Are Some People More Consistent? Examining the Stability andUnderlying Processes of Personality Profile Consistency
Amanda J. Wright and Joshua J. JacksonDepartment of Psychological and Brain Sciences, Washington University in St. Louis
Personality traits are relatively consistent across time, as indicated by test–retest correlations. However,ipsative consistency approaches suggest there are individual differences in this consistency. Despite this, itis unknown whether these differences are due to person-level characteristics (i.e., some people are just moreconsistent) or exogenous forces (i.e., lack of consistency is due to environmental changes). Moreover, it isunclear whether the processes promoting long-term consistency are the same across people. We examinethese two questions using item-level profile correlations across four to nine waves of data with four data sets(N = 21,616) with multilevel asymptotic growth models. Results indicated that there were, on average, highlevels of profile consistency. However, there were notable individual differences in initial profile correlationvalues as well as in changes in levels of consistency across time, indicating that some people are more stablyconsistent than others. Moreover, the directions of people’s trajectories across increasing time intervalssuggest that the mechanisms responsible for reinforcing personality consistency vary across people. Theseeffects were typically moderated by age at 30 years old, maturity-related traits, and education level. Overall,findings indicate some people are more consistent than others, such that this stable level of (in)consistency isa dispositional factor. Additionally, individual differences in profile consistency are shaped by differentlevels of three processes. On average, stochastic factors are not impactful for most individuals, andtransactional processes have an important role in increasing consistency for a sizable amount of people—nuances not previously revealed when focusing on rank-order stability.
Keywords: personality development, Big Five, ipsative consistency, profile correlations, person-centered
Supplemental materials: https://doi.org/10.1037/pspp0000429.supp
Throughout the lifespan, a general pattern of personalityconsistency and stability emerges within and across individuals(Asendorpf, 1992; Beck & Jackson, 2020; Donnellan et al., 2007;Roberts et al., 2006). However, personality traits are also adaptable,malleable qualities that can fluctuate over the short-term (Beck &Jackson, 2020) as well as over longer time frames (e.g., Bleidorn etal., 2013; Conley, 1984a). Test–retest correlations are a popularchoice to index the amount of consistency in personality traits across
time (Robins et al., 2001). However, using only two time points andaggregating across people with rank-order correlations obscures twoopen questions about personality consistency.
First, test–retest correlations are typically applied to aggregatelevels of between-person, rank-order stability, inevitably maskingany individual differences in personality consistency (Aldwin et al.,1989; Lamiell, 1981). In contrast, profile correlations take a person-centered approach in which the consistency of one’s personalityprofile (i.e., scores on personality traits or items) is relative to one’spast self, not in comparison with others, thus allowing for individualdifferences to be estimated. While individual differences in profileconsistency are well established (e.g., Asendorpf & van Aken, 1991;Ozer & Gjerde, 1989; Terracciano et al., 2010), it is unclear whetherthese individual differences in consistency persist across time. Thatis, are some people more or less consistent in general, as some formof a dispositional individual difference? Or are individual differ-ences in profile consistency ephemeral, perhaps due to externalcircumstances or measurement error?
Second, the pattern of multiple consistency estimates can provideinsights into the processes driving consistency more than singletest–retest correlations can (Fraley & Roberts, 2005). While pastwork suggests that stochastic, transactional, and enduring factors areall necessary to explain patterns of personality consistency (Anusic& Schimmack, 2016; Fraley et al., 2013; Fraley & Roberts, 2005),these tests were aggregated across people, making the assumptionthat the processes are the same for everyone. In contrast, person-centered approaches may identify different patterns of continuityacross people (Beck & Jackson, 2020), with different processes
Thisdocumentiscopyrightedby
theAmerican
Psychological
Associatio
nor
oneof
itsalliedpublishers.
Thisarticle
isintended
solely
forthepersonal
useof
theindividual
user
andisnotto
bedissem
inated
broadly.
This article was published Online First July 11, 2022.Amanda J. Wright https://orcid.org/0000-0001-8873-9405Amanda J. Wright played lead role in formal analysis and writing of
original draft and equal role in conceptualization, methodology, and writingof review and editing. Joshua J. Jackson played lead role in supervision,supporting role in writing of original draft, and equal role in conceptualiza-tion, methodology, and writing of review and editing.All code to reproduce analyses, the data codebook, and Supplemental
Materials are available at: https://osf.io/23u9b/. All data are freely availablevia application and/or data use agreements at the links specified in each dataset’s “Participants” subsection in the Method section. Neither author has anyconflicting interests to report. Some findings from this article were presentedat the 2021 Association for Research in Personality conference during adata blitz.Correspondence concerning this article should be addressed to Amanda
J. Wright, Department of Psychological and Brain Sciences, WashingtonUniversity in St. Louis, 1 Brookings Drive, St. Louis, MO 63130, UnitedStates. Email: [email protected]
Journal of Personality and Social Psychology:Personality Processes and Individual Differences
© 2022 American Psychological Association 2023, Vol. 124, No. 6, 1314–1337ISSN: 0022-3514 https://doi.org/10.1037/pspp0000429
1314
driving consistency. As such, it is unclear whether the processes thatdrive person-centered consistency are shared across people and ifthey mirror the processes responsible for between-person stability.The current article addresses these two questions using multiple
waves of person-centered (ipsative) consistency data. Multiple datasets spanning up to 12 years are used to examine longitudinal trendsacross at least three waves of consistency data. Trajectories inwithin-person profile consistency are viewed from two perspectives.First, we examine individual differences in levels of personalityprofile similarity and consistency across sequential waves to exam-ine if those with high or low levels of consistency continue to behighly consistent or inconsistent across subsequent assessments.That is, are people dispositionally stable or unstable in theirmagnitude of profile consistency? Second, the processes underlyinglong-term trends of profile consistency from an individual’s initialtime point are viewed across increasing time intervals. That is, doindividuals show similar long-term trajectories of personality profileconsistency? These increasing intervals of consistency allow us toask whether people have similar patterns in their long-term person-ality consistency trajectories, which inform whether the processesunderlying personality consistency are shared across individuals.
Stability and Consistency of Personality
Despite general trends of stability and consistency in personalitythat emerge between and within people, personality traits are alsomalleable qualities that can vary over time (Bleidorn et al., 2013;Roberts & Mroczek, 2008). Importantly, there are many ways toconceptualize the ways in which personality can change (Robertset al., 2008). One helpful way to think about the ways personalitycan change is in terms of a two-by-two table: If the changes occur atthe population or the individual level, crossed by if it is an absolutechange or if it changes relative to something else (e.g., anotherperson or another trait; Roberts et al., 2008). Below, we describeeach of the four ways personality can change according to thiscross-table.Commonly examined are population mean-level changes. These
are absolute changes that reflect normative, or typical, changes inaverage levels of traits across time (Roberts et al., 2006). Similarly,individual-level and mean-level changes are often examined andreflect absolute changes that occur at the individual level (e.g.,Mroczek & Spiro, 2003). These individual differences in changereflect changes in a person’s own average levels of a trait that candiffer from the normative changes that are observed at the popula-tion level.In contrast to absolute change, relative change is often used to
describe the stability and consistency of personality. Rank-orderstability, which occurs at the population level and represents therelative ranking of individuals for a single trait, is typically used todemonstrate the consistency of personality across the lifespan(Roberts & DelVecchio, 2000). Less examined is ipsative consis-tency or person-centered personality consistency. Consistency oc-curs at the individual level where the configuration of trait indicatorswithin a single person across time is examined, hence person-centered. This person-centered approach to personality develop-ment can take several forms, such as comparing personality profilesof different individuals/groups (such as different age groups orgenders; e.g., De Fruyt et al., 2006; Ozer & Gjerde, 1989), compar-ing a group of individuals with a profile of certain characteristics to a
group with another distinct profile (e.g., Block, 1971; Gilbert et al.,2021; Meeus et al., 2012; Specht, Luhmann, & Geiser, 2014; Stecaet al., 2010), or comparing an individual’s own personality profile atdifferent time points (e.g., Ibáñez et al., 2016; Jackson & Beck,2021; Robins et al., 2001; Terracciano et al., 2010). Notably, out ofthe four ways outlined above, it is the only perspective of consis-tency that considers multiple aspects of someone’s personality—asopposed to being limited to examining a single trait.
An extension of relative personality change that incorporatesabsolute change is examining how people tend to change in theirlevel of ipsative consistency across time. That is, are there absoluteincreases or decreases in levels of person-centered profile consis-tency or are people stable in their level of profile consistency?Furthermore, the individual differences in these average trends canbe examined, such that some individuals may increase in profileconsistency, whereas others decrease across time. It is this blend ofrelative and absolute personality change that our article focuses on.
Indices of Personality Profile Similarity and Consistency
Profile correlations (Q) provide an index of the degree of continuityof a configuration of indicators over time for an individual(Asendorpf, 1992; De Fruyt et al., 2006; Klimstra et al., 2009;Ozer & Gjerde, 1989; Roberts et al., 2008; Robins & Tracy,2003). That is, they provide a measure of personality profile consis-tency across two time points. Profile correlations can be calculated atvarying levels of breadth and depth, ranging from trait-level (e.g.,Robins et al., 2001), facet-level (e.g., De Fruyt et al., 2006), or item-level examinations (e.g., Ozer & Gjerde, 1989). In general, peopleshowmoderate-to-high levels of profile consistencywith the Big Fivetraits (De Fruyt et al., 2006; Ibáñez et al., 2016; Robins et al., 2001;Terracciano et al., 2010). Despite these high average profile consis-tency estimates, there are considerable individual differences aroundthem (Asendorpf & van Aken, 1991; Jackson & Beck, 2021; Ozer &Gjerde, 1989; Terracciano et al., 2010), suggesting some people arevery consistent, whereas others are much unlike their previous selves.
Closely related to profile correlations are the D indices proposedby Cronbach and Gleser (1953). These three indices—D2, D′2, andD″2—quantify the degree to which personality profiles change inelevation (i.e., mean-level), scatter (i.e., spread of scores), andshape, respectively. That is, they quantify the degree to whichprofiles are (dis)similar across two time points. Each of theD indices can be calculated from one another and D″2 is perfectlyinversely related toQ, so similar yet complementary information canbe obtained by examining all indices of personality profile similarityand consistency. Past research examining these three D indices hasfound that, when personality profiles change, most of it tends to bedue to changes in elevation and scatter, with typically a smallerproportion of individuals showing changes in profile shapes (DeFruyt et al., 2006; Robins et al., 2001). Although, one study did findthe largest proportion of change in their sample being due to changesin shape of profiles (Ibáñez et al., 2016).
Why are some people consistent in their personality profiles whileothers are inconsistent? In this article, we focus on two potentialreasons. First, outside factors may be a driving factor in thisheterogeneity. Life experiences are associated with changes inrank-order consistency (Specht et al., 2011), suggesting that ex-periences may drive individual differences. Engagement in maturesocial roles such as becoming a parent or starting a career could
Thisdocumentiscopyrightedby
theAmerican
Psychological
Associatio
nor
oneof
itsalliedpublishers.
Thisarticle
isintended
solely
forthepersonal
useof
theindividual
user
andisnotto
bedissem
inated
broadly.
PERSON-CENTERED PERSONALITY CONSISTENCY 1315
facilitate stability in one’s personality so long as these roles aremaintained across time (Roberts et al., 2003). Evidence for thismaturity–stability hypothesis has indeed been found in past work(e.g., Donnellan et al., 2007). Furthermore, the predictability ofone’s environment (Asendorpf & van Aken, 1991), in addition to thegreater likelihood of some people experiencing certain life eventsmore than others (Headey & Wearing, 1989; Magnus et al., 1993),all serve as possible routes by which external factors could bolster ordecrease personality consistency across time. Life experiences mayalso be shaped by health or biological changes (e.g., dementia;Terracciano et al., 2018), thereby mediating the effect betweenbiology and personality continuity.Second, some people may be inherently more consistent than
others (Bem & Allen, 1974). This alternative pattern has not oftenbeen discussed as it fails to cleanly align with either of the twopredominant schools of thought within personality—dispositionalistsversus situationists. Dispositionalists assume relatively constant traitsguide behavior, whereas situationists assume behavior is largelyguided by the situations we are in (Mischel & Shoda, 1998).However, some people may be more consistent in general, not dueto situations per se, but as an inherent individual difference that itselfis a stable factor, much like for traits themselves (Anusic &Schimmack, 2016). For example, much like people have trait levelsof Neuroticism and Conscientiousness, which remain relatively stableto that person over time, a person’s level of personality profileconsistency might be another dispositional characteristic. Mostwork examining whether people are stable in their level of profileconsistency has looked at behaviors across short periods of time, suchas days or weeks (Epstein, 1979). As such, it is hard to know whetherthere are stably consistent people over longer periods of time. Somepeople may be relatively unstable in how much they change theirpersonality profile across the years, whereas others may be quitestable in their profile consistency.Teasing apart these two perspectives requires multiple ipsative
consistency estimates. If people are stable in their level of person-ality profile (in)consistency, there should be a strong correspon-dence between ipsative consistency estimates across time. That is,the magnitude of someone’s personality profile consistency (asquantified by a profile correlation), regardless of if it is Q = .30or .80, would remain similar if calculated at two or more separateoccasions. In contrast, if individual differences in personality profileconsistency are due to external forces, there should be less corre-spondence between ipsative consistency estimates across time. Thatis, the magnitude of a single person’s personality consistency couldvary fromQ = .60 to .30 across two separate occasions, perhaps dueto some environmental change occurring between them, whichresults in lower profile consistency.
