A Longitudinal Twin Study of General Cognitive Ability over Four Decades

Acknowledgement: William S. Kremen and Hong Xian contributed equally to this article.
The content of this article is the responsibility of the authors and does not represent official views of NIA/NIH, or the Veterans’ Administration. Numerous organizations provided invaluable assistance in the conduct of the VET Registry, including the following: U.S. Department of Veterans Affairs, Department of Defense; National Personnel Records Center, National Archives and Records Administration; Internal Revenue Service; National Opinion Research Center; National Research Council, National Academy of Sciences; the Institute for Survey Research, Temple University. The authors gratefully acknowledge the continued cooperation of the twins and the efforts of many staff members. The study was supported by awards from the National Institutes of Health/National Institute on Aging [R01s AG018386, AG022381, AG022982, AG050595 to W.S.K.; R01 AG018384 to M.J.L.; R03 AG 046413 to C.E.F, and K08 AG047903 to M.S.P].

Studies of general cognitive ability (GCA) have emphasized its substantial stability through midlife (Gold et al., 1995; Larsen, Hartmann, & Nyborg, 2008; Lyons et al., 2009; Schwartzman, Gold, Andres, Arbuckle, & Chaikelson, 1987; Tucker-Drob & Briley, 2014). However, fewer studies have examined the genetic and environmental influences on individual differences in stability and change over time. Using biometrical latent growth curve modeling (LGCM; McArdle, Prescott, Hamagami, & Horn, 1998), some twin studies have examined cognitive change in older adults over 6- to 13-year intervals. Two studies found significant genetic influences on initial ability level (intercept), but not on change (slope; McGue & Christensen, 2002; Reynolds, Finkel, Gatz, & Pedersen, 2002). One study found significant heritability for linear change (McArdle et al., 1998) and one study found significant heritability for quadratic, but not linear, change (Reynolds et al., 2005). Thus, results from prior work are inconsistent. It is worth noting, however, that when some of the cited studies have fitted a linear-linear spline model, for example, with a knot-point at age 65, the second linear slope, representing change during later life, differential change patterns and etiological contributions have emerged (e.g., Finkel et al., 2003; Reynolds et al., 2005). Indeed, multivariate models of slopes across cognitive domains suggest greater reliability of change after age 65 years than prior to age 65, and indeed that a general factor of change across abilities was moderately heritable (Tucker-Drob et al., 2014). As noted by Tucker-Drob and Briley (2014), although there are considerable developmental data on cognition during childhood and old age, there is very sparse coverage during the middle portion of the life span.

We are aware of two longitudinal nontwin studies that used mixture modeling to identify different cognitive trajectories of GCA in old age (Hayden et al., 2011; Yu et al., 2015). In follow-ups averaging 7 and 8 years, both found a large group with little or no decline and smaller groups with moderate and severe declines. A study of midlife adults at 39, 46, and 53 years of age also showed heterogeneity of cognitive trajectories after defining three groups based on an a priori definition of decline and gain (Willis & Schaie, 2005). Like the other studies, this study found that the majority of the sample showed no change in cognitive ability. However, both decreases and increases in cognitive ability were observed. Although twin studies using LGCM address heterogeneity of stability and change in cognitive ability by considering genetic and environmental influences on individual differences in intercept and slope, to our knowledge no twin study has examined heterogeneity through identification of trajectories of GCA over time.

The goals of the present study are to examine stability and change in GCA across four decades from young adulthood to late middle age by addressing the following questions: (a) Is there change in mean GCA and genetic and environmental influences on GCA over time? (b) To what extent are individual differences in change in GCA influenced by genetic and environmental factors? and (c) Do reliably different trajectory subgroups suggest heterogeneity in patterns of change in GCA over time?

