In my discussion of major methodological issues addressed in the Seattle Longitudinal Study (SLS) (see chapter 2), I emphasized the desirability of obtaining separate estimates of age, cohort, and period effects. In this chapter, I report our findings regarding cohort and period differences in cognitive abilities as well as on other variables included in our study. Data regarding these matters were previously reported through the seventh study cycle (Schaie, 1983a, 1996b, 2005a). Note that expanded operational definitions have been made more explicit and are provided for the computation of cohort and period effects. Moreover, cumulative findings through the eighth study cycle are expanded and updated where appropriate.
It is not possible to disaggregate cohort and period effects unambiguously unless relevant information has been obtained independently. However, it is possible from data such as ours to estimate cohort differences between any two cohorts over fixed time periods by comparing the performance of successive cohorts over the age ranges for which both cohorts have been observed. The cohort effects estimated in this manner, of course, will be confounded with period effects, but if series of cohort differences are computed across the same time period, each estimate will be equally affected. In the SLS, it is possible to generate twelve cohort differences for thirteen 7-year birth cohorts with mean birth years from 1889 to 1973.1
To obtain the most stable estimates available, the average level difference between two cohorts is defined as the average of unweighted mean differences at all ages for which observations are available for the two cohorts. Thus, a cohort difference Cd is defined as:
where Mij is the unweighted mean for cohort i at age j, and where a indicates the number of common ages for which observations are available for each cohort pair. Given the seven measurement occasions of the SLS, this means that cohort differences for those cohorts entering the study at an early stage can be compared at as many as six different ages; the most recently entered cohort can be compared only at the one age at which it was measured.
It should be recognized that cohort differences reported here reflect the comparison of unrelated groups of individuals. For estimates of intrafamily cohort effects, see chapter 15.
Table 6.1 gives mean differences in T-score points computed for all cohort combinations in our study. This table should be read as follows: A positive value indicates that the performance of the cohort identified by the column exceeds, on average, the performance of the cohort identified by the row. A negative value means that the performance of the row (earlier-born cohort) exceeds that of the column (later-born cohort).
Comparative cohort gradients for the five abilities and the composite indices are graphed in figure 6.1. It is interesting to note that the composite Index of Intellectual Ability will tend to obscure cohort differences because of differential cohort trends in the subtests; for this composite index, only the five earliest-born cohorts differ significantly from any later-born cohort, although there is a recent trend for further gain. On the other hand, when the abilities are considered separately it becomes clear from these data that there are systematic advances in cohort level for Verbal Meaning, Spatial Orientation, and Inductive Reasoning. A significant advantage of the later-born cohorts is apparent throughout for Spatial Orientation and Inductive Reasoning. However, the cohort gradients for Verbal Meaning begin to decline slightly with the cohort born in 1959.
Very different findings, however, are seen for Number and Word Fluency. The former shows positive cohort differences up to about the 1910 cohort. Then there is a plateau and a shift to a successive lowering of performance level. Hence, the 1924 cohort exceeds both earlier- and later-born cohorts; the youngest cohorts are therefore currently at a disadvantage when compared with the older cohorts. For Word Fluency, there is a successive lowering of cohort level up to the 1938 cohort, but improvement for subsequent cohorts. Hence, for this ability, earlier cohorts have a slight advantage over the later-born ones, but beginning with the cohort born in 1945, there are successive positive cohort differences for this variable also.
Perhaps of considerable significance in terms of policy implications are the findings for cohort differences in the composite Index of Educational Aptitude. This index shows systematic positive cohort shifts, with a significant disadvantage for all cohorts born in 1931 or earlier. This finding would seem to be another convincing demonstration of the importance of taking generational differences into account when planning present and future adult education activities and programs.
TABLE 6.1. Mean Advantage of Later-Born Cohorts Over Earlier-Born Cohorts in T-Score Points
FIGURE 6.1. Cohort gradients for the basic ability test battery.
In this section, I report cohort gradients for the additional ability markers used in the 1984, 1991, and 1998 testing cycles. In contrast to the rather firm data provided above, caution should be exercised in that the cohort estimates for the expanded battery provided in table 6.2 and charted cumulatively in figure 6.2 were based on a maximum of only three estimates for each cohort from mean birth years 1903 to 1973.
TABLE 6.2. Cohort Differences for the Expanded Ability Battery in T-Score Points
FIGURE 6.2. Cohort gradients within ability domains from the expanded test battery. ETS, Educational Testing Service; PMA, Primary Mental Abilities.
Cohort differences for the added markers of the Inductive Reasoning factor show the same positive linear shape observed for the original marker test. Over the cohort range from 1903 to 1973, there is a gain from 0.8 to 1 SD, with the largest for the ADEPT (Adult Development and Enrichment Project) Letter Series test and the least for the Number Series test.
Two of the new markers for the Spatial Orientation factor show significantly lower cohort differences from the original marker. Perhaps because of the more concrete nature of the stimulus material, cumulative cohort differences amount to only 0.1 SD for the Object Rotation test and to 0.3 SD for the Alphanumeric Rotation test. However, the new marker introducing three-dimensional rotation, the Cube Comparison test, does show a cumulative cohort difference of approximately 1.5 SD, which is of a magnitude similar to the cohort effect shown for the original marker.
The cohort differences for the added markers of the Perceptual Speed factor show a positive and accelerating profile. The cumulative cohort difference for Identical Pictures is only half the magnitude of cohort differences for the other measures, 0.6 SD as compared with 1.8 and 1.3 SD, respectively, for Number Comparisons and Finding A’s.
The cohort differences for the added markers of Verbal Comprehension peak for the 1924 cohort, remain stable through the 1945 cohort, and show a steady decline thereafter, with modest recovery for the most recent cohort. The decline is greater for the easier vocabulary test.
The new markers for the Numeric Facility factor also attain a peak for the early cohorts, with modest decline thereafter. The declines are somewhat more pronounced than for the original marker, but there is an upward trend for the most recently born cohort on Subtraction and Multiplication.
After an initial rise from the oldest to the second-oldest cohort, there seems to be a plateau for the new vocabulary tests until the most recent baby boom cohorts, for whom a negative trend can be noted, which may be reversing in the post–baby boomers.
Finally, for the measures of Verbal Memory, we observe positive cohort trends amounting to 0.6 SD for the Immediate Recall and 0.9 SD for the Delayed Recall parts of the word list memorized in this test.
Having considered differences among alternative markers of mental abilities, cohort differences at the latent construct level are now considered. Differences between adjacent cohorts were computed for the factor scores for the six latent constructs for which cross-sectional and longitudinal age differences were reported in chapters 4 and 5. The cohort difference estimates at the latent construct level are provided in table 6.3 and graphed cumulatively in figure 6.3.
For the factor scores describing the latent constructs, substantial positive and linear cohort differences are observed for the Inductive Reasoning and Perceptual Speed factors (approximately 1 SD). A similar, although less steep, positive difference pattern occurs for Spatial Orientation (0.6 SD) and Verbal Memory (0.7 SD); a modest negative gradient (approximately 0.5 SD) is found for Numeric Facility, and there is a modest concave gradient with recent declines for Verbal Comprehension.
TABLE 6.3. Cohort Differences for the Latent Construct Factor Scores in T-Score Points
FIGURE 6.3. Cohort gradients for the six latent ability constructs.
Our initial measure of practical intelligence, the Educational Testing Service (ETS) Basic Skills test, is an expression of combinations and permutations of the basic abilities in particular practical situations. It is therefore not surprising that the cohort pattern for this ability (estimated over a single 7-year interval) is rather similar to that observed for the measures of Inductive Reasoning. Indeed, inductive reasoning is the ability that has the highest correlation with the practical intelligence measure (see Willis & Schaie, 1986a). Figure 6.4 shows substantial increments in performance level for our earlier-born cohorts up to the cohort born in 1938; thereafter, the cohort gradient for practical intelligence reaches a virtual asymptote.
