Basic or Adaptation: The Assessment and Heritability of a Brief Measure of Agency

The interpersonal circumplex describes two major axes of personality that guide much of social behavior. Agency, one half of the interpersonal circumplex, refers to relatively stable behavioral patterns that center on self-focused dominance and assertiveness. Past empirical work on agency tends to treat the dimension as a characteristic adaptation, rather than a basic component of personality, in part due to the relatively large gender difference in agency with masculine individuals tending to behave more agentic. However, the psychometric overlap between agency and the most closely linked big five dimension, extraversion, is not well-established, and no behavior genetic work has documented evidence concerning the role of genetic and environmental influences. It is unclear whether agency is more similar to a personality trait, with no evidence of shared environmental influence and moderate heritability, or a characteristic adaptation, with some evidence for shared environmental influence and possibly lower heritability. We used the Midlife Development in the United States study to examine agency, big five, and generativity with replication and robustness check (Nnon-twins = 5,194; Ntwins = 1,914; NMilwaukee = 592). Results indicated that agency was higher in men (d = −.24), moderately heritable (44.4%), strongly correlated with extraversion (r = .51), moderately correlated with generativity (r = .36), and that approximately 40% of the variance in agency was shared with the big five. Agency also changed strongly with extraversion and openness, but less so generativity. Altogether, these results indicate that agency functions similar to other basic personality dimensions but is not clearly a dispositional trait.


Introduction
The interpersonal circumplex (Bakan, 1966;Wiggins, 1979) describes two major axes that guide interpersonal behavior, communion (i.e., getting along) and agency (i.e., getting ahead).Agentic individuals tend to engage socially in a selffocused dominant manner without attention to others.Individuals expect (Eagly & Karau, 2002;Eagly & Wood, 2012) and self-report (Badura et al., 2018;Feingold, 1994;Hsu et al., 2021) higher levels of agency among men.Given evidence for social learning of gender roles (Lott & Maluso, 1993) and the arbitrary nature of many gender-based expectations (Wood & Eagly, 2009), agency may re ect a characteristic adaptation (McAdams & Pals, 2006).Alternatively, agency may re ect a dispositional trait toward certain thoughts, feelings, or behaviors, similar to other major dimensions of personality like the big ve (John et al., 2008).The conceptual distinction between a characteristic adaptation and dispositional trait centers on the role that social learning plays in developing behavioral scripts.For example, individuals may have a dispositional trait toward politeness and respect for others that does not rely on experience or learning, but cultures differ in the sorts of actions that qualify as polite, leading individuals to develop different characteristic adaptations to express their basic level of politeness.
Most previous work has used rational or theoretical approaches to categorize constructs as a dispositional trait or characteristic adaptation.In contrast, Kandler and colleagues (2022) performed biometric decompositions of 50 presumed dispositional traits and 43 presumed characteristic adaptations drawn from the domains of goals, interests, morality, values, and self-schemas.They found that dispositional traits displayed substantially larger evidence for genetic in uences, and characteristic adaptations displayed substantially larger environmental in uences (correcting for any differences due to measurement error).Here, we test the psychometric and biometric properties of a short measure of agency in the Midlife Development in the United States study.Speci cally, we evaluated the phenotypic overlap between agency and measures of dispositional traits (i.e., the big ve) and a measure of a characteristic adaptation (i.e., generativity) and test whether the distribution of genetic and environmental effects is more similar to a dispositional trait or characteristic adaptation.