Participants were drawn from the nationally representative Vietnam Era Twin Registry (VETR; Eisen, True, Goldberg, Henderson, & Robinette, 1987). The Registry contains male twins who served in the United States military at some time between 1965 and 1975. To be eligible to participate in VETSA 1, twins had to be between 51 and 59 years old when recruited and both members of a twin pair had to agree to participate. VETSA participants are a community-dwelling sample, not a patient or VA clinical sample. Most (78%) were not exposed to combat. Health (e.g., diabetes, hypertension) and demographic characteristics (e.g., education, median income, employment) of the VETSA sample were comparable to men in the general population of the United States in their age group (Schoenborn & Heyman, 2009). Mean ages of participants at each time point were 19.7 years (SD = 1.2) at baseline, 55.9 years (SD = 2.4) at VETSA 1, and 61.55 years (SD = 2.4) at VETSA 2. We refer to these study waves by their mean ages: 20, 56, and 62 years.

Age 20 GCA scores were obtained from military records by the VETR for nearly all (N = 1273; 98.8%) participants. Age 56 GCA scores were collected from all 1,288 men in VETSA 1 and age 62 scores were collected from 1,008 men in VETSA 2. There were 1,291 unique individuals with an AFQT score from at least one age. Participants in VETSA 1 and VETSA 2 were assessed at one of the two testing sites (Boston University [BU] and the University of California, San Diego [UCSD]) or, in rare circumstances, elected to have a research assistant travel to them. Written informed consent was obtained from all participants. The studies were approved by the BU and UCSD Institutional Review Boards. Zygosity was determined by analysis of 25 microsatellite markers. The sample included 349 complete monozygotic (MZ) pairs and 265 complete dizygotic (DZ) pairs. At age 56 years 78.7% were married, 5.5% single, 1.5% widowed, not remarried, 1.1% separated, and 13.2% were divorced, not remarried; 86.8% were non-Hispanic white. The mean education level at age 56 was 13.9 years (SD = 2.1; range 8 – 20 years).

The AFQT is a 50-min paper-and-pencil test consisting of 100 multiple-choice items that was originally administered just before military induction (Bayroff & Anderson, 1963). Items include the four domains of vocabulary, arithmetic word problems, knowledge and reasoning about tools and mechanical relations, and visual-spatial processing (Uhlaner & Bolanovich, 1952). The AFQT is highly correlated (∼.85) with traditional measures of IQ or GCA in VETSA and in other samples (Lyons et al., 2009) and served as our measure of GCA. We administered the identical AFQT version given to participants at the time of military induction. AFQT scores were recorded as percentiles based on military norms. We used scores that had been transformed to achieve a normal distribution for statistical analyses. However, for ease of interpretation we report both untransformed percentile scores for descriptive purposes (e.g., tables) and the transformed scores used for statistical analyses.

Genetically informed latent growth curve modeling

A genetically informed latent growth curve (LGC) model was fit to the data using the maximum likelihood-based structural equation modeling software package OpenMx in an R environment (Boker et al., 2011). As with standard phenotypic LGC models, latent Intercept and Slope factors are fit to the repeated AFQT measures (see Figure 1). In addition, the model allows for the estimation of the additive genetic (A), common environmental (C), and unique environmental (E) variance components of these factors, as well as the genetic and environmental correlations between them. The model is a variation of the widely used common pathways model (Kendler et al., 1987; McArdle et al., 1998), with specific constraints imposed to fulfill the traditional assumptions of an LGC analysis: (a) factor loadings from each observed variable to the Intercept factor are constrained to unity (1.0); (b) factor loadings from each observed variable to the Slope factor are fixed such that the parameter for the first time-point is zero, whereas the second and third parameters reflect the interval between assessments; and (c) residual (variable specific) variance is constrained to be equal across assessments. To account for individual variability in the duration of time between assessments, factor loadings from the second and third assessments (age 56 AFQT and age 62 AFQT) were specified as definition variables, allowing us to use the precise person-specific duration or interval value rather than an average value for the sample. The interval between assessments was rescaled so that average interval value between induction and VETSA 1 (≈35 years) was equal to 1 [that is, interval01i = (age1i—age0i)/36], and the average interval value between induction and VETSA 2 was 1.161 [that is, interval02i = (age2i—age0i)/36]. To account for variability in the age at which the AFQT was first taken (age at induction), we regressed Intercept and Slope on age at induction centered at 20 years (i.e., age0i-20). Hence the Intercept reflects the expected AFQT performance at age 20 and the Slope reflects change in AFQT performance across an interval of 36 years. These analyses include all available twins (n = 1,291); missing data (assumed missing at random) were handled in the analyses with maximum likelihood estimation.

Latent class growth analysis (LCGA)

LCGA, a type of Growth Mixture Modeling, was applied using MPLUS 7.4 (Muthen & Muthen, 2015) to derive the trajectory patterns of AFQT score change over the three times; times fixed at 0 for baseline, 1 for time point 2, and 1.161 for time point 3. Missing data were dealt with through maximum likelihood estimation. The number of starts was varied between 100 to 200 for higher class solutions, and number of iterations was set to 10,000. Changes in AFQT score were represented by three latent variables: a latent class representing the unobserved subgroup with a common pattern of growth; and for each latent class, initial level or intercept, and slope of AFQT score change. The latent intercept and slopes were adjusted for entry age, with regressions equated across classes. Latent residual variables accounted for time-specific variation at the three waves, and were fixed to be equal. Twin pair dependency was accounted for by employing the clustering option in MPLUS. The initial levels and slopes of latent variables were characterized by their mean levels, with no random variation in the initial levels or slope estimated given that individual differences in AFQT growth were captured by class membership (Muthen, 2004; Jung & Wickrama, 2008). Trajectories were identified by starting with a single trajectory and, subsequently, models with increasing number of trajectories were estimated. The Bayesian Information Criterion (BIC), Entropy, and the Vuong-Lo-Mendell-Rubin Likelihood Ratio Test (VLMR-LRT) values and Lo-Mendell-Rubin Likelihood Adjusted LRT Test (Adj-LRT) were calculated for each model and utilized as fit indices for model comparisons. In general, models with the smallest BIC, Entropy of .80 or higher (range 0 to 1; indexes the quality of classification, or degree of separation of latent class posterior probabilities), and a significant VLMR-LRT and Adj-LRT tests when adding a latent class (cf. Jung & Wickrama, 2008; Muthen, 2004; Nylund et al., 2007) indicate the model with the best balance between parsimony and fit to the data.

The mean AFQT score at age 20 was 61.2 (SD = 22.2; 95% CI [59.85, 62.42]). The mean AFQT score at age 56 was 64.1 (SD = 20.9; 95% CI [62.96, 65.24]) and the mean AFQT score at age 62 was 62.9 (SD = 21.7; 95% CI [61.56, 64.24]). The differences between 95% CIs were quite small. The phenotypic correlations among the AFQT scores at the three ages ranged from 0.73 to 0.85 (see Table 1).

The mean time between Wave 1 (Induction) and Wave 2 (VETSA 1) was 36.16 years (SD = 2.37; variance = 5.60). The mean time between Wave 1 (Induction) and Wave 3 (VETSA 2) was 41.81 years (SD = 2.30; variance = 5.27).

Prior to fitting the genetically informed LGC model, we first estimated the genetic and environmental variance components for each observed variable, as well as the genetic and environmental correlations among the variables, using a trivariate Cholesky decomposition. Results from this analysis are presented in Table 2. There was a nonsignificant increase in heritability over time and the 95% CIs were overlapping. The heritability of GCA at age 20 was .59, at age 56 the heritability was .66, and heritability at age 62 was .64. Estimates of common environmental influences were small at all three times, and all 95% CIs for these estimates included zero. Genetic correlations among the variables ranged from .95 to .99, and all confidence intervals possessed an upper bound of 1.0, suggesting near complete genetic overlap among the measures. Common environmental correlations were not significant; however, unique environmental correlations did reach statistical significance, ranging from .21 to .47.

