| HOME | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|---|
| ||||||||||||||||||||||||||||||||
RESEARCH ARTICLE |
a Department of Sociology and Institute for Social Research, University of Michigan, Ann Arbor
Duane F. Alwin, Institute for Social Research, University of Michigan, 426 Thompson Street, Ann Arbor, MI 48106-1248 E-mail: dfa{at}umich.edu.
 |
Abstract |
|---|
 TOP Abstract Theory and Research on...MethodsResultsDiscussionReferences |
|---|
Methods. We examined age-related patterns in a measure of verbal ability using 14 repeated cross-sectional surveys from the General Social Survey (GSS) over a 24-year period.
Results. The raw GSS data show the expected age-related growth and decline in vocabulary knowledge, but these age differences are reduced when adjusted for cohort differences. There is evidence of small age-related patterns in vocabulary knowledge within cohorts, but the curvilinear contributions of aging to variation in verbal scores account for less than one-third of 1% of the variance in vocabulary knowledge, once cohort is controlled. Cohort differences in schooling contribute substantially to this effect.
Discussion. Within-age-group variation in vocabulary knowledge is vastly more important than age differences per se, and the complexities of the relationship of verbal skills to historical differences in the experience of schooling present an interesting avenue for future research.
IT is often hypothesized that age-related patterns on cognitive test scores may be due to differences in levels of cognitive aging (
Palmore 1998
;
Park 1998;
Wilson and Gove 1999;
Wolfle 1980). A deeper examination of research on the relationship between age and cognitive performance, however, suggests that the issue is more complex. The first thing one needs to appreciate in this context is the fact that there may be little uniformity across domains of cognitive functioning, such that some, but not all, abilities show age-related declines (
Hertzog and Schaie 1988). There are several possible trajectories of the life-span stability of cognitive abilities, especially with regard to when the development of abilities seems to peak and level off, prior to any significant decline. Thus, it is important to be careful not to generalize from one form of ability to another; naturally, one of the most important research questions in this area of investigation involves the nature of the differences among various ability domains in their trajectories of growth and decline. Still, in general, longitudinal research on the life-span stability of intellective test scores indicates that they are developed relatively early in adult life and are among the most stable components of human behavior (
Bloom 1964;
Brim and Kagan 1980); however, because of the onset of dementia of various forms in older adulthood, the aggregate scores of older adults on many cognitive tests show some decline (
Schaie 1996).
At the same time, there is also evidence of declines in perceptual speed and memory in old age, and recent evidence suggests that declines in measures of cognitive performance in the older age ranges are probably linked to declines in processing and sensory abilities (
Salthouse 1996). Furthermore, a convincing argument can be made that neurological function is the more fundamental latent variable reflected in these kinds of data and is responsible for any observed age differences in scores on cognitive measures; thus, net of differences in processing and sensory abilities, there may be few if any true age differences in test scores (
Baltes and Lindenberger 1997;
Lindenberger and Baltes 1994,
Lindenberger and Baltes 1997). It is important, therefore, to attempt to understand age-related trajectories on cognitive test scores in terms of what they measure. In this regard, it is generally thought that measures of processing abilities manifest the earliest decline, whereas the more crystallized aspects of ability, which depend on cultural stimulation, decline much later (
Cattell 1971a,
Cattell 1971b). Verbal skills, for example, which are highly dependent upon the completion of formal schooling, seem to reflect the kinds of abilities that grow through adolescence and by early adulthood stabilize throughout most of the life span until very old age. Given the relation of most crystallized forms of ability to the amount of schooling, it is therefore important to be able to separate aging processes from those that interact with schooling experiences.