Nomological Network of Agency
Research domains differ on the conceptualization and measurement of agency (Abele, 2003;Alden et al., 1990;Bem, 2011;Hogan, 1983;McAdams, 1988;Spence & Helmreich, 1979;Wiggins, 1979).A connection to masculinity is common across most research domains.In a meta-analysis of 100,915 participants from 409 studies, Badura and colleagues (2018) found that men score signi cantly higher on self-report measures of agency than women (δ = .48).
Importantly, the effect sizes derived to estimate gender differences in agency were based on measures explicitly measuring masculinity (e.g., Bem Sex role Inventory; Personal Attributes Questionnaire), highlighting the explicit con ation of these constructs in the literature.Situating agency within the larger construct space of psychological individual differences can aid interpretation and translation across research domains.
Abele and colleagues (2016) compared a newly constructed Agency-Communion scale to the big ve in samples from Germany (N = 476), France (N = 250), and Australia (N = 140).The agency-assertive facet was most strongly associated with extraversion (r = .37− .57)and neuroticism (-.28 -− .56),less strongly with conscientiousness (.19 − .34)and openness (.13 − .27),and not consistently associated with agreeableness (-.2 − .16).Recently it has been suggested that overlap between agency and common measures of the big ve might be substantial enough to extract reliable estimates of agency from the big ve (Entringer et al., 2021).However, few empirical examples exist in the literature of direct comparisons between the narrow construct of agency and the big ve.
Bene ts of the big ve model of personality are that the factors are identi able cross-culturally, broadly capture many more narrowly de ned facets, and are related to and predictive of many real-world outcomes of interest, such as health, well-being, academic achievement, and job performance (John et al., 2008;Roberts et al., 2007).Further, these associations are highly replicable (Beck & Jackson, 2022;Soto, 2019).
Agency is also related to life outcomes and other indicators of social status.For example, agentic individuals tend to experience more career success (Abele, 2003) From a motivational point of view, generativity (i.e., the inner desire or social expectation to contribute something lasting to the world before death; McAdams & de St. Aubin, 1992) may be a logical link between agency and these outcomes.Highly agentic individuals tend to also express higher levels of generativity (Doerwald et al., 20121).Although generativity is also correlated with the big ve (Blatný et al., 2019), generativity shares more conceptual similarity with a characteristic adaptation, rather than a dispositional trait.By situating an explicit measure of agency within a longitudinal study with the big ve and a characteristic adaptation, such as generativity, we expand the nomological network of agency.

Lifespan Trends in Agency
Much is known concerning the development of the big ve (Bleidorn et al., 2022;Roberts et al., 2006;Roberts & DelVecchio, 2000).Personality development continues throughout the lifespan, with mean-levels shifting about a standard deviation in more "mature" directions with increasing age (i.e., the maturity principle).This change is concentrated in late adolescence and early adulthood (Roberts & Davis, 2016).Social dominance, conceptually similar to agency, increased substantially across the lifespan, with most of the increases occurring in early adulthood.However, it is not entirely clear whether social dominance is an appropriate marker for agency.For example, Diehl and colleagues (2004) found that younger adults were much more likely to self-attribute agency descriptors (r = − .2022) found for characteristic adaptations, with the higher end more consistent with results for dispositional traits.This ambiguity leaves unaddressed whether agency is better thought of as a characteristic adaptation or dispositional trait.To the best of our knowledge, no direct biometric estimates of agency have been reported.Here, we provide an estimate of heritability for agency, as well as estimates for stability and individual differences in change, and compare the magnitude of this value to that found for dispositional traits and characteristics adaptations (Kandler et al., 2022).

The Current Study
We sought to assess the psychometric and biometric features of a brief measure of agency available from the National Survey of Midlife Development in the United States (MIDUS; Brim et al., 1999) using three waves of longitudinal data.We do so by analyzing gender differences, measurement invariance, associations with generativity, and relationship to the big ve personality traits among unrelated individuals cross-sectionally and longitudinally (N = 5,194).We then demonstrate consistency of the results using the twin subsample and estimate genetic and environmental (co)variance (N = 1,914).The present research tested whether agency conforms to theoretical expectations of a characteristic adaptation or dispositional trait by addressing its relationships to gender, the big ve, and generativity within and across time.