The genetically informed LGC model provided a good fit to the data (−2LL = 7043.372; DF = 3543; AIC = −42.6283; RMSEA = 0.0111; CFI = 0.9969; TLI = 0.9965) and revealed evidence for slight, but nevertheless significant changes in GCA across time. Without accounting for the effects of age at military induction, we observed a significant slope estimate of 0.073 (95% CI: 0.034 to 0.111), suggesting very slight increases in GCA between the ages of 20 and 62. Age at induction, however, was found to have significant effects on both the Intercept and Slope factors. Age at Induction had a positive association (β = 0.130) with Intercept, indicating that older participants performed better than younger participants. In contrast, a negative effect of age at induction was observed for slope (β = −0.111). The nature of the effect was such that younger participants had positive slope estimates (indicating improvement in performance over time), whereas older participants had negative slope estimates (indicating performance declines). The genetic and environmental influences of Intercept and Slope are reported in Table 3. The Intercept factor was found to have a heritability of .68 (95%CI = .47 to .87), indicating that 68% of the variance in GCA in early adulthood could be explained by genetic influences. The shared environment explained 16% (95% CI = .00 to .35) of the variance, but was not statistically significant. The nonshared environment was significant and explained 17% (95% CI = .12 to .23) of the variance in the Intercept. Slope was found to have a heritability estimate of .23 (95% CI = .00 to .60), however this estimate was not statistically significant. Common environmental influences accounted for 19% (95% CI = .00 to .50) of the variance in Slope, but once again were not statistically significant. The nonshared environment explained 59% (95% CI = .39 to .78) of the variance in Slope. The Intercept and Slope factors had a significant phenotypic correlation of −.31 (95% CI = −.39 to −.23), and there was no evidence for genetic overlap between the two factors rg = .01. Indeed the observed relationship between Intercept and Slope appeared to be overwhelmingly due to unique environmental influences shared between the two factors.

A four-class solution provided the best fit to the data in the LCGA considering the four indices, BIC, Entropy, VLMR-LRT, and Adj-LRT. BIC values dropped with each subsequently added class, however, the drop in BIC lessened notably after the 4-class solution. Entropy was equivocal for the 2-, 3-, and 4-class solutions hovering at .80 [range = .794–.801], but fell afterward <0.77. The VLMR-LRT and Adj-LRT were significant for 2- though 6-class solutions indicating improved fit, however class sizes became smaller with some <10% (see Table 4). Hence, with parsimony in mind we chose the 4-class solution as showing optimal fit across all indices. Figure 2 illustrates the four trajectories and Table 5 gives quantitative information about trajectory groups. To help quantify these classes, we provide the estimated Intercepts and Slopes in the chart legend. The trajectories are differentiated mainly by differences in overall AFQT performance level at entry as indexed by the Intercepts, which were uniformly significant. The Slope estimate for class 2, the largest class (39%), was positive and statistically significant (0.052, SE = 0.025, p = .034), but indicated only a slight positive growth in AFQT for these individuals. Thus, classes 1, 3 and 4 can be considered to have had stable trajectories by comparison and the very slight positive slope of class 2 is very modest. Notably, the patterns of mean intercept values coupled with the larger distributions of participants in Classes 2 and 3 compared with Classes 1 and 4 is consistent with cut points or binning a normal distribution (cf., Bauer, 2007).

Consistent with other studies of GCA spanning midlife (Gold et al., 1995; Larsen et al., 2008; Schwartzman et al., 1987) and the meta-analysis of Tucker-Drob and Briley (2014), mean GCA scores in the present study showed little change and demonstrated substantial rank-order stability. There was also very little change in the amount of genetic variance and nonshared environmental variance in GCA from age 20 to age 62 (cf. Lyons et al., 2009). However, growth curve and trajectory analyses indicated significant, but modest variability in age-related changes in GCA.

