scored approximately 2.12 standard deviations lower (p < .001), whereas Whites in this subgroup scored 1.36 standard deviations lower (p < .001) and Latinos only 0.89 standard deviations lower (p < .01), after adjusting for net worth and household income.
|
Discussion |
|---|
|
TOPAbstractMethodsResultsDiscussionReferences |
|---|
What Are the Implications of These Findings for Research Examining the Social Factors that Predict Cognitive Function at Later Ages?
First, similar to a number of previous studies in both the social science and neuroscience literatures (Brayne and Calloway 1990; Evans et al. 1993; Leibovici et al. 1996; Ott et al. 1995; Scherr et al. 1988; Stern et al. 1994; White et al. 1994), we found that early life education has a strong and significant impact on cognitive performance late in life. In addition, we found this effect robust to controls for current income and lifetime wealth, suggesting that (a) education and these other SES indicators are not interchangeable with respect to cognitive function and (b) education is contributing something unique to cognitive function. These findings offer somewhat stronger support to some of the potential mechanisms of an education–cognitive function association than to others. The similar magnitude of the association for Whites and Blacks and for men and women is consistent with a causative association such as either the brain reserve capacity theory or a life course explanation for which education leads, in a graded fashion, to increased mental stimulation throughout life.
Second, although wealth behaves similarly across race/ethnic groups in bivariate analyses (indicated by the magnitude of the estimates), it no longer significantly predicts cognitive function for Blacks or Latinos once education is controlled. Although this does not mean that wealth and household income do not have an effect, it suggests that they share the same causal pathway as education. Particularly for Blacks and Latinos, it appears that education overwhelmingly determines the pathway by which wealth could enhance cognitive function. For Whites, though, there are additional gains in cognitive function associated with wealth that are not related to educational attainment. This suggests that Whites may have greater opportunity to accumulate wealth apart from the pathway that formal education provides. Alternatively, once wealth has been accrued, Whites may be able to more easily translate this wealth into better environmental circumstances or less stressful living situations, further contributing to cognitive health in later life.
Results for the Latino respondents merit additional discussion and underscore the complexity of studying the Latino population within the United States. The finding of lower returns to educational attainment and the plateau effect after the completion of only a few years of secondary school are inconsistent with the patterns observed for other groups. The Latino respondents in this study consisted primarily of two subgroups owing to the sampling strategy of the AHEAD survey: a foreign-born population residing in Florida and a predominantly U.S.-born population residing in the Southwest. Although the size of these subsets is not large enough to conduct more detailed analyses, examining cell counts reveals that there are differences between these groups. The data suggest that Latinos born outside the United States who currently reside in the South Atlantic United States have the same cognitive returns to education as Whites and Blacks in this sample; the effect is markedly weaker for those in the Southwest. Nativity, language of education, and ethnicity differences between those residing in the South Atlantic and those in the Southwest may help to explain differences in the education–cognition relationship, but these data do not allow us to adequately test the hypotheses. In addition to differences in respondent characteristics, the measures themselves may have contributed to these results. Although the AHEAD survey was also administered in Spanish, there is cultural content in the cognitive status measures. For example, naming the vice president may be a more difficult question for a respondent who has recently emigrated to the United States versus a native-born respondent.
There are a number of caveats to this analysis. First, the sample is cross-sectional so it is not possible to disentangle an age effect from a cohort effect. The availability of future waves of AHEAD data will allow us to examine change within individuals over time. Second, the Latino sample, although much larger than comparable secondary data sets, is still fairly small, limiting the number of covariates, such as geography, that can be introduced into the model with sufficient power to detect differences. Third, the measures used to assess cognitive function in AHEAD may be too coarse to detect the more subtle effects of increased educational attainment. Especially in the case of knowledge, language, and orientation, there is an apparent ceiling effect. Nonetheless, we were still able to detect enough variation in these cognitive-status measures to explore the education–wealth–cognition relationship. Finally, it is well known that wealth and income data are difficult to collect. However, the bracketing techniques used to determine wealth and income categories in the AHEAD study are considered by survey methodologists to be among the best approaches to date (Soldo, et al. 1997). On a related note, the quality of the wealth and income data may vary by cognitive performance. Because these respondents did not require a proxy to answer questions for them, however, we believe that this variation would be minimal and, further, would not result in systematic under- or overreporting of wealth or household income.
