In this article we will provide an overview of the Seattle Longitudinal Study (SLS). Because this study has provided a major model for longitudinal-sequential studies of aging we emphasize the basic conceptual model, the design, and the measures utilized. But we also provide a summary of the major findings thus far obtained, including the major references to more detailed descriptions of these findings. We then turn to the topic of the relationship between personality traits and cognition and report results of analyses of new data collected during our most recent assessment wave as well as findings resulting from the integration of the new data with data previously collected.

The Seattle Longitudinal Study began as the senior author’s doctoral, dissertation at the University of Washington in 1956 {Schaie, 1958, 2000b). Results of previous work on the SLS have been widely disseminated in the psychological and gerontological literature. Comprehensive reports of the study can be found in Schaie (1983[1956–1977]; 1996 [1956–1991], 2004, [1956–1998]).

In brief, the SLS has charted the course of selected psychometric abilities from young adulthood through old age. It has investigated individual differences and differential patterns of change. In so doing it has focused not only on demonstrating the presence or absence of age-related changes and differences but has attended also to the magnitude and relative importance of the observed phenomena. More recent phases of the study have identified contextual, health, and personality variables that offer explanations for differential change and that provide a scientific basis for possible interventions. We have also studied cognitive similarity within parent-offspring sibling pairs, and have recently begun to acquire data on a third generation (participants who have both a parent and a grand-parent participating in the SLS). Cognitive interventions have been designed that have been successful in remediating statistically reliable declines and in improving the cognitive functions of older persons who have remained stable. Age changes and age differences in cognitive ability structure have also been studied at the latent construct level, the relative effect of speed and accuracy in age decline and training gain has been examined, and the relevance of cognitive training to real life tasks has been investigated. Work in progress seeks to relate our behavioral data to neuropsychological diagnostic procedures, to study behavior-stractural-anatoraical relation-via data collected at autopsy and to investigate the cognitive change correlates of the Apolipoprotein-E marker gene and of serum cholesterol level. A schematic conceptual model for the study is shown in Figure 1.

Fig. 1 Conceptual model for the Seattle Longitudinal Study (from Schaie, 200la).

Fig, 2 Basic Design of the Seattle Longitudinal Study (SLS).

All of our study participants are or were originally members of an HMO, the Group Health Cooperative of Puget Sound, in the Seattle, Washington metropolitan area, or are family members of these individuals. The initial sampling frame in 1956 consisted of approximately 18,000 potential adult participants. These were stratified by age and sex, with 25 men and 25 women randomly selected for each year of birth from 1889 to 1939. Testing proceeded in small groups from ten to thirty persons until a total of 500 persons (25 men and 25 women in each 5-year age interval from 21 to 70 years) had been, tested (cf. Schaie, 1958).

In the 1963 cycle, 301 members of the original sample were retested. In addition, approximately 3000 new names were drawn randomly from the original, sampling frame, deleting those individuals who had been tested in 1956. In the second wave of the study 996 persons aged 22–77 years were (Schaie & Strother, 1968). A similar procedure was followed in 1970; retesting as many survivors as possible from the first two cycles, and initially examining a new randomly selected panel of 705 persons, aged 22–84 years (Schaie & Labouvie-Vief, 1974). Since the original sampling frame had been substantially depleted, a collateral study determined that it would be feasible to shift to a sampling with replacement (1974; Schaie, 1996). Thus, in the 1977 cycle, in addition to retesting survivors of the first three waves, we sampled 3000 persons from what had by then become a 210,000 member organization. During the fourth wave 612 new participants (aged 22–84) were tested (Schaie, 1983). During the 1984 cycle retests were conducted for 839 surviving participants of the first four waves as well as a subset of 160 participants of the 1974–75 collateral studies. Again a random sample was drawn from the now 300,000 member health plan and 628 new participants were tested (Schaie, 1988, 1996). In the 1991 cycle 1117 surviving participants were reassessed and 693 new participants were tested from a new random sampling of the plan that now exceeds 400,000 members (Schaie, 1996). Finally, in the 1998 cycle we reassessed 1051 participants and added a new random sample of 724 participants.