There is substantial evidence that genetic factors influence level of cognitive functioning in adulthood (McArdle et al., 2002; McGue & Christensen, 2002; Reynolds et al., 2005), but evidence regarding genetic influences on the dynamic process of cognitive aging is less consistent (review by Kremen et al., 2014; McArdle et al., 2002; McGue & Christensen, 2002; Reynolds et al., 2002, 2005). Only McGue and Christensen found significant linear change. Reynolds et al. only found significance for quadratic change and suggested that linear and nonlinear change may have different genetic and environmental determinants. However, there are at least two important differences between the current study and previous twin studies. The current study covered a longer time interval, 42 years compared with from 1 to 13 years in the other studies. VETSA participants’ age at the first assessment was ∼age 20 while most participants in previous twin studies were at least 65 years old at their first assessment. Our study, spanning the period from young adulthood to late middle age did not find a significant genetic influence on change. As mentioned above, there is evidence for genetic influence on change among individuals older than those in the current study (Tucker-Drob et al., 2014). Growth curve analyses revealed a significant negative phenotypic correlation between change and initial level of GCA. This indicates a tendency for individuals in the upper part of the distribution of GCA at age 20 to score lower on subsequent assessments and a tendency for individuals in the lower part of the distribution at age 20 to score higher on subsequent assessments. Thus, the only systematic trend for Slope (i.e., change) during this period appears to be regression to the mean (Galton, 1886). This negative phenotypic correlation reflects a very substantial negative correlation between aspects of the nonshared environment that influence initial level of GCA (Intercept) and aspects of the nonshared environment that influence change over time (Slope). Error variance in a biometric model is assigned to the nonshared environment and regression to the mean is a type of error variance. There was no relationship between genetic factors that influence level of GCA at age 20 and genetic factors that influence change over the next four decades. Our results suggest that, to a great extent, genetic factors are responsible for stability in GCA and aspects of the environment that differ between members of a twin pair are responsible for change. In the biometric analysis of twin data, the variance attributed to the nonshared environment includes the error variance. Regression to the mean, one type of error variance, is reflected in the nonshared environmental variance.

There was an interesting difference in the effect of age at induction on Slope versus Intercept. Individuals who were older at the time of induction were more likely to have attended college whereas those who were younger at the time of induction were more likely to only have a high school education. This educational difference at induction may be related to differences in GCA at the time of military induction. Individuals who were younger at the time of military induction may have been more likely to pursue additional education after leaving the military and this might contribute to the positive Slope associated with younger age at induction.

Our study does not demonstrate meaningful heterogeneity in patterns of change in GCA. Approximately 39% of our sample was in the only trajectory group that had a significant Slope; the modest magnitude of the group’s Slope suggests that it is not meaningful. The other three trajectory groups did not have significant Slopes. Although requiring replication, our results suggest that there is little heterogeneity in patterns of change in GCA over the course of middle adulthood.

There are several limitations to the current study. Because the sample includes only men and is primarily Caucasian, it is unknown whether our findings are applicable to women or racial/ethnic minorities. Because we measured GCA, we cannot determine whether and how specific cognitive abilities change. GCA tends to demonstrate more phenotypic stability compared to less generalized measures of cognitive ability (Tucker-Drob & Briley, 2014). Specific cognitive abilities change differently with age, so it is possible that we would observe more heterogeneity of trajectories for some specific abilities.

