Explaining the Role of Sex on Disability: A Population-Based Study

Abstract

Purpose: There is no clear consensus about how sex influences prevalent disability and through what mechanisms. We investigated whether sex had a direct inde-pendent effect on disability or whether sex had an interactive effect on the relationship between chronic diseases/conditions and disability, and whether these effects differed in middle-aged versus older adults. Design and Methods:We used baseline data from two nationally representative health interview surveys, the Health and Retirement Study (HRS) and the Study of Asset and Health Dynamics Among the Oldest Old (AHEAD), and disability and covariate measures that were nearly identical in both surveys. Logistic regression models tested the contributions of diseases, impairments, and demographic and social characteristics on difficulties with prevalent activities of daily living (ADLs), mobility, and strength. Results:Models demonstrated no direct sex effect for ADL disability in either age group after adjusting for key covariates. However, sex did exert an indirect effect on ADL disability in older adults via musculoskeletal conditions and depressive symptoms. In contrast, female sex remained strongly associated with mobility and strength disability in both age groups, net of covariates. Major interactions were also significant, including a female sex/body mass index (BMI) interaction for mobility difficulty and several sex–disease interactions for strength disability in the middle-aged group. Implications: The effect of sex on ADL difficulty is largely explained by social and health-related covariates in middle-aged and older adults. In contrast, the independent association of female sex with decreased strength and mobility in both groups cannot be explained by our models' social or health-related variables. In addition, the positive association of BMI with mobility difficulty is significantly worse for women than for men.

Decision Editor: Laurence G. Branch, PhD

Although the relationship between sex and disability has been studied for more than 20 years, there is no consensus about how sex influences disability and through what mechanisms. It is well known that there is a consistently higher prevalence of self-reported disability in women than in men, yet the independent effect of sex varies with the measures of disability examined and the covariates considered (Fried, Ettinger, Lind, Newman, and Gardin 1994; Johnson and Wolinsky 1994; Verbrugge 1985, Verbrugge 1989). The higher prevalence of disability reported by women can be explained: Women and men differ in the prevalence or severity of diseases causing disability, in their reporting of disability, and in their rates of mortality (Fried et al. 1994; Merrill, Seeman, Kasl, and Berkman 1997; Verbrugge 1985, Verbrugge 1989; Wingard 1984). However, it is not clear why the independent effect of sex on disability differs across studies. Three reasons are plausible. First, researchers measure disability in different ways, evaluating some or all of the activities of daily living (ADLs), instrumental activities of daily living (IADLs), or functional limitation measures, or using separate or composite measures, factors, or scales combining all three types of disability. Second, researchers use different covariates to adjust for confounding effects. Third, disability transitions differ by disease and sex (Crimmins and Saito 1993; Peek and Coward 1999). Although increased disability incidence in women may help to explain women's higher prevalence of mobility disability (Leveille, Penninx, Melzer, Izmirlian, and Guralnik 2000), the reasons for the higher incidence remain speculative.

While some researchers concentrate on explaining the association of sex with disability, others assume interaction effects by modeling associations between covariates and disability separately for men and women (Ettinger, Fried, Harris, Shemanski, and Schulz 1994; Johnson and Wolinsky 1994; Manton 1988). Few have systematically tested an interactive model in which sex moderates the link between disease and disability. Such analyses could help determine whether processes leading from diseases to disabilities differ between men and women. Finally, most studies on sex and disability have examined data on older adults alone. Comparisons of older versus younger age groups (Clark, Stump, and Wolinsky 1997) may also provide insights into the underlying mechanisms and generalizability of the sex effect on different measures of disability and across age groups.