Processes Underlying Patterns of PersonalityConsistency
The patterns of consistency in personality profiles can also giveinsights into processes driving the consistency. A single profilecorrelation cannot describe what the expected test–retest profilecorrelation should be when time increases, as it only provides asingle point estimate. Should it decrease to the minimum valueexpected by the normativeness of profiles? Should it stay relativelythe same over increasing time periods? Most work with longer test–retest intervals (i.e., decades long; Atherton et al., 2021; Jones &
Meredith, 1996; Roberts & DelVecchio, 2000; Terracciano et al.,2006) stems from between-person, rank-order stability. Rank-orderstability in personality traits increases throughout early and middleadulthood, peaks in midlife, and decreases again in old age and veryold age (Ardelt, 2000; Bleidorn & Hopwood, 2019; Briley &Tucker-Drob, 2014; Lucas & Donnellan, 2011; Roberts &DelVecchio, 2000; Wortman et al., 2012). From this past workon rank-order stability, relatively high levels of profile stabilityacross longer stretches of time might be expected as well. Indeed, aspreviously noted, there are high levels of profile stability acrossnearly 20 years (Terracciano et al., 2010). Yet, these studies do notdirectly examine whether the consistency decreases as a functionof time.
Importantly, by examining the patterns of profile consistency, onecan gain insight into the processes driving personality consistency.Fraley and Roberts (2005) investigated what drives between-person,rank-order stability by using multiple test–retest estimates aggre-gated across different studies of young individuals that varied inlength of test–retest intervals. They found a pattern whereby stabil-ity decreased relatively dramatically and then plateaued to remainstable regardless of the time interval. That is, even when the lengthof the test–retest interval continued to increase, people were ex-pected to have a similar level of consistency in reference to theoriginal time point. Their trajectory reached an asymptote at a test–retest correlation value of approximately .25, suggesting that ifpersonality was assessed many years in the future, the associationwould at least be .25, not zero (Fraley & Roberts, 2005).
Furthermore, they found that three processes are necessary toexplain personality stability: stochastic-contextual processes(e.g., Lewis, 1997, 2001; Revelle & Wilt, 2020), developmentalconstancy factors (e.g., Roberts & Wood, 2006), and person–environment transactions (e.g., Caspi & Bem, 1990; Neyer &Asendorpf, 2001). Each of these three processes was associatedwith a particular pattern of stability trajectories: Stochastic-contextual processes led to decreasing levels of stability, devel-opmental constancy factors led to stable levels of stability, andperson–environment transactions were the only process that couldlead to increasing levels of stability. Thus, under the assumptionthat these three processes reliably lead to these predicted patterns,the direction of observed stability trajectories allows inferences tobe made about the processes underlying personality stability.
The first of these mechanisms, stochastic-contextual processes,indicates that a largely influential force in personality consistency ischance—or unpredictable noise—that serves to decrease consis-tency (Lewis, 1997). For example, a chance run-in with an oldcoworker while someone is looking for a job could precede both anew employment opportunity and rekindling of an old friendship.These unforeseen experiences result in novel opportunities andenvironments, the implication being that these random influenceslead to decreases in consistency across time. Without these mechan-isms at play, the consistency of personality would be quite high.
The second process is developmental constants. In contrast tofactors that may largely ebb and flow in importance, there areconstancies that promote stability in personality (Roberts &Caspi, 2003). These are most traditionally thought of as geneticfactors (Blonigen et al., 2008; McGue et al., 1993; Reiss et al.,2000), though they also may be transformative early life experiences(Fraley et al., 2013) whose impact is constant across the lifespan.These constant factors are assumed to provide a form of long-term
Thisdocumentiscopyrightedby
theAmerican
Psychological
Associatio
nor
oneof
itsalliedpublishers.
Thisarticle
isintended
solely
forthepersonal
useof
theindividual
user
andisnotto
bedissem
inated
broadly.
1316 WRIGHT AND JACKSON
stability. Past work examining the influence of stable factors on test–retest correlations has found that personality traits are slightly lessstable than intelligence (Conley, 1984b) but relatively more stablethan evaluative individual differences (e.g., self-esteem, life satis-faction; Anusic & Schimmack, 2016; Conley, 1984b). This suggeststhat, while some individual differences are indeed more subject tofluctuations due to changing factors, personality traits may bestrongly influenced by stable factors.The third and final mechanism is person–environment transac-
tions. Although someone can certainly be shaped by their environ-ment, characteristics of and actions by the individual in thatenvironment can in turn influence the degree to which the environ-ment impacts them (Caspi & Bem, 1990; Caspi & Roberts, 1999).That is, people are not passive receptacles awaiting environmentalinfluence. Rather, individuals are capable of proactively placingthemselves in environments that align with their personality(Asendorpf & van Aken, 1991; Caspi & Bem, 1990). To the extentthat people are able to actively niche-seek and select environmentsthat align with their personality—and thus mitigate random orstochastic processes—personality should remain relatively consis-tent or even increase in consistency. While transactional processeswith the environment do not necessarily always lead to increases inconsistency, such that there could be no changes or even decreases,they are the only processes of the three that could lead to increases.Fraley and Roberts (2005) concluded that all three processes—
developmental constants, stochastic factors, and transactionalprocesses—are necessary to explain patterns of stability in person-ality. When systematically removing the effect of each of these threeprocesses, though, they found that the removal of stochastic factorshad the largest impact on the predicted trajectories, followed by theremoval of developmental constants. The removal of transactionalprocesses with the environment had the smallest impact, suggestingthese have less of a role in bolstering personality stability comparedto the other two processes.The pattern described by Fraley and Roberts (2005) is thought to
extend to all individuals, as the type of change employed capturesthe population. However, it should be noted that this meta-analyticsample does not include people over the age of 30, making it unclearwhether these trends occur for older individuals. Moreover, thisassertion was based on population-level, between-person traits, notindividual-level, person-centered consistency. As such, it isunknown whether person-centered consistency is guided by similarmechanisms in the same degree. That is, it is unclear if each of thesethree factors contributes similarly to personality consistency indifferent individuals, which would be revealed by examining thedirection of many individual-level trajectories of person-centeredprofile consistency. It is likely that people differ on which of and thedegree to which these factors are influencing them. If, for example, aperson is in a novel environment because they moved to a new city,they may have more stochastic factors affecting consistency relativeto their time in their previous city. Moreover, it could be possiblethat stable trajectories are due to a combination of stochastic factorsand transactional processes, such that their decreasing and increas-ing effects on stability, respectively, cancel each other out and thusno change is observed. When aggregating across many individuals,this pattern of change would be indistinguishable from a purelystable trajectory. However, by examining individual-level trajecto-ries, nuances such as these can be revealed and not lost due toaggregation across people. Overall, the different factors by which
the (in)consistency of personality can be reinforced or elicited giverise to countless ways individuals can be idiosyncratically impacted.That is, these factors do not have to impact everyone similarly norimpact a single person similarly across time. Assuming similarestimates of consistency across individuals who differ in theirenvironments, intrinsic stable tendencies, and the degree to whichthese factors interact can lead to misleading conclusions about thenature of personality consistency.
Potential Factors Impacting Person-CenteredProfile Consistency
The processes influencing personality consistency may differacross people, but other factors may uniquely vary as well. Ageis the most routinely examined factor believed to influence person-ality stability. Personality traits typically become increasingly stablewith increasing age, and this stability is thought to plateau in middleto late adulthood, around age 50 (Ardelt, 2000; Bazana & Stelmack,2004; Roberts & DelVecchio, 2000; Schuerger et al., 1989).Although, the effect of age might be curvilinear such that declinesin stability are observed among older adults (Lucas & Donnellan,2011; Terracciano et al., 2018; Wortman et al., 2012). Cross-agegroup differences in test–retest correlations have also been found inwork on ipsative continuity in young (Donnellan et al., 2007;Roberts et al., 2001; Robins et al., 2001) and middle-aged adults(Terracciano et al., 2010) relative to children and adolescents (DeFruyt et al., 2006; Klimstra et al., 2009). It appears that youngerpeople have lower levels of profile stability, and stability perhapsplateaus around age 30 (McCrae & Costa, 2003; Terracciano etal., 2010).
Additionally, gender differences could be present. In terms ofgender differences for individual personality consistency in the BigFive traits, there appear to be inconsistent effects (e.g., Ozer &Gjerde, 1989; Robins et al., 2001), likely a result of methodologicaland sample differences in past studies (Klimstra et al., 2009). Moststudies report no gender differences (Ozer & Gjerde, 1989; Robinset al., 2001), whereas at least one study examining adolescentsreported large gender differences in both initial values and trendsacross time (Klimstra et al., 2009). Roberts et al. (2001) found slightgender differences, with women being slightly more consistentthan men.
Furthermore, past work has also found that the traits themselvesare important to consider. Single traits can vary with their typicalstability estimates across time and can further interact with oneanother, such that those high or low on some combination of traits atbaseline change in similar ways across time compared to individualswho began with different initial levels (Block, 1971; Morizot & LeBlanc, 2005). For example, with single traits, one study found thatstability coefficients for Extraversion did not differ with age butwere slightly higher in older participants for Neuroticism (Vikenet al., 1994). Levels of the traits themselves also appear to have someinfluential role in the degree to which they change alone or intandem with other traits across time. For instance, studies ofadolescents and young adults have found that lower levels ofConscientiousness and higher levels of Neuroticism, or a somewhat“immature” personality profile, predict lower profile stability acrosstime (Donnellan et al., 2007; Roberts et al., 2001). Traits can alsopredict exposure to certain experiences or life events that precede orserve as catalysts for personality change (e.g., Kendler et al., 2006),
Thisdocumentiscopyrightedby
theAmerican
Psychological
Associatio
nor
oneof
itsalliedpublishers.
Thisarticle
isintended
solely
forthepersonal
useof
theindividual
user
andisnotto
bedissem
inated
broadly.
PERSON-CENTERED PERSONALITY CONSISTENCY 1317
thus leading to lower stability over time (Costa et al., 2005; McCrae,1993; Ozer & Benet-Martínez, 2006; Terracciano et al., 2006).Alternatively, it could lead to enhanced stability across time due toincreased levels of person–environment fit (Beck & Jackson, 2021a;Caspi & Moffitt, 1993). This could be the case, for example, withtraits predicting who goes onto finish their education (Lüdtke et al.,2011), having a degree, which leads someone to a career, and acareer turning into a decades-long stabilizing environmental force.Lastly, another important consideration is the length of the test–
retest intervals. Stability estimates will decrease with longer timeintervals (Fraley & Roberts, 2005; Roberts & DelVecchio, 2000;Terracciano et al., 2006). Notably, this factor is important to note notonly for comparisons across individuals but also for a singleindividual. Within individuals, the increased lapse of time betweenmeasures also allows for more numerous life events and experiencesto be accumulated (Lüdtke et al., 2011). Thus, aside from the generalfinding of decreased stability estimates across greater lengths oftime, there could be individual differences in precisely what occursduring this time that could lead to either enhanced stability, such asthe continuation of a lifelong role or stable environment, ordecreased stability, such as the loss of a lifelong role or dramaticchange in environment.
The Present Study
The consistency of personality has primarily been examined at thegroup level. However, consistency at the between-person levelmight not reflect the patterns of personality consistency withinthose individuals. Instead, ipsative consistency, a neglected aspectof personality development, allows for the examination of anindividual’s own personality configuration across time. Withinthis article, we investigate personality development through thelens of individual-level indices of personality profile similarity andconsistency for Big Five trait items across multiple waves in fourlongitudinal panel studies. We address two primary questions.First, are some people more consistent in their personality
profiles? Previous studies identify individual differences in profileconsistency, but are these individual differences due to person-levelattributes or short-term fluctuations brought on by environmentalchange? Much like individual differences in slope estimates fromgrowth models provide an estimate of individual differences inpeople’s mean-level changes, we investigate individual differencesin profile consistency and similarity. This question cannot beaddressed using only two waves of consistency data, as thosedata do not provide the overall pattern of consistency acrosstime. To investigate this question, we used personality profilesimilarity and consistency indices calculated across sequentialwaves (i.e., Wave 1 to Wave 2, Wave 2 to Wave 3, etc.).Second, are the processes promoting long-term personality con-
sistency and change (stochastic, transactions, stable factors) similarfor each person? Can the general conclusions based on between-person consistency estimates using younger samples (i.e., Fraley &Roberts, 2005) be ascribed to person-centered consistency, and doesthe process occur for everyone? To address this question of theprocesses underlying changes in personality profile consistency, weused profile correlations calculated across increasing time intervals(i.e., Waves 1 and 2, Waves 1 and 3, etc.), so that long-termtrajectories can be examined.
We hypothesize that the general trajectory of ipsative correlationswill be nonlinear in form, decreasing, and asymptote at some modestlevel (Fraley & Roberts, 2005; Terracciano et al., 2006). For our firstresearch question, we are noncommittal whether there are meaning-ful individual trajectories of ipsative consistency across time—thatis, whether some people are stably (in)consistent or whether indi-vidual differences in ipsative are mostly time-specific error. Further,for the second question, we hypothesize that all three mechanisms—stochastic, transactional, and stable factors—will be important.Evidence of stochastic mechanisms can be found via nonperfectestimates of consistency, whereas evidence of stable factors is theopposite, such that relatively unchanging, nonzero associations arefound. Evidence for transactional factors would be observed viaincreasing ipsative trajectories, as these increases can only beexplained by transactional factors.
Method
Participants
In this article, we use data fromN= 21,616 total participants fromfour longitudinal panel data sets (Table 1). The number of partici-pants with four waves was 17,311; five waves was 778; six waveswas 898; seven waves was 834; eight waves was 1,783; and ninewaves was 12. Participants were included in the present study if theyhad at least four waves of personality data for the Big Five traits.Results from attrition analyses can be found in Supplemental FileS1. The institutional review board (IRB) at Washington Universityin St. Louis deemed this project exempt from IRB approval becauseit involves accessing publicly available data sets and thus does notmeet federal definitions under the jurisdiction of an IRB (IRB ID:202201080).
German Socioeconomic Panel Study
The German Socioeconomic Panel (GSOEP) study (Socio-Economic Panel, 2018) is an ongoing longitudinal study conductedby the German Institute of Economic Research (DIW Berlin)collecting data on individuals in more than 11,000 German house-holds. Data are freely available by application at https://www.diw.de/soep. Data collection began in 1984 and continues annually,with the latest release in 2019. The sample from this data setconsisted of N = 6,771 individuals.
Household Income and Labour Dynamics inAustralia Study
The Household Income and Labour Dynamics in Australia(HILDA) Study (Watson & Wooden, 2012) is an ongoing longitu-dinal study collecting data on more than 17,000 individuals inAustralian households. Data are freely available by application athttps://melbourneinstitute.unimelb.edu.au/hilda/for-data-users. Datacollection began in 2001 and has continued annually, with the latestrelease in 2020. The sample from this data set consisted of N = 6,518individuals.
Health and Retirement Study
Health and Retirement Study (HRS; Juster & Suzman, 1995) is anongoing longitudinal study of more than 35,000 individuals from in
Thisdocumentiscopyrightedby
theAmerican
Psychological
Associatio
nor
oneof
itsalliedpublishers.
Thisarticle
isintended
solely
forthepersonal
useof
theindividual
user
andisnotto
bedissem
inated
broadly.
1318 WRIGHT AND JACKSON
households in the United States. Data are freely available at https://hrs.isr.umich.edu. Data collection began in 1992 and continuesbiennially, with the latest release in 2020. The sample from thisdata set consisted of N = 2,688 individuals.
Longitudinal Studies for the Social Sciences
Longitudinal Studies for the Social Sciences (LISS; Scherpenzeel& Das, 2010) is an ongoing longitudinal study of approximately8,000 Dutch-speaking individuals from 5,000 households in theNetherlands. Data are freely available through application at https://statements.centerdata.nl/liss-panel-data-statement. Data collectionbegan in 2007 and have continued annually, with the latest releasein 2021. The sample from this data set consisted of N = 5,639individuals.
Measures
Big Five
The primary variables in this study are Big Five trait items(Goldberg, 1990). All items were scored such that higher scoresindicated greater levels of the trait and lower scores indicated lowerlevels (Neuroticism was keyed as emotional instability1). Thenumber of items and specific content of items varied across studies(see Supplemental Table S1, for psychometric information perstudy), but full content for all items per study can be found inSupplemental File S2.GSOEP. All items were scored on a 1–7 Likert scale (1= “does
not apply” to 7 = “applies fully”). An example item for Extraver-sion, translated to English, is, “I am sociable”; for Agreeableness, “Iam able to forgive”; for Conscientiousness, “I tend to be lazy”(reverse scored); for Neuroticism, “I deal well with stress” (reversescored); and for Openness, “I have a lively imagination.”HILDA. All items were scored on a 1–7 Likert scale (1= “does
not describe me at all” to 7= “describes me very well”). An exampleitem for Extraversion is, “Talkative”; for Agreeableness,
“Sympathetic”; for Conscientiousness, “Orderly”; for Neuroticism,“Calm” (reverse-scored); and for Openness, “Creative.”
HRS. All items asked how well an adjective applied to theparticipants and were scored on a 1–4 Likert scale (1 = “a lot” to4= “not at all”). An example item for Extraversion is, “Talkative”; forAgreeableness, “Sympathetic”; for Conscientiousness, “Organized”;for Neuroticism, “Calm” (reverse-scored); and for Openness,“Creative.”
LISS. All items asked participants to rate how well the descrip-tion applied to themselves and were scored on a 1–5 Likert scale(1 = “very inaccurate” to 5 = “very accurate”). An example item forExtraversion, translated to English, is, “Am the life of the party”; forAgreeableness, “Feel little concern for others” (reverse scored);for Conscientiousness, “Make a mess of things” (reverse scored); forNeuroticism, “Often feel blue”; and for Openness, “Am not inter-ested in abstract ideas” (reverse scored).
Moderators
We examined the effect of four moderators: gender, age, person-ality traits, and education level. For all data sets, gender was adummy variable coded such that 0 =male and 1 = female. Age wascalculated from a participant’s date of birth and considered in 1-yearincrements. For the personality traits, we used the composited BigFive traits (i.e., averaged across the items for each trait) as separatemoderators. Last, the level of education variable was operationa-lized according to if a participant had a college degree from a 4-year
Thisdocumentiscopyrightedby
theAmerican
Psychological
Associatio
nor
oneof
itsalliedpublishers.
Thisarticle
isintended
solely
forthepersonal
useof
theindividual
user
andisnotto
bedissem
inated
broadly.
Table 1Descriptive Information by Study
Variable description GSOEP HILDA HRS LISS Total
Sample size (N ) 6,771 6,518 2,688 5,639 21,616Age (M ) 53.72 50.40 69.40 52.27 54.06Age (SD) 15.23 16.20 9.48 16.97 16.58Age (range) 17–100 15–100 30–102 16–97 15–102Female (%) 53.20 55.00 60.60 53.66 55.00Years of education (M ) 12.80 — 13.35 — —
Years of education (SD) 2.81 — 2.70 — —
Years of education (range) 7–18 — 0–17 — —
Percentage of sample with university degree (%) 21 30 29 44 31No. of waves of personality data (M ) 4.00 4.00 4.00 6.17 4.57No. of waves of personality data (SD) 0.00 0.00 0.04 1.58 1.25No. of waves of personality data (range) 4–4 4 4–5 4–9 4–9Years between waves (M ) 4.00 4.00 3.99 1.78 3.16Years between waves (SD) 0.00 0.00 0.13 0.81 1.19Length of test–retest interval (in years) 4–12 4–12 2–12 1–12 1–12
Note. HILDA and LISS did not have an explicit variable for years of education and thus these cells do not contain values. GSOEP = German SocioeconomicPanel Study; HILDA=Household Income and Labour Dynamics in Australia Study; HRS=Health and Retirement Study; LISS= Longitudinal Studies for theSocial Sciences; M = mean; SD = standard deviation; Age = age across all available waves.
1 It should be noted that, depending on how the traits are coded, profilecorrelation values will nearly always change. As a relative measure ofpersonality consistency, comparisons of raw profile correlation values acrossdifferent datasets/studies are not advisable because of their dependency onseveral study-specific factors, such as the direction of how traits are coded.To confirm our results and inferences did not depend on how we codedNeuroticism, the same calculations and baseline models were run for profilecorrelations where Neuroticism was conceptualized as emotional stabilityinstead. The findings remained unchanged. These results can be found inSupplemental Tables S28–S31.
PERSON-CENTERED PERSONALITY CONSISTENCY 1319
university (or equivalent) or higher. Having a college degree wascoded as 1, whereas not having the degree was coded as 0.
Transparency and Openness
Within this Method section, we report how we determined ourfinal sample size through inclusion criteria, all measures used alongwith their psychometric properties, and we follow the APA StyleJournal Article Reporting Standards (JARS; Kazak, 2018). Data arefreely accessible at all links specified in each respective study’s“Participants” subsection. All code for all data cleaning, dataanalysis, and the codebook for each data set are available athttps://osf.io/23u9b/. Data were analyzed using R, Version 4.0.3(R Core Team, 2021) and the package brms (Bürkner, 2017). Thisstudy’s design and its analyses were not preregistered.
Analytic Plan
The analytic plan consisted of first calculating intraindividualindices of personality profile similarity and consistency and thenconducting analyses to examine interindividual trends in thesevalues. The indices of personality profile similarity calculatedwere the three D metrics first proposed by Cronbach and Gleser(1953), which describe the degree to which a personality profile maychange in elevation (i.e., mean levels), spread (i.e., scatter of scoresaround the mean), and shape of the profile. The index of personalityprofile consistency calculated was the within-person, item-level2
profile correlation (Q). To begin, all data were downloaded directlyfrom the data repositories for each study and cleaned/reverse scoredas necessary. Cronbach’s alpha estimates were calculated using thepsych package (Revelle, 2021).
Intraindividual Indices of Personality ProfileSimilarity and Consistency
D Metrics. Three indices to assess profile similarity are thoseproposed by Cronbach and Gleser (1953): D2, D′2, and D″2. Thefirst of these, D2, is sensitive to differences in elevation, scatter,and shape between two profiles. It can be considered a dissimilarityindex, such that larger values indicate more dissimilar profiles andsmaller values (with the minimum value being 0) indicate moresimilar profiles. Mathematically, it quantifies the squared differ-ence between item responses (xi) across two time points (1 and 2)for all items per person ( j). It is calculated using the followingformula:
D2j =
Xðx1ij − x2ijÞ2: (1)
The second metric, D′2, is sensitive to differences in scatter andshape. It quantifies the squared difference between profiles aftereach item in the profile at the two different time points has beencentered around its mean (hence the insensitivity to changes inelevation), summed across all items per person. The same formula asabove can be used, but with a slight modification:
D′2j =X
½ðx1ij − x1jÞ – ðx2ij − x2jÞ�2: (2)
The third metric, D″2, is sensitive only to differences in shapes ofprofiles. It quantifies the squared difference between profiles at two
time points after each profile has been standardized, thus removingthe influence of the scatter of scores. Again, a similar formula can beused, but this time after substituting standardized scores (zij) for thecentered scores:
D″2j =X
½ðz1ijÞ – ðz2ijÞ�2 (3)
Notably, D″2 is perfectly inversely related to the within-personprofile correlation (Q), such thatD″2j = 2 (1 −Qj). Similarly, each ofthe D metrics can be calculated from each other, such that:
D′2j = D″2j ðs1s2Þ + Δ2s , (4)
D2j = D′2j + kΔ2
m, (5)
whereby, s is the square root of the scatter of scores around the meanat that time point (i.e., the sum of the squared differences), Δ2
s is thesquared difference between the two scatter quantities at the two timepoints, k is the number of items, and Δ2
m is the squared meandifference between the two time points. From these equations, itbecomes clear that these values are not necessarily comparableacross data sets, as they are a function of the range of possibleresponse options and the number of items in the measure. Thus,similar to profile correlations, they are a relative measure of profilesimilarity and should only be interpreted relative to the data set fromwhich the values were obtained.
The D metrics were calculated across sequential waves for eachdata set. To interpret the magnitude of the initial empiricalDmetrics(i.e., theDmetrics calculated from the first two waves of personalitydata per data set), we simulated scores for the Big Five items in eachdata set for 50,000 individuals that had equivalent levels of eleva-tion, scatter, and shape in their personality profiles across two timepoints. This approach parallels what has been done in past research(De Fruyt et al., 2006; Ibáñez et al., 2016; Robins et al., 2001). Thesimulated item scores were constructed such that they had the samemeans, variances, and reliability from the real data for each data set.Then, the Dmetrics were calculated from these simulated scores foreach data set. To determine if our participants showed meaningfulchanges in their elevation, scatter, or shape of their profiles acrossthe first twowaves of personality data, they were classified as havingchanged in these three indices if the probability associated with theirvalues for each D metric was less than 5% (i.e., any other changesnot beyond this point would be due to measurement error alone), asindicated by the distribution of D metrics calculated from thesimulated item scores.
Overall Profile Correlations. Then, individual test–retestprofile correlations for all Big Five trait items were calculatedwithin each study. The multicon package (Sherman & Serfass,2015) in R statistical software was used for calculating all profilecorrelations. There were two formulas used for calculating profilecorrelations. First, overall profile correlations were computed;these are “overall” in the sense that the grand mean for eachitem is not subtracted out from each individual’s scores prior tocalculating the profile correlations. The formula for calculating an
Thisdocumentiscopyrightedby
theAmerican
Psychological
Associatio
nor
oneof
itsalliedpublishers.
Thisarticle
isintended
solely
forthepersonal
useof
theindividual
user
andisnotto
bedissem
inated
broadly.
2 Some past research has also calculated profile correlations at the traitlevel, whichwill lead to different profile correlation values compared to thosecalculated at the item level. If comparisons of these values are of interest forour data sets, profile correlation values calculated at the trait level areavailable in Supplemental Tables S32–S33.
1320 WRIGHT AND JACKSON
overall profile correlation (Qj) can be represented via the followingequation:
Qj =
Pðxij1 − xj1Þðxij2 − xj2ÞffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiP ðxij1 − xj1Þ2P ðxij2 − xj2Þ2
q , (6)
where, xij1 represents an individual’s score for a personality item atone wave; xj1 represents the average of their scores at that wave; xij2represents an individual’s score for a personality item at a secondmeasurement wave; and xj2 represents the average of their scores atthat second measurement wave.Importantly, the expected value of an overall profile correlation is
not 0 (Cronbach & Gleser, 1953; Ozer & Gjerde, 1989). Thus, todetermine if the profile correlation estimates were larger than wouldbe expected by chance, scores on personality profiles were randomlymatched for different participants for the first and second waves ofpersonality data.Distinct Profile Correlations. The second formula used for
calculating profile correlations was that for distinct profile correla-tions. These differ from the above overall profile correlations in thateveryone’s scores for every item are centered around the grandmeanfor that item, thus capturing unique stability separate from that of thestability to be expected from showing a normative personalityprofile. These profile correlations should then be smaller in valuethan overall profile correlations, depending on the degree to whichnormative profiles inflate the overall profile correlations in each dataset. The formula for calculating a distinct profile correlation can berepresented via the following equation:
Qdj =
Pðxij1 − xi1 − xj1Þðxi2 − xi2 − xj2ÞffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiP ðxij1 − xi1 − xj1Þ2P ðxij2 − xi2 − xj2Þ2
q , (7)
where, xij1 represents an individual’s score for a personality item atone wave; xi1 represents the grand mean for an item at one wave; xj1represents the average of a participant’s scores at that wave; xj2represents an individual’s score for a personality item at a secondmeasurement wave; xi2 represents the grand mean for an item at asecond measurement wave; and xj2 represents the average of aparticipant’s scores at that second measurement wave.As previously mentioned, with regards to the waves used in the
calculation, the profile correlations in our study were computed intwo ways. First, test–retest profile correlations were calculatedacross sequential waves, such that the first correlation was betweenthe item scores for Waves 1 and 2, the second correlation wasbetween the scores for Waves 2 and 3, and so forth. These served tohelp examine individual differences in average levels of and changesin person-typical personality profile consistency across waves.Second, profile correlations were calculated from the referencepoint of Wave 1, such that there were always increasing timeintervals across waves. For example, the first test–retest correlationfor an individual would be between their item scores for Waves 1and 2, and their second correlation would be betweenWaves 1 and 3.These allowed us to examine patterns that give insight into theprocesses underlying patterns of consistency across time. In therest of the article, we will refer to the first method of obtainingprofile correlations as sequential and the secondmethod as increasingintervals.
Interindividual Differences in Indices of PersonalityProfile Similarity and Consistency
D Metrics. In addition to evaluating the magnitude of the initialempiricalDmetrics relative to those calculated from simulated trait itemscores, we also examined trends in how the three D metrics changedacross sequential waves using Bayesian multilevel models. Thesemodels were fit as linear3 multilevel models, with measurements nestedwithin individuals. Separate models were run for each D metric. Thegeneric form of our model specification can be seen with the following:
Level 1
Yij = b0j + b1jtimeij + eij: (8)
Level 2
b0j = γ00 + U0j (9)
b1j = γ10 + U1j (10)
The Yij is one of the Dmetrics. The time variable was scaled suchthat a participant’s D metric value (always the D metric between aparticipant’s first and second waves of data) had a value of 0. Thenext set of waves used to calculate the Dmetric (i.e., the second andthird waves of data) had a value of time = 1, the following set had avalue of time = 2 (i.e., the third and fourth waves of data), and soforth. All priors were regularizing priors such that the intercept priorwas centered around the average value of each metric for that dataset and all other priors were centered around 0.
Profile Correlations. Next, we used a Bayesian multilevelmodeling framework to examine the interindividual trends in profilecorrelations (Q) within each data set. All models were fit asnonlinear4 multilevel models, specifically asymptotic nonlinearmodels, with measurements nested within individuals. We choseasymptotic nonlinear models a priori as these would best model thepatterns identified in Fraley and Roberts (2005). The generic form ofour model specification can be seen with the following:
Level 1
Yij = aj − ðaj − bjÞ � eð−cjtimeijÞ: (11)
Level 2
bj = γb0 + Ubj (12)
cj = γc0 + Ucj (13)
where, aj is the maximum possible Yij value (i.e., +1.00); bj is Ywhen timeij = 0 (i.e., the intercept); and cj is the proportional rate ofchange in Yij while timeij increases. The outcome variable Yij was
Thisdocumentiscopyrightedby
theAmerican
Psychological
Associatio
nor
oneof
itsalliedpublishers.
Thisarticle
isintended
solely
forthepersonal
useof
theindividual
user
andisnotto
bedissem
inated
broadly.
3 Nonlinear models were also run for the D metrics to mirror the modelform used for the within-person profile correlations (Q). However, modelcomparisons indicated the linear models were the superior models for thesemetrics. Due to the results of model comparisons and the ease of interpretinglinear models relative to nonlinear models, these are the only models weinclude in the article. Both model types yielded similar conclusions.
4 We additionally ran linear models to ensure conclusions did not changebased on our model form. Results from the linear models can be found inSupplemental Tables S35–S54. However, model comparisons indicated thenonlinear models were the far superior models compared to the linear models(see Supplemental Table S34). Thus, the nonlinear models are the onlymodels we report in the article.
PERSON-CENTERED PERSONALITY CONSISTENCY 1321
the test–retest profile correlation for each individual across waves.Again, the profile correlations were calculated in two ways: acrosssequential waves and across increasing time intervals. The timevariable was scaled such that a participant’s first test–retest correla-tion (always the correlation between a participant’s first and secondwaves of data) had a value of 0. The next set of waves used tocalculate the correlations, which varied across our two researchquestions, had a value of time = 1, the following set had a value oftime = 2, and so forth. The resulting parameters (bj and cj) aresimilar to standard growth model intercept and slope terms, with themain difference that the slope is allowed to be nonlinear andinconstant. All priors were regularizing priors with a Student’s tdistribution centered around 0, with the degrees of freedom set to 3and scale parameter set to 2.5.The effects of potential moderating variables—age, gender,
personality traits, and education level—were also tested. Givenpast research (Lucas & Donnellan, 2011; McCrae & Costa,2003; Terracciano et al., 2010, 2018; Wortman et al., 2012), thelikely effects of age could take multiple forms: linear, curvilinear, ordiscontinuous. Thus, age was tested in these three ways. The agevariable was derived by taking the average age of a participantacross all waves and centering it around the average age among all ofthe participants in their data set. Thus, a participant’s age variablewas representative of how far they deviated from the average age intheir data set across all waves. This particular age variable was testedboth linearly and quadratically and will be referred to as numericage. The third way age was operationalized will be considereddichotomous age, and this variable was obtained by splittingparticipants according to their numeric age variable. If their averageage was 30 or below, they were coded 1 for this variable; if theiraverage age was above 30, they were coded 0 for this variable.Notably, age was only tested in this manner for three of our four datasets, as no participants in HRS had an average age of 30 or younger.For gender, males were coded 0 and females were coded 1. Foreducation level, participants were coded 1 if they had a collegedegree from a 4-year university (or equivalent) or higher and werecoded 0 if they did not.Then, the effects of traits were examined two ways. First, a
between-person trait variable captured the effect of a person deviat-ing from their sample’s average initial levels of each Big Five trait.The initial time point (i.e., baseline) was used instead of the averageacross all waves so that changes occurring across the study wouldsolely be captured as within-person changes. Second, a within-person trait variable captured those within-person changes. Thesewithin-person lag variables captured the difference in a person’sown level of each trait between the two waves used to calculate theirprofile correlation. For example, if we were to determine this valuefor Wave 3, then for the sequential wave models, a lagged variablefor Extraversion would be calculated as the difference between aperson’s level of Extraversion from Wave 2 to Wave 3, whereas inthe increasing interval models, it would be calculated as the differ-ence from Wave 1 to Wave 3.
Results
The D Metrics of Profile Similarity
First, we examined the distributions of the three D indices withinour data sets: D2, D′2, and D″2. Descriptive statistics for each of the
empirically derived D metrics for each data set can be found inSupplemental Tables S2–S7 across all sequential waves. To evalu-ate the magnitude of our empirical D values between the first twowaves of personality data, we simulated data for 50,000 individualsbased on the initial wave of personality data for each data set. Fromthese simulations, we obtained two waves of simulated personalityitem data based on the characteristics of each of our data sets (i.e.,means, standard deviations, reliability of measures) in which eachperson’s profile had equivalent levels of elevation, scatter, and shapeacross the two waves. Similar to one study (Ibáñez et al., 2016), wefound that most changes occurred at the profile level (D″2) asopposed to changes in mean levels (D2) or scatter of scores (D′2)within profiles (Table 2). The percentage of people showing changesin each of the three indices varied across data sets, with GSOEPhaving the fewest overall percentage of people change in any index,whereas LISS had the largest percentage of people changing.
Then, to supplement the above simulations, we further examinedmean-level trends in each data set for the three D metrics5 (Table 3)across all sequential waves. This allowed us to potentially makeinferences about how profile similarity, as quantified by the Dmetrics, was changing across all sequential waves—as opposedto examining only those values calculated from the first two wavesof personality data.
For models predicting D2, GSOEP, HILDA, and LISS hadnegative trends in this metric, such that the sum of squared differ-ences in scores from the mean of a person’s profile decreased acrosstime (i.e., less of an elevation change in one’s profile). HRS, incomparison, had a positive trend in this metric, suggesting thatchanges in elevation slightly increased across time. However, in themodels predicting D (i.e., the square root of D2), the change in thismetric was not meaningful such that its credible interval included 0(Supplemental Table S7). For models predicting D′2, the samepattern emerged such that all data sets except HRS showed negativetrends and HRS showed a positive trend. This suggests that changesin the scatter of individuals’ profiles decreased across time inGSOEP, HILDA, and LISS, whereas changes in scatter increasedacross time for individuals in HRS. Lastly, for models predictingD″2, GSOEP, HILDA, and LISS, all had slightly negative trendsemerge—indicating that changes in the shape of individuals’ pro-files decreased across time. In comparison, HRS had null changes inthis metric across time.
In addition to the mean-level trends in each metric, there wereconsiderable individual differences around these average trends(Table 3). These random effects indicate that not only did indivi-duals have very different starting values for each D metric, but thatthey further greatly varied in how they changed in these metricsacross time.
What Are Typical Levels of Within-PersonProfile Consistency?
Next, we examined if the initial levels of personality profileconsistency, as quantified by within-person, item-level profile cor-relations (Q), were larger than what would be expected by chance.
Thisdocumentiscopyrightedby
theAmerican
Psychological
Associatio
nor
oneof
itsalliedpublishers.
Thisarticle
isintended
solely
forthepersonal
useof
theindividual
user
andisnotto
bedissem
inated
broadly.
5 To ensure the skewed distributions expected from squared values did notproduce biased results, models predicting the unsquared version of eachD metric (i.e., D, D′, and D″) were also run. Results were similar (seeSupplemental Table S7).
1322 WRIGHT AND JACKSON
The empirical initial overall profile correlation estimates, or theprofile correlations between the first and second waves of personal-ity data, are provided in Table 4 (see also Figure 1). Profilecorrelations have an expected value that is larger than zero.Thus, we calculated the expected profile correlation for each sampleby randomly pairing profiles across individuals in each data set forthe first two waves of personality data. These values are presentedin Table 4 and are marked for each data set via the dashed line inFigure 1. For each data set, the empirical profile correlation wasconsiderably larger than the simulated profile correlation value,suggesting there is, on average, notable consistency in item-levelprofiles for our participants.As for the initial distinct profile correlations (Table 4), the values
were somewhat smaller in value, similar to a decrease found in pastresearch as well (Klimstra et al., 2009). This is expected to the extentthat the magnitudes of the overall profile correlations are inflated dueto normative responding. Additionally, it should be noted thatbecause these profile correlations reflect an individual’s own uniqueconsistency, separate from any average-level of consistency ex-pected from the normative profile, that the expected value fordistinct profile correlations is 0.
Are Some People More Consistent ThanOthers in Their Personality Profiles?
Overall and distinct item-level profile correlations averagedacross all sequential waves are presented in Table 5 for each study.The average overall profile correlation values were moderate tolarge by conventional standards, with considerable variabilityaround these estimates. Notably, across all data sets, values from−1.00 to +1.00 were obtained, indicating that there were both
perfectly opposite patterns of responding as well as perfectly similarpatterns of responding. These values indicate that people in generalare relatively consistent in their personality profiles.
Next, we used these profile correlations in a series of nonlinear,asymptotic Bayesian multilevel models.6 These were used to exam-ine individual differences in the average trends of personality profileconsistency in each data set, showing the relative changes acrosstime in person-centered consistency from wave to wave.
Table 6 and Figure 2 describe the results. The intercepts, whichreflect average profile consistency at the initial waves, were similaracross data sets, ranging from .58 (GSOEP) to .69 (HRS). Theslopes, which in these models are the proportional rates of change inthe profile correlations across sequential waves, were positive andhad nonzero values ranging from .03 (HRS/LISS) to .07 (GSOEP/HILDA). This suggests that, on average, levels of profile consis-tency were slightly increasing as opposed to individuals maintainingstable levels of consistency (Figure 2A–D). See Supplemental TableS8, for a comparison of results for overall and distinct profilecorrelations. Notably, these results are consistent with the findingsfrom the mean-level trends in the D metrics, such that, on average,changes in profiles decreased across waves (i.e., profile similaritythus increased). Also consistent with the D metric models is themagnitude of the changes across time (Table 6), such that HRS andLISS show the smallest amount of change.
To illustrate how to interpret the parameters, we will describe thisprocess for two data sets. The overarching equation guiding theinterpretation of these models is Yij = 1 − ð1 − interceptijÞ �eð−slopeij ×timeijÞ. Since these models do not include other predictors,the additional equations for the intercept and slope calculations arenot needed (see the “Analytic Plan” section for full equations). Forthe average effect for GSOEP, the predicted profile correlation attime = 1 (i.e., the second wave of profile correlations), is thus 1 −(1 − .58) × e(−.07×1) = .608 and for time = 2, it is .635. From thesecalculations, we see that the average difference between theintercept and the second wave of profile correlations is .028 unitsand the average difference between the second and third waves ofprofile correlations is .027 units. As for another example, for HILDA,the intercept value is .66, the next predicted profile correlation value attime= 1 is .683 (difference of .023 units), and the following predictedvalue at time = 2 is .704 (difference of .021 units).
These above examples demonstrate a few things. First, theamount of change across waves is proportional to the interceptvalue, such that slope values of −.03 (HRS, LISS) and −.07(GSOEP, HILDA) do not result in the exact same magnitude ofchange between profile correlations across data sets. In general, thelarger the intercept value, the more attenuated future change is.Second, the larger the slope value, the larger the changes will be(holding intercept values constant)—in this regard, this is identicalto the interpretation of the magnitude of slope parameters intraditional growth models. Third, the magnitude of changes willattenuate over time, such that change is not constant across waves.This attenuation is not large but is important to note, as it is one ofthe key features, along with the proportional rate of change relative
Thisdocumentiscopyrightedby
theAmerican
Psychological
Associatio
nor
oneof
itsalliedpublishers.
Thisarticle
isintended
solely
forthepersonal
useof
theindividual
user
andisnotto
bedissem
inated
broadly.
Table 2Descriptive Information for the D Metrics Obtained From Simu-lated Item Scores in Each Data Set
D index Metric
Data set
GSOEP HILDA HRS LISS
D″2
M 0.67 0.42 0.44 0.37SD 0.32 0.13 0.17 0.1095th percentile value 1.28 0.65 0.75 0.56% changed 17.29 41.79 28.27 60.45
D′2
M 46.79 66.79 13.13 28.92SD 18.54 16.53 4.00 6.0895th percentile value 80.87 96.21 20.40 39.64% changed 3.09 19.13 9.56 30.48
D2
M 50.15 68.70 13.65 29.51SD 19.27 16.78 4.10 6.1595th percentile value 85.69 98.62 21.09 40.35% changed 3.77 20.39 11.5 31.97
Note. GSOEP = German Socioeconomic Panel Study; HILDA =Household Income and Labour Dynamics in Australia Study; HRS =Health and Retirement Study; LISS = Longitudinal Studies for the SocialSciences; M = mean; SD = standard deviation; % changed = the percentageof the sample that had D values greater than the value that was the 95thpercentile for that data set, indicating significant changes in that metric (i.e.,the percentage of people in that data set that had values of that metric thatwere greater than would be expected by measurement error alone).
6 The conclusions between the overall and distinct profile correlationswere similar. Thus, for succinctness, we elaborate on the results for theoverall profile correlations in the article, but results for the distinct profilecorrelations are available in Supplemental Tables S8–S17.
PERSON-CENTERED PERSONALITY CONSISTENCY 1323
to the intercept values, that distinguishes these models from linearmodels.Furthermore, the average trend of slightly increasing levels of
profile consistency across waves masks a large amount of variabilityin profile consistency. This is evidenced by the random effects forthe slope, with values ranging from .08 (LISS) to .23 (GSOEP). ForGSOEP, the average of the top 5% of person-level slopes was .36(95% CI [.04, .70]) and the average of the bottom 5% was −.27(95% CI [−.56, .01]). For HILDA, the average of the top 5% ofperson-level slopes was .36 (95% CI [.09, .63]) and the average ofthe bottom 5% was −.28 (95% CI [−.51, −.06]). For HRS, theaverage of the top 5% of person-level slopes was .31 (95% CI [.04,.61]) and the average of the bottom 5% was −.33 (95% CI [−.58,−.09]). For LISS, the average of the top 5% of person-level slopeswas .13 (95% CI [.02, .24]) and the average of the bottom 5% was−.11 (95% CI [−.20, −.01]). These random effects indicate thatsome people decreased in their levels of consistency across waves,whereas others increased more than average. In general, thesefindings indicate that there are individual differences in how con-sistent personality is and how this consistency changes acrosswaves. Some people increase in consistency across time, somepeople become more inconsistent, and many people are stablyconsistent (see Figure 2A–D). For the latter, this does not meanthat their personality does not change, but the extent to which theirpersonality does or does not change is similar from wave to wave—they are stable in their level of consistency.Next, we ran a series of models to examine possible moderators of
these trends (Table 7; see Supplemental Tables S9–S17, for allmoderator results with distinct profile correlations as well). First, we
examined the potential effects of age and gender. In all modelswhere it was included, gender inconsistently moderated interceptvalues, varying both in direction and magnitude across the data sets,and never had any effect on slopes. In the models with the lineareffect of numeric age, there were no effects of age on the interceptvalues nor slopes, meaning numeric age did not impact averagelevels of personality profile consistency nor changes in consistencyacross waves (Table 7; Supplemental Table S9). In the models withthe additional quadratic effect of numeric age, there were again noage effects (Table 7; Supplemental Table S10).
In the models including dichotomous age—that is, distinguishingbetween participants with average ages 30 and younger versus thoseolder than 30—age impacted both the intercept values and the slopes(Table 7; Supplemental Table S11). The effect of age on interceptvalues indicated that participants aged 30 or younger had loweraverage values of personality profile consistency, with profilecorrelations ranging from .05 (GSOEP) to .11 (HILDA/LISS) unitslower. For participants with average ages over age 30, slope valuesranged from .02 (LISS) to .06 (GSOEP/HILDA). The interaction ofage and slope indicated that participants with average ages 30 andunder increased at a greater rate in their levels of profile consistencyrelative to older participants. Specifically, the younger groupexperienced increases ranging from .03 (LISS) to .08 (GSOEP/HILDA) units.
For the models with personality traits, results of the greatestmagnitude and that emerged most frequently across data sets werefound for Agreeableness, Conscientiousness, and Neuroticism—
traits associated with maturity (Table 7; see SupplementalTables S12–S16, for all trait results). For Agreeableness and
Thisdocumentiscopyrightedby
theAmerican
Psychological
Associatio
nor
oneof
itsalliedpublishers.
Thisarticle
isintended
solely
forthepersonal
useof
theindividual
user
andisnotto
bedissem
inated
broadly.
Table 3Linear Trends in the D Metrics Across Sequential Waves
Model parameter
GSOEP HILDA HRS LISS
Est CI Est CI Est CI Est CI
D″2
Person levelIntercept SD 0.38 [0.37, 0.39] 0.36 [0.35, 0.37] 0.32 [0.31, 0.34] 0.31 [0.30, 0.31]Time SD 0.18 [0.17, 0.19] 0.13 [0.13, 0.14] 0.14 [0.13, 0.15] 0.05 [0.05, 0.05]Correlation −0.53 [−0.56, −0.49] −0.39 [−0.42, −0.35] −0.32 [−0.38, −0.26] −0.32 [−0.36, −0.28]
Sample levelIntercept 0.83 [0.82, 0.84] 0.67 [0.66, 0.68] 0.61 [0.60, 0.63] 0.71 [0.70, 0.72]Time −0.04 [−0.04, −0.03] −0.03 [−0.03, −0.02] 0.00 [−0.01, 0.01] −0.01 [−0.01, −0.01]
D′2
Person levelIntercept SD 16.46 [16.00, 16.93] 41.70 [40.71, 42.72] 6.10 [5.85, 6.36] 18.46 [18.03, 18.91]Time SD 7.89 [7.51, 8.26] 17.09 [16.29, 17.88] 2.79 [2.58, 2.99] 2.62 [2.47, 2.77]Correlation −0.53 [−0.56, −0.49] −0.34 [−0.38, −0.31] −0.36 [−0.42, −0.30] −0.56 [−0.60, −0.52]
Sample levelIntercept 30.13 [29.64, 30.61] 69.15 [68.00, 70.31] 11.62 [11.34, 11.89] 36.22 [35.67, 36.76]Time −1.53 [−1.82, −1.24] −1.58 [−2.18, −0.98] 0.23 [0.07, 0.39] −0.61 [−0.73, −0.49]
D2
Person levelIntercept SD 18.71 [18.19, 19.23] 44.09 [43.05, 45.16] 7.59 [7.27, 7.91] 19.64 [19.18, 20.12]Time SD 8.80 [8.37, 9.23] 18.27 [17.44, 19.09] 3.73 [3.49, 3.98] 2.78 [2.61, 2.94]Correlation −0.54 [−0.57, −0.51] −0.34 [−0.37, −0.30] −0.40 [−0.45, −0.33] −0.57 [−0.61, −0.54]
Sample levelIntercept 33.90 [33.35, 34.45] 73.29 [72.05, 74.54] 13.15 [12.80, 13.50] 38.04 [37.46, 38.62]Time −1.98 [−2.32, −1.66] −2.07 [−2.75, −1.42] 0.26 [0.04, 0.47] −0.68 [−0.81, −0.55]
Note. GSOEP=German Socioeconomic Panel Study; HILDA=Household Income and Labour Dynamics in Australia Study; HRS=Health and RetirementStudy; LISS = Longitudinal Studies for the Social Sciences; Est = the maximum a posteriori (MAP) estimate; CI = 95% credible intervals; SD = standarddeviation. Bolded values indicate parameter estimates that do not include 0 in the credible intervals.
1324 WRIGHT AND JACKSON
Conscientiousness, having higher than average levels at Wave 1 ofpersonality data led to considerably higher levels of initial consis-tency, whereas the opposite effect was found for Neuroticism. Thoseindividuals that had higher than average initial levels of Agreeable-ness and sometimes Conscientiousness experienced less of a relativeincrease in their wave-to-wave profile consistency. In comparison,those who had higher than average levels of Neuroticism experi-enced greater relative increases in consistency across waves. Last,individuals who increased in Agreeableness and Conscientiousnessshowed larger increases in their consistency across waves, whereasthose who increased in Neuroticism decreased in their consistencyacross waves.Last, we examined the effect of education on individual differ-
ences in person-typical levels of profile consistency (Table 7;Supplemental Table S17). Across all data sets, having a collegedegree was associated with higher initial values of profile consis-tency, ranging from .04 (HRS/LISS) to .06 (GSOEP) units. Then,for all data sets except GSOEP, having a college degree wasassociated with greater increases in consistency across waves rela-tive to those in the data set that did not have a degree, with effectsranging from .01 (LISS) to .04 (HILDA/HRS) units.
Which Processes Contribute to Personality Consistency?
Next, profile correlations were calculated across increasing timeintervals. Average item-level profile correlations are presented inTable 8 for overall and distinct correlations for each data set.Average values were lower than the previous profile correlationsbut still high given they were calculated across much longer timespans.
Then, we again used the above profile correlations in a series ofnonlinear, Bayesian multilevel models (Table 9; Figure 3A–D).These models served to examine the interindividual differences inthe trends that represent the processes underlying consistency withineach data set.
The slopes, which now represent the proportional rate of changein the profile correlations while the length of the test–retest intervalalways increases, were now negative and had nonzero valuesranging from −.05 (HRS) to −.03 (all but HRS). This showsthat, on average, levels of profile consistency decreased as thislength of time increased, consistent with the idea that stochasticfactors accumulate across time. However, the declines were notlarge, suggesting that people do not drastically drop in their profilecorrelations within their lifetime, even to the minimum level ofexpected profile consistency due to profile normativeness. SeeSupplemental Table S18 for a comparison of results for overalland distinct profile correlations.
There was again variability around the slopes (Figure 3A–D), butto a lesser degree than there was with the sequential wave models,despite the longer time intervals. The standard deviation valuesaround the slope estimates ranged from .04 (LISS) to .16 (HRS),indicating that some people were more greatly decreasing in theirlevel of consistency while others were increasing at more rapid ratesacross time. To illustrate these differences, we estimated an extremetop and bottom group. For GSOEP, the average value within the top5% of person-level slopes was .12, whereas the bottom 5% was−.22. Similar findings for the top 5% and bottom 5% of person-levelslopes were found for HILDA (.12 and −.22), HRS (.13 and −.28),and for LISS (.02 and −.09), respectively. Surprisingly, manypeople evidenced positive trajectories. In GSOEP, 29.9% of in-dividuals had positive slopes; 27.9% in HILDA had positive slopes;23.9% in HRS had positive slopes; and 8.6% in LISS had positiveslopes. Importantly, these increasing trajectories can only occur dueto transactional processes with the environment, as neither stochas-tic nor constant factors can lead to increases (Fraley & Roberts,2005). Overall, these differences in trajectories indicate a sizeableamount of people increase, whereas a large number also decrease inconsistency. This variability highlights that people differ in theprocesses that promote personality consistency. Furthermore, theproportion of individuals with positive slopes suggests that transac-tional factors have a larger role in promoting within-person consis-tency for some individuals—a nuance not revealed from averagefindings of between-person consistency.
Then, we examined possible moderators of interindividual dif-ferences in the trends representative of the processes underlyingprofile consistency within each data set (Table 10; see SupplementalTables S19–S27, for all moderator results with distinct profilecorrelations as well). The effects of each moderator on interceptvalues are equivalent to those in the sequential wave models, as theintercept remains the same (i.e., it is always the within-person profilecorrelation between waves one and two of personality data). Thus,we restrict our discussion of these effects to the impact of themoderators on changes in profile consistency over the longer termstretches of time. Similar to the models with profile correlationscalculated across sequential waves, gender never had any effect onthe slopes (Table 10). Age effects for the numeric age variables werealso similar to the sequential wave models (Supplemental TablesS19–S20). In the models including dichotomous age, there wereagain effects of age on the slopes, although the effects were now
Thisdocumentiscopyrightedby
theAmerican
Psychological
Associatio
nor
oneof
itsalliedpublishers.
Thisarticle
isintended
solely
forthepersonal
useof
theindividual
user
andisnotto
bedissem
inated
broadly.
Table 4Comparison of Initial Empirical Values and Values Obtained FromRandomly Pairing Profiles for Overall and Distinct Item-LevelProfile Correlations
Method ofcalculation
Type of profile correlation
Overall Distinct
M SD Range M SD Range
GSOEPEmpirical 0.59 0.25 −0.78 to 1.00 0.44 0.31 −0.77 to 1.00Random 0.27 0.29 −0.79 to 0.93 0.00 0.32 −0.86 to 0.92
HILDAEmpirical 0.66 0.21 −0.67 to 0.98 0.51 0.23 −0.56 to 1.00Random 0.35 0.27 −0.69 to 0.93 0.00 0.26 −0.75 to 0.78
HRSEmpirical 0.69 0.20 −0.98 to 1.00 0.49 0.24 −0.64 to 1.00Random 0.40 0.25 −0.74 to 0.96 0.00 0.26 −0.80 to 0.73
LISSEmpirical 0.65 0.18 −0.32 to 0.95 0.51 0.20 −0.50 to 1.00Random 0.29 0.23 −0.65 to 0.84 0.00 0.22 −0.70 to 0.79
Note. Empirical profile correlations are those obtained from the actual datasets and calculated between the first and second waves of personality data.Random profile correlations are those obtained from randomly shufflingpersonality item profiles for the first and second waves of personality data ineach data set. GSOEP = German Socioeconomic Panel Study; HILDA =Household Income and Labour Dynamics in Australia Study; HRS = Healthand Retirement Study; LISS = Longitudinal Studies for the Social Sciences;M = mean; SD = standard deviation.
PERSON-CENTERED PERSONALITY CONSISTENCY 1325
smaller in magnitude and inconsistently present across data sets(Table 10; Supplemental Table S21). For participants over age 30,slope values ranged from −.03 (GSOEP) to −.02 (HILDA/LISS),indicating that their levels of consistency slightly decreased as timewent on. For HILDA and LISS, younger participants had greaterrates of decline in their levels of consistency as time increasedrelative to older participants.For the models with personality traits, the largest and most frequent
results were again found for Agreeableness, Conscientiousness, andNeuroticism (Table 10; see Supplemental Tables S22–S26, for all traitresults). Individuals that had higher than average initial levels ofAgreeableness and Conscientiousness often bucked the average trendand experienced increases in their consistency across time or showedno changes. Similarly, those who increased in Agreeableness andConscientiousness relative to their initial values also showed
increases in their long-term levels of personality profile consistency.In comparison, those who had higher than average levels of Neuroti-cism at Wave 1 and/or increased in Neuroticism relative to their traitlevel at Wave 1 experienced greater decreases in long-term consis-tency across time, thus exacerbating the average declining effect.Last, and in contrast to the sequential wave models, there were noeffects of having a college degree on longer term changes inconsistency (Table 10; Supplemental Table S27).
Discussion
We tested two primary questions using indices of personalityprofile similarity and consistency across four large-scale data sets.First, are there stable individual differences in levels of person-typical personality profile consistency? Results indicated that people
Thisdocumentiscopyrightedby
theAmerican
Psychological
Associatio
nor
oneof
itsalliedpublishers.
Thisarticle
isintended
solely
forthepersonal
useof
theindividual
user
andisnotto
bedissem
inated
broadly.
Figure 1Density Plots of Initial, Overall Profile Correlations Calculated in Each Data Set
Note. Average values for the initial, overall profile correlations in each data set are presented on the x axis. The solid line represents the average empiricalvalue in each data set. The dashed line represents the profile correlation value obtained from randomly pairing profiles in each data set (i.e., the expectedminimum value due to some degree of normativeness). Panel A is for GSOEP, Panel B for HILDA, Panel C for HRS, and Panel D for LISS. GSOEP =German Socioeconomic Panel Study; HILDA = Household Income and Labour Dynamics in Australia Study; HRS = Health and Retirement Study; LISS =Longitudinal Studies for the Social Sciences.
1326 WRIGHT AND JACKSON
were stable in how (in)consistent they are, with some people moregreatly changing in their personality, on average, whereas otherstend not to change much wave after wave. How consistent some-one’s personality is differs across people, with some people chang-ing often (and consistently so), whereas others are stably consistentacross time. These findings suggest that (in)consistency is bestconsidered as a person-level attribute rather than one that isenvironmentally induced or due to measurement error.Second, are the processes promoting long-term personality profile
consistency and change (stochastic, transactions, stable factors)similar for each person? Results indicated that there were, onaverage, high levels of long-term consistency across increasingtime intervals. These findings seem to suggest that there aremore stable and transactional factors at play rather than stochasticfactors, thus differing from the relative impact of each of thesefactors for between-person consistency (Fraley & Roberts, 2005).Moreover, individual differences in the longitudinal profile
consistency trajectories suggest that the processes that promotepersonality consistency differ across people.
People Differ in Dispositional PersonalityProfile Consistency
We found profile correlations across time were consistent within aperson rather than different across each time period. Although somepeople may be inherently more consistent than others (i.e., havinglarger test–retest profile correlations), people in general are stable intheir own person-typical level of consistency. Those people that areless consistent are less consistent wave-to-wave, and those that aremore consistent tend to be highly consistent across waves. Yet, theseconsistency levels are not set in stone, as changes in them exist suchthat people can increase and decrease in their consistency levelsacross time. These findings suggest important features drivingpersonality development.
Thisdocumentiscopyrightedby
theAmerican
Psychological
Associatio
nor
oneof
itsalliedpublishers.
Thisarticle
isintended
solely
forthepersonal
useof
theindividual
user
andisnotto
bedissem
inated
broadly.
Table 5Descriptive Information for the Overall/Distinct Item-Level Profile Correlations Calculated Across Sequential Waves in Each Data Set
Study Average
Waves used for profile correlation
1–2 2–3 3–4 4–5 5–6 6–7 7–8 8–9
GSOEPM .60/.46 .59/.44 .60/.46 .62/.48SD .25/.30 .25/.31 .25/.30 .24/.30Range −.81 to 1.00 −.78 to 1.00 −.75 to .99 −.81 to 1.00
HILDAM .68/.52 .66/.51 .69/.53 .69/.54SD .21/.23 .21/.23 .21/.23 .21/.23Range −.84 to 1.00 −.67 to 1.00 −.54 to 1.00 −.84 to 1.00
HRSM .69/.50 .69/.49 .70/.51 .69/.50 .61/.30SD .20/.24 .20/.24 .19/.23 .21/.24 .27/.20Range −.98 to 1.00 −.98 to 1.00 −.66 to 1.00 −.77 to 1.00 .16 to .87
LISSM .66/.52 .65/.51 .65/.51 .66/.51 .67/.53 .67/.53 .69/.53 .71/.55 .67/.56SD .18/.20 .18/.20 .18/.20 .19/.20 .18/.20 .19/.20 .17/.20 .17/.20 .20/.19Range −1.00 to 1.00 −.50 to 1.00 −1.00 to 1.00 −.66 to 1.00 −.69 to 1.00 −.63 to 1.00 −.63 to 1.00 −.47 to 1.00 .15 to .87
Note. Average overall and distinct profile correlations calculated across sequential waves are presented for each data set with the standard deviation and range.The averages and standard deviations for the profile correlations are presented such that the overall values are presented first and followed by the distinct valuesafter the/(i.e., overall/distinct). For the ranges, the largest values across both types of profile correlations are presented. GSOEP=German Socioeconomic PanelStudy; HILDA=Household Income and Labour Dynamics in Australia Study; HRS=Health and Retirement Study; LISS= Longitudinal Studies for the SocialSciences; M = mean; SD = standard deviation.
Table 6Models for Individual Differences in Personality Consistency Across Sequential Waves
Model parameter
GSOEP HILDA HRS LISS
Est CI Est CI Est CI Est CI
Person levelIntercept SD .19 [.18, .19] .18 [.18, .19] .16 [.16, .17] .15 [.15, .16]Time SD .23 [.22, .24] .21 [.20, .22] .22 [.21, .23] .08 [.08, .08]Correlation −.41 [−.45, −.37] −.26 [−.29, −.22] −.27 [−.32, −.21] −.19 [−.23, −.15]
Sample levelIntercept .58 [.57, .58] .66 [.65, .66] .69 [.68, .70] .64 [.64, .65]Time .07 [.06, .08] .07 [.07, .08] .03 [.02, .04] .03 [.02, .03]
Note. GSOEP=German Socioeconomic Panel Study; HILDA=Household Income and Labour Dynamics in Australia Study; HRS=Health and RetirementStudy; LISS = Longitudinal Studies for the Social Sciences; Est = the maximum a posteriori (MAP) estimate; CI = 95% credible intervals; SD = standarddeviation. Bolded values indicate parameter estimates that do not include 0 in the credible intervals.
PERSON-CENTERED PERSONALITY CONSISTENCY 1327
First, given the long timeline of our studies, these findings aresome of the strongest to date that identify consistency as anindividual difference (cf., Bem & Allen, 1974). There is a richdebate of understanding personality coherence across time frames asshort as weeks and up to a year (Beck & Jackson, 2021b), but mostresearch on individual differences in consistency addresses issuesstemming from the personality–situation debate (such as consis-tency across contexts) rather than directly investigating whethersome people are more or less consistent over long periods of time.As such, consistency has typically been viewed as being one of twoextremes: whether personality is or is not consistent across situa-tions. Viewed this way, situations were the focus of what does ordoes not contribute to consistency rather than consistency poten-tially being an intrinsic property of the person. These findingsdemonstrate that over long swaths of time, some people are moreconsistent than others in their personality profiles, suggesting thatone’s level of consistency is a dispositional characteristic. Impor-tantly, these effects do not mean that personality does not change,but rather that the extent that someone’s personality does or does notchange is similar from wave-to-wave or year-to-year. Some peopleare more changeable, whereas others are more immutable.
Second, these findings argue against external forces leading todifferent levels of consistency. The preponderance of two-waveipsative profile correlation designs in previous investigations ofindividual level consistency (e.g., Asendorpf & van Aken, 1991;Terracciano et al., 2010) makes it possible that individual differ-ences in consistency are due to differences in environments orrandom error. Instead, our multiwave findings indicate that internalmechanisms are partially responsible for driving personality consis-tency, as examining profile correlations across multiple wavesacross multiple years averages out environmental and transientfluctuations. In any one year or so, different life events could yieldchanges in personality (Specht et al., 2011), but repeated assess-ments over roughly a decade suggest that environmental upheaval isless likely to be the root cause of these individual differences. Thisidea, in conjunction with our above first point, possibly alludes to anidiosyncratic set point of personality consistency for individuals. Setpoint theories in personality have previously been discussed in thecontext of the interplay of factors such as genetics, states, traits, andlife events (Headey, 2008; Lucas et al., 2003), with life eventshaving only a temporary effect (Roberts, 2018). Our findings alignwith the idea that environmental influences, if and when present, do
Thisdocumentiscopyrightedby
theAmerican
Psychological
Associatio
nor
oneof
itsalliedpublishers.
Thisarticle
isintended
solely
forthepersonal
useof
theindividual
user
andisnotto
bedissem
inated
broadly.
Figure 2Sample- and Person-Level Trends in Personality Profile Consistency Across Sequential Waves
Note. The dashed black line represents the average, sample-level effect. For the person-level trends, a random subset of 100 participants is plotted for eachdata set. The wave number represents the wave of the profile correlation and is centered around 0 such that wave = 0 is the profile correlation between the firstand second measurement occasions. GSOEP =German Socioeconomic Panel Study; HILDA =Household Income and Labour Dynamics in Australia Study;HRS = Health and Retirement Study; LISS = Longitudinal Studies for the Social Sciences.
1328 WRIGHT AND JACKSON
not appear to strongly alter individuals’ typical personality consis-tency trajectories (see also Jackson and Beck, 2021).Third, in addition to the relatively large individual differences in
average levels of consistency, there were individual differences in thetrajectories of change in consistency across waves. While on averagemost people are stably consistent in their personality profiles, noteveryone showed this pattern. Some people became relatively moreconsistent to a greater degree while others became relatively lessconsistent in their personality profiles from wave to wave. These
changes in consistency for the latter individuals may point to theinfluence of life experiences for those people. It is becoming increas-ingly clear that life events are not simply the slings and arrows thatpush around personality consistency. Life events appear to haveminimal impact on personality development at the aggregate level(Jackson & Beck, 2021). Instead, it is possible that life experiencesmay be better viewed as a person-centered issue, one where the effectsof events are unique to each person. Those who showed increases ofgreater magnitude could be growing more accustomed to their
Thisdocumentiscopyrightedby
theAmerican
Psychological
Associatio
nor
oneof
itsalliedpublishers.
Thisarticle
isintended
solely
forthepersonal
useof
theindividual
user
andisnotto
bedissem
inated
broadly.
Table 7Models for Individual Differences in Personality Consistency Across Sequential Waves With Moderators
Model parameter
GSOEP HILDA HRS LISS
Est CI Est CI Est CI Est CI
Numeric ageIntercept .59 [.58, .60] .63 [.62, .63] .67 [.66, .68] .64 [.63, .64]Num. age .00 [−.00, 00] .00 [.00, .00] .00 [.00, .00] .00 [.00, .00]Gender −.02 [−.03, −.01] .06 [.05, .07] .04 [.02, .05] .01 [.00, .02]Time .07 [.06, .08] .07 [.06, .09] .03 [.01, .05] .03 [.02, .03]Time × Num. age −.00 [−.00, −.00] −.00 [−.00, −.00] −.00 [−.00, −.00] −.00 [−.00, −.00]Time × Gender −.01 [−.02, .01] −.00 [−.02, .01] .00 [−.03, .03] −.00 [−.01, .01]
Numeric age (quadratic)Intercept .59 [.58, .60] .68 [.67, .68] .69 [.68, .70] .67 [.67, .68]Age .00 [−.00, .00] .00 [.00, .00] .00 [−.00, .00] .00 [.00, .00]Age2 −.00 [−.00, −.00] −.00 [−.00, −.00] −.00 [−.00, .00] −.00 [−.00, −.00]Time .06 [.05, .07] .07 [.06, .08] .03 [.02, .05] .03 [.02, .03]Time × Num. age −.00 [−.00, −.00] −.00 [−.00, −.00] −.00 [−.00, −.00] −.00 [−.00, −.00]Time × Num. age2 .00 [.00, .00] .00 [−.00, .00] −.00 [−.00, .00] −.00 [−.00, .00]
Dichotomous ageIntercept .59 [.58, .60] .64 [.64, .65] .65 [.64, .66]Dich. age −.05 [−.07, −.03] −.11 [−.12, −.09] −.11 [−.12, −.09]Gender −.02 [−.03, −.01] .06 [.05, .07] .01 [−.00, .02]Time .06 [.05, .08] .06 [.05, .08] .02 [.02, .03]Time × Dich. age .08 [.05, .11] .08 [.06, .10] .03 [.02, .04]Time × Gender −.01 [−.02, .01] −.00 [−.02, .01] −.00 [−.01, .01]
AgreeablenessIntercept .58 [.57, .59] .66 [.65, .66] .69 [.68, .69] .64 [.64, .65]Trait T1 .04 [.04, .05] .10 [.10, .11] .15 [.14, .17] .10 [.09, .10]Time .07 [.06, .08] .08 [.07, .08] .03 [.02, .04] .03 [.02, .03]Time × Trait T1 −.03 [−.04, −.02] −.03 [−.04, −.02] −.05 [−.07, −.02] −.02 [−.02, −.01]Time × Trait lags .01 [.01, .02] .03 [.03, .04] .05 [.04, .07] .00 [−.00, .01]
ConscientiousnessIntercept .58 [.57, .58] .66 [.65, .66] .69 [.68, .69] .64 [.64, .65]Trait T1 .06 [.05, .06] .08 [.08, .09] .16 [.14, .17] .10 [.09, .11]Time .07 [.06, .08] .08 [.07, .08] .03 [.02, .05] .03 [.02, .03]Time × Trait T1 −.04 [−.05, −.03] −.02 [−.03, −.01] .00 [−.02, .03] −.01 [−.01, −.00]Time × Trait lags .03 [.03, .04] .02 [.01, .02] .07 [.05, .08] .01 [.00, .01]
NeuroticismIntercept .58 [.57, .58] .66 [.66, .66] .69 [.68, .69] .64 [.64, .65]Trait T1 −.05 [−.05, −.04] −.11 [−.11, −.11] −.14 [−.15, −.13] −.10 [−.11, −.09]Time .07 [.06, .08] .08 [.07, .08] .03 [.02, .04] .03 [.02, .03]Time × Trait T1 .02 [.01, .02] .03 [.03, .04] .04 [.01, .06] .01 [.01, .02]Time × Trait lags −.02 [−.02, −.01] −.02 [−.02, −.01] −.06 [−.07, −.05] −.01 [−.01, −.01]
EducationIntercept .57 [.57, .58] .65 [.64, .65] .68 [.67, .69] .63 [.62, .63]Education .06 [.05, .07] .05 [.04, .06] .04 [.03, .06] .04 [.02, .04]Time .06 [.05, .07] .06 [.05, .07] .01 [−.00, .03] .02 [.01, .02]Time × Education .03 [.00, .05] .04 [.02, .06] .04 [.01, .07] .01 [.01, .02]
Note. To reduce table length, only the sample-level effects (i.e., fixed effects) for each moderator are presented in this table. Please see Supplemental TablesS9–S11, S13–S15, and S17, for full results including the person-level effects (i.e., random effects). GSOEP = German Socioeconomic Panel Study; HILDA =Household Income and Labour Dynamics in Australia Study; HRS =Health and Retirement Study; LISS = Longitudinal Studies for the Social Sciences; Est =the maximum a posteriori (MAP) estimate; CI = 95% credible intervals; Trait T1 = between-person trait variable, indicating the difference between a person’slevel of a trait atWave 1 relative to the data set’s average level atWave 1; Trait lags=within-person trait variable, indicating a person’s deviation in their level ofa trait between the waves used to calculate a profile correlation. Bolded values indicate parameter estimates that do not include 0 in the credible intervals.
PERSON-CENTERED PERSONALITY CONSISTENCY 1329
environment, perhaps through niche-seeking, habituation of a routine,or increasing familiarity with an environment. In comparison, thosewho showed decreases in wave-to-wave consistency could haveexperienced long-term disruptions in their environment that offsettheir intrinsic level of consistency. These possibilities highlight thecomplex relationship between environmental contexts and their re-inforcing or detrimental effects on consistency.Fourth, we found no evidence of numeric age contributing to any
meaningful between-person differences in personality profile consis-tency, despite being associated with between-person estimates ofstability (e.g., Bazana & Stelmack, 2004; Roberts & DelVecchio,2000). However, we did find that those younger than 30 had differentprofile consistency than those older than 30, again establishing thatconsistency seems to plateau after age 30 for most people(Terracciano et al., 2010). It is possible that age is more importantin younger samples for both profile correlations (e.g., Klimstra et al.,2009) and rank-order correlations (Roberts & DelVecchio, 2000), as
our samples were relatively older. The average age for all of oursamples was above the age that personality stability is considered toplateau (i.e., age 30; McCrae & Costa, 2003). Thus, the majority ofparticipants in our samples could have been overwhelmingly “stable”in their personalities—or stably consistent—such that age effects forparticipants who were outliers for age were masked. It is for thisreason that our results should not be interpreted as age never affectsprofile consistency; rather, at some point, age appears to no longermoderate the degree of consistency in one’s personality profile or thechanges in it over time.
Combining the age 30 results with those of the trait results, pastwork on personality development, particularly that pertaining to thematurity–stability hypothesis (Roberts et al., 2003), can be usedto speculate on the reasons both effects emerge. First, “mature”changes in some traits are to be expected with increasing age.Normative age trends found in past research indicate that Agree-ableness and Conscientiousness tend to increase with age, whereas
Thisdocumentiscopyrightedby
theAmerican
Psychological
Associatio
nor
oneof
itsalliedpublishers.
Thisarticle
isintended
solely
forthepersonal
useof
theindividual
user
andisnotto
bedissem
inated
broadly.
Table 8Descriptive Information for Overall/Distinct Item-Level Profile Correlations Calculated Across Increasing Intervals
Study Average
Waves used for profile correlation
1–2 1–3 1–4 1–5 1–6 1–7 1–8 1–9
GSOEPM .57/.40 .59/.44 .57/.40 .54/.38SD .26/.31 .25/.31 .26/.32 .26/.32Range −.87 to 1.00 −.78 to 1.00 −.84 to .99 −.87 to 1.00
HILDAM .65/.48 .66/.51 .64/.48 .63/.46SD .22/.24 .21/.23 .22/.24 .23/.24Range −.72 to 1.00 −.67 to 1.00 −.72 to 1.00 −.71 to 1.00
HRSM .67/.46 .69/.49 .64/.47 .64/.44 .69/.42SD .21/.24 .20/.24 .20/.24 .22/.25 .17/.35Range −.98 to 1.00 −.98 to 1.00 −.63 to 1.00 −.67 to 1.00 .50 to .93
LISSM .62/.52 .65/.51 .63/.51 .61/.51 .61/.53 .60/.53 .60/.54 .61/.55 .61/.56SD .19/.20 .18/.20 .19/.20 .19/.20 .20/.20 .20/.20 .20/.20 .19/.20 .17/.19Range −.84 to 1.00 −.50 to 1.00 −.84 to 1.00 −.66 to 1.00 −.69 to 1.00 −.63 to 1.00 −.47 to 1.00 −.35 to 1.00 .15 to .88
Note. Average overall and distinct profile correlations calculated across increasing time intervals are presented for each data set with the standard deviation andrange. The averages and standard deviations for the profile correlations are presented such that the overall values are presented first and followed by the distinctvalues after the/(i.e., overall/distinct). For the ranges, the largest values across both types of profile correlations are presented. GSOEP=German SocioeconomicPanel Study. HILDA=Household Income and Labour Dynamics in Australia Study; HRS=Health and Retirement Study; LISS= Longitudinal Studies for theSocial Sciences; M = mean; SD = standard deviation.
Table 9Models for Processes Underlying Personality Consistency
Model parameter
GSOEP HILDA HRS LISS
Est CI Est CI Est CI Est CI
Person levelIntercept SD .18 [.18, .19] .18 [.18, .19] .16 [.16, .17] .16 [.15, .16]Slope SD .15 [.14, .16] .13 [.12, .14] .16 [.15, .17] .04 [.04, .05]Correlation −.21 [−.27, −.15] −.12 [−.17, −.08] −.19 [−.25, −.12] −.07 [−.12, −.02]
Sample levelIntercept .58 [.57, .59] .66 [.65, .66] .68 [.68, .69] .64 [.63, .64]Slope −.03 [−.04, −.02] −.03 [−.04, −.02] −.05 [−.06, −.03] −.03 [−.03, −.02]
Note. GSOEP=German Socioeconomic Panel Study; HILDA=Household Income and Labour Dynamics in Australia Study; HRS=Health and RetirementStudy; LISS = Longitudinal Studies for the Social Sciences; Est = the maximum a posteriori (MAP) estimate; CI = 95% credible intervals. Bolded valuesindicate parameter estimates that do not include 0 in the credible intervals.
1330 WRIGHT AND JACKSON
Neuroticism decreases, with these changes likely due to bothbiological and environmental influences (Lucas & Donnellan,2011; Terracciano et al., 2005; Wortman et al., 2012). The neoso-cioanalytic theory highlights the impact of social roles on personal-ity development (Roberts &Wood, 2006). Our younger participantscould be transitioning into environments or life roles, such asbeginning careers after finishing their education or a starting afamily, that are serving to increase their consistency across timeby either impacting their levels of relevant personality traits orproviding a stable environment. The “investment” in these newsocial roles offers not only a source of stability for individuals butcan also lead to reinforcing changes in relevant traits across time thatare associated with the development and maintenance of thesemature roles (i.e., the corresponsive principle; Denissen et al.,2019; Neyer & Lehnart, 2007; Roberts & Wood, 2006; Specht etal., 2011). Some evidence of this can be seen with the effects ofuniversity degree attainment on consistency—participants withhigher education levels had higher levels of profile consistency.Importantly, our work extends previous research (see Terracciano etal. 2018, for another similar finding) on the maturity–stabilityhypothesis by showing it holds in samples other than young adults,as these trait findings were evident in all of our mostly middle-aged
samples (Blonigen et al., 2008; Donnellan et al., 2007; Lönnqvistet al., 2008; Roberts et al., 2001).
There Is Variability in the Processes UnderlyingPersonality Consistency
We next examined if the underlying processes driving personalityconsistency are similar to those observed at the between-personlevel (i.e., via rank-order consistency) and if there were individualdifferences in these trends. In our study, when later patterns ofresponding were compared to people’s initial responses—showingchanges in personality consistency across increasing timeintervals—we found that people became increasingly less similarfrom their original selves across time. Moreover, while averagetrends did emerge across all data sets, there was also variability interms of the magnitude and direction of absolute changes. Thisvariability in the trajectories of personality consistency suggests thatthe degree to which these processes matter vary across people(Fraley & Roberts, 2005). Importantly, however, our findingsfurther differ in two primary ways from previous between-personinvestigations of personality consistency (i.e., Fraley & Roberts,2005): (a) a lack of support for stochastic factors having a large
Thisdocumentiscopyrightedby
theAmerican
Psychological
Associatio
nor
oneof
itsalliedpublishers.
Thisarticle
isintended
solely
forthepersonal
useof
theindividual
user
andisnotto
bedissem
inated
broadly.
Figure 3Distribution of Person-Level Slope Estimates for Changes in Consistency Across Increasing Intervals
Note. The posterior distributions for the model-derived, person-level slopes from the models in Table 9 are presented above. The vertical black lineconstitutes the portion of slopes with values ranging from−.01< x< .01 (i.e., represents .00 with rounding). The black dot represents the median person-levelslope estimate, the thicker black horizontal line indicates the 68% quartile interval, and the thin black horizontal line indicates the 95% quartile interval.GSOEP=German Socioeconomic Panel Study; HILDA=Household Income and Labour Dynamics in Australia Study; HRS=Health and Retirement Study;LISS = Longitudinal Studies for the Social Sciences.
PERSON-CENTERED PERSONALITY CONSISTENCY 1331
impact on personality stability and (b) support for transactionalprocesses being impactful and even bolstering consistency.Although there were average slight decreases in absolute levels of
profile consistency across longer stretches of time, the asymptoticvalues were nowhere near the lowest values found in past meta-analytic research examining these processes in younger samples(i.e., r = .25–.30; Fraley & Roberts, 2005), but rather were in linewith the larger values previously found in adult samples for bothrank-order stability (Atherton et al., 2021; Jones & Meredith, 1996;Terracciano et al., 2006) and within-person ipsative consistency
(Terracciano et al., 2010). As time increases, people’s personalityprofile consistency does not decay quickly, suggesting stable factors(as opposed to changing factors) have an appreciable role inpromoting this stability (Anusic & Schimmack, 2016). These find-ings suggest that stochastic influences, while certainly present tosome degree on average, appear to be less important for adult-levelpersonality consistency. Even if it were the case that the stability intrajectories of profile consistency across longer stretches of time isdue to a mixture of stochastic and transactional processes rather thandue to the sole influence of some constant factor, our results indicate
Thisdocumentiscopyrightedby
theAmerican
Psychological
Associatio
nor
oneof
itsalliedpublishers.
Thisarticle
isintended
solely
forthepersonal
useof
theindividual
user
andisnotto
bedissem
inated
broadly.