This study aimed to address three questions about GCA across four decades from young adulthood to late middle age by addressing the following questions. (a) Is there change in mean GCA and genetic and environmental variance in GCA over time? Although our genetically informed LGC model demonstrated slight changes in GCA across time, the magnitude of these changes in terms of absolute level and the overlapping confidence intervals suggests the change is not meaningful. Moreover, the very substantial correlations among GCA on each occasion indicated very substantial rank-order stability. Estimates of the heritability of GCA at the three ages were similar, ranging from .59 to .66 with overlapping confidence intervals. The shared environment did not significantly influence GCA at any age. The nonshared environment explained from 25% and 29% of the variance in GCA over the three ages and there was substantial overlap in the confidence intervals. Overall, the levels of GCA, the rank-order of individuals, and the genetic and environmental determinants of GCA did not change meaningfully during the 42 years of observation. (b) To what extent are individual differences in change in GCA influenced by genetic and environmental factors? Our results demonstrated that genetic factors are primarily responsible for the stability of GCA and the nonshared environment is primarily responsible for individual differences in change. Our results suggest that there is no overlap between genetic factors that influence one’s cognitive ability at age 20 and genetic influences on how those abilities are maintained or changed over the course of middle adulthood. (c) Do reliably different trajectory subgroups suggest heterogeneity in patterns of change in GCA over time? We identified four distinct trajectory groups. The groups differed in initial levels of GCA. However, the four trajectories were essentially parallel over time, with each of the four groups demonstrating a flat trajectory; there were no differences in ‘pattern.’

Bauer, D. J. (2007) Observations on the use of growth mixture models in psychological research. Multivariate Behavioral Research, 42, 757–786.

Bayroff, A. G., & Anderson, A. A. (1963). Development of the Armed Forces Qualification Tests 7 and 8. (Vol. Technical Research Report 1122). Arlington, VA: U.S. Army Research Institute.

Boker, S., Neale, M., Maes, H., Wilde, M., Spiegel, M., Brick, T., . . .Fox, J. (2011). OpenMx: An open source extended structural equation modeling framework. Psychometrika, 76, 306–317. 10.1007/s11336-010-9200-6

Eisen, S., True, W., Goldberg, J., Henderson, W., & Robinette, C. D. (1987). The Vietnam Era Twin (VET) Registry: Method of construction. Acta Geneticae Medicae et Gemellologiae, 36, 61–66. 10.1017/S0001566000004591

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. 10.1037/0012-1649.39.3.535

Galton, F. (1886). Regression towards mediocrity in hereditary stature. Journal of the Anthropological Institute of Great Britain and Ireland, 15, 246–263. 10.2307/2841583

Gold, D. P., Andres, D., Etezadi, J., Arbuckle, T., Schwartzman, A., & Chaikelson, J. (1995). Structural equation model of intellectual change and continuity and predictors of intelligence in older men. Psychology and Aging, 10, 294–303. 10.1037/0882-7974.10.2.294

Hayden, K. M., Reed, B. R., Manly, J. J., Tommet, D., Pietrzak, R. H., Chelune, G. J., . . .Jones, R. N. (2011). Cognitive decline in the elderly: An analysis of population heterogeneity. Age and Ageing, 40, 684–689. 10.1093/ageing/afr101

Jung, T., & Wickrama, K. A. S. (2008). An introduction to latent class growth analysis and growth mixture modeling. Social and Personality Psychology Compass, 2, 302–317. 10.1111/j.1751-9004.2007.00054.x

Kendler, K. S., Heath, A. C., Martin, N. G., & Eaves, L. J. (1987). Symptoms of anxiety and symptoms of depression. Same genes, different environments?Archives of General Psychiatry, 44, 451–457. 10.1001/archpsyc.1987.01800170073010

Kremen, W., Moore, C., Franz, C., Panizzon, M., & Lyons, M. (2014). Behavior genetics of cognition in middle adulthood. In D.Finkel & C.Reynolds (Eds.), Behavior genetics of cognition across the lifespan (pp. 105–134). New York, NY: Springer. 10.1007/978-1-4614-7447-0_4

Larsen, L., Hartmann, P., & Nyborg, H. (2008). The stability of general intelligence from early adulthood to middle-age. Intelligence, 36, 29–34. 10.1016/j.intell.2007.01.001

Lyons, M. J., York, T. P., Franz, C. E., Grant, M. D., Eaves, L. J., Jacobson, K. C., . . .Kremen, W. S. (2009). Genes determine stability and the environment determines change in cognitive ability during 35 years of adulthood. Psychological Science, 20, 1146–1152. 10.1111/j.1467-9280.2009.02425.x

McArdle, J. J., Ferrer-Caja, E., Hamagami, F., & Woodcock, R. W. (2002). Comparative longitudinal structural analyses of the growth and decline of multiple intellectual abilities over the life span. Developmental Psychology, 38, 115–142. 10.1037/0012-1649.38.1.115

McArdle, J. J., Prescott, C. A., Hamagami, F., & Horn, J. L. (1998). A contemporary method for developmental-genetic analyses of age changes in intellectual abilities. Developmental Neuropsychology, 14, 69–114. 10.1080/87565649809540701

McGue, M., & Christensen, K. (2002). The heritability of level and rate-of-change in cognitive functioning in Danish twins aged 70 years and older. Experimental Aging Research, 28, 435–451. 10.1080/03610730290080416

Muthen, B. (2004). Latent variable analysis: Growth mixture modeling and related techniques for longitudinal data. In D.Kaplan (Ed.), Handbook of quantitative methodology for the social sciences (pp. 345–368). Newbury Park, CA: Sage Publications.

Muthén, L. K., & Muthén, B. O. (1998–2015). Mplus user’s guide (7th ed.). Los Angeles, CA: Author.

Nylund, K. L., Asparouhov, T., & Muthén, B. O. (2007). Deciding on the number of classes in latent class analysis and growth mixture modeling: A Monte Carlo simulation study. Structural Equation Modeling, 14, 535–569. 10.1080/10705510701575396

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. 10.1080/03610730290103104

Reynolds, C. A., Finkel, D., McArdle, J. J., Gatz, M., Berg, S., & Pedersen, N. L. (2005). Quantitative genetic analysis of latent growth curve models of cognitive abilities in adulthood. Developmental Psychology, 41, 3–16. 10.1037/0012-1649.41.1.3

Schoenborn, C. A., & Heyman, K. M. (2009). Health characteristics of adults aged 55 years and over: United States, 2004–2007. National Health Statistics Report, 16, 1–31.

Schwartzman, A. E., Gold, D., Andres, D., Arbuckle, T. Y., & Chaikelson, J. (1987). Stability of intelligence: A 40-year follow-up. Canadian Journal of Psychology/Revue canadienne de psychologie, 41, 244–256. 10.1037/h0084155

Tucker-Drob, E. M., & Briley, D. A. (2014). Continuity of genetic and environmental influences on cognition across the life span: A meta-analysis of longitudinal twin and adoption studies. Psychological Bulletin, 140, 949–979. 10.1037/a0035893

Tucker-Drob, E. M., Reynolds, C. A., Finkel, D., & Pedersen, N. L. (2014). Shared and unique genetic and environmental influences on aging-related changes in multiple cognitive abilities. Developmental Psychology, 50, 152–166. 10.1037/a0032468

Uhlaner, J. E., & Bolanovich, D. J. (1952). Development of the Armed Forces Qualification Test and Presecesser Army Screening Tests. Washington, DC: Personnel Research Section, Department of the Army.

Willis, S. L., & Schaie, K. W. (2005). Cognitive trajectories in midlife and cognitive functioning in old age. In S. L.Willis & M.Martin (Eds.), Middle adulthood: A lifespan perspective (pp. 243–276). Thousand Oakes, CA: Sage. 10.4135/9781452225999.n8

Yu, L., Boyle, P. A., Segawa, E., Leurgans, S., Schneider, J. A., Wilson, R. S., & Bennett, D. A. (2015). Residual decline in cognition after adjustment for common neuropathologic conditions. Neuropsychology, 29, 335–343. 10.1037/neu0000159

Submitted: January 25, 2016 Revised: December 14, 2016 Accepted: December 20, 2016