Setting aside the question of what is measured by cognitive tests, and regardless of the type of cognitive measures involved, there are several problems with drawing any conclusions about the relationship between aging and cognitive abilities based on much of the available research to this point. Age-related differences in cognitive abilities observed in samples of different-aged individuals may in part be spurious because of the effects of cohort differences in schooling and related factors. Due to the confounding of age and cohort in cross-sectional samples of individuals, any attempt to draw inferences regarding the effects of aging must control in the most advantageous manner for any differences in cohort-based social experiences of different age groups (
Riley 1973).
We investigated age differences in one aspect of the broad spectrum of cognitive abilities, namely scores on a test of vocabulary knowledge, seeking to understand the nature and source of age differences in this dimension of verbal ability. Several earlier analyses, using data on nationally representative samples from the General Social Survey (GSS) have suggested that there are independent effects of aging on vocabulary test-score data (
Hauser and Huang 1997;
Huang and Hauser 1998;
Wilson and Gove 1999). We argue that some of these findings may result from not adequately controlling for between-cohort variation in schooling and related experiences. In the following analysis we use the diachronic data from the GSS to systematically examine patterns of age variation independent of cohort factors. Our analysis—an extension of the results presented in
Alwin and McCammon 1999—shows that in assessing age differences in cognitive scores, it is very important to control variation in cohort-linked factors, such as differences by cohort in levels of formal schooling.
|
Theory and Research on Cognitive Aging |
|---|
|
TOPAbstractTheory and Research on...MethodsResultsDiscussionReferences |
|---|
, p. 13). Crystallized intelligence, on the other hand, has it origins in experience but is not expected to be independent of other capacities because it "arises as the result of the investment of fluid intelligence, over the years, in whatever higher-level cultural skills the individual is exposed to" (p. 13). Similarly,
Salthouse 1991(p. 34) distinguished between measures that are more oriented toward "process" abilities and those that tap "product" abilities (see also
Cattell 1971b;
Denny 1982;
Horn 1982a,
Horn 1982b;
Horn and Cattell 1967;
Horn and Donaldson 1980). These two distinct aspects of cognitive functioning potentially differ in their relationships to aging.
Cattell 1971a argued that their development was collinear through adolescence (or age 15), whereupon fluid intelligence systematically declines with age, and crystallized intelligence increases ever so slightly, or otherwise remains relatively stable with age (see
Horn 1982a).
Some of the best evidence for the life-span development of various cognitive abilities is
Schaie 1996 Seattle Longitudinal Study (SLS; see
Salthouse 1991), which began tracking a cross-section of the adult population in 1956 at 7-year reinterview intervals.
Schaie 1983 concluded, based on 21 years of the SLS, that "reliably replicable age changes in psychometric abilities of more than a trivial magnitude cannot be demonstrated prior to age 60" (p. 127), and if anything, a decrement is shown only in old age, noting that a "reliable decrement can be shown to have occurred for all abilities by age 74" (p. 127).
Schaie 1989,
Schaie 1990,
Schaie 1994,
Schaie 1996(see also
Hertzog and Schaie 1986,
Hertzog and Schaie 1988;
Schaie and Hertzog 1983), thus, has made a strong case for stability in measured abilities over most of the adult life span. During the 1970s, he, along with Paul Baltes, championed the argument for the life-long stability of mental abilities and challenged what they called the "myth of intellectual decline" (
Baltes and Schaie 1976), against the arguments of
Horn and Donaldson 1976,
Horn and Donaldson 1977,
Horn and Donaldson 1980. Rather, Schaie and colleagues have argued that there is relative stability of mean performance levels throughout most of the life span with some decline in old age, although the "decline began later for the PMA [Primary Mental Abilities] subtest Verbal Meaning (a test of recognition vocabulary)," See Hertzog & Schaie, 1988, p. 122.
Schaie 1996 most recent evaluation of the age-related trajectories of cognitive abilities based on 35 years of the SLS reinforces his prior conclusions.