What Next Steps Could Enhance our Understanding of the Education–Cognition Relationship?
This analysis has demonstrated that there is very little independent effect of wealth or income on cognition once education is considered. Although we have had the opportunity, afforded by the AHEAD data, to test the impact of wealth on cognition in later life, the question still remains as to the mechanism by which education is contributing to cognitive health. This mechanism may be the educational process, the content of the material, mental stimulation in the workplace (Schooler et al. 1999), an active lifestyle (Hultsch, Hertzog, Small, and Dixon 1999), or a combination of these factors that helps to maintain cognitive function later in life. One insight gained through these results is that although educational opportunity was very different for Blacks and Whites, and for men and women, at the time when these respondents were educated, we still observed a similar relationship between education and cognitive function across race and gender groups. This underscores the notion that the educational process itself contributes to the maintenance of cognitive health. In addition, the strong monotonic association between education and cognition after adjusting for wealth makes it unlikely that the association could be largely explained were some other, better, measure of lifetime economic status available.
Finally, examining other correlates of wealth and education would be instructive; factors such as parental education, occupation, and neighborhood characteristics may have a unique contribution to cognitive function at older ages, creating or contributing to a mentally stimulating environment and thus maintaining cognitive function in later years. Analyses that examine factors over the life course and interactions among these factors, and that incorporate a longitudinal approach, could help us to understand how education affects not just current cognitive function but cognitive decline as well.
|
Acknowledgments |
|---|
Received for publication December 7, 1999. Accepted for publication November 16, 2000.
|
References |
|---|
|
TOPAbstractMethodsResultsDiscussionReferences |
|---|
This article has been cited by other articles:
 |
|
 |
|
  S. C. Brown, C. A. Mason, A. R. Spokane, M. C. Cruza-Guet, B. Lopez, and J. Szapocznik The Relationship of Neighborhood Climate to Perceived Social Support and Mental Health in Older Hispanic Immigrants in Miami, Florida J Aging Health, June 1, 2009; 21(3): 431 - 459. [Abstract] [PDF]
|
|
|
|
 |
|
 Z. Zhang, D. Gu, and M. D. Hayward Early Life Influences on Cognitive Impairment Among Oldest Old Chinese J. Gerontol. B. Psychol. Sci. Soc. Sci., January 1, 2008; 63(1): S25 - S33. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
J. W. Elias and M. V. Wagster Developing Context and Background Underlying Cognitive Intervention/Training Studies in Older Populations J. Gerontol. B. Psychol. Sci. Soc. Sci., June 1, 2007; 62(suppl_Special_Issue_1): 5 - 10. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 R. A. Shih, T. A. Glass, K. Bandeen-Roche, M. C. Carlson, K. I. Bolla, A. C. Todd, and B. S. Schwartz Environmental lead exposure and cognitive function in community-dwelling older adults Neurology, November 14, 2006; 67(9): 1556 - 1562. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 R. G. Wight, C. S. Aneshensel, D. Miller-Martinez, A. L. Botticello, J. R. Cummings, A. S. Karlamangla, and T. E. Seeman Urban Neighborhood Context, Educational Attainment, and Cognitive Function among Older Adults Am. J. Epidemiol., June 15, 2006; 163(12): 1071 - 1078. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
F. A. Sloan and J. Wang Disparities Among Older Adults in Measures of Cognitive Function by Race or Ethnicity J. Gerontol. B. Psychol. Sci. Soc. Sci., September 1, 2005; 60(5): P242 - P250. [Abstract] [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||
| HOME | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|---|