The response rates of potential participants have ranged from 16% for the 1956 data collection to 11% for the most recent data collection, with best response for the middle aged and worst response for young adult males. However, the demographic characteristics of our sample do not differ significantly from their parent population.

Although the successive random draws from our sampling frame have been quite representative of the population described above, there has clearly been non-random attrition. We have studied the effects of attrition systematically, have found, as is true in other studies (e.g., Palmore, Busse, Maddox. Nowlin, & Siegler, 1985; Rott, 1993; Sliwinski & Buschke, 1999), that those who return for retest outperform those who do not return. However, we have also noted that these effects do not seem to be systematically related to the age of the participants, although reasons for drop-out may change across the age span. Furthermore, dropout effects are of greatest magnitude subsequent to the first retest occasion, but lessen progressively the longer participants remain in the study. Less than 5% of our participants were lost through refusal to return. We have reported attrition effects for each of our study cycles, and have proposed a number of corrections that correct for the effects of attrition and other confounds from our estimates of cognitive age changes (cf. Schaie, 1996).

The main SLS database currently consists of the 9,476 complete records on 4,857 participants, of whom. 36 were tested seven times, 122 were six times, 223 were tested five times, 281 four times, 527 were tested three times, 1,004 were tested twice, and 2,664 participants were tested only once. Cumulatively this results in a total of 2,193 participants followed over 7 years, 1,189 over 14 years, 662 over 21 years, 381 over 28 years, 158 over 35 years, and 36 over 42 years. Five hundred and eighty participants of the training study received an additional test administration after a one-month interval (also see Schaie, 1996, 2004).

The family study archive contains 776 adult offspring of longitudinal participants (465 daughters and 311 sons). It also contains 400 siblings of the longitudinal participants (248 sisters and 152 brothers). These participants were first tested with the basic test battery in 1989–90. Of these participants, 886 persons were successfully retested in 1996–97 (385 daughters, 229 sons; 177 sisters, 95 brothers). In addition, 672 relatives not previously tested were added to the archives during 1996–97 (239 daughters, 134 sons; 162 sisters, 107 brothers).

During all seven cycles of the SLS, our principal dependent variables were the measures of Verbal Meaning, Space, Reasoning, Number and Word Fluency, identified by Thurstone as accounting for the major proportion of variance in the abilities domain in children and adolescents (Thurstone, 1938) contained in the 1948 version of the Thurstone’s SRA Primary Mental Abilities Test (Form AM 11–17; Schaie, 1985; Thurstone & Thurstone, 1949). The second set of variables that has been collected consistently includes the rigidity-flexibility measures from, the Test of Behavioral Rigidity (Schaie & Parham, 1975), which also include a modified version of the Gough social responsibility scale (Gough, McCloskey, & Meehl, 1952). Limited demographic were collected during the first three cycles. The above measures are referred to as the “Basic Test Battery,” and have been supplemented since 1974 with a more complete personal data inventory, the Life Complexity Inventory (LCI; Gribbin et al., 1980), which includes topics such as major work circumstances (with home-making defined as a job), friends and social interactions, daily activities, travel experiences, physical environment and life-long educational pursuits. The battery was expanded in 1991 by adding the Moos Family Environment and Work Scales (Moos, 1981, 1986; Schaie & Willis, 1995), and a family contact scale. A Health Behavior Questionnaire (Maier, 1995; Maitland, 1997) was in 1993.

In the 1975 collateral study (Schaie, 1996) a number of measures from the ETS kit of factor-referenced tests (Ekstrom, French, Harman, & Derman, 1976) as well as the 1962 revision of the PMA (Thurstone, 1962) were added. Of these the Identical Picture, Finding A’s and Hidden Pattern tests (Ekstrom et al., 1976) were included in the fourth (1977) SLS cycle.