While the SLS is one of the most important and influential studies of cognitive aging, and its findings should not be ignored in the investigation of life-span stability, even those who work closely with it realize some of its limitations. The SLS is based on a nonprobability sample of unknown representativeness, and there are serious potential threats to internal validity due to problems of attrition that
Schaie 1996 himself acknowledges as a potential problem. However, even if one is willing to generalize from patterns of the Verbal Meaning score in the SLS, the longitudinal mean-level differences reported by
Schaie 1996 depict strong support for a story of stability from young adulthood to old age, with very little change with respect to age until at least age 70 and beyond. Moreover, other evidence from the literature on cognitive aging suggests that, whereas there are clear declines in old age in measures of processing abilities, there is little relationship of age to verbal skills (
Park 1998). Using a small sample of 310 college-educated adults aged 20–90,
Park and colleagues 1996 found evidence for systematic differences in performance across age groups on speed of processing, working memory, and free and cued recall tasks; however, measures of vocabulary knowledge did not show an age-related difference, suggesting that measures of knowledge or more crystallized measures are relatively stable across the life span.
|
Methods |
|---|
|
TOPAbstractTheory and Research on...MethodsResultsDiscussionReferences |
|---|
). In surveys beginning in 1974 and extending through 1998 (1974, 1976, 1978, 1982, 1984, 1987, 1988, 1989, 1990, 1991, 1993, 1994, 1996, and 1998), the GSS interview included a short 10-item vocabulary test developed for survey measurement (
Miner 1957;
Thorndike 1942;
Thorndike and Gallup 1944). Thorndike and Gallup described it as a "test of verbal intelligence ... [assessing] the nature of past learnings and not the ability to make novel adaptations" (pp. 78–79). Measures of vocabulary knowledge correlate highly with tests of general intelligence—usually 0.8 or higher (
Miner 1957)—and are good indicators of the verbal component of standard tests of general intelligence.
In the GSS, 10 vocabulary words were given to respondents; for each word, they were given 5 other word choices and asked to select the one that was the "closest to the meaning." These data permit an assessment of age-related patterns in verbal ability as well as variations among subgroups of the population. The GSS vocabulary measure is the number of correct answers on these 10 vocabulary items, a score that has an internal consistency reliability of 0.72. We expect the GSS vocabulary test score to behave much more like a crystallized ability than a fluid or processing kind of ability, say compared to
Schaie 1996 measure of Thurstone's PMA (Primary Mental Abilities) Verbal Meaning score, which is a "highly speeded test with a significant loading on Perceptual Speed," and which therefore clearly taps processing abilities to a much greater extent (p. 52).
Because of the limitations in sampling of minority immigrant populations, we generalize only to the native-born English-speaking population of persons aged 24 and older living in households in the United States. We focus only on native-borns because of the association between nativity and cohort membership (
2 = 51.43, 19 degrees of freedom, (df ), p = .00008) and the vast differences between natives and nonnatives on the GSS vocabulary test (F ratio with 1 and 14,510 df = 157.36, p = .0000). We focus on persons aged 24 and older for two reasons: (1) until about age 24 the amount of schooling attained is censored for vast numbers of the population aged 24 and under, and (2) because of sampling and field procedures the GSS has a great deal of difficulty obtaining an unbiased sample of younger members of the population in any cross-sectional survey. Finally, in order to generalize to a population of people rather than households, the analyses reported here weight the data by the number of adults in the household. The way in which we define the population may account for some differences between our findings and those of others.
Hauser and Huang 1997 and
Huang and Hauser 1998 include the youngest members of the GSS samples but exclude those respondents over age 65.
Wilson and Gove 1999 do not restrict the sample in any way. However, we suspect that the differences in results among these studies and ours probably have more to do with the treatment of the data than the definition of the population.
|
Results |
|---|
|
TOPAbstractTheory and Research on...MethodsResultsDiscussionReferences |
|---|
). This is revealed in part by the fact that the data for the youngest age groups, those under 30, come primarily from people born after World War II, and the data for the oldest age groups, those over 60, come principally from those born during the early quarter of this century. Age and cohort membership are fundamentally confounded in data of this sort, and without being able to separate the effects of aging from cohort, the source of the patterns in the data is ambiguous. The entries in this table also illustrate the phenomenon of cohort replacement.
|
|
;
Alwin and McCammon 1999;
Chall and Conard 1991;
Chall 1983;
Chall, Conard, and Harris 1977;
Hayes, Wolfer, and Wolfe 1996;
Stedman 1996). Thus, each cohort trajectory potentially begins at a different location on the scale, which results from cohort-related differences in experiences relevant to verbal skills. In addition, there could be period factors shaping the scores of all cohorts at various times (
Alwin 1991).
To summarize, the diachronic data within cohorts confound age and period, while the synchronous data within time confound age and cohort. In this situation it is difficult to draw any inferences about the effects of aging, unless one is willing to make some simplifying assumptions about the nature of the effects of these various factors. One approach to begin separating the influences of aging from other factors is to examine the diachronic data more carefully while controlling for cohort, making the assumption that period influences are minimal. Most of the available evidence suggests that, although people may be reading less, the nature of the standard vocabulary of modern American English has not changed within the period covered by the GSS, 1974 to 1998, which is the time interval within which such potential period effects would have to be operating (see
Glenn 1994;
Hauser and Huang 1997;
Huang and Hauser 1998;
Weakliem, McQuillan, and Schauer 1995; see
Wilson and Gove 1999, for an alternate view.) Given the unlikelihood that serious errors will result from assuming minimal period effects during this time on the aspects of vocabulary assessed by the GSS vocabulary test, it is then more readily possible to separate aging and cohort influences (
Alwin and McCammon 1999).
Following this approach, we can control for cohort by plotting each cohort's age variation in the GSS vocabulary test on a common scale, thus achieving what might be viewed as a synthesis of the various age trends. The numeric results obtained in this case are given in Table 2 . We obtained these results by deviating the GSS vocabulary test score for each case from its cohort mean and then recentering this result about the grand mean. In parallel to Table 1 , these results present the age trajectory for each cohort, but after having removed the case's cohort mean from its score. In the terminology of the analysis of variance (ANOVA), we have partitioned the GSS vocabulary test score into two parts, a within-cohort part and a between-cohort part. In Table 2 we have displayed the within-cohort portion of the GSS vocabulary score, centered around the grand mean (i.e., we have removed the between-cohort part of the variation in the score). Note that because of the way in which we have defined this score, all cohort categories have the same mean, namely the grand mean for the entire sample. Fig. 2 presents the pattern of within-cohort adjusted GSS vocabulary test means plotted in terms of age differences (see the dotted line for adjusted GSS vocabulary test scores). Each of the two curves plotted is a different age trajectory of GSS vocabulary test scores, given a particular model. The solid line is the best-fitting model for a nonlinear function of age based on the raw vocabulary score data, as in Fig. 1. The second curve is the original score after having been adjusted for cohort membership. These results suggest that a substantial part of the differences potentially attributable to age in Table 1 are removed upon controlling for differences among cohorts in factors shaping vocabulary knowledge. The curvilinear relationship of vocabulary knowledge to age is still apparent in the adjusted vocabulary score data, although the dramatic nature of the age trajectory we observed in the raw GSS test-score means is tempered considerably. Rather than a decrement from age 50 to 80 of three quarters of a word on the ten-item scale, the cohort-adjusted decline is in the neighborhood of one quarter of a word. Between ages 40 and 70 the cohort-adjusted vocabulary scores are relatively flat, although a slight curvilinear pattern is still apparent. There is a noticeable decline in old age, say after age 70, but the changes in the cohort-adjusted scores are gradual and slight, compared to the raw unadjusted figures.
|
|
Age Patterns, Cohorts, and Schooling
If aging and cohort effects on cognitive test performance are confounded due to the relationship of education to test performance, and if there are cohort differences in schooling experiences, then some of the effects of controlling for cohort (see
Fig. 2) may be attributable to schooling. To the extent that cohort differences in schooling account for these effects, it may be possible to correct for the problem in cross-sectional data by controlling statistically for schooling. In nonexperimental research, schooling can be controlled either through selection or through statistical controls (
Kish 1987). In the present case we rely on statistical adjustments. In Table 3 we demonstrate the effects of controlling for level of education in our estimation of the parameters in the regression of the GSS vocabulary test scores on age variation. We use a set of polynomials to represent the effects of age, Age and Age2. Due to the definition of our sample, we have defined Age as Age - 24. Schooling is measured as the number of years of schooling completed, centered at the grand mean.Although it may be advisable to specify the nature of the functional form more precisely (e.g.,
Goldberger 1964, pp. 214–215), we should point out that this form of the model fits the general curve of the data quite well (see
Fig. 1). Also, a model that includes a set of dummy variables representing 1-year age intervals provides a marginally better fit to the data, as measured by the improvement to the explained variance (data not presented). However, given the large sample sizes for most of the cohorts represented here, we suspect that these departures from a more or less continuous nonlinear form for the age function do not represent substantive differences worthy of retaining in the model. Thus, we rely solely on the Age and Age2 terms to describe the age function here.
|
). In either case, the magnitude of the relationships suggests that schooling should be controlled in the assessment of effects of other variables, such as aging, on GSS test score data. In the second model we estimate the coefficients for Age and Age2 without any controls. This is the model that generated the raw means for the GSS vocabulary test scores in
Fig. 1. These results suggest that aging has significant effects on vocabulary knowledge, such that vocabulary knowledge increases with age up to a point where it levels off, and later begins to decline and deteriorate in old age (
Wilson and Gove 1999). However, it should be pointed out that aging contributes only a small proportion—about one percent—to the explained variance in vocabulary knowledge, although the results are statistically significant. In model 3 we examine these same effects controlling for schooling. These results indicate that an adjustment for age differences in schooling does not modify the conclusions one would reach about the effects of aging on vocabulary knowledge, and if anything, the effects of aging are ever so slightly stronger. Although we have included schooling in the equation, we have not explicitly included cohort levels of schooling (i.e., to this point we have assessed only the reduced-form of the effects of schooling on vocabulary knowledge).
In order to assess the extent to which cohort factors are important in explaining variation in GSS test scores, in model 4 we include cohort differences as well as schooling in the model. This indicates that cohort effects on vocabulary knowledge contribute significantly to the explained variance. By adding Age and Age2 to this equation (see model 5), we can assess within this framework the extent to which aging contributes to vocabulary knowledge, net of both schooling and cohort. Note that in this regression, because we have centered the schooling variable at the grand mean and define Age as Age - 24, the intercept is the expected value of the GSS vocabulary score for persons 24 years of age. Changing the zero point on an interval scale rearranges the information contained in this regression involving interaction terms involving that variable, but in this case it does not change the basic interpretation of the effects of age or its contribution to the explained variance (see
Allison 1977).
These results reinforce those presented earlier with respect to the consequences for the aging hypothesis of controlling for cohort, namely that by controlling for cohort in these regressions, the effects of aging are substantially reduced—the increment to the R2 for model 5 compared to model 4 is .0027, which is less than one third of 1% of the variance. Compared to the unadjusted R2 of .0112, this seems quite small. As suggested earlier, there is support here for the conclusion that aging contributes to vocabulary knowledge, but the cohort-adjusted differences are much smaller. By including cohort differences in the model, we have de facto removed any and all differences among cohorts from the slopes for Age and Age2, although it is not clear from this what aspects of cohort differences account for this effect.
That the apparent effect of aging is significantly reduced by taking into account cohort differences is reflected in the comparisons between the two age functions in
Fig. 2. This is also reflected in the magnitude of the Age and Age2 coefficients in model 5. For example, controlling for cohort reduces the linear age coefficient by 46%—compare model 3 versus model 5 in Table 3 . These results show rather convincingly that a substantial portion—roughly half—of the effects of the linear increase in the GSS vocabulary score associated with aging operates via the between-cohort component of the GSS test-score data. Although we draw this comparison here, we think it is risky to make absolute comparisons across models involving quadratic terms, that is, models including product terms like Age2. The fact is that the magnitude of the linear age coefficient is dependent entirely upon where the age distribution is centered and is therefore somewhat arbitrary (see
Allison 1977, pp. 145–148). In a similar case, Allison recommends that, due to this limitation of not being able to interpret the linear coefficient independently of the intercept, the best strategy in assessing the fit of the model is to evaluate the increment to R2 (
Allison 1977, p. 149).
Assuming there is some validity to these claims, the question arises as to whether it is cohort differences in level of schooling that account for these findings. To address this, in models 6 and 7 we include a measure of between-cohort differences in schooling—the cohort-specific mean years of schooling completed. This variable is included to assess the effects of the composition of schooling on GSS test scores in order to help us decide whether it is cohort differences in schooling that account for the effect of cohort differences in patterns of age-related variation in vocabulary knowledge. As in model 5, the addition of the Age and Age2 variables to the equation containing cohort differences in schooling (see model 7) results in very little improvement in variance explained, namely .0064. By comparing the proportions of variance explained by Age and Age2 in GSS test scores in model 2 prior to any controls for cohort, and the increments to R2 linked to the addition of Age and Age2 (model 5 vs model 4, .0027, and model 7 vs model 6, .0064), we see that cohort differences in schooling account for approximately 43% of the effect of controlling cohort on estimates of aging parameters. It is relatively clear, then, that a substantial portion of the cohort differences responsible for the spurious association of aging and test scores can be summarized by intercohort differences in average level of schooling attained.
In order to investigate the robustness of the age effects in these analyses, we also included controls for race, gender, and family background (results not reported here). Specifically, the inclusion of controls for race, gender, parental education, father's occupational prestige, maternal employment, sibship size, rural upbringing, Southern upbringing, and family intactness as a child did not change any of the above results. In these analyses there were significant main effects of race (favoring Whites), gender (favoring women), parental education (favoring those with more educated parents), father's occupational prestige (favoring those with higher prestige occupations), sibship size (favoring small sibships), rural upbringing (favoring those from urban backgrounds), and Southern upbringing (favoring non-South origins). The effects of these variables on verbal test scores are beyond the scope of the present analysis (but see
Alwin 1991). However, we examined the interaction effects of gender and race with the age functions estimated in these models and found no significant differences—both in the case of gender [F (df = 2 and 16,044) = 2.268, p = .104] and race [F (df = 2 and 16,044) = .581, p = .560]. Thus, our conclusions about the consequences of aging for vocabulary knowledge seem to be quite general with respect to categories of gender and race.
Cohort Differences in Effects of Schooling
In these analyses we examined a within-cohort model for the effects of Age and Age2 (see model 2 in Table 3 ), indicating that aging seems to have only minimal effects on verbal knowledge net of cohort. Although these effects are statistically significant, they explain very little variance in GSS vocabulary test scores. Including schooling in this model did not change the results for the effects of aging, over and above the effects of controlling for cohort. In that model we were assuming that the effect of schooling was constant over cohorts. In fact this may not be the case, because there may have been some form of decline in the value of schooling with regard to verbal ability. Here we address this question by examining more closely the differential effects of schooling on vocabulary knowledge cohort by cohort. Specifically, we address the possibility that the effects of years of schooling on vocabulary knowledge differ significantly across cohorts. Within each birth-year category, we regressed the GSS vocabulary test score on the amount of schooling. In order to provide all relevant information for the examination of this hypothesis, Table 4 presents: each cohort category (1), the mean age range covered (2), the sample size (3), the correlation between age and birth year within the category (4), the mean GSS vocabulary score (5), mean years of schooling (6), the intercept (7) and slope for schooling in the prediction of the GSS vocabulary score (8), the R2 for schooling in the prediction of the GSS vocabulary score within cohorts (9), the slope for Age when added to the previous model (note that in the total sample we include both Age and Age2) (10), and the increment to the R2 for Age (and Age2 in the total sample) over and above that contributed by schooling (11). Note that in these within-cohort regressions we center the schooling variable at the grand mean of the population.
|
;
Alwin and McCammon 1999). These are unconstrained by the imposition of the "common slope" as in the analysis of covariance. By contrast, the effect of the covariance adjustment on these intercepts (i.e., the adjusted means [not presented in Table 4 ]) does impose a common slope; these two values are virtually identical (r = .99 over cohort categories), indicating that the within-cohort slopes for schooling do not vary a great deal around the common slope. In either case, the observed difference among these intercepts probably reflects differences in cohort experiences. Here we see that, net of schooling, there are systematic declines in GSS vocabulary test performance from earlier born to more recently born cohorts. The differences seem particularly pronounced for cohorts born after 1946, although these patterns are also consistent with the possibility that vocabulary knowledge increases with age. It is also possible to consider differences in the slopes in the regression of the GSS vocabulary test score on schooling as cohort effects as well, and we investigate this here. The question of whether there is a common slope for schooling amounts to whether units of education have changed their meaning over time with respect to vocabulary knowledge. Does it take more or less schooling now than it used to in order to bring about the same amount of learning? Did a year of schooling obtained in the 1920s produce the same amount of learning, for example, in students' knowledge of vocabulary, as it did in the 1960s? In other words, in addition to differentials in exposure to amount of school, is it possible to argue that there were differences in school experiences as well? This hypothesis is actually about a second-order cohort effect, that is, an effect on the relationship among variables rather than their mean levels. We can examine this question by inspecting the numbers in column 8 of Table 4 , which suggest some systematic variation in these slopes across cohorts. It appears that the within-cohort slopes are smaller for earlier born cohorts (those born before 1939), suggesting that schooling produced less learning per year. Also, it appears that the effect of schooling on vocabulary knowledge is also systematically smaller in the cohorts born after 1958. However, despite the curvilinear patterns to these slopes, the differences in these slopes are not all that large. Regardless of cohort, a 4-year schooling advantage results in a gain of about 1.5 words on the GSS vocabulary score scale. The standard F test for the differences among these slopes results in a significant value (F = 2.173 with 19 and 16,100 df ; p = .0022). However, despite this high degree of statistical significance, the model including different cohort-specific slopes increases the R2 by .0018—less than two tenths of 1% of the variance. On this basis, it is probably reasonable to assume homogeneity of slopes for purposes of estimating aging effects and cohort differences, as we have here, although we recognize there may be a substantive interpretation of the differences in these slopes.
Returning explicitly to the examination of the aging hypothesis, the results in Table 4 also present the estimation of the parameters of a within-cohort regression model including, in addition to schooling, coefficients for Age (and Age2 in the total sample) as defined earlier. These results (shown in column 11 of Table 4 ) indicate fairly clearly that there is only marginal support for effects of aging. Note that in the total sample there are systematic patterns to the age variation, as we reported earlier, but the age differences, though statistically significant, simply do not explain a substantial amount of variance. We realize that, given the nature of the data, there may be problems of redundancy between age and cohort that could bias their assessed effects; however, note that using 4-year spans of birth years to define cohorts (as in Table 1 ) in the typical case within cohorts, age and cohort are correlated about -.13 in the total 1974–1998 GSS data, so the potential bias is probably minimal. Indeed, the results presented in Table 4 are not affected by this level of correlation, because in the extreme case of setting this correlation to zero (see the Total (2) line of Table 4 ), the results are unchanged. This further reinforces the tentative conclusion drawn from Fig. 2 that when cohort differences are removed from the comparisons, there are only very small "pure" age differences in vocabulary knowledge. In only a few cases are the cohort-specific Age coefficients significant; however, the signs of these coefficients are generally consistent with the patterns predicted by Cattell, namely positive slopes, or growth in verbal ability in young adulthood coupled with declines in test performance with age in the older age ranges. It is worth noting, however, that while these coefficients are significant in the total sample, where the sample size is quite large, the effects are of very small magnitudes—together they explain only three-tenths of 1% of the variance. Given the small magnitude of these effects of aging, examined here within cohorts, we seriously doubt that differences in aging can explain the cohort effects inferred from the differences among intercepts in column 7 of Table 4 .
|
Discussion |
|---|
|
TOPAbstractTheory and Research on...MethodsResultsDiscussionReferences |
|---|
Because of the complex nature of the relationship between aging, cohorts, and schooling, we engaged in a detailed analysis of the effects of aging and school attendance on verbal skills using diachronic GSS data within cohorts. These analyses further reinforce the conclusion that when cohort differences are removed from the comparisons, there are only very small "pure" age differences in vocabulary knowledge. Greater attention should be paid to an investigation of the ways in which access to schooling and its benefits vary across cohorts. Too little is known about these processes. Such results should be examined across a wider variety of measures of cognitive ability. At the same time, while it is interesting to pursue further the question of how intercohort sources of variation may inform differences among age groups in verbal scores, it remains the case that within-age-group variation in vocabulary knowledge is vastly more important than age differences per se.
We would also conclude that it is important to better understand the ways in which schooling develops resources that can be translated into successful cognitive test performance and how these processes may change historically. And finally, given the prevailing wisdom that aging is linked to declines in cognitive skills and job performance, further research is necessary, not only to investigate the links between aging and test performance, but more importantly, on the relationship between cognitively linked declines in work performance and decisions to disengage or retire from work. As already mentioned, available cognitive tests cover a broad range of function, and it is difficult to obtain pure measures of theoretically interesting aspects of ability. Schooling and other cohort-related experiences may be differentially linked to different types of cognitive scores, and therefore one must be careful in generalizing too widely regarding the relationship of aging and cognitive abilities on the basis of measures of a restricted range of function.
|
Acknowledgments |
|---|
Received for publication July 17, 2000. Accepted for publication November 10, 2000.
|
References |
|---|
|
TOPAbstractTheory and Research on...MethodsResultsDiscussionReferences |
|---|
This article has been cited by other articles:
 |
|
 |
|
  S. L. Hatch, L. Feinstein, B. G. Link, M. E. J. Wadsworth, and M. Richards The Continuing Benefits of Education: Adult Education and Midlife Cognitive Ability in the British 1946 Birth Cohort J. Gerontol. B. Psychol. Sci. Soc. Sci., November 1, 2007; 62(6): S404 - S414. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
D. Finkel, C. A. Reynolds, J. J. McArdle, and N. L. Pedersen Cohort Differences in Trajectories of Cognitive Aging J. Gerontol. B. Psychol. Sci. Soc. Sci., September 1, 2007; 62(5): P286 - P294. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
R. P. Bowles, K. J. Grimm, and J. J. McArdle A Structural Factor Analysis of Vocabulary Knowledge and Relations to Age J. Gerontol. B. Psychol. Sci. Soc. Sci., September 1, 2005; 60(5): P234 - P241. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
K. F. Ferraro and J. A. Kelley-Moore A Half Century of Longitudinal Methods in Social Gerontology: Evidence of Change in the Journal J. Gerontol. B. Psychol. Sci. Soc. Sci., September 1, 2003; 58(5): S264 - 270. [Abstract] [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||
| HOME | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|---|