Cohort differences (estimated over a single 7-year interval) are also reported for the Everyday Problems Test (EPT; Willis, 1996). Positive cohort differences show a virtually linear increase from the 1910 to the most recent 1973 cohort.
The cohort gradients for the measures of cognitive style previously reported in Schaie (1983a, 1996b, 2005) have been updated where appropriate. Table 6.4 provides the cumulative cohort difference estimates, which are graphed in figure 6.5. All of these measures show positive cohort effects. The Motor-Cognitive factors show a virtually linear increase in flexibility across cohorts amounting to approximately 1.2 SD. The Attitudinal Flexibility factor shows a smaller increment of about 0.8 SD but becomes virtually asymptotic with the cohort born in 1938. The cohort gradient for Psychomotor Speed shows a modest decline from the first- to the second-oldest cohort; after that, it parallels the cohort gradients for the other cognitive style measures, but beginning with the 1938 cohort, shows somewhat steeper positive increment. For this measure, the cumulative increment from the oldest to the youngest cohorts amounts to 1.4 SD.
FIGURE 6.4. Cohort gradient for the Basic Skills and Everyday Problems tests.
TABLE 6.4. Mean Advantage of Later-Born Cohorts Over Earlier-Born Cohorts in T-Score Points
FIGURE 6.5. Cohort gradients for the rigidity-flexibility factor scores.
Some data have been accumulated in the SLS on cohort shifts in the demographic characteristics of our sample. Of particular interest are data on educational level, age at first marriage, and age when the study participants’ first child was born (see table 6.5 and figure 6.6). For these variables, we report both cohort differences for the entire group, and separately by gender. Over the cohort range represented in our study, there has been a steady increase in years of education, amounting to a difference in education of about 5.5 years between the earliest and latest cohorts studied. The increase has been approximately 1 year greater for men than for women.
Averages declined by almost 4 years from our earliest cohort to those born in the 1930s (the lowest level was reached by men for the 1938 cohort and by women for the 1931 cohort). From that point, there has been a steady rise, which is most pronounced for women. Age of marriage of our youngest cohort had returned to the level of their grandparents and for the women had actually exceeded that of our oldest cohort.
TABLE 6.5. Cohort Differences for Years of Education, Age at First Marriage, and Age at Birth of First Child for Total Sample and Separately by Gender
There was a decrease in age at birth of first child of about 2 years from the oldest to the 1931 cohort, followed by a steady increment of approximately 4.5 years from the lowest point to our most recent (1973) cohort. A similar pattern was found for both sexes, although both the decrease from the oldest cohort and the increase to current cohorts were approximately 1 year greater for women.
Other demographic characteristics that may be important in understanding cohort differences in the cognitive variables include our measures of mobility (changes in the location of one’s home, changes of job, and changes in occupation). Average data over the 5 years preceding each reporting date were employed for these measures, which are reported across gender (table 6.6 and figure 6.7). Note that there is a very modest drop in residential and job mobility from the oldest cohort to that born in 1931; over the same cohort range, there are few cohort differences in occupational mobility. Mobility characteristics, however, increase again for the baby boomer cohorts for all three measures, with residential and job mobility changes the most pronounced.
Just as we were able to estimate cohort differences by matching across age and assuming equivalence of period effects across cohorts, so the data can be used to estimate period (time-of-measurement) effects by matching across age and assuming equivalence of cohort effects across periods. The theoretical basis of period or historical effects has been discussed in Schaie (2011) and in Schaie, Willis, & Pennak (2005). Estimates of period effects can be obtained from the same cross-sectional sequence used for the cohort differences estimates. However, the time-sequential analysis strategy must now be used. A time-of-measurement (period) difference (Td) is defined as
FIGURE 6.6. Cohort gradients for the demographic variables of education age at first marriage and age at birth of first child.
where Mjk is the unweighted mean for time k at age j, and a indicates the number of common ages for which observations are available for each pair of times of measurement. In the SLS, these effects can be estimated over seven ages for the first period, eight ages for the second period, and nine ages for all subsequent periods.
TABLE 6.6. Cohort Differences for the Mobility Measures of Changes in Homes, Jobs, and Occupational Pursuits Over the Previous 5 Years
FIGURE 6.7. Cohort gradients for the mobility variables of number of changes in job, occupation, and place of residence.
The formulas for estimating average cohort and period differences seem superficially similar. However, they involve quite different data. Cohort differences are averaged differences for a given cohort pair that occurs at different ages that must by definition span many time periods. The emphasis here is on obtaining the best estimate for the difference between a particular cohort pair. By contrast, period effects are averaged differences between samples of the same age over two adjacent time periods.
Seven period effects for the primary mental abilities are shown in table 6.7 for the total sample because no significant Period × Gender interactions were found. Significant positive period trends are observed from 1956 to 2005 for Verbal Meaning, Spatial Orientation, and Inductive Reasoning. For Number, there is a positive trend from 1956 to 1970, but a negative trend from 1970 through 2005. For Word Fluency, a significant negative period trend occurs from 1956 to 1977, reversing to a slightly positive trend from 1977 to 2005. Both the indices of Intellectual and Educational Aptitude show positive trends from 1956 through 2005.
Similar to the period effects for the ability data, significant period effects are also observed for the measures of cognitive style. These effects are shown in table 6.8. Significant positive period effects are found for Motor-Cognitive Flexibility from 1956 to 2005. However, magnitudes of effects from 1977 through 1998, although still positive, do not reach statistical significance.
For Attitudinal Flexibility, positive period effects also occur that are statistically significant for effects from 1956 to 1984 and beyond, from 1963 through 2005, from 1970 to 1991 and beyond, and from 1977 to 1991 and beyond. For the Psychomotor Speed factor, a negative period difference is found from 1956 to 1963, but statistically significant positive period effects occur from 1963 through 2005, from 1970 to 1984 and beyond, from 1977 through 2005, and from 1984 to 1991 and beyond.
Finally, I provide data on period effects for the demographic variables for which cohort data are given above. Significant period effects are found for all, except for the variable of age at birth of first child. Table 6.9 provides the estimates for educational level and age at first marriage. Statistically significant period effects (in a positive direction) are found throughout except between the 1956 and 1963, between the 1984 and 1991, and between the 1991 and 1998 data collections. Period effects for age at first marriage, however, reach statistical significance only for the difference between the 1956 and all other data collections. In all these instances, there is a negative period effect from our first to the later assessment points.
TABLE 6.7. Period (Time-of-Measurement) Effects for the Primary Mental Abilities in T-Score Points
TABLE 6.8. Period (Time-of-Measurement) Effects for the Measures of Cognitive Style in T-Score Points
Statistically significant period effects are also found for the mobility measures (table 6.10). A shift toward lower residential and occupational mobility occurs between 1956 and 1963. However, period effects in the direction of greater mobility occur for residential change between the 1970 and 1977 data collections. Greater job mobility is observed from 1970 and 1977 as well as between 1963 and 1970 with respect to the data collections from 1977 to 1991. Greater occupational mobility is also seen between the 1963 to 1977 and all later data collections, as well as between 1970 to 1984 and all later data collections. However, no mobility differences were found to be significant beginning with the 1977 cohort with respect to later cohorts.
Several alternative explanations can be offered for the observed period effects. They may simply represent testing effects, that is, inadvertent small but systematic changes in test administration and scoring procedures that, even with the best documentation, can easily slip into long-term longitudinal studies. Although unlikely for large samples, it is nevertheless possible that these differences represent systematic selection effects attributable to changes in the composition of the pool from which the successive samples were drawn.
TABLE 6.9. Period (Time-of-Measurement) Effects for the Demographic Variables
An alternate explanation would be the occurrence of a systematic cohort trend, although cohort differences should only minimally affect the period estimates obtained from SLS data because, for each period difference estimate, all but two of the cohorts used are identical. Finally, of course, these findings might represent true period effects caused by systematic positive environmental impacts such as the improvement of media, increased utilization of adult education opportunities, improved nutrition, and increased participation in preventive health care programs or, in the case of negative period effects, the neglect of drill in number skills or writing exercises in educational practice.
These matters are far from trivial because longitudinal data should be adjusted for period effects to obtain age functions that can be generalized across time. In particular, the matter of period effects becomes an important problem when age functions are constructed from data gathered in short-term longitudinal studies that apply sequential data-gathering strategies.
Fortunately, however, data from cross-sectional sequences allow consideration of certain adjustments to these short-term longitudinal age functions. As indicated in chapter 2, a net age change can be obtained by subtracting the average period effect from the observed longitudinal age difference (Ad = Lod − Td). If one assumes that there are no significant Age × Time interactions, then it is possible to adjust longitudinal change estimates and period effect estimates such as those presented in table 6.7.
TABLE 6.10. Period (Time-of-Measurement) Effects for the Mobility Variables
If Age × Time interactions are presumed to exist, then more complicated corrections are needed. In that case, one would compute age/time-specific time lags from cross-sectional data such as those in table 4.2 and use the resultant age/ time-specific estimates of period effects to adjust the longitudinal age-change estimates. The first correction would be most appropriate for use in dealing with testing effects or true period effects occurring across all age/time levels. The second correction is appropriate for dealing with age/time-specific fluctuations. This correction would be appropriate whenever it is suspected that period effects may differentially impact different age/cohort groups. Detailed numeric examples of the adjustment procedure have previously been provided by Schaie (1988d).
What are the consequences of making these adjustments for our longitudinal data? In a multivariate analysis of variance for the five basic abilities, I determined that, when holding age constant, there were significant main effects for time (period) for all abilities as well as significant Age × Time interactions for all abilities except Spatial Orientation. I therefore made the adjustments described above for the 7-year longitudinal estimates graphed for the total sample in figure 5.1. The revised estimates are presented in figure 6.8, adjusted for (a) the average period effects and (b) age-specific period effects. These graphs have fairly similar slopes. However, the adjustments have the effect of reducing increases in young adulthood, reducing decrement in early old age, and estimating steeper decrement slopes in advanced old age. They also decrease the separation between the five abilities in young adulthood and increase the separation in old age. At the oldest age studied, the adjusted estimates result in greatest decline for Verbal Meaning and least decline for Word Fluency.
This chapter reports findings on systematic cohort trends, which generally favor later-born cohorts for variables such as Verbal Meaning, Spatial Orientation, and Inductive Reasoning. But different cohort patterns do occur, including a convex pattern favoring the middle cohorts for Number, with a currently negative trend favoring the earlier-born cohorts and a concave trend for Word Fluency, which attains a low point for the middle generations, with a recent favorable reversal.
The implication of the positive cohort differences is that when older persons are compared with their younger peers, they will, on average, show lower performance even if they have experienced little or no age decrement. On the other hand, when negative cohort differences occur, such as on Number, older persons may compare favorably with younger persons even though they may actually have declined from previous performance levels. Whether older persons are at a disadvantage in occupations requiring certain basic skills will therefore depend markedly on their relative position in the cohort succession.
At the latent construct level, positive cohort gradients (favoring more recently born cohorts) are found for Inductive Reasoning, Perceptual Speed, Verbal Memory, and Spatial Orientation. Verbal and Numeric abilities had concave cohort gradients, showing lower levels for the baby-boomer cohorts. Positive cohort gradients are also observed for our measure of practical intelligence, the measures of cognitive style, level of education, and measures of mobility. These findings are important because they suggest that as future cohorts age they will be better positioned to respond to an increasingly complex environment, given their greater education and ability to respond in a more flexible manner.
Estimates of period effects are also provided. These effects show a positive time trend for Verbal Meaning, Spatial Orientation, and Inductive Reasoning. Such secular trends imply that performance levels over time have improved for persons at all adult ages.
FIGURE 6.8. Longitudinal age gradient adjusted for period and Age × Period effects.
Finally, possible applications of the period effect estimates are considered. An example is provided of how corrections for cohort and period effects can be applied to adjust longitudinal estimates to obtain increased generalizability.