Non-twin Sample
The rst wave of the National Survey of Midlife Development in the United States (MIDUS 1) collected a general population sample from 1995-1996 yielding 7,108 participants.The collaborative project investigated patterns, predictors, and consequences of midlife development in terms of physical health, psychological well-being, and social responsibility.The full sample consisted of unrelated individuals, siblings, and twins (Brim et al., 1999).We used all available non-twins for the rst set of phenotypic analyses, totaling 5,194 participants with mean age 46.93 years (range from 20 to 75, SD = 13.27years).The sample was 50.2% female and 49.8% male.The self-reported races of participants were White (N = 4,016), Black and/or African American (N = 255), Native American or Aleutian Islander (N = 28), Asian or Paci c Islander (N = 60), Multiracial (N = 40), and 106 identi ed as "other".Study variables were chosen for their expected associations with Agency.For a negative control (i.e., showing not everything is correlated) we included BMI of the participants pre-calculated in the MIDUS data in the correlation matrices.
Participants were followed longitudinally with a time lag between assessments of roughly a decade.Response rates for phone interviews from wave 1 to wave 2 were 75% [1] and from wave 2 to wave 3 were 77%.The self-administered questionnaires yielded a total of 3,479 at wave 2 (52.3% female and 47.7% male) and 2,276 at wave 3 (53.3%female and 46.7% male).The mean age for the at wave 2 was 56.01 years (range from 28 to 84, SD = 12.72 years) and 64.17 years (range from 39 to 93, SD = 11.57years) at wave 3. The wave 3 non-twin sample had low racial diversity (White/Caucasian = 2,012; Black and/or African American = 85; Asian or Paci c Islander = 16; Multiracial = 22; Other and Native American or Aleutian Islander = 41).All data used in this report are publicly available [2].

Twin Sample
Next, we used the twin sample to decompose variance in the phenotypes into genetic and environmental components.There was a total of 1,914 individuals (31 missing zygosity information), and the sample was 55.3% female and 44.7% male.The self-reported race of the participants was as follows: White (N = 1,632), Black and/or African American (76), Native American or Aleutian Islander/Eskimo (11), Multiracial (12), and 18 identi ed as "other".The twin sample had mean age 44.89 years (range from 25 to 75, SD = 12.07 years).Means, standard deviations, and Cronbach's alpha reliabilities in the twin sample are reported in Table 1.
Zygosity was determined via an eight-item self-report screener which asked about physical similarity.Although zygosity classi cation was not veri ed with genotyping, similar studies using the same items have found accuracy to be over 90% when con rmed via genotyping (Lykken et al., 1990).Triplets were included pairwise and weighted to correct for the same individual appearing in multiple pairs.The resulting twin sample included 347 MZ pairs, 322 DZ pairs, and 252 DZOS pairs (total = 921 pairs).The gender breakdown for same-sex pairs was 185 MZF, 162 MZM, 200 DZF, and 122 DZM.At the nal wave, there were 107 MZF, 86 MZM, 109 DZF, 53 DZM and 145 DZOS pairs.

Milwaukee Sample
In order to better examine health issues in minority populations, the Midlife in the United States study sampled Milwaukee, Wisconsin based on areas of the city with high concentrations of African American residents.The crosstime correlations will be reported as a robustness check of the generalizability of the cross-time correlations in nontwins and twins.

Measures Personality Traits
Personality was measured by the Midlife Development Inventory (MIDI; Lachman & Weaver, 1997).The MIDI personality inventory contains 30 adjectives that assess neuroticism, conscientiousness, extraversion, agreeableness, openness to experience, and agency.Each trait is measured by 4 to 7 items.Participants rated items on a 4-point scale indicating whether the adjective described them "not at all" to "a lot."Means, standard deviations, and Cronbach's alpha reliability for the items and scales are reported in Table S2.

Loyola Generativity Scale-contributions domain
Participants completed six items rated on a 4-point scale from the Loyola Generativity Scale (McAdams & de St. Aubin, 1992) to measure the speci c generativity domain of "contributions."Example items were: "Many people come to you for advice" and "You have had a good in uence on the lives of many people."Psychometric properties of this scale are reported in Table S2.

Statistical Analyses
Analytic scripts are available at the OSF link: https://osf.io/qsj9u/?view_only=5794f5d6b414474a953000337de38b95.Analyses were conducted using R (R Core Team, 2022) and the packages lavaan, semTools, effsize, uMx, and psych (Bates et al., 2019;Jorgensen et al., 2021;Revelle, 2020;Rosseel, 2012;Torchiano, 2020).Given the large sample size and aims of the current study, we focus on effect size estimates and precision throughout.To correct for non-independence of observations due to the familial structure of the data, we used cluster-robust standard errors (McNeish & Harring, 2017).All models use full information maximum likelihood estimation to handle missing data.

Measurement Invariance
Measurement invariance was examined between men and women and across time by testing the thresholds, loadings, and intercepts for equivalence (Svetina et al., 2020;Wu & Estabrook, 2016).We used standard cutoffs for determining whether the assumptions of measurement invariance hold (i.e., ΔCFI ≤ 0.01).Model comparisons indicated that factor scaling was consistent across sex and time (see Table S3, S6, and S7 for complete details).After establishing measurement invariance, subsequent analyses were carried out with the mean scores.

Attrition Analysis
We estimated Cohen's d effect sizes comparing baseline scores for those who remained in the study compared to those who did not participate at the second or third wave.All differences were small (|d| < .21;see Table S4).

Phenotypic Analyses
We estimated bivariate correlations and multiple regressions between agency and the other phenotypes crosssectionally.These analyses indicate the extent to which agency shares common variance with dispositional traits and the characteristic adaptation of generativity.To estimate longitudinal trends, we used linear growth curve models and correlated intercepts and slopes across phenotypes.Substantial variance in slopes indicates that individuals follow varying trajectories of development.Correlated slopes indicate that individual differences in change in one construct overlap with individual differences in change in another construct.Put differently, if the extent to which someone changes in agency is similar to their change in generativity, a positive slope-slope correlation would be found.Such models can estimate the extent to which the phenotypes co-develop, possibly because one phenotype causally shapes the development of the other, reciprocal causal processes, or some other variable causally impacts the development of both phenotypes.

Behavior Genetic Analyses
We decomposed the variance in each phenotype using the genetically informative twin subsample (Neale & Cardon, 1992).Because MZ twins are more genetically similar than DZ twins, larger MZ twin correlations compared to DZ twin correlations indicate additive genetic in uences (A).If twins are more psychologically similar than would be expected due to additive genetic in uences alone, this result indicates that shared environmental factors (C) also in uence the phenotype.However, if MZ twin similarity is more than double DZ twin similarity, this result indicates that dominant genetic in uences (D) are plausible.The classical twin design does not provide su cient information to identify C and D simultaneously, and therefore are not estimated in the same model.Finally, unless MZ twins are psychologically identical, nonshared environmental in uences (E) lead to differences between twins.
We formalized the assumptions of this model in a structural equation modeling framework where variance in the phenotypes was decomposed into A, C, D, and E factors, depending on the pattern of twin correlations.For MZ twins, the correlation between factors representing genetic and shared environmental in uences were xed to 1, re ecting that MZ twins share nearly identical genotypes and shared environmental in uences.For DZ twins, the correlation between the A factors were xed to .5, re ecting the assumption that DZ twins share, on average, 50% of segregating genetic material.The correlation between D factors was xed to .25, re ecting the probability that the twins share the dominant allele.The correlation between C factors was xed to 1 as all siblings, by de nition, have the same shared environment.For all twin pairs, the correlation between E factors was xed to 0 due to these effects being individualspeci c.Multivariate extensions of these models are premised on the same logic and allow for decomposing covariance between phenotypes into genetic and environmental components.Because we included opposite-sex pairs, all phenotypes were residualized for age and sex, as is standard practice (McGue & Bouchard, 1984).
Using these behavior genetic techniques, we estimated a series of univariate models to identify the best tting set of variance components to represent each phenotype (see Table S10 for cross-twin cross-trait correlations).These analyses indicate the extent to which genetic and environmental in uences contribute to the development of agency.Then, we estimated bivariate models to decompose the covariance between agency and the other phenotypes.These analyses indicate the extent to which genetic and environmental in uences link agency with dispositional traits and a characteristic adaptation.We used a multivariate Cholesky decomposition, the behavior genetic analogue of multiple regression, to estimate associations with agency controlling for the other included phenotypes.In contrast to regression, the order in which phenotypes are entered into the model impacts interpretation.We speci ed phenotypes with the weakest association with agency to take precedent in the model to limit convergence issues.Finally, we decomposed the (co)variance of intercepts and slopes of growth curves for each pair of phenotypes to evaluate the extent to which genetic and environmental in uences guide individual differences in trajectories of development.In one instance, we encountered convergence di culties when decomposing the intercept and slope latent factors.To provide estimates, we performed the decomposition on factor scores derived from the phenotypic model.

Descriptive Statistics
After establishing measurement invariance, we calculated descriptive statistics (Table 1) and bivariate correlations (Table 2) split by gender for variables standardized with respect to the baseline mean and standard deviation.Itemlevel descriptive statists are reported in Table S1.Because descriptive information concerning the big ve have been reported elsewhere using these data (Graham et al., 2020; Olaru & Allemand, 2022), instead we focus on agency, generativity, and their links to the big ve.
At baseline, men reported modestly higher levels of agency than women in the non-twin (d = -0.24,99%CI [-0.317, -0.162]) and twin samples (d = -0.24,99%CI [-0.368, -0.116]; see Figure S1-S3 for distributions and item information curves).Women were modestly more variable in agency relative to men.Despite typical descriptions of agency centering on masculinity, distributions of agency were largely overlapping.Generativity showed even smaller gender differences in mean and variance.Results were largely similar across waves (Table S5).Note.Variables were standardized with respect to the grand mean and standard deviation at wave 1.
Consistent with expectations, agency was moderately-to-strongly correlated with extraversion and moderately correlated with generativity.Openness was also moderately-to-strongly correlated with agency.Agentic individuals reported being outgoing, active, imaginative, creative, and desiring to leave a mark on the world to a greater extent than less agentic individuals.Generative individuals also reported higher levels of extraversion and openness, similar to agentic individuals.In contrast, generative individuals reported higher levels of agreeableness.This association was much weaker for agency, indicating that being warm, caring, and sympathetic are personological distinguishing factors between agency and generativity.Associations with the other personality dimensions were more modest.Age was uncorrelated with Agency (-0.012).
Multiple regression results (Table 3) further supported agreeableness as playing a differential role for agency and generativity.For both phenotypes, extraversion and openness retained moderate, positive associations even when controlling for the other big ve.For agency, agreeableness was estimated to have a negative association when controlling for the big ve.Put differently, among individuals with similar levels of extraversion and openness, more agreeable individuals would be expected to have lower levels of agency.At the zero-order level, this negative association was masked due to the positive associations among agreeableness, extraversion, and openness.Overall, approximately 40% of the variance in agency was associated with the big ve.Thus, although agency and generativity share variance with the big ve, neither phenotype primarily re ects typical dispositional traits.Next, we evaluated the extent to which the longitudinal development of agency and generativity were associated with trajectories of big ve development.All models t the data well (see Table S8-S9 for complete parameter estimates and t statistics).On average, agency decreased slightly across time (-.326 baseline-SD units), but there was considerable variance in trajectories (slope variance = .031).Put differently, these results imply that someone following a more positive (i.e., 1.5 slope SDs above the mean) or more negative (i.e., 1.5 slope SDs below the mean) would show − .062and − .590baseline-SD units of change in agency across waves.
Individual differences in change for agency were weakly-to-modestly related to baseline levels of the big ve (Table 4).The largest associations were with extraversion and openness, both of which were negative.These results indicate that individuals who began the study relatively higher on extraversion and openness tended to display smaller increases or larger decreases in agency across time.Correlations among slopes were much stronger.Individuals who followed more positive trajectories of growth in extraversion (r = .90)and openness (r = .93)also followed very similar trajectories of agency development, relative to other participants in the sample.Trajectories of agreeableness (r = .49)and conscientiousness (r = .50)were also moderately correlated with trajectories of agency.
Results for generativity showed generatively only moderately changed with agency.Starting levels of either generativity or agency were weakly correlated with change in the other.Despite stronger within-time correlations, agency changed more positively with agreeableness or conscientiousness than generativity.
Taken together, these results imply that agency codevelops along with dispositional traits like the big ve, with nearly identical relative trajectories among agency, extraversion, and openness.Co-development was weaker with generativity, a characteristic adaptation.How are levels and trajectories of agency, generativity, and the big ve?
the supplemental materials (Tables S11-S16), we report time-speci c behavior genetic decompositions, as well as multivariate extensions.Broadly, these results indicate that agency is moderately heritable (44.4%), that shared variance with dispositional traits tended to re ect a mix of primarily genetic and some nonshared environmental sources, and that residual variance in agency was primarily nonshared environmental (including measurement error) with modest evidence of unique genetic effects (~ 5% of the total variance).Generativity, on the other hand, was estimated 32.4% due to genetic contribution in line with Kandler and colleagues' (2022) ndings Behavior genetic results for the big ve in these data have been reported in greater detail elsewhere (Luo et al., 2017;South et al., 2018).
Here, we focus on results derived from growth curve models.
We estimated the heritability and environmentality of the intercept and slope factors for each phenotype (Table S18-S19).For the big ve, the heritability of the intercept and slope factors ranged from 44.4-70.9%for the intercepts and 0-62.9% for the slopes.Genetic in uences accounted for 63.7% and 65.8% of the variance in the intercept and slope of agency, respectively.Results for generativity showed much lower genetic in uences for the intercept, 39.6%, and no genetic in uences for the slope factor.
Finally, we decomposed the estimated phenotypic correlations among intercepts and slopes into genetic and environmental contributions.The genetic contribution to a phenotypic correlation indicates the extent to which the correlation is due to genetic sources of variance, in correlation units.For example, a phenotypic of .8 could be decomposed into a genetic contribution of .6 and a nonshared environmental contribution of .2, which sums to the phenotypic correlation of .8.These statistics can be calculated from the heritability, environmentality, and genetic and environmental correlations for the phenotypes (see Briley & Tucker-Drob, 2014 for further details).
Table 5 reports genetic and environmental contributions to correlations across intercepts and slopes among the phenotypes.Focusing rst on the intercepts of agency and generativity, the genetic correlation was .416and environmental correlation was .316which was in line with the phenotypic correlations being third largest.In terms of the slopes of agency and generativity, there was correlated shared environment suggesting that living in environments that function to make people more similar on agency are partially overlapping with environments that shape people on generativity.Extraversion and openness were also strongly correlated genetically and environmentally with agency intercept and slope.The starting level of extraversion was strongly genetically correlated with the starting level of agency as well as the change in agency.The starting level of openness was strongly genetically correlated with the starting level of agency and weakly genetically correlated with the slope of agency.Which is to say that the starting levels of dispositional traits were strongly genetically related to starting levels of agency.Change in agency was strongly genetically related to change in extraversion and shared environmentally correlated with generativity.

Discussion
Personality psychologists have relied on the distinction between dispositional traits and characteristic adaptations to distinguish aspects of persons that develop largely without or with contextual inputs, respectively.Following Kandler et al. (2022), we used genetically informative data to test whether agency, a personality attribute centered on assertiveness and self-con dence in interpersonal relations, is more similar to a dispositional trait, like the big ve, or a characteristic adaptation, like generativity.Taken as a whole, the results were somewhat ambiguous.Agency was moderately-to-strongly correlated with the big ve and displayed nearly identical relative growth trajectories compared to extraversion and openness.These results were found in non-twins, replicated in twins, and cross-time correlations of the Milwaukee oversample all converged on similar ndings.Further, the magnitude of genetic and environmental in uences was similar and the covariance between trajectories of agency and the big ve were disproportionately due to genetic sources of variance.Each of these pieces of evidence would point to the construct of agency operating as a dispositional trait.However, many of these pieces of evidence were also found for generativity, a construct that much more clearly meets the de nition of a characteristic adaptation.As with many binary categorization systems (e.g., Morgenroth & Ryan, 2020), this deceptively simple distinction may hide an underlying continuous dimension of input from external sources.
Early descriptions of the big ve in terms of dispositional traits were explicit and clear that the big ve were unable to be altered by the environment (e.g.

Limitations and Directions
study was the rst investigation of the heritability of agency in MIDUS.Only one recent study was found on the genetic and environmental correlations of personality and loneliness that included agency (Freilich et al., 2022).However, our focus was on the correlates of agency cross-sectionally and longitudinally and they were interested in loneliness and personality correlates cross-sectionally.One study was found to study agency in MIDUS longitudinally with well-being (Haas & vanDellen, 2020) but not among personality traits.
One limitation was the available measures of agency and the big ve.The MIDI personality scales are an extra short personality inventory that is not often used by personality researchers.Because of its inclusion in a wide-ranging population study, the MIDI inventory necessarily values brevity over breadth; therefore, the question of whether agency encompasses speci c facets under such traits as extraversion and openness or whether agency explains the differences between men and women on the big ve traits remains unanswered.The current measure, also, lacked a distinct communion scale.That being said, even the brief measure of agency showed results that aligned with expectations and falls somewhere between a dispositional trait and characteristic adaptation.
The second limitation was lack of generalizability across cohorts, race, and gender diverse people.The current study, which used cohorts who were born 1920 to 1975, was not equipped to test for cohort effects we would expect to see in younger cohorts or those born nearer the turn of the 21st century.Whether gender differences in agency remain stable over generations and whether the pattern of associations persist from one generation to the next are two questions that would be useful to test with more recent cohorts.Next, although the sample was collected to be population representative, racial diversity was still insu cient to provide appropriate sample sizes to speci cally test equality of effects sizes across races or ethnicities.Finally, there were no gender diverse people included in MIDUS by its reporting , higher judgments of status (Fiske et al., 2007), more positive selfperceptions (Wojciszke et al., 2011), and greater achievement (Trapnell & Paulhus, 2012).
McAdams et al. (2006)(2000)ions decrease across age.Given thatRoberts et al. (2006)'s meta-analytic evidence was based on longitudinal studies which estimated within-person change, the discrepancy may be due to cohort effects.Test-retest stability increases across the lifespan (i.e., cumulative continuity principle;Bleidorn et al., 2022).Roberts & DelVecchio (2000)synthesized longitudinal studies of personality, including masculinity/femininity (k = 17).Results were consistent with moderate-to-strong rank-order stability across time, particularly in adulthood.Stability was not substantially different between measures of the big ve and masculinity/femininity.Further,McAdams et al. (2006)found 3-year test-retest stability of .43 for agency derived from narrative interviews which draw more heavily on the motivation and identity components of agency.We extend these previous studies by tracking stability and change of agency across two decades of midlife development.

Table 2
Zero-order correlation matrix from the non-twin (upper triangle) and twin (lower triangle) MIDUS wave 1 samples.
Note.Correlations greater than 0.30 are bolded.

Table 3
Multivariate regression results for associations betweenAgency and the Big Five factors in the non-twin and twin samples.

Table 4
Model implied correlation matrix from the correlated latent growth models in the non-twin (left) and twin (right) samples.

Table 5
The correlated genetic and environmental in uences on growth to the genetic and environmental in uences on stability parameters from bivariate longitudinal models with agency.A refers to genetic in uences; C refers to shared environment; and, E refers to unique environment.
Kandler et al. (2022) 1999), but as evidence inconsistent with such strong claims accumulated (e.g., Bleidorn et al., 2013; Bühler et al., 2023; Dugan et al., 2023; Gurven et al., 2013; Hudson & Fraley, 2017; Niehoff et al., 2017; Obschonka et al., 2018; Roberts et al., 2017; Tucker-Drob & Briley, 2014), the distinction became fuzzy (e.g., McCrae & Costa Jr., 2008).It seems obvious that some aspects of human development are more di cult to change than others.Height may be quite di cult to actively control given adequate nutrition, whereas shortterm states such as low positive affect could be actively raised through socializing or exercising.Highly stable constructs, such as cognitive ability, may at rst glance seem to re ect a dispositional trait, but education is a volitional action that causally raises general ability (Ritchie & Tucker-Drob, 2018).Rather than attempting to t constructs and messy developmental processes into neat categories, it may make more sense to collect a range of psychometric, developmental, and behavioral aspects of constructs in order to place them along a continuum.Importantly, it is not clear what sorts of properties would be necessary and su cient to identify this continuum.Heritability estimates have been used, improperly, to justify claims about insensitivity to environmental input.However, heritability estimates include gene-environment interplay processes which could lead to high heritability estimates with many modi able environmental risk factors, or novel interventions could be designed to work in tandem with genetically in uenced characteristics (e.g., wearing eyeglasses).We favor the approach taken byKandler et al. (2022)which uses behavior genetic estimates as one data point among many to guide inferences.In this context, agency appears to re ect an intermediate variable.