The goal of this study was to answer two questions about the effect of sex on prevalent disability: (1) Can the association of sex with disability be explained by social and health-related covariates; and (2) Does sex have an interactive effect on the relationship between chronic diseases and disability? We hypothesized that the association of increased disability with female sex would not be fully explained by covariates and that the associations between chronic diseases/conditions and disability would not differ by sex. We examined these issues in two age groups using data from two nationally representative surveys: the Health and Retirement Study (HRS) of adults aged 51–61 and the Study of Asset and Health Dynamics Among the Oldest Old (AHEAD) of adults aged 70 and older. We developed a conceptual model linking chronic diseases, conditions, and impairments to disability, and examined three composite measures of disability that were identical in both surveys.

Methods

Conceptual Model and Covariates

Our basic model included documented covariates of disability: demographics, chronic diseases and impairments, physiological characteristics, and social risk factors. We estimated models in both older and middle-aged age groups and tested whether sex modified the relationship of chronic diseases and impairments with disability. Further, we analyzed disability measures that were identical or nearly so in the two surveys and grouped them into three domains: measures of ADLs, mobility, and strength. These measures are similar to classic disability and functional limitation measures (Katz, Ford, Moskowitz, Jackson, and Jaffe 1963; Nagi 1979; Rosow and Breslau 1966) and have been discussed in detail elsewhere. Their factor structure has been reported to be similar in the HRS and AHEAD (Clark et al. 1997).

Because demographic and social variables are known to covary with disability, we controlled for age, race/ethnicity, educational level, and marital status as well as sex in all models. Similarly, we included measures of chronic diseases and conditions (cardiopulmonary, musculoskeletal, diabetes, stroke, vision, and hearing) that have different prevalences by sex and are strongly associated with functional difficulty (Boult, Kane, and Louis 1994; Fried et al. 1994; Manton 1989; Rudberg, Furner, and Dunn 1993).

Declines in cognition are also associated with difficulties in daily functioning (Diehl, Willis, and Schaie 1995; Scherr et al. 1988; Wray, Herzog, Park, and Alwin 2001). Although women generally score higher on cognitive performance tests, net of age and education, few studies have evaluated the effect of sex on disability controlling for cognitive performance. Further, although the causal ordering of the depression/disability relationship is not well established (i.e., whether depression is the cause or consequence of functional difficulties), depressive symptoms have also been found to be significantly associated with disability, particularly in older versus middle-aged adults (Bruce, Seeman, Merrill, and Blazer 1994; Mirowsky and Ross 1992). Some studies have found that depressive symptoms are even stronger correlates of functional limitations than are the underlying diseases (Hays, Wells, Sherbourne, Rogers, and Spritzer 1995). In addition, the prevalence of depressive symptoms is generally reported to differ among men and women, but few studies have looked at the sex effect on disability while controlling for depressive symptoms. We controlled for both cognitive performance and depressive symptoms in our models.

Finally, although physiological tests were not administered in our surveys, we used self-reported height and weight to construct body mass index (BMI, or weight corrected for height), which is known to be significantly and independently associated with disability (Launer, Harris, Rumpel, and Madans 1994; Verbrugge, Gates, and Ike 1991). Because the relationship is not linear but, rather, associated with high and low BMI (Ferraro and Booth 1999; Galanos, Pieper, Cornoni-Huntley, Bales, and Fillenbaum 1994), we evaluated the effect of BMI in the highest tertile.

Age Group Differences

There are several ways in which age-group comparisons help validate findings about sex and disability in older adults. The association of sex with disability may differ in middle-aged versus older people because middle-aged men and women experience diseases and impairments in different relative prevalences than do older adults (Wingard 1984). Other covariates of disability (e.g., cognitive performance, depression, or education) may also differ between middle-aged and older men and women. Finally, selective survival and the sex difference in active life expectancy may confuse interpretation of sex effects on disability, particularly in older adults.

Data

Data were analyzed from the baseline waves of two population-based surveys: the 1992 HRS and the 1993 AHEAD. The HRS surveyed 12,652 Americans who were aged 51–61 in 1992 (or their spouses), and the AHEAD surveyed 8,222 Americans who were at least aged 70 in 1993 (or their spouses). The current study samples include the 9,824 HRS nonproxy respondents who were aged 51–61 in 1992 and the 6,660 AHEAD nonproxy respondents who were at least aged 70 in 1993. Additional information on the design and administration of the HRS and AHEAD is described in detail elsewhere (Juster and Suzman 1995; Soldo, Hurd, Rodgers, and Wallace 1997).

Variable Measurement

The dependent variables measured the extent to which HRS or AHEAD respondents reported any difficulties with three categories of activities:

1. Difficulties with ADLs—difficulties with any of four personal care activities (e.g., transferring in/out of bed, dressing, bathing, or eating);

2. Difficulties with mobility activities—difficulties with any of three mobility (or lower body) activities (e.g., walking across a room, walking several blocks, climbing a flight of stairs); and

3. Difficulties with strength activities—difficulties with any of two strength (or primarily upper body) activities (e.g., pushing a piece of furniture, lifting 10 pounds).

The study's independent variables included:

1. Demographic risk factors: Age (chronological age, 51–61 in the HRS, 70 and above in the AHEAD, centered on the grand mean in each data set); Female; Black (African American);

2. Life-threatening or disabling chronic conditions: Cardiopulmonary conditions—history of heart problems or lung disease; Musculoskeletal conditions—active arthritis or history of fractures or broken bones; history of Stroke; active Diabetes; Other (history of cancer or high blood pressure) (see Appendix, Note 1);

3. Physiology: Body Mass Index (BMI in the highest tertile for the HRS or AHEAD respondents);

4. Sensory conditions: Impaired vision (fair/poor/blind); Impaired hearing (fair/poor/deaf);

5. Low cognitive performance—a total score from the immediate and delayed word recall tests within the lowest quartile for scores in the HRS or AHEAD (Herzog and Wallace 1997) (see Appendix, Note 2);

6. High depressive symptoms—a total score from the Center for Epidemiologic Studies–Depression scale (Radloff 1977) within the highest quartile for the HRS or AHEAD respondents (see Appendix, Note 3); and

7. Social risk factors: Education (years of completed education 0–17, centered on 12 years); being Married/partnered.

Data Analysis

Characteristics of HRS and AHEAD respondents were compared using standard descriptive methods. Logistic regression models tested the relative contributions of diseases, conditions, and impairments as well as demographic and social factors on three measures of functional difficulties. The models estimated the effects of four sequentially entered blocks of variables on functional difficulties in both age groups: (1) sex and other demographic risk factors; (2) sex, demographic factors, diseases, and impairments; (3) the measures in (2) plus the social risk factors; and (4) the measures in (3) plus interaction terms for sex by each of the diseases, conditions, and impairments. We weighted all models for selection probability and evaluated them by standard techniques to assess significance and goodness of fit using SAS for Windows (SAS Institute 1990).

Results

Table 1 presents percentages and means on demographic and social risk factors as well as diseases, conditions, and impairments for HRS and AHEAD respondents, overall for each age group and by sex within each age group. Values that are significantly different for men and women (at p ≤ .05) are marked with asterisks. Table 2Table 3Table 4Table 5Table 6Table 7 follow with results of the multivariate analyses (presenting odds ratios and p values) on the models for ADL, mobility, and strength difficulties in the HRS and AHEAD age groups. The three factors most strongly associated with the outcomes (based on Wald chi-square statistics) are bolded. As shown in Table 1 , the mean age in the two age groups differed by 20-plus years, and older AHEAD respondents had higher proportions of women and African Americans. Among HRS respondents, educational levels were similar for men and women, and men were more frequently married. Middle-aged HRS men and women differed on specific health conditions: Women reported more musculoskeletal conditions; men reported stroke, cardiopulmonary disease, and impaired hearing more often than did women. Women tested higher in cognitive performance and reported greater difficulties with ADLs, mobility, and strength, compared to men.

Neither marital status nor education differed in AHEAD men and women. However, disproportionately greater numbers of older men reported cardiopulmonary conditions, strokes, and hearing im-pairments; more women reported musculoskeletal problems and other diseases. Although BMI was similar in older men and women, both means were less in the AHEAD than in the HRS. Similar to the middle-aged women, the older women performed somewhat better on the word recall tests. More than half of the older women reported difficulties with function. Although the prevalence of functional limitations was greater in the AHEAD group, the sex differences showed the same patterns as in the HRS.

Difficulties With ADLs

Table 2 indicates that being a woman was not significantly associated with difficulties in personal care or ADLs in middle-aged adults. For example, older age and Black race/ethnicity status significantly increased the odds of reporting difficulties in Model 1 but female sex did not. Once diseases, conditions, and impairments as well as demographic factors were controlled in Model 2, age and race/ethnicity lost significance; instead, each disease, condition, and impairment was positively associated with ADL difficulties. Based on Wald chi-square statistics, depressive symptoms, musculoskeletal conditions, and impaired vision (bolded) were particularly important. After social risk factors were added in Model 3, higher levels of education were strongly significant. Finally, no interactions (Model 4) were significant or changed the previous pattern of associations.

Table 3 illustrates that being an AHEAD woman was not directly and significantly associated with difficulty with ADLs; however, sex did exert indirect effects. Although all demographic variables were associated with ADL difficulties in Model 1, when diseases and impairments were added in Model 2, only the age effect remained significant and important and the direct sex effect was borderline. In contrast, all diseases, conditions, and impairments other than hearing and miscellaneous conditions were significant, the strongest associations being musculoskeletal conditions and depressive symptoms where sex differences have been demonstrated in previous studies. As shown in Model 3, social variables were not significantly associated with ADL difficulties nor did they add to the model. Notably, education did not change the association of low cognitive performance to ADL difficulties in the older age group (AHEAD).

When the interaction terms were added in Model 4, the female by musculoskeletal conditions and female by depressive symptoms terms were significant for ADL difficulties. In addition, the main effect for female (from Model 3) disappeared, but the main effects for musculoskeletal conditions and depressive symptoms remained. One can calculate joint effects to distinguish how musculoskeletal problems influenced ADL disability differently in men and women: Women were about three times as likely to report ADL difficulties if they had musculoskeletal problems, and men with musculoskeletal problems were about twice as likely to report ADL difficulties compared to men without musculoskeletal problems (the referent group). In the absence of musculoskeletal problems, there was no direct sex effect.

Age-Group Differences.

HRS and AHEAD differences included the effects of older age, highest tertile BMI, lower cognitive performance, and sex/musculoskeletal and sex/depressive symptoms interactions in the older age group as well as significant associations of education and hypertension/cancer in the younger group. Age-group similarities included the disappearance of the sex effect on ADL difficulties when covariates were considered, and the strong associations of depressive symptoms and most major diseases and conditions.

Difficulties With Mobility

Being a middle-aged woman was significantly associated with difficulties in mobility, even after controlling for covariates (Table 4 ). In Model 1, being female increased the odds of reporting mobility problems by 63%; and age and race/ethnicity were also positively associated. When diseases, conditions, and impairments were added (Model 2), the effect of age disappeared and the effect of sex increased. All of the health factors—particularly cardiopulmonary, musculoskeletal, and depressive symptoms—were significantly associated with mobility difficulties. When education and marital status were controlled in Model 3, impaired cognition lost significance and the associations of other diseases and conditions were attenuated; however, the effect of being female persisted.

In Model 4, the sex/BMI interaction was highly significant, and the independent effect of sex also remained. The joint effect of sex and high BMI nearly tripled the odds of mobility difficulties for women with high BMI, compared with a 53% increase for men with high BMI.

Table 5 presents parallel models for mobility difficulties for AHEAD respondents. Sex and age were consistently significant, but race/ethnicity dropped out after health factors were added. All diseases, conditions, and impairments (except hearing) were significant in Model 2. Education became significant when added in Model 3, but impaired cognitive performance lost significance. Musculoskeletal conditions, depressive symptoms, and age were most strongly associated with mobility difficulties. A significant interaction between sex and BMI (Model 4) indicated that women with a high BMI were more than twice as likely as were men without high BMI to report mobility difficulty. Men with high BMI were about equally likely as women without high BMI to report mobility difficulty.

Age-Group Differences

Differences between the age groups were few: Age was associated with mobility difficulty in the older group and hearing in the younger group. In contrast, the significant direct associations of sex with education, depressive symptoms, most diseases and conditions, and a highly significant interaction between sex with BMI were of similar magnitude in both age groups.

Difficulties With Strength

As illustrated in Table 6 , the effect of being female was strong for strength difficulties in middle-aged adults. Net of age and race/ethnicity, women were three times as likely as men to report difficulties with pushing furniture or lifting 10 pounds (Model 1). The effect increased once diseases, conditions, and impairments were added in Model 2, despite the fact that many health factors were also significant and positively associated with strength problems. As shown in Model 3, the effects of sex, diseases, and impairments remained strong even after controlling for education and marital status. Finally, after adding the interaction terms to the model (Model 4), the sex effect was even more pronounced. Similar to the mobility findings, the main effect for being female remained strong (odds ratio [OR] = 4.09), and multiple interactions were also significant (cardiopulmonary conditions, diabetes, high BMI, sensory impairments, and impaired mental health). For example, women with cardiopulmonary conditions were about nine times, women without over four times, and men with cardiopulmonary conditions over three times as likely as men without such conditions to report strength difficulties.

Table 7 illustrates the models for strength difficulties in the older group. The impact of sex and age was strong in all models and, as with the middle-aged group, women were nearly four times more likely than men to report strength difficulties even after all diseases, conditions, and impairments were considered. In Model 2, all health factors except diabetes and BMI were positively associated with strength difficulties. In Model 3, education was significantly associated, but cognitive impairment lost significance. No significant Sex × Condition interactions were noted for the older respondents although the Sex × BMI interaction was marginally significant.

Age-Group Differences.

The differences between the age groups were most evident in strength. For the middle-aged group, cardiopulmonary conditions were highly significant and multiple significant Sex × Condition interactions were found. In contrast, age was highly significant in the older group, and only the Sex × BMI interaction showed borderline significance. For both groups, the independent association of female sex remained strong, and there were significant associations of many conditions, depressive symptoms, and lower education.

Discussion

Our results demonstrated that the association of sex with disability depends on the measure of disability examined. For example, the association of female sex with increased ADL disability was completely (in middle-aged adults) or nearly completely (in older adults) explained by our disease and social covariates. However, sex did exert an indirect effect on ADL disability in older adults through the sex differentials in musculoskeletal conditions and depressive symptoms. In contrast, the direct significant associations of female sex with mobility and strength disability in both age groups were either increased or only mildly attenuated when the disease and social covariates were considered, partially supporting our first hypothesis.

Hypothesis 2 was largely disproved. Several interactions of sex with diseases, conditions, and impairments were significantly associated with all of our measures of disability and in both age groups. However, the more significant interactions were not generalized; rather, they were quite specific to covariates or age group. For example, highly significant interactions of female sex and BMI on mobility difficulties were nearly identical in both age groups, indicating that obesity had a more deleterious effect on mobility for women than for men. We found multiple significant interaction effects between sex and diseases or impairments on strength difficulties in the middle-aged group but only one (Sex × BMI) in the older age group. Previous studies indicating that middle-aged adults manifest more heterogeneity in chronic conditions than do older adults may underlie this finding. Strength was affected more strongly by the presence of specific conditions in middle-aged men than it was in women, a somewhat surprising finding. It may be that men's reports of difficulties in middle age are relatively more sensitive to their experiences with most of the diseases, conditions, and impairments that we tested. In contrast, women's greater strength difficulties overall in middle age may be due to other sex-associated factors not tested in our models (e.g., sedentary lifestyles, hormonal changes). Regardless, the differences between men and women in these relationships were quite pronounced for strength difficulties in middle age. This difference virtually disappeared in the older cohort.

It is notable that few of the diseases, conditions, and impairments or the demographic and social risk factors explained the strong associations of female sex with mobility and strength difficulties. These sex associations were of strikingly similar magnitudes in both age groups, suggesting common physiological or biomechanical mechanisms that persist over the life course and may be unmeasured in health interview surveys. The association of female sex with mobility and strength decreased only when a female Sex × BMI interaction was added in the older age group, supporting a physiological hypothesis. Alternatively, it is interesting to revisit the finding that the association of female sex with increased ADL disability was apparently accounted for (or nearly so) by our measured covariates, and, in particular, by the disease and impairment measures. Perhaps the crude direct association between sex and ADL disability is less strong to begin with than are the associations between sex and other disability measures, increasing the importance of indirect sex effects. Alternatively, ADL disability may occur so late in the disability trajectory—with mobility and strength difficulties occurring prior (Johnson and Wolinsky 1993; Verbrugge and Jette 1993; Lawrence and Jette 1996)—that sex differences earlier in the process are not as important.

Although not the primary focus of this study, comparisons of covariates in the HRS and AHEAD models are worth noting. Older age was generally significantly associated with disability for older adults, regardless of controls. This significance may be due in part to the fact that age in the younger group spanned only 10 years (ages 51–61), whereas age in the older group was broader (ages 70–103). Older age may proxy unmeasured diseases, conditions, and impairments for which older adults are increasingly at risk. Lower education was also significant in all but the ADL models in the older group, where we found an independent association with our measure of cognitive performance. Overall, education was a stronger factor in disability for HRS respondents than for AHEAD respondents. The differences between the age groups may be an artifact of age group, where education had already exerted most of its effect on health characteristics over the 70-plus years of the AHEAD respondents. Cardiopulmonary diseases and vision impairment also tended to be highly associated with all measures of disability, more in the younger than in the older age group. Although important for both groups, musculoskeletal conditions had stronger associations for older adults. In addition, mental health impairment was always among the top three associations for both age groups and across all disability measures.

Our study findings may be limited by the cross-sectional data and self-reported disease and disability measures we used to take advantage of nationally representative data, large sample sizes in key age groups, and specific methodologies. Thus, our analytic approach limited our ability to examine the roles of disability transitions and mortality. In addition, our survey data did not include biological, physiological, or clinical measures of health or other measures relating to the person–environment fit. Thus, there are many potential missing variables that may underlie our inability to explain increased mobility and strength disability in women. For example, data on diseases that present differently in men and women (e.g., Parkinson's, some types of arthritis) are not appropriately measured. Biomechanical and hormonal differences are also unmeasured. Finally, key details of disease management (e.g., rehabilitation, replacement surgery, and treatment) as well as social and lifestyle covariates (e.g., activity levels, environmental barriers) are not available in these data sets.

In summary, this research addressed hypotheses regarding the effect of sex on disability prevalence within an empirical framework that may be used for future research on sex and disability. We found that the relatively weak relationship of sex to ADL disability was explained largely by disease and social covariates, but that the sex effect on mobility and strength disability could not be explained by those same covariates. Unmeasured disability correlates and disability transitions that differ by sex may explain those differences. We found persistent and strong patterns of Sex × Condition interactions that were relatively specific (particularly female sex and BMI) and suggest different physiological determinants of disability in women. The pathophysiological substrates of potential sex differences in the disease to disability pathway as well as the social and psychological variation between the sexes may be key to disability prevention. Ferreting out those variations may mean the difference between effective and ineffective clinical or social interventions and, in turn, the ability of middle-aged and older adults experiencing diseases, conditions, and impairments to live independently in the community.

Practice Concepts

The Forum

Book Reviews

Notes

1. Although HRS and AHEAD respondents were not always asked the same battery of questions for all diseases and conditions, we were able to construct our measures of five categories of life-threatening or disabling chronic conditions based on identical or nearly identical questions in the two surveys. First, we categorized eight common chronic conditions as cardiopulmonary (heart problems, lung disease), musculoskeletal (arthritis, fractured/broken bones/hips), active diabetes, history of stroke, or other (cancer, high blood pressure). We expected that these categories of conditions would impact different capacities (e.g., cardiopulmonary conditions and aerobic capacity, musculoskeletal conditions and strength) and, in turn, different domains of physical functioning. The categories reflected self-reported history or active status in the eight individual conditions. For example, both groups were asked about physician diagnosis of a range of heart problems (e.g., heart attack, congestive heart failure), lung disease, stroke, cancer, and high blood pressure, so the category measures that include these conditions reflect history of diagnosis rather than current status for the conditions. In contrast, because HRS respondents were asked about history of diabetes and active diabetes, whereas AHEAD respondents were only asked about active diabetes, our measure of diabetes is based on active status. Similarly, although HRS respondents were asked about physician diagnosis of arthritis, AHEAD respondents were not; however, because both groups were asked if they had seen a physician within the past year about their arthritis, the latter question was the basis for our measure of active arthritis. Related questions that were somewhat different but tapped the same dimension of health were those associated with fractures or broken bones. HRS respondents were asked if they had fractured or broken any bones since age 45, and AHEAD respondents were asked if they had ever broken a hip.

2. Although there is no consensus on cognitive performance tests that influence a broad range of functioning, we based our measure of low cognitive performance on the sum of the immediate and delayed word recall test scores for two reasons. First, word recall tests largely tap the cognitive dimension of fluid abilities—processing capabilities that affect the ability to acquire new information. Because most studies document that fluid abilities are resource-intensive and decline somewhat with aging, such declines might represent a disadvantage in daily functioning. Second, word recall tests are available in both HRS and AHEAD although the range of scores differed across the studies in the first two waves of data collection. The total word recall score ranged between 0–40 in the HRS respondents, based on 20 nouns to be recalled in each of the immediate and delayed recall tests. In contrast, the total word recall score ranged between 0–20 in the AHEAD respondents, based on 10 nouns to be recalled in each test. We defined the lowest quartile of the self-respondents as the low cognitive performance group (LCP25). This categorization assesses low cognitive performance rather than cognitive impairment or clinical dementia, although the latter (impaired) respondents would be included. Both age groups had significant item nonresponse for the word recall tests. Because AHEAD respondents who could not answer or refused to answer were similar on several characteristics to respondents whose performances were poor (Herzog and Wallace 1997), we classified refusers and nonrespondents in the word recall tests for both surveys in the lowest 25th percentile. For other variables, there was very little item nonresponse. Because of scaling differences between HRS and AHEAD, cognitive performance cannot be compared across the two groups. Rather, the findings indicate effects of low cognitive performance relative to cognitive performance in their age peers.

3. The Center for Epidemiologic Studies–Depression scale (CES-D), a 20-item self-reported index of current depressive symptoms that is often used in survey research settings, was derived from earlier validated clinical depression scales to tap major feelings or behaviors associated with clinical depression (Kohout, Berkman, Evans, and Cornoni-Huntley 1993; Radloff 1977). Studies have found that summary scores from the original 20-item CES–D scale and its shorter 8-item relative correlate reasonably well with clinical ratings, although some studies suggest that the CES–D scale may not correlate as well with diagnosed depression in much older persons, compared with middle-aged or young-old adults (Kohout et al. 1993). The depressive symptoms measured in the 8-item CES–D scale included feeling depressed, lonely, sad, unhappy, or that everything done required an effort, not enjoying life, not being able to "get going," or having restless sleep.

References