To be able to explore age changes and differences in factor structure, we included multiple markers for most abilities during the fifth (1984) cycle. We also added measures of Verbal Memory (Zelinski, Gilewski, & Schaie, 1993). This now permits us to measure the primary abilities of Verbal Comprehension, Spatial. Orientation, Inductive Reasoning, Numerical Facility, Perceptual. Speed and Verbal Memory at the latent construct level (cf. Schaie, Dutta, & Willis, 1991; Schaie, Maitland, Willis, & Intrieri, 1998). The expanded cognitive battery is described referenced in Table 1. Also added were a criterion measure of “real life tasks,” the ETS Basic Skills test (Educational Testing Service, 1977), and a scale for measuring participants’ subjective assessment of ability changes between test cycles (Schaie, Willis, & O’Hanlon, 1994). Beginning in 1997 we substituted the Everyday Problems Test (EPT, Diehl, Willis, & Schaie, 1995; Willis, 1992, for the Basic Skills test, since the more recent test was specifically constructed for work with adults and has been related to measures of the Instrumental Activities of Daily Living (IADL; Lawton & Brody, 1969).

Table 1 Psychometric Intelligence Measurement Battery.

We have abstracted the health histories of most of our participants who were retested in 1991 and/or 1998, and who have been tested two or more times, over the entire period they have been in the study using the International Classification of Diseases (U.S. Public Health Service, 1968), coding each outpatient visit or hospital day by diagnosis and by constructing annual illness counts by illness incidents (single visits) and illness episodes (continuous series of visits for a specified diagnosis). Records of drugs used concurrently with the psychological testing were also obtained and coded (American Society of Hospital Pharmacists, 1985). These data have been used to study the relation of health, cognition and mortality (e.g., Bosworth & Schaie, 1997; Bosworth, Schaie, & Willis, 1999).

Throughout the history of the SLS, an effort now extending over 48 years, we have focused on five major questions with regard to the nature and antecedents of cognitive decline across adulthood, which we have attempted to ask with greater clarity and increasingly sophisticated methodologies at each successive stage of the study:

• Does intelligence change uniformly through adulthood or are there different life course ability patterns?
• At what age are there reliably detectable age decrements in ability and what is the magnitude of these decrements?
• What are the patterns of generational differences and what is the magnitude of these differences?
• What accounts for individual differences in age-related change in cognitive abilities in adulthood?
• Can intellectual decline with increasing age be reversed by educational interventions?

More recently we have considered three other questions. The first concerns family similarity in level and rate of change of cognitive functioning. The second begins to bridge behavioral science biology by inquiring into structural anatomical and physiological changes that correlate with behavior. The third topic concerns the bridge between the study of normal aging and the precursors of dementia. This summary will review what we have learned from the SLS up to now to answer the five basic questions as well as provide some information on the family studies.

Fig. 3 Longitudinal estimates of within participant age changes on the latent ability constructs (from 7-year longitudinal data).

While cohort-related differences in the rate magnitude of age changes in intelligence remained fairly linear for cohorts entering old age during the first three cycles in our study, they have since shown substantial shifts. For example, of decremental age change have abated, while at the same time negative cohort trends are appearing as we begin to study members of the baby-boom generation. Patterns of socialization unique to a given sex role within a specific historical period may be a major determinant for the pattern of change in abilities. More finegrained analyses suggest that there may be substantial gender differences for those who remain stable as well as differential changes for those who decline, when age changes are decomposed into those due to accuracy or speed (cf. Willis & Schaie, 1988). We have also demonstrated substantial relationships between the psychometric abilities and real life tasks (Willis & Schaie, 1986).

With multiple markers of abilities first available for the fifth cycle, we have conducted cross-sectional and longitudinal analyses of ability structure invariance over a wide age range (Maitland, Intrieri, Schaie, & Willis, 2000; Schaie et al., 1998). Two types of invariance may be distinguished. The first, configured invariance, implies that the same measurement variables load on the same factors across time or different groups. This is a minimal requirement that assures that the constructs of interest remain stable over time. Metric invariance implies the additional requirement that the factor loadings retain the same magnitude across time or groups. In this case we can be assured that the measurement metric remains constant. Our results suggest that it is possible to demonstrate configural but not metric factor invariance across a wide age/cohort range, but that metric invariance within cohorts over seven years prevails at all but the oldest ages.

The data from the SLS have attained increasing importance in providing a normative base to determine at what ages declines reach practically significant levels of importance for public policy issues related to mandatory retirement, age discrimination in employment or for cases of population proportions that can live independently in the community. From the SLS data we were able to show both level of performance and rate of decline show significant age by cohort interactions (Schaie, 1983, 1996, 2000a).

Fig. 4 Cross-sectional differences on five ability measures for cohorts born from 1889 to 1973.

A revision of the Moos Family Environment and Work Environment scale was constructed. Separate forms were constructed to survey family environment in the family of origin and the current family. Psychometric analyses of these forms have demonstrated retention of construct validity for our revised forms, equivalence of structure across the family of origin and current family versions, as well as comparability across different age levels throughout adulthood. Substantial changes in perception of family environments over time was found to occur in both parent and adult offspring, with greater similarity found between the parent’s current family and offspring’s family of origin (same family) than between current families or families of origin for either generation (Schaie & Willis, 1995).

The effects of structural and functional social support as well as age and previous health status on health outcomes was examined (Bosworth & Schaie, 1997). The latent outcome variable is marked by number of physician visits, hospital stays, and number of illness diagnoses over a 7 year period. Structural social support is marked by demographic variables and functional social support is measured by the Moos family environment scales (Moos, 1986). Results of this study suggest that social support variables account for only very small amounts of individual differences variance (primarily in those individuals with greatest disease incidence), while previous health status accounts for substantial, variance.

Changes in Rate of Aging We now have available 7-year longitudinal change data for parents and adult offspring assessed at approximately the same ages. These data have allowed us to supplement the findings on generational differences in level of performance to address the question of whether rates of aging have changes. The findings suggest that during young old age, the slopes of decline for several of the primary mental abilities have significantly flattened. This is the case for Verbal Meaning, Inductive Reasoning, Spatial Orientation, and Psychomotor Speed. (Schaie, 2004).

Early Detection of Risk for Dementia Data on the CERAD neuro-psychology measures have been accumulated for over 500 study participants over the age of 60. From these data individuals were rated by neuro-psychologists as normal, to be monitored, probably demented, or demented. Using the methods of extension analysis (describe above) neuro-psychology measures were estimated from psychometric tests collected seven and fourteen years previously. We found that a significant proportion of study participants who were eventually diagnosed as demented could have been predicted from data collected 7 and 14 years earlier, (cf. Schaie, Caskie, et al., in press).

In summary, the Seattle Longitudinal Study has provided a model for longitudinal-sequential studies of cognition over the adult life course. It introduced the concept of cohort into cognitive aging research, and it has pioneered family studies of cognitive aging. It has substantively contributed to our understanding of the varied indigenous and exogenous influences on cognitive aging, including health, life styles, and personality characteristics. Most recently it has expanded to studies of early identification of risk for dementia and policy relevant studies of generational differences in rates of cognitive aging.

Throughout the SLS we have collected limited personality data that have been derived from, the questionnaire part of the Test of Behavioral Rigidity (TBR; Schaie & Parham, 1975). Most recently we have also collected data on the NEO Personality Inventory (Costa & McCrae, 1992). In this section we report new analyses that speak to the relationship between personality traits and performance on tests of cognitive abilities.

Personality Factor Scores A factor analysis was conducted on the 75 questionnaire items in the Test of Behavioral Rigidity (TBR; Schaie & Parham, 1975) using 4326 test records accumulated over the 1963–1984 study cycles. Initial analyses considering the number of factors unambiguously represented in the data resulted in an acceptable 13-factor model with good fit (χ²[df = 1,191] = 3,548.16, p<.00l; GFI = .945, RMSR = .007). The 13-factor model was then tested by means of confirmatory factor analyses on the participants assessed in 1977 and 1984, and it continued to show an acceptable fit (χ²[df = 1,191] = 4,302.98, p<.001; GFI = .941, RMSR = .007). A two-group analysis further investigated factorial invariance across time by constraining factor loadings and factor variance–covariance matrices to be equal across the two data sets. This analysis also yielded an acceptable fit (χ²[df = 2,512] = 6,910.00, p<.00l; GFI = .945, RMSR = .007). (Maitland, Dutta, Schaie, & Willis, 1992).

The 13-factor model includes 8 factors that can be mapped upon the Cattell (1957; Cattell, Eber, & Tatsuoka, 1970) taxonomy of personality dimensions: Affectothymia, Superego Strength, Threctia, Premsia, Untroubled Adequacy, Conservatism of Temperament, Group Dependency, and Low Self-Sentiment. The remaining five factors are best described as attitudinal traits and were labeled Honesty, Interest in Science, Inflexibility, Political Concern, and Community Involvement.

The factors that were mapped upon one end of the trait continuum described by Cattell, have been described as follows (Cattell, Eber, & Tatsuoka, 1970):

Affectothymia – Outgoing, warmhearted, easygoing, participating tendencies.

Superego Strength – Conscientious, persistent, moralistic, staid.

Threctia — Shy, timid, restrained, threat-sensitive.

Premsia – Tender-minded, sensitive, clinging over-protected.

Untroubled Adequacy – Self-assured, placid, secure, complacent, serene.

Conservatism – Respecting traditional ideas, tolerant of traditional difficulties.

Group Dependency – A “joiner” and sound follower, group adherence.

Low Self-Sentiment – Uncontrolled, lax, follows own urges, careless of social rules.

The additional five attitudinal traits may be described as follows:

Honesty – Endorsement of items that reflect personal beliefs of honesty

Interest in science – Endorsement of an item couplet that reflects interest in science

Inflexibility — Endorsement of items that reflect lack of tolerance for disruption of routines

Political Concern – Reflects attitudes toward other countries

Community Involvement — Endorsement of positive attitudes about citizenship and civic responsibilities.

The NEO Personality Inventory The scales in this inventory (Costa & McCrae, 1992) are described as follows:

Neuroticism (N) This scale contrasts adjustment or emotional stability with maladjustment or neuroticism.

Extraversion (E) Extraverts are sociable but also assertive, active and talkative. Introverts are reserved, independent and they prefer to be alone.

Openness (O) Open individuals are curious, willing to entertain novel ideas and unconventional, values. They experience positive and negative emotions more intensely than do closed individuals.

Agreeableness (A) Agreeable persons are altruistic, sympathetic to other and eager to help, expecting others to be equally helpful in return. Disagreeable persons are egocentric, sceptical of others’ intentions and competitive rather than cooperative.

Conscientiousness (C) High scorers are scrupulous, punctual and reliable. Low scores do not necessarily lack moral principles, but are less exacting in a-plying them, more hedonistic, and more lackadaisical in working towards their goals.

Extension Analysis In order to permit postdiction of past standing on the NEO we conducted an extension analysis that was designed to project the NEO scores into the TBR 13 factor space. An important application of confirmatory factor analysis is to use this procedure to implement the Dwyer (1937) extension method. As Tucker (1971) has demonstrated, it is not necessarily optimal to use factor scores on a latent variable to estimate their regression on an observed variable. However, confirmatory factor analysis permits the estimation of the location of some new observed variable or variables of interest within a previously known factor (latent construct) space. This is a situation that frequently arises in aging studies as samples are followed over long lime periods.

In the extension analysis, the 75 × 75 TBR personality item correlation matrix for the 1998 sample was augmented by the 5 NEO scores converted into z-score metric. The factor loadings for the 13 TBR factors were constrained to the values obtained from the original confirmatory factor analysis solution for the 13 personality factors for this sample. Factor loadings for the NEO scales were then freely estimated providing information on the projection of these measures into the previously established 13 personality factor space. This procedure produced an acceptable fit for the extended model (χ²[df = 1,453] =5924.22, p<.001; GFI = 920, RMSR = .04).

Projections of the NEO into the 13 factor space have a + sign if a NEO scale is positively correlated with the personality factor or a − sign if it is negatively correlated.

Significant projections of the NEO (p < .01) into the 13 factor space were found for Neuroticism with Affectothymia (+), Superego Strength (−), Untroubled Adequacy (−), Conservatism (+), Inflexibility (+), and Community Interest (−). Extraversion projected significantly to Superego Strength (+), Premsia (+), Untroubled Adequacy (+), Conservatism (−), Low Self Esteem (−), and Community Interest (+). Openness projected to Premsia (+), Low Self Esteem (−) and Inflexibility (−). Agreeableness projected to Superego Strength (+) Threctia (+), Premsia (+), Untroubled Adequacy (+), Low Self Esteem (−), Honesty (+) and Community Interest (+). Finally, Conscientiousness projected to Affectothymia (−), Superego Strength (+), Untroubled Adequacy (+), Conservatism (−). Low Self Esteem (−), Inflexibility (−), and community Interest (+). Table 2 gives the concurrent correlations between the NEO scales and the 13 personality factors. Table 3 provides standardized factor loadings for the projection of the NEO into the 1.3 personality factor space.

Table 2 Concurrent Correlations of Personality Factor Scores and NEO Scales (N= 1417).

Table 3 Standardized Factor Loadings for the NEO on the 13 Personality Factors.

TBR Personality Factors

Stability of personality factor scores As an initial step we conducted an analysis of the stability of the thirteen personality factors over time. Included in this analysis were the 1055 participants who had retest data over at least seven years. All stability coefficients were statistically significant (p < .01). The seven-year stabilities ranged from 0.32 for Affectothymia to 0.71 for Group Dependency. Fourteen-year stability ranged from 0.33 for Affectothymia to 0.69 for Interest in Science. Twenty-one year stabilities ranged from 0.20 for Affectothymia to 0.67 for Group Dependency. Twenty-eight year stabilities ranged from 0.24 for Political Concern to 0.65 for Interest in Science. And the 35 year stabilities ranged from 0.29 for Honesty to 0.66 for Group Dependency. Average stability coefficients were 0.59 over 7 years; 0.54 over 14 years; 0.49 over 21 years; 0.46 over 28 years; and 0.45 over 35 years (see Table 4).

Table 4 Stability of the Personality Factor Scores.

We also examined the 7-year stabilities separately for four age groups: Young adult (age 29–49; N= 182): middle-aged (age 50–63; N = 276); young–old (age 64–77; N = 379); and old–old (age 78+; N = 182). All ages are given for the second measurement occasion. Average stability coefficients over seven years were 0.54 for the young adults, 0.58 for the middle-aged, 0.60 for the young–old, and 0.57 for the old–old. These stabilities range from somewhat lower to comparable values frequently seen in the personality literature (cf. Roberts & DelVecchio, 2000).

Concurrent relation between personality and ability factors The concurrent correlations are provided in Table 5. Correlations range from small to modest. Consistently highest relationships for all abilities occurred with Conservatism (−), Untroubled Adequacy (+), and Group Dependency (−). Additional correlations significant at or beyond the .001 level of confidence were found for Inductive Reasoning with Affectothymia (+), Threctia (−), Premsia (+), Low Self Esteem (−), Honesty (+), Interest in Science (+), Inflexibility (−), Political Concern (+), and Community Interest (−); for Spatial Orientation with Affectothymia (+), Threctia (−), Premsia (+), Low Self Esteem (−), Interest in Science (+), Inflexibility (−), and Community Interest (−); for Perceptual Speed with Affectothymia (+), Premsia (+), Low Self Esteem (−), Interest in Science (+), Inflexibility (−), Political Concern (+), and Community Interest (−); for Verbal Comprehension with Affectothymia (+), Superego Strength (+), Premsia (+), Low Self Esteem (−), Interest in Science (+), Inflexibility (−), Political Concern (+), and Community Interest (+); and for Verbal Memory with Affectothymia (+), Premsia (−), Inflexibility (−), Political Concern (+), and Community Interest (−).

Table 5 Concurrent Correlations of Personality Factor Scores and Cognitive Abilities.

We also computed OLS regressions of the ability factor scores on the personality factor scores (see Table 6). Multiple Rs range from 0.27 for Numeric Facility to 0.49 for Verbal Comprehension. Proportions of variance accounted for by personality in the ability factors are approximately 20% for Inductive Reasoning, 11% for Spatial. Orientation, 18% for Perceptual Speed, 7% for Numeric Facility, 24% for Verbal Comprehension, and 14% for Verbal Memory.

Table 6 Concurrent OLS Regression of Personality Factor Scores on Cognitive Abilities.

The NEO Concurrent correlations were computed also between the five scales of the NEO personality inventory and our six ability factors. These correlations are provided in Table 7. Again, correlations range from small to modest. Consistently highest relationships for all abilities (except Numeric Facility) occurred with Openness. Additional correlations significant at or beyond the .001 level of confidence were found for Inductive Reasoning with Extraversion (+) and Agreeableness (−); for Spatial Orientation with Agreeableness (−); for Perceptual Speed with Extraversion (+); for Numeric Facility with Extraversion (+) and Conscientiousness (+); and for Verbal Memory with Extraversion (+).

Table 7 Concurrent Correlations of NEO Scores and Cognitive Abilities.

OLS regressions of the ability factor scores on the NEO scales are shown, in Table 8. Multiple Rs range from 0.12 for Numeric Facility to 0.40 for Verbal Comprehension. Proportions of variance accounted for by the NEO personality factors in the ability factors are. approximately 10% for Inductive Reasoning, 6% for Spatial Orientation, 12% for Perceptual Speed, 2% for Numeric Facility, 16% for Verbal Comprehension, and 12% for Verbal Memory.

Table 8 OLS Regressions of Cognitive Abilities on NEO Scores.

TBR Personality Factors OLS regressions of the ability factor scores obtained in 1998 on each of the personality factor scores were competed using personality factor scores obtained in 1963, 1970, 1977, 1984, and 1991. The pattern of statistically significant predictors remained fairly constant across increasing time intervals, although the p levels declined with shrinking sample sizes. Group dependency (−) was the strongest personality predictors for most abilities, followed by Conservatism (−). Untroubled adequacy (+), Premsia (+), and Low Self Esteem (−). Table 9 reports regression coefficients and proportions of variance accounted for in the 1998 cognitive ability factors by earlier standing on the 13 personality factors. The proportions did not vary markedly across the increasing length of the prediction interval. Average proportion of cognitive ability variance predicted was 15.8% over 7 years, 13.6% over 14 years, 13.0% over 21 years, 15.5% over 28 years, and 14.8% over 35 years. The predictability was highest for Verbal Comprehension (20–37%) and lowest for Spatial Orientation (7–13%) and Numeric Facility (6–15%).

Table 9 Predictive OLS Regression of Personality Factor Scores on Cognitive Abilities.

Postdicting the NEO Our final analyses are concerned with obtaining postdicted NEO scores by bootstrapping via the 13 personality factor extension analysis, thereby obtaining an estimated longitudinal NEO data set for study participants in 1991 and 1998. Estimation procedures involved computing NEO scores by multiplying the 13 factor scores by weights obtained from the orthonormalized factor loadings in Table 3 and restandardizing resulting scores to a mean of 50 and SD of 10. Within participant changes over 7 years were then computed and aggregated across successive 7-year age cohorts. The resulting longitudinal age gradients were then centered on average scores at age 53 and are depicted in Figure 5. Considerable caution is in order in interpreting the findings using these NEO proxy estimates. However, they do represent within subject change data across much of the adult life span. While cross-sectional data usually depict few personality differences adulthood, these data suggest much more dramatic developmental trends. For neuroticism we see a sharp increase until midlife with virtual stability thereafter. Openness shows a modest until age 46, a plateau until the late sixties and decline thereafter. Extraversion shows steady decline from the forties, Agreeableness shows steep increment with age, finally, Conscientiousness declines until the fifties followed by a virtual plateau.

Fig. 5 Longitudinal estimates of within participant age changes on the NEO scales (from predicted 7-year longitudinal data).

There have been many suggestions that the study of cognition and aging might be advanced by introducing personality constructs as possible covariates that might explain some proportion of age-related changes and differences in cognitive performance. In the earlier literature (cf. Mischell, 1973) it was often argued that most measures of personality traits were not sufficiently stable nor showed, high enough correlations with cognitive measures to make it likely they could account for substantial proportions of age-related variance. The former criticism has largely been addressed by more recent rigorous measurement development (e.g., Costa & MacCrae, 1992), but the latter concern requires further empirical investigations. In this article, we try to lay the necessary groundwork for studying these questions by examining the relationship between two personality trait measures and measures of cognitive performance in a large sample covering the adult age range.

Interestingly enough we can show that there are modest but significant concurrent relationships between personality trait measures and ability construct that account for up to 20% of shared variance. Both our 13 personality factor measures and the NEO could be related to the cognitive ability constructs, albeit the 13-factor measure accounted for more of the shared variance than did the NEO. The personality dimensions that were found to be most substantively related to high performance on cognitive ability factors were high Untroubled Adequacy, low Conservatism and low Group Dependency from the 13 PF measure, and high scores of Openness on the NEO.

We were also able to show that there is moderate stability across time for the personality measures that is fairly comparable with the stability found in much of the personality literature (cf. Roberts & DelVecchio, 2000). It might be argued therefore, that prediction of cognitive change over age would benefit from the inclusion of personality traits as predictors of distal levels of cognitive performance. This argument is bolstered by the fact that some of the personality-cognition relations could be established over as long as a 35-year interval.

Given suitable longitudinal data, we also show that it is possible to utilize methods of extension analysis to bootstrap older to newer measurement domains, and thus try to reconstruct what changes on the newer measures would have been like had they be available at the earlier measurement points. Findings from the analyses using the estimated NEO data must be taken with caution since the multiple correlations of the personality factor scores with the NEO scales range from 0.39 to 0.59 (highest for the estimated scores for Openness and Conscientiousness, lowest for Agreeableness). Nevertheless, it is noteworthy that the estimated longitudinal data suggest greater developmental changes in personality over the adult life course than has previously been suspected as well as being generally consistent in changes from young adulthood documented in other studies (cf. Costa & McCrae 1993, Soldz & Vaillant, 1999).

Perhaps, most importantly, we also demonstrated even though the Seattle Longitudinal Study did not originally focus on the assessment of personality traits, it was possible to utilize suitable estimation procedures that permit longitudinal, analyses bearing upon the contributions of personality constructs in understanding adult cognition.

The authors wish to express their gratitude to the members and staff of the Group Health Cooperative of Puget Sound without whose long-term support this study would have been impossible. The study has been supported by grants from the National Institute of Child Health and Human Development (HD00367, 1963–65; and HD04476, 1970–73) and by the National Institute on Aging (AG00480, 1973–79; AG03544, 1982–86; AG04770, 1984–88 and AG08055, 1989–2005).

The following colleagues, students and support staff (in alphabetical order) participated in one or more of the various data collections and analyses and/or contributed to the resultant scholarly products: Christopher Adams, David Adams, Diane Backschies, Margret Baltes, Paul Baltes, Thomas Barrett, Ute Bayeo, Timothy Benner, Gisela Bertulis, Julie Blaskevicz, Joy Bodnar, Hayden Bosworth, Barbara Buech, Michael Cady, Heather Chipaer, Soyeon Cho, Theresa Cooney, Jean Day, Cindy DeFrias, Robin Dunlap, Ranjana Dutta, Walter Eden, Charles Pick, Carrie Freeh, Michael Gilewski, Judith Gonda, Kathy Gribbin, Ann Gruber-Baldini, Cheryl Guyer, Brian Hallett, Elaine Hardin, Gene Hardin, Sarah Haessler, Charlene Herold, Christopher Hertzog, Judy Higgins, Robert Intrieri, Gina Jay, Christine Johnson, Heather Johnson, John Just, Alfred Kaszniak, Iseli Krauss, Eric Labouvie, Gisela Labouvie-Vief, Tamra Lair, Karen Lala, Karen Laughlin, Christine Lehl, Helen Leisowitz, Jackie Levine, Holly Mack, Heiner Mater, Scott Maitland, Hiroko Makiyama, Renee Marquardt, Dean Melang, Sherry Murr, Ann Nardi, John Nesselroade, Ha Nguyen, Shirley Paton Norleen, Ann O'Hanlon, Phyllis Olson, Holly Overman, Sara Paneck, Ms Parham, Julie Parmentier, Cherill Petera, Robert Peterson, Robert Plomin, Samuel Popkin, Alan Posthomer, Margaret Quayhagen, Andrew Revell, Sarah Rosen, Amy Roth, Christine Roy, Pat Sand, Coloma Harrison Schaie, Carolyn Seszniak, John Schulenberg, Anna Shuey, Michaei Singer, Anita Stolov, Yield Stone, Charles Strother, Alejandra Suarez, Linda Ten, Richard Vigesaa, Nathaniel Wagner, Faika Zanjani, and Elizabeth Zelinski.