Table 10Moderators of the Processes Underlying Consistency
Model parameter
GSOEP HILDA HRS LISS
Est CI Est CI Est CI Est CI
Numeric ageIntercept .59 [.58, .60] .62 [.62, .63] .66 [.65, .67] .63 [.62, .64]Num. age .00 [−.00, .00] .00 [.00, .00] .00 [.00, .00] .00 [.00, .00]Gender −.02 [−.03, −.01] .06 [.05, .07] .04 [.02, .05] .01 [.00, .02]Time −.03 [−.04, −.02] −.02 [−.03, −.01] −.05 [−.07, −.03] −.03 [−.03, −.02]Time × Num. age −.00 [−.00, .00] .00 [−.00, .00] −.00 [−.00, −.00] .00 [.00, .00]Time × Gender .00 [−.01, .02] −.01 [−.02, .00] .00 [−.02, .03] .00 [−.00, .00]
Numeric age (quadratic)Intercept .59 [.58, .60] .68 [.67, .68] .68 [.68, .69] .68 [.67, .69]Age .00 [−.00, .00] .00 [.00, .00] .00 [−.00, .00] .00 [.00, .00]Age2 −.00 [−.00, −.00] −.00 [−.00, −.00] .00 [−.00, .00] −.00 [−.00, −.00]Time −.03 [−.04, −.02] −.02 [−.02, −.01] −.04 [−.05, −.03] −.02 [−.03, −.02]Time × Num. age −.00 [−.00, .00] .00 [−.00, .00] −.00 [−.00, −.00] .00 [.00, .00]Time × Num. age2 .00 [−.00, .00] −.00 [−.00, −.00] −.00 [−.00, .00] −.00 [−.00, −.00]
Dichotomous ageIntercept .59 [.59, .60] .64 [.63, .65] .65 [.64, .65]Dich. age −.06 [−.08, −.04] −.11 [−.12, −.10] −.11 [−.12, −.09]Gender −.02 [−.03, −.01] .06 [.05, .07] .01 [.00, .02]Time −.03 [−.04, −.02] −.02 [−.03, −.01] −.02 [−.03, −.02]Time × Dich. age .00 [−.02, .03] −.02 [−.04, −.01] −.01 [−.02, −.01]Time × Gender .00 [−.01, .02] −.01 [−.02, .00] −.00 [−.00, .00]
AgreeablenessIntercept .58 [.57, .59] .66 [.65, .66] .68 [.68, .69] .63 [.63, .64]Trait T1 .04 [.04, .05] .10 [.10, .11] .15 [.14, .17] .10 [.09, .11]Time −.02 [−.03, −.02] −.03 [−.03, −.02] −.04 [−.05, −.02] −.02 [−.03, −.02]Time × Trait T1 .03 [.02, .04] .05 [.04, .05] .08 [.05, .10] .02 [.01, .02]Time × Trait lags .04 [.04, .05] .08 [.07, .08] .16 [.15, .18] .05 [.04, .05]
ConscientiousnessIntercept .58 [.57, .58] .66 [.65, .66] .68 [.68, .69] .64 [.63, .64]Trait T1 .06 [.05, .06] .08 [.08, .09] .16 [.15, .18] .10 [.09, .11]Time −.02 [−.02, −.01] −.03 [−.04, −.02] −.03 [−.04, −.02] −.03 [−.03, −.02]Time × Trait T1 .04 [.03, .05] .04 [.03, .04] .08 [.06, .11] .02 [.02, .03]Time × Trait lags .07 [.06, .07] .05 [.05, .06] .15 [.14, .17] .04 [.04, .04]
NeuroticismIntercept .58 [.57, .59] .66 [.65, .66] .68 [.68, .69] .64 [.63, .64]Trait T1 −.05 [−.05, −.04] −.11 [−.11, −.11] −.14 [−.15, −.13] −.11 [−.11, −.10]Time −.03 [−.04, −.02] −.03 [−.04, −.02] −.05 [−.06, −.04] −.03 [−.03, −.03]Time × Trait T1 −.02 [−.03, −.01] −.05 [−.06, −.04] −.07 [−.09, −.06] −.02 [−.03, −.02]Time × Trait lags −.03 [−.04, −.03] −.08 [−.09, −.08] −.12 [−.14, −.11] −.04 [−.04, −.04]
EducationIntercept .57 [.57, .58] .65 [.64, .65] .68 [.67, .69] .62 [.62, .63]Education .06 [.05, .07] .05 [.04, .06] .04 [.03, .06] .03 [.03, .04]Time −.03 [−.04, −.02] −.03 [−.04, −.02] −.06 [−.07, −.04] −.03 [−.03, −.02]Time × Education −.00 [−.02, .02] .01 [−.01, .02] .03 [.00, .05] .00 [−.00, .01]
Note. To reduce table length, only the sample-level effects (i.e., fixed effects) for each moderator are presented in this table. Please see Supplemental TablesS19–S21; S23–S25; and S27, for full results including the person-level effects (i.e., random effects). GSOEP=German Socioeconomic Panel Study; HILDA=Household Income and Labour Dynamics in Australia Study; HRS =Health and Retirement Study; LISS = Longitudinal Studies for the Social Sciences; Est =the maximum a posteriori (MAP) estimate; CI = 95% credible intervals; Trait T1 = between-person trait variable, indicating the difference between a person’slevel of a trait atWave 1 relative to the data set’s average level atWave 1; Trait lags=within-person trait variable, indicating a person’s deviation in their level ofa trait between the waves used to calculate a profile correlation. Bolded values indicate parameter estimates that do not include 0 in the credible intervals.
1332 WRIGHT AND JACKSON
that these stochastic factors do not greatly outweigh the impact oftransactional processes. This could be due to adults being moreresilient to the impact of stochastic factors due to ingrained life-styles/routines or perhaps a stronger sense of self that is not as easilyaltered compared to younger individuals. That is, even with our upto 12-year test–retest interval, there were no indications that peoplebegan to critically decline and asymptote at low values in theirprofile consistency. Thus, even over longer time spans (i.e., 20–30years), it is expected that stochastic factors do not accumulate suchthat people will still mostly maintain their person-centered level ofpersonality profile consistency across time.Next, our results provide strong evidence of there being devel-
opmental constant forces at play, emphasizing the consistency ofpersonality at quite high levels, even over longer time spans (Anusic& Schimmack, 2016). There was an average effect of slight declinesin absolute consistency across time, but the points from whichpeople declined varied considerably. This suggests that althoughconstant factors are likely at play to reinforce a person’s unique setpoint of personality consistency, the influence of these need not bethe same for everyone.Moreover, it is possible that some individualsshowing stable profile consistency trajectories instead had a mixtureof stochastic and transactional processes at play, rather than devel-opmental constant processes impacting their long-term stability.That is, stochastic processes leading to decreases in consistency andtransactional processes leading to increases in consistency caneffectively cancel each other out and result in high levels ofconsistency. Notably, these possible combinations of processesunderlying our observed trajectories of consistency highlight thepotentially idiosyncratic nature of this type of personality develop-ment. On the one hand, this seems obvious, such that people differ inthe processes that promote consistency and change. On the otherhand, few designs are capable of addressing this question.These findings continue to point to taking a person-specific view
of change and stability (Beck & Jackson, 2021c). Much of theresearch done thus far has focused on shared characteristics thoughtto similarly impact the development of personality for everyone(e.g., life events; Bleidorn et al., 2018). However, the evidence forprocesses driving change and stability is often nonspecific anddiscussed in a manner that is applicable to groups or averagesrather than individuals. It is important to note that there could beindividual differences in these processes, though. For instance,while heritable estimates change over time (Briley & Tucker-Drob, 2013; Plomin & Spinath, 2004), one cannot state that thesechanges apply to all individuals. Additionally, although an indivi-dual’s genotype does not vary across time, the influence of thosegenetic factors may vary and could further be a function ofenvironmental (Reiss et al., 2000) or age-related maturation pro-cesses (McCrae & Costa, 2008; Roberts & Jackson, 2008). Simi-larly, the variable activation or timing of the activity of certain genesacross individuals may give rise to individual differences by affect-ing the extent to which and length of time they exert their influence(Plomin, 1986). This suggests that even this constant factor (i.e.,one’s genetics) could contribute to within-person variability acrosstime and further differ across individuals. As for a nonbiologicalexample, attachment theories suggest that the early-life mentalrepresentations a child develops for themselves and their attachmentfigures are enduring factors throughout their lives (Bowlby, 1973;Fraley, 2002). Even though this attachment may vary in expressionor even in the opportunity to exhibit it at any given time, it is
typically assumed that it serves as an enduring and influential factorin a person’s life that itself is largely invariant across time (Fraley &Brumbaugh, 2004; Roisman et al., 2005). However, even thisenduring factor can change throughout the course of one’s life,with there further being individual differences in the extent to if andwhat it changes in response to (Fraley et al., 2021).
Lastly, one prominent example in how the processes promotingconsistency differ among people is that some people increase inpersonality profile consistency across longer stretches of time. Theseincreases we found in a substantial proportion of participantsindicate the presence of transactional factors at work, as these arethe only processes that can lead to increases in consistency and nocombination of other processes can mimic their effects (Fraley &Roberts, 2005). Whereas stochastic factors are expected to result indeclines in consistency, transactional processes often serve toreinforce and potentially even increase personality consistencyacross time (Caspi & Roberts, 1999; Fraley & Roberts, 2005).Given that transactional processes with the environment can be stillpresent and not necessarily lead to increases, such that individualswith stable or declining trajectories can also have these, the amountof individuals with positive trajectories is perhaps an underestimateof the number of individuals with these factors at play. However, thepositive slopes indicate that, for these individuals, transactionalprocesses not only appear to be the most impactful force, but furtherthat they have a bolstering effect on their consistency—as opposedto serving to help them maintain their person-typical levels. Thesefindings highlight the complex role that environmental factors canhave on personality development. Despite the prevalence of trans-actional theory within the personality development domain (Specht,Bleidorn, et al., 2014), Fraley and Roberts (2005) did not find thatthe removal of the influence of transactional processes had adetrimental effect on models of personality development. Ourfindings identify that the impact of transactional effects is veryimportant for some individuals as this is the only process that canlead to increases in consistency (Fraley & Roberts, 2005). Again,this underscores the utility of taking a person-centered approach tounderstanding personality development.
Limitations
While our study had a number of advantages in terms of a largesample, multiple data sets across different countries, and four to ninewaves of data spanning many years, there are still limitations worthconsidering. First, all data sets were from European-descent coun-tries. Thus, the generalizability of these results to countries that donot fit this criterion could be limited. Second, the average age acrossour data sets was well above adolescence or even young adulthood.That is, our samples might not have been young enough to, onaverage, show effects of personality maturation, which is oneprocess thought to drive changes in consistency (Donnellanet al., 2007). It has been found that profile correlations reach aplateau by late adolescence (Branje et al., 2007; Klimstra et al.,2009), suggesting that, at the very least, stark changes shouldperhaps not be expected in our middle age to older samples. Third,it was indeed the case that individuals who did not qualifyfor inclusion in our study had lower average levels of profileconsistency, as quantified by the within-person profile correlations(Supplemental File S1). While this would have affected the interceptvalues in our models, likely by decreasing their magnitude to an
Thisdocumentiscopyrightedby
theAmerican
Psychological
Associatio
nor
oneof
itsalliedpublishers.
Thisarticle
isintended
solely
forthepersonal
useof
theindividual
user
andisnotto
bedissem
inated
broadly.
PERSON-CENTERED PERSONALITY CONSISTENCY 1333
extent, our conclusions did not rest on the magnitude of the interceptvalues. However, we cannot confirm that our conclusions wouldhave held across the excluded individuals, so it should be noted thatthese inferences may not generalize to individuals with lower-than-average consistency values or that our inferences may have changed ifmore individuals with lower consistency estimates were included.Fourth, the three processes underlying personality consistency werenot directly tested but were inferred based on the direction of theobserved long-term trajectories and past work linking these factorswith certain directions (Fraley & Roberts, 2005). Future work shouldincorporate testable factors representing these processes to examine ifthey have their expected effects and if the assumptions hold. Lastly,the measures used across studies were not equivalent in their exactcontent, number of items, or psychometric properties. This limitsconclusions regarding generalizability to other studies to somedegree, as these correlations do vary by these factors (Schmidt &Hunter, 1996), and thus the raw profile correlation values should becompared to other studies with recognition that these factors doperhaps preclude direct comparisons. However, the model-derivedconclusions drawn across our four data sets were extremely similar,suggesting the different measures did not affect the study’s mainfindings.
Conclusion
This study showed that there were considerable individual differ-ences in levels of person-typical personality consistency and, to aslightly lesser extent, changes in these levels across time, indicatingthat some people are more stably consistent in their personalityprofiles than others. Individual differences in profile consistency arenot short-term perturbations but rather inherent properties of anindividual, such that some people are more stable in their personalitythan others. Our findings also highlight that, at the individual level,the processes underlying personality consistency need not affecteveryone to the same degree nor equivalently impact a single personto the same extent across time. Between-person investigations canconceal these nuanced individual differences and give too muchweight to any single process being associated with one trend for allindividuals. This person-centered investigation of personality con-sistency highlights the benefit of taking a within-person approach.
References
Aldwin, C.M., Spiro, A., Levenson,M. R., &Bossé, R. (1989). Longitudinalfindings from the normative aging study: 1. Does mental health changewith age? Psychology and Aging, 4(3), 295–306. https://doi.org/10.1037/0882-7974.4.3.295
Anusic, I., & Schimmack, U. (2016). Stability and change of personalitytraits, self-esteem, and well-being: Introducing the meta-analytic stabilityand change model of retest correlations. Journal of Personality and SocialPsychology, 110(5), 766–781. https://doi.org/10.1037/pspp0000066
Ardelt, M. (2000). Still stable after all these years? Personality stabilitytheory revisited. Social Psychology Quarterly, 63(4), 392–405. https://doi.org/10.2307/2695848
Asendorpf, J. B. (1992). Beyond stability: Predicting inter-individual differ-ences in intra-individual change. European Journal of Personality, 6(2),103–117. https://doi.org/10.1002/per.2410060204
Asendorpf, J. B., & van Aken, M. A. (1991). Correlates of the temporalconsistency of personality patterns in childhood. Journal of Personality,59(4), 689–703. https://doi.org/10.1111/j.1467-6494.1991.tb00927.x
Atherton, O. E., Grijalva, E., Roberts, B. W., & Robins, R. W. (2021).Stability and change in personality traits and major life goals from collegeto midlife. Personality and Social Psychology Bulletin, 47(5), 841–858.https://doi.org/10.1177/0146167220949362
Bazana, P. G., & Stelmack, R. M. (2004). Stability of personality across thelife span: A meta-analysis. In R. M. Stelmack (Ed.),On the psychobiologyof personality: Essays in honor of Marvin Zuckerman (pp. 113–144).Elsevier Science. https://doi.org/10.1016/B978-008044209-9/50009-9
Beck, E. D., & Jackson, J. J. (2020). Consistency and change in idiographicpersonality: A longitudinal ESM network study. Journal of Personality andSocial Psychology, 118(5), 1080–1100. https://doi.org/10.1037/pspp0000249
Beck, E. D., & Jackson, J. J. (2021a). Idiographic personality coherence: Aquasi experimental longitudinal ESM study. European Journal of Per-sonality, 36(3), 391–412. https://doi.org/10.1177/08902070211017746
Beck, E. D., & Jackson, J. J. (2021b). Within-person variability. In J. F.Rauthmann (Ed.), The handbook of personality dynamics and processes(pp. 75–100). Elsevier Academic Press. https://doi.org/10.1016/B978-0-12-813995-0.00004-2
Beck, E. D., & Jackson, J. J. (2021c). Detecting idiographic personalitychange. Journal of Personality Assessment, