The Journals of Gerontology Series B: Psychological Sciences and Social Sciences 62:P286-P294 (2007)
© 2007 The Gerontological Society of America
Cohort Differences in Trajectories of Cognitive Aging
Deborah Finkel,
Chandra A. Reynolds,
John J. McArdle and
Nancy L. Pedersen
1 School of Social Sciences, Indiana University Southeast, New Albany.
2 Department of Psychology, University of California, Riverside.
3 Department of Psychology, University of Southern California, Los Angeles.
4 Department of Medical Epidemiology and Biostatistics, The Karolinska Institutet Stockholm, Sweden.
Address correspondence to Deborah Finkel, PhD, Indiana University Southeast, 4201 Grant Line Road, New Albany, IN 47150. E-mail: dfinkel{at}ius.edu
 |
Abstract
|
|---|
To separate age and cohort effects on decline in normal cognitive aging, we applied growth curve models to longitudinal data from the Swedish Adoption/Twin Study of Aging. Data from up to five measurement waves covering a 16-year period were available from 806 participants (age 50 to 88 at the first wave). We divided the sample into two cohorts by birth year: 1900–1925 and 1926–1948. We generated components to tap four cognitive domains: verbal and spatial ability, memory, and speed. We tested cohort differences by using two growth models: quadratic and two linear slopes. Results indicated significant cohort differences in average performance at age 67.5 for all components except speed. When we compared linear slopes during the same age range (age 62–78), we found no cohort differences. Trajectories of change with age in these four domains were fundamentally the same in middle-old age for individuals born during the first half of the 20th century.
One of the primary challenges facing researchers in cognitive aging is to isolate true age changes from the confounding effects of time and cohort. It has long been observed that cross-sectional and longitudinal studies of cognitive aging do not always suggest the same patterns of change with age in cognitive abilities (e.g., Schaie, 2005
; Schaie & Hofer, 2001; Sliwinski & Hofer, 1999). Moreover, time-lag studies indicate significant cohort differences in intelligence. Dramatic increases in IQ test scores over the past century, known as the Flynn effect, have been documented in many countries (Flynn, 1987; Kaufman, 2001; Neisser, 1998; Rodgers, 1999; Wicherts et al., 2004). Depending on the trait in question, younger cohorts exhibit both superior and inferior performance on measures of cognitive functioning (Bowles, Grimm, & McArdle, 2005; Rönnlund, Nyberg, Bäckman, & Nilsson, 2005; Uttle & Van Alstine, 2003).
One method for advancing our understanding of cohort effects on cognitive aging is to examine cohort differences in aging trajectories for various cognitive abilities. In addition to testing cohort differences in average performance, latent growth curve models allow us to investigate cohort differences in average rates of decline with age (McArdle & Anderson, 1990; Muthen & Khoo, 1998). For example, a recent analysis compared average performance and rates of decline in basic and advanced vocabulary skills across three cohorts: 1920–1939, 1940–1959, 1960–1979 (Bowles et al., 2005). The authors reported faster decline in older cohorts for advanced vocabulary knowledge, but faster decline in younger cohorts for basic vocabulary knowledge. Using data from a cohort-sequential study of aging, Reynolds, Finkel, Gatz, and Pedersen (2002) compared time-based growth curve models in adults older or younger than 65 years at the first measurement occasion. Estimates of intercept and linear slope over time differed markedly between age groups for all three variables: block design, picture memory, and digit symbol. The conclusions that researchers make about cohort differences in cognitive aging may differ depending on whether they use time-based or age-based models, even if they use entry age as a covariate, because the time predictor may reflect additional factors beyond aging processes, including retest and attrition effects (Sliwinski & Buschke, 1999).
Our goal in the current analysis is to expand on previous investigations of cohort differences in cognitive aging. Data from the Swedish Adoption/Twin Study of Aging (SATSA; Finkel & Pedersen, 2004), an ongoing cohort-sequential study of aging twins, allows us to examine cohort differences in trajectories of aging in multiple domains of cognitive functioning by using an age-based growth curve model. The sample can be divided into two cohorts to allow comparisons of aging trajectories for individuals born during the first and second quarters of the 20th century. Five testing occasions provide sufficient data to support estimation of both linear and quadratic rates of decline in the growth curve models. Because SATSA includes extensive cognitive assessment, cohort comparisons can be made for cognitive functioning in four domains: verbal and spatial ability, memory, and processing speed.
Many researchers and theorists have highlighted the range of social, historical, and cultural forces that can and often do differ between cohorts (e.g., Riley, Foner, & Riley, 1999; Schaie, 2005; Settersten, 2006). Clearly, these forces can impact the cognitive aging process, resulting in cohort differences in the trajectories of aging. Although identifying the specific factors that impact cohort differences in cognitive aging is important, the first step is to describe cohort differences in aging trajectories. Within the growth curve model, there are two ways in which cognitive aging could differ between cohorts: intercept and rate of decline. Cohort differences in the forces that impinge on cognitive abilities could produce an initial cohort difference in cognitive performance that is then maintained as aging progresses. In other words, although average performance may differ, the rate of decline could be the same across cohorts. Salthouse (2006) refers to this phenomenon as preserved differentiation: The difference between cohorts in average performance is maintained throughout the aging life span, resulting in parallel aging trajectories. In contrast, the multiple forces that influence the aging process could impact both average performance and rate of decline. It is possible that the same factor that contributes to higher average performance also contributes to more successful maintenance of that performance over the life span. The result is differential preservation (Salthouse): Cognitive abilities may be better preserved in one cohort than in the other cohort. The difference in preservation of cognitive functioning leads to differential rates of decline and thus divergent aging trajectories.
The results of previous cross-sectional comparisons allow us to make predictions about cohort differences in average performance; therefore, we expect cohort differences in the intercept of the growth curve model in all four cognitive domains, with higher average performance in the younger cohort (Bowles et al., 2005; Flynn, 1987; Reynolds et al., 2002; Schaie, 2005). Conclusions from cross-sectional data may not transfer to predictions about cohort differences in cognitive decline, however (e.g., Riley et al., 1999; Schaie & Hofer, 2001; Sliwinski & Hofer, 1999), and few studies of have reported analyses of cohort differences in rates of decline in cognitive performance. Two-component theories of intelligence (Horn, 1988; Horn & Cattell, 1966; Lindenberger, 2001) predict greater cultural impact on aging trajectories for aging-resilient (i.e., crystallized) abilities than for age-sensitive (i.e., fluid) abilities. Given cohort differences in cultural experience (cf., Riley et al.; Settersten, 2006), two-component theories of intelligence suggest that crystallized abilities (including verbal ability and memory) would follow the differential preservation hypothesis and exhibit faster rates of decline in the older cohort. Previous cohort comparisons of rates of decline in vocabulary provide only partial support for the differential preservation hypothesis, however (Bowles et al.).
Spatial abilities and processing speed are generally considered components of fluid intelligence (e.g., Birren & Fisher, 1995; e.g., Lindenberger, Mayr, & Kleigl, 1993; Salthouse, 1996), although processing speed can be conceptualized as a separate construct (e.g., Cattell, 1971; Horn & Hofer, 1992; Schaie, 2005). Previous growth curve comparisons report significant age group differences in rates of decline for processing speed and spatial ability (Reynolds et al., 2002). However, two-component theories of intelligence hypothesize small cultural effects for measures of fluid abilities (Lindenberger, 2001). Therefore, it is possible that any cohort differences in experience will have no impact on rates of decline for measures of fluid abilities, providing support for the preserved differentiation hypothesis.
|
METHODS
|
|---|
Participants
Ascertainment procedures for SATSA have been described previously. In brief, the sample is a subset of twins from the population-based Swedish Twin Registry (Finkel & Pedersen, 2004). The base population comprises all pairs of twins who indicated that they had been separated before the age of 11 and reared apart, and a sample of twins reared together who were matched on the basis of gender and date and county of birth. Twins were mailed questionnaires, and a sample of those twin pairs who were aged 50 years or older in which both twins responded was invited to participate in an additional in-person examination of health and cognitive abilities (Pedersen et al., 1991). In-person testing (IPT1) took place in a location convenient to the twins. Testing was completed during a single 4-hour visit. The second and third waves of in-person testing (IPT2 and IPT3) occurred after 3-year intervals. In-person testing did not occur during Wave 4; therefore, the next wave of in-person testing is labeled IPT5 and occurred after a 7-year interval (see Finkel & Pedersen). The fifth wave of in-person testing (IPT6) took place 3 years after IPT5.
Dementia status was determined by clinical diagnosis based on current diagnostic criteria (Gatz, et al., 1997), and in the current analyses we did not include participants who developed dementia at any point during their participation. The number of participants at each in-person testing occasion who remained free of dementia as of IPT5 is 594, 558, 538, 516, and 441. In total, 806 nondemented individuals had cognitive data available from at least one testing occasion. In the older cohort, 59% of participants are women and 41% are men; in the younger cohort, 58% are women and 42% are men.
The median birth year of the individuals in the sample was 1924. However, participants were recruited into the Swedish Twin Registry in two waves: twins born in 1925 and earlier and twins born after 1925 (Lichtenstein et al., 2002). Therefore, we defined two cohorts on the basis of the recruitment phases: individuals born between 1900 and 1925 (n = 425) and individuals born between 1926 and 1948 (n = 381; see the end note at the end of this article). Sample sizes and age ranges in each cohort at each testing wave are presented in Figure 1. On the left side of the figure, truncation at age 50 is evident in the younger cohort. On the right side, subject mortality is evident in the upper limits of the age ranges in the older cohorts. In SATSA, education is rated on a 4-point scale from 1 (elementary school) to 4 (university or higher). Mean education was 1.75 in the younger cohort and 1.48 in the older cohort. This is a statistically significant difference, at t(804) = –3.95, p <.01.
View larger version (18K):
[in this window]
[in a new window]
|
Figure 1. Age range of individuals assessed in younger (born 1926–1948) and older (born 1900–1925) cohorts at each wave of measurement. Sample sizes are indicated in each bar. IPT denotes the wave of in-person testing. Note that the fourth wave of measurement did not include an IPT component
|
|
In our current analyses we focus on cohort comparisons of individual performance, making it necessary for us to eliminate any bias resulting from the inclusion of twins. We fit all models to a sample that included a randomly selected member of each twin pair; this is Sample A. We then replicated analyses in a sample consisting of the other member of each twin pair; this is Sample B. We randomly assigned individuals from incomplete pairs to either Sample A or Sample B.
Measures
Four cognitive domains are represented in the SATSA cognitive test battery (see Nesselroade, Pedersen, McClearn, Plomin, & Bergeman, 1988; Pedersen et al., 1992): verbal, spatial, memory, and processing speed abilities. Verbal abilities are tapped by tests on Information, Synonyms, and Analogies. Figure Logic, Block Design, and Card Rotations tests assess spatial abilities. Memory tests include Digit Span, Picture Memory, and Names and Faces. Finally, Symbol Digit and Figure Identification tests measure processing speed. Reliabilities for these tests range from.82 to.96 (Pedersen et al.). We used principal components analysis to construct latent components from the individual tests within each domain: verbal, spatial, memory, and speed. For the verbal, spatial, and speed components, loadings ranged from.78 to.92 and the components explained 74%, 67%, and 85% of the variance among the individual measures. The memory component was more diverse, including measures of short-term, long-term, and picture memory. Loadings ranged from.64 to.78 and the component explained 53% of the variance. Previous comparisons of component structure between cohorts and across testing occasions indicate that the structure does not vary systematically across age or time (Finkel, Reynolds, McArdle, & Pedersen, 2005a). To avoid the issue of measurement variance (cf. Wicherts et al., 2004), we created an invariant definition of components at each testing occasion by standardizing the cognitive measures relative to the respective means and variances at IPT1, and we used the loadings from the principal components analyses conducted at IPT1 to construct the verbal, spatial, and memory components. We combined the speed measures into a speed component by using unit weighting. Finally, for ease of interpretation, we translated all component scores into T scores, using means and variances from IPT1.
Statistical Method
We used a growth curve model to examine cohort differences in cognitive aging. The structural model can be considered as a multilevel random coefficients model (Bryk & Raudenbush, 1992; Laird & Ware, 1982; McArdle & Anderson, 1990). The model provides estimation of fixed effects, that is, fixed population parameters as estimated by the average growth model of the entire sample, and random effects, that is, interindividual variability in intraindividual change in growth model parameters. Growth curve models take into account missing data by giving more weight to individuals with the most time points.
Figure 2 presents a general form of the growth models that we used in these analyses. Individual scores at any one time are a function of a latent intercept (I), practice or retest effects (P), the first age basis (B1), the second age basis (B2), and random error (u0–u4). Note that I*, B1*, and B2* refer to the standardized scores of I, B1, and B2. Standardized practice effects can also be included; however, previous analyses have indicated no interindividual variance around mean practice effects (Finkel et al., 2005a). The model-fitting procedure entails fitting individual growth models to all available data; repeated measurements are indicated by the y0 through y4 variables. Paths from practice to the observed scores indicate that the entire practice effect was assumed to occur at the first retest. The paths from the latent age basis parameters to the observed scores are the age basis coefficients: B1(t) and B2(t). The age basis serves as a marker for the age of the subject at each time of measurement, adjusted for the centering age. The random errors or uniquenesses (u0–u4) represent unaccounted variation from fitting the growth model to the cognitive measures; we constrained these time-specific residual variances to be equal over time. The means (Mi = mean intercept; Mp = mean practice; Mb1 = mean first age basis; Mb2 = mean second age basis) are the estimates of the average performance and average amount of change. Standard deviations of the interindividual differences in the intercept and change parameters are indicated by Di, Db1, and Db2. Finally, the relationships among the intercept and rates of change are represented by the correlations Rib1, Rib2, and Rb1b2.
We used two versions of the growth model in the present analyses; each model is defined by the manner in which the two age bases were parameterized. First, we fit a quadratic model to the data, setting B1 equal to the linear slope (age) and B2 equal to accelerating changes with age (age2). We compared parameter estimates for intercept, slope, and quadratic terms across cohorts. We estimated quadratic growth models and tested cohort differences by using PROC Mixed with maximum likelihood estimation in SAS 8.0 (SAS Institute, 1999). However, comparing quadratic growth models across the different age ranges encompassed by each cohort does not adequately capture the overlapping age range that is covered by both cohorts (i.e., ages 62–78). Therefore, in a second approach, we expanded the standard linear growth model to include two linear growth rates (Bryk & Raudenbush, 1992; Finkel, Reynolds, McArdle, Gatz, & Pedersen, 2003): one linear slope before the centering age and a separate linear slope after the centering age, with the centering age acting as a pivot point. We estimated and tested two-slope growth models in Mx by using maximum likelihood estimation (Neale, Boker, Xie, & Maes, 2003).
It is important to note that one of the fundamental assumptions of growth curve models is that data are missing at random (MAR). Without complete mortality data, it is difficult to establish whether the MAR assumption is met, although previous investigations of SATSA data suggest that participants who continue in the study are significantly different from those who drop out (e.g., personality ratings; see Pedersen & Reynolds, 1998). Of most importance in a comparison of growth models across cohorts is to demonstrate that the pattern of missing data does not differ between cohorts. Table 1<--?1--> presents the participation rates in the two cohorts. We can see that although the patterns are fairly similar, the older cohort is somewhat more likely than the younger cohort to have at least 3 time points simply as a result of a greater number of opportunities for participation. However, in the overlapping age interval where the two cohorts can be compared most directly (age 62–78; see Figure 1), there was no significant difference between the cohorts in the rate of missing data.
|
RESULTS
|
|---|
Quadratic Model
We used quadratic growth curve models to examine cohort differences in both average performance and rate of change with age. Age 67.5 was the centering age that we used for the quadratic growth curve analyses, which we selected because it crosses three of the five waves of measurement for each cohort. First we tested three general models that examine cohort differences in fixed effects (means) and random effects (variances). In Model 1, we fit the quadratic growth model to Sample A. In Model 2, we entered cohort as a covariate in the model, allowing the fixed effects (mean intercept, mean practice, mean slope, and mean quadratic terms) to differ between cohorts. In Model 3, we also allowed random effects to differ between cohorts: intercept, slope, and quadratic variance and correlation terms, as well as residual variance. We compared the three nested models by using the difference chi-square test obtained by taking the difference between the model fits [i.e., –2 ln(likelihood)] and testing its significance with the degrees of freedom equal to the difference in the number of parameters of the two models. Comparing Models 1 and 2 provided an omnibus test of cohort differences in average performance and average rate of change, and comparing Models 2 and 3 provided an omnibus test of cohort differences in interindividual differences in the parameters. Parameters estimates from Model 2 and model fit statistics are presented in Table 2. We repeated all analyses and model comparisons in Sample B; the results of model testing in Sample B were consistent with the results presented here. For all four cognitive components, model comparison indicated a significant cohort difference in the fixed effects of the quadratic model. In contrast, model comparison failed to indicate any cohort differences in the interindividual differences in the quadratic model. Therefore, we constrained random effects to be equal across cohorts for the remaining analyses of quadratic growth curves.
Following the omnibus test of cohort differences in fixed effects, we tested the significance of cohort differences in individual growth curve parameters in an additive manner. In Model 4, only the intercept differed between cohorts, and we compared model fit with Model 1. In Model 5, both intercept and practice differed between cohorts, and we compared model fit with Model 4. Model 6 included separate cohort estimates for intercept, practice, and linear slope; Model 7 included separate cohort estimates for intercept, practice, linear slope, and the quadratic term. (Model 7 is identical to Model 2; both allow cohort differences in all fixed effects. Note that testing the growth parameters independently produced the same results as did the additive method described.)
The resulting changes in fit between these nested models are provided in Table 3. The quadratic growth curves estimated by Model 2 (allowing cohort differences in all fixed effects) are presented in Figure 3. For the verbal, spatial, and memory components, the results follow the same pattern: significant cohort differences in the intercept and quadratic term, but no significant differences in practice or linear slope. The intercept estimates the average performance at the centering age, and performance on the verbal, spatial, and memory components at age 67.5 is significantly lower in the older cohort. The older cohort also demonstrated a more negative quadratic estimate (faster decline) than the younger cohort for these three components. We found a very different pattern of results for the speed component. The only significant cohort difference was in the significantly smaller practice term in the older cohort: The average performance and rates of age-related decline for the processing speed component did not differ significantly between the two cohorts.
View larger version (12K):
[in this window]
[in a new window]
|
Figure 3. Quadratic latent growth curve models estimated by Model 2 (allowing cohort differences in all fixed effects): younger cohort (dashed line) and older cohort (solid line)
|
|
One concern with comparing quadratic growth curves across the two cohorts is that the quadratic term estimates accelerating decline across the entire age range. Given the differences between cohorts in the completeness of the data, especially at the ends of the age range (see Figure 1), the precision of the growth curve models may not be comparable across cohorts. Therefore, testing cohort differences in the quadratic parameter may be functionally equivalent to comparing accelerating rates of decline in middle and late adulthood. To examine cohort differences in rates of decline, a more direct comparison of decline in the same age range is required.
Two-Slope Model
The purpose of applying the two-slope model was to compare rates of decline for the two cohorts precisely where the age ranges overlap: ages 62 to 78. In the two-slope model, the first age basis consists of the linear slope prior to the centering age and the second age basis is the linear slope after the centering age. To establish a linear slope in the same age range in the separate cohorts for the purposes of cohort comparison, we selected different centering ages on the basis of the endpoints of the overlapping age range (62 and 78). We centered the data at age 62 in the younger cohort. Consequently, Slope 1 indicates average linear change with age between ages 50 and 62, and Slope 2 indicates average linear change with age between ages 62 and 78. In the older cohort, we centered the data at age 78; therefore, Slope 1 covered ages 62 to 78 and Slope 2 covered ages 78 to 94. As a result, Slope 2 in the younger cohort and Slope 1 in the older cohort both covered the overlapping age range of 62 to 78 years. We fit the two-slope growth models separately in the two cohorts, and we compared two models. First, we fit the full model, estimating separate intercept, practice, Slope 1, and Slope 2 in each cohort, to the data. In the second model, we constrained Slope 2 in the younger cohort to be equal to Slope 1 in the older cohort. Model fitting in Mx was required to model this incongruous constraint between the two cohorts. Given the differences in centering ages, interindividual variances were not constrained to be equal across cohorts. We again compared the two nested models by using the difference chi-square test.
The parameter estimates from the full two-slope model and the results of model fitting are presented in Table 4. As the focus of the two-slope model was comparison of mean slopes, the 11 variance, covariance, and residual variance estimates in each cohort are not reported here but are available from D. Finkel upon request. Parameter estimates for intercept and practice will differ between Tables 2 and 4 because of the difference in centering ages: intercept and practice parameters indicate the average performance and average practice effects at the centering age. The two-slope growth curves estimated for each cohort are presented in Figure 4. We fit all models in Sample A and we repeated model fitting in Sample B; results in the two samples were consistent and the results from Sample A are reported. The differences in intercept indicated by the quadratic growth curve analysis are still evident in the two-slope models presented in Figure 4. For verbal, spatial, and memory variables, there is a clear separation of the decline trajectories for the two cohorts. In addition, all slope estimates reported in Table 4 are significantly different from zero, with the exception of slope estimates in the 50–62 age interval in the younger cohort for the verbal and memory components. However, as the model-fitting results presented at the bottom of Table 4 indicate, there are no significant cohort differences in the average rate of decline in the overlapping age range.
View this table:
[in this window]
[in a new window]
|
Table 4. Results of Fitting the Two-Slope Model to the Cognitive Factors: Parameter Estimates From the Full Model.
|
|
View larger version (13K):
[in this window]
[in a new window]
|
Figure 4. Two-slope latent growth curve models estimating independent slopes for the younger cohort (dashed line) and older cohort (solid line)
|
|
|
DISCUSSION
|
|---|
Our goal in the current analysis was to examine possible cohort differences in trajectories of decline for cognitive variables, as separate from cohort differences in average performance. We applied two parameterizations of a growth model to the data: a quadratic model estimating linear and accelerating decline and a two-slope model estimating decline in two linear components before and after a midpoint. We estimated cohort differences in the average performance in the quadratic approach, and we examined cohort differences in average age-related rates of decline by using both the quadratic and two-slope growth models.
Average Performance
By comparing the intercept of the quadratic growth model across cohorts, we found that the average performance was one third to one half of a standard deviation lower in the older cohorts in three of the four cognitive domains: verbal ability, spatial ability, and memory. We found no significant cohort differences in average performance for processing speed. It is important to note that the magnitude of cohort differences in average performance will depend on the centering age used in the growth model; general conclusions may be drawn but interpretations of the size of the cohort difference should be made with caution. Previous analyses of the Flynn effect suggest small cohort effects for verbal ability (Flynn, 1987), but investigations focusing on vocabulary typically report large cohort differences in samples with 40- to 80-year ranges in birth years (Bowles et al., 2005; Schaie, 2005; Uttle & Van Alstine, 2003), and Kaufman (2001) reported large cohort differences for both the Wechsler Adult Intelligence Scale (WAIS) verbal IQ and a verbal comprehension index. Similarly, in the present analyses, cohort differences in average performance were significant for the verbal component. It is likely that the cohort differences in education found in this sample are responsible for at least part of the cohort difference in verbal ability, but previous analyses indicate that 20th-century trends in education do not fully explain cohort differences in verbal ability (e.g., Alwin & McCammon, 2001; Kaufman; Finkel, Reynolds, McArdle, & Pedersen, 2005b).
On the basis of previous investigations, we predicted significant cohort effects for memory and spatial ability, and these predictions were supported. We found no cohort differences in average performance for processing speed. A comparison of WAIS-R and WAIS-III data also found no cohort differences for performance IQ or for indices of perceptual organization and processing speed (Kaufman, 2001). Results from the Seattle Longitudinal Study also report some cohort differences for measures of perceptual speed, although the magnitude of those differences is smallest for Identical Pictures, the task that is most similar to the Figure Identification component of the processing speed component used here (Schaie, 2005). In addition, other tasks with a substantial speed component (e.g., Word Fluency) exhibit little or no cohort differences. Two-component theories of cognitive aging predict smaller cohort differences for age-sensitive abilities such as processing speed (Horn, 1988; Horn & Cattell, 1966; Lindenberger, 2001). These results may indicate that slowing of perceptual speed is a component of primary aging, fundamental to the aging process and shaped very little by cohort or cultural factors.
Rates of Decline
Comparing average performance across cohorts fails to capture the dynamic processes of change that occur with cognitive aging. With cohort-sequential data, we can focus not only on interindividual differences in performance but on cohort differences in intraindividual change over time (Baltes & Nesselroade, 1979). Therefore, the primary focus of the current analysis was to investigate cohort differences in trajectories of decline for cognitive measures. We applied both quadratic and two-slope growth models to the data and, as is often the case in aging research, the form of the question affected the outcome.
Although the quadratic model estimates accelerating age changes across the entire age range of the sample, the cohorts differed in the distribution of data available to support the parameter estimates. As a result, examining cohort differences in quadratic parameters in the two cohorts functionally compares aging trajectories defined by young-old and old-old age. It is not surprising, then, that significant cohort differences in accelerating decline were found for verbal, spatial, and memory abilities. In contrast, no cohort differences in the quadratic parameter were found for processing speed. Therefore, we can conclude that the average rate of decline in processing speed is fundamentally the same across cohorts.
Applying the two-slope growth model to the issue of cohort differences in rates of decline allowed us to make direct cohort comparisons of decline in the same age range: ages 62 to 78. In that age range, no significant cohort differences in rates of decline were identified for any of the cognitive domains assessed. In their comparison of longitudinal trends in vocabulary across three cohorts, Bowles and colleagues (2005) did not constrain their models to compare aging trajectories in the same age range. Even so, their results indicate cohort differences in average performance, but not in the rates of decline in the age ranges for which the cohorts overlap (ages 35–45 and 55–65). Taken together, these results are in direct opposition to theories of cohort differences in cognitive functioning based primarily on investigations of average performance. Cohort differences in average performance on cognitive tasks have been strongly supported by several studies, including the present quadratic growth curve analysis. The absence of cohort differences in linear aging trajectories, however, suggests preserved differentiation: Initial cohort differences in cognitive performance are maintained throughout the aging process. There is a growing body of evidence for preserved differentiation in cognitive aging. For example, although researchers find the expected sex differences in average performance, they report few sex differences in rates of decline for measures of cognitive ability (Aartsen, Martin, & Zimprich, 2004; Caskie, Schaie, & Willis, 1999; Finkel, Reynolds, Pedersen, & Berg, 2006; MacDonald, Hultsch, Strauss, & Dixon, 2003). Similarly, Salthouse (2006) reports that, although mental exercise impacts average performance, there is no evidence to date that it results in differential preservation of cognitive abilities (i.e., differential rates of decline).
In the current investigation, the fundamental nature of these conclusions is restricted by the limited amount of overlap in the age ranges covered by the two cohorts (only 16 years) and by the focus on linear decline when the life-span aging trajectories are primarily quadratic in nature. Furthermore, the use of time-based, as opposed to event-based, cohorts may have limited substantive interpretation of the results. Although it is easy to pinpoint the date of events that may impact functioning (e.g., the 1918 influenza epidemic in Sweden), determining the point in the life span when the impact of the event will be strongest is not a simple matter (Alwin, Hofer, & McCammon, 2006). Therefore, with regard to the impact on cognitive decline in the second half of the life span, event-based cohort definition may be as arbitrary as time-based cohort definition. Additional support for the absence of cohort differences in aging trajectories would be provided by comparisons of growth models from cohorts that differ more dramatically than the cohorts used in the present analyses, such as the first and second halves of the 20th century. In summary, the current analyses indicate that the dramatic social, historical, and cultural differences between cohorts (Riley et al., 1999; Schaie, 2005; Settersten, 2006) impact average performance on cognitive tasks, but they do not result in cohort differences in the preservation of cognitive functioning during the second half of the life span. Clearly, interindividual differences in cognitive aging exist (e.g., Finkel et al., 2003), and identifying the factors that contribute to slower rates of decline and thus more successful aging continues to be of paramount importance.
|
Acknowledgments
|
|---|
The Swedish Adoption/Twin Study of Aging is supported by the National Institute on Aging (under Grants NIA AG04563 and AG10175), The MacArthur Foundation Research Network on Successful Aging, and the Swedish Council for Social Research (under program 97:0147:1B).
|
Footnotes
|
|---|
Decision Editor: Thomas M. Hess, PhD
Received for publication October 26, 2006.
Accepted for publication May 15, 2007.
|
References
|
|---|
- Aartsen, M. J., Martin, M., Zimprich, D. (2004). Gender differences in level and change in cognitive functioning. International Journal of Experimental, Clinical, and Behavioural Gerontology, 50,35-38.
- Alwin, D. F., Hofer, S. M., McCammon, R. J. (2006). Modeling the effects of time: Integrating demographic and developmental perspectives. In R. H. Binstock & L. K. George (Eds.), Handbook of aging and the social sciences (pp. 20–38). New York: Academic Press.
- Alwin, D. F., McCammon, R. J. (2001). Aging, cohorts, and verbal ability. Journal of Gerontology: Social Sciences, 56B,S151-S161.[Abstract/Free Full Text]
- Baltes, P. B., Nesselroade, J. R. (1979). History and rationale of longitudinal research. In J. R. Nesselroade & P. B. Baltes (Eds.), Longitudinal research in the study of behavior and development (pp. 1–40). New York: Academic Press.
- Birren, J. E., Fisher, L. M. (1995). Aging and speed of behavior: Possible consequences for psychological functioning. Annual Review of Psychology, 46,329-353.[Medline]
- Bowles, R. P., Grimm, K. J., McArdle, J. J. (2005). A structural factor analysis of vocabulary knowledge and relations to age. Journal of Gerontology: Psychological Sciences, 60B,P234-P241.[Abstract/Free Full Text]
- Bryk, A. S., Raudenbush, S. W. (1992). Hierarchical linear models. London: Sage.
- Caskie, G. I. L., Schaie, K. W., Willis, S. L. (1999). Individual differences in the rate of change in cognitive abilities during adulthood [Abstract]. The Gerontologist, 39,398.
- Cattell, R. B. (1971). Abilities: Their structure, growth, and action. Boston: Houghton Mifflin.
- Finkel, D., Pedersen, N. L. (2004). Processing speed and longitudinal trajectories of change for cognitive abilities: The Swedish Adoption/Twin Study of Aging. Aging, Neuropsychology, and Cognition, 11,325-345.
- Finkel, D., Reynolds, C. A., Pedersen, N. L., Berg, S. (2006). Surprising lack of sex differences in normal cognitive aging in twins. International Journal of Aging and Human Development, 62,335-357.[Medline]
- Finkel, D., Reynolds, C. A., McArdle, J. J., Gatz, M., Pedersen, N. L. (2003). Latent growth curve analyses of accelerating decline in cognitive abilities in late adulthood. Developmental Psychology, 39,535-550.[Medline]
- Finkel, D., Reynolds, C. A., McArdle, J. J., Pedersen, N. L. (2005a). The longitudinal relationship between processing speed and cognitive ability: Genetic and environmental influences. Behavior Genetics, 35,535-549.[Medline]
- Finkel, D., Reynolds, C. A., McArdle, J. J., Pedersen, N. L. (2005b, November). Cohort differences in patterns of decline in cognitive abilities. Paper presented at the annual meeting of the Gerontological Society of America, Orlando, FL.
- Flynn, J. R. (1987). Massive IQ gains in 14 nations: What IQ tests really measure. Psychological Bulletin, 101,171-191.
- Gatz, M., Fiske, A., Reynolds, C. A., Wetherell, J. L., Johansson, B., Pedersen, N. L. (2003). Sex differences in genetic risk for dementia. Behavior Genetics, 33,95-105.[Medline]
- Horn, J. L. (1988). Thinking about human abilities. In J. R. Nesselroade & R. B. Cattell (Eds.), Handbook of multivariate experimental psychology. New York: Plenum Press.
- Horn, J. L., Cattell, R. B. (1966). Age differences in primary mental ability factors. Gerontology, 21,210-220.
- Horn, J. L., Hofer, S. M. (1992). Major abilities and development in the adult period. In R. J. Sternberg & C. A. Berg (Eds.), Intellectual development (pp. 44–99). New York: Cambridge University Press.
- Kaufman, A. S. (2001). WAIS-III IQs, Horn's theory, and generational changes from young adulthood to old age. Intelligence, 29,131-167.
- Laird, N. M., Ware, H. (1982). Random-effects models for longitudinal data. Biometrics, 38,963-974.[Medline]
- Lichtenstein, P., deFaire, U., Floderus, B., Svartengren, M., Svedberg, P., Pedersen, N. L. (2002). The Swedish Twin Registry: A unique resource for clinical, epidemiological and genetic studies. Journal of Internal Medicine, 252,184-205.[Medline]
- Lindenberger, U. (2001). Lifespan theories of cognitive development. In N. J. Smelser & P. B. Baltes (Eds.), International encyclopedia of the social and behavior sciences (pp. 8848–8854). Oxford, England: Elsevier.
- Lindenberger, U., Mayr, U., Kleigl, R. (1993). Speed and intelligence in old age. Psychology and Aging, 8,207-220.[Medline]
- MacDonald, S. W., Hultsch, D. F., Strauss, E., Dixon, R. A. (2003). Age-related slowing of digit symbol substitution revisited: What do longitudinal age changes reflect? Journal of Gerontology: Psychological Sciences, 58B,P187-P194.[Abstract/Free Full Text]
- McArdle, J. J., Anderson, E. (1990). Latent variable growth models for research on aging. In J. E. Birren & K. W. Schaie (Eds.), Handbook of the psychology of aging (3rd ed., pp. 21–44). New York: Academic Press.
- Muthen, B. O., Khoo, S. T. (1998). Longitudinal studies of achievement growth using latent variable modeling. Learning and Individual Differences, 10,73-101.
- Neale, M., Boker, S. M., Xie, G., Maes, H. H. (2003). Mx: statistical modeling (6th ed.). Richmond, VA: Virginia Commonwealth University, Department of Psychiatry.
- Nesselroade, J. R., Pedersen, N. L., McClearn, G. E., Plomin, R., Bergeman, C. S. (1988). Factorial and criterion validities of telephone-assessed cognitive ability measures: Age and gender comparisons in adult twins. Research on Aging, 10,220-234.[Abstract]
- Neisser, U. (1998). The rising curve: Long term gains in IQ and related measures. Washington, DC: American Psychological Association.
- Pedersen, N. L., McClearn, G. E., Plomin, R., Nesselroade, J. R., Berg, S., de Faire, U. (1991). The Swedish Adoption/Twin Study of Aging: An update. Acta Geneticae Medicae et Gemellologiae, 40,7-20.[Medline]
- Pedersen, N. L., Plomin, R., Nesselroade, J. R., McClearn, G. E. (1992). Quantitative genetic analysis of cognitive abilities during the second half of the lifespan. Psychological Science, 3,346-353.
- Pedersen, N. L., Reynolds, C. A. (1998). Stability and change in adult personality: Genetic and environmental components. European Journal of Personality, 12,365-386.
- Reynolds, C. A., Finkel, D., Gatz, M., Pedersen, N. L. (2002). Sources of influence on rate of cognitive change over time in Swedish twins: An application of latent growth models. Experimental Aging Research, 28,407-433.[Medline]
- Riley, M. W., Foner, A., Riley, J. W., Jr.. (1999). The aging and society paradigm. In V. L. Bengtson & K. W. Schaie (Eds.), Handbook of theories of aging (pp. 327–343). New York: Springer.
- Rodgers, J. L. (1999). A critique of the Flynn effect: Massive IQ gains, methodological artifact, or both? Intelligence, 26,337-356.
- Rönnlund, M., Nyberg, L., Bäckman, L., Nilsson, L.-G. (2005). Stability, growth, and decline in adult life span development of declarative memory: Cross-sectional and longitudinal data from a population-based study. Psychology and Aging, 20,3-18.[Medline]
- Salthouse, T. A. (1996). The processing-speed theory of adult age difference in cognition. Psychological Review, 103,403-428.[Medline]
- Salthouse, T. A. (2006). Mental exercise and mental aging: Evaluating the validity of the "use it or lose it" hypothesis. Perspectives on Psychological Science, 1,68-87.
- SAS Institute Inc. (1999). SAS online documentation for Windows (Release Version 8). Cary, NC: Author.
- Schaie, K. W. (2005). Developmental influences on adult intelligence: The Seattle Longitudinal Study. New York: Oxford University Press.
- Schaie, K. W., Hofer, S. M. (2001). Longitudinal studies in aging research. In J. E. Birren & K. W. Schaie (Eds.), Handbook of the psychology of aging (5th ed., pp. 53–77). New York: Academic Press.
- Settersten, R. A., Jr.. (2006). Aging and the life course. In R. H. Binstock & L. K. George, Handbook of aging and the social sciences (pp. 3–19). New York: Academic Press.
- Sliwinski, M., Buschke, H. (1999). Cross-sectional and longitudinal relationships among age, cognition, and processing speed. Psychology and Aging, 14,18-33.[Medline]
- Sliwinski, M., Hofer, S. (1999). How strong is the evidence for mediational hypotheses of age-related memory loss? International Journal of Experimental, Clinical, and Behavioural Gerontology, 45,351-354.
- Uttle, B., Van Alstine, C. L. (2003). Rising verbal intelligence scores: Implications for research and clinical practice. Psychology and Aging, 18,616-621.[Medline]
- Wicherts, J. M., Dolan, C. V., Hessen, D. J., Oosterveld, P., van Baal, G. C. M., Boomsma, D. I., et al. (2004). Are intelligence tests measurement invariant over time? Investigating the nature of the Flynn effect. Intelligence, 32,509-537.
TOP
Abstract
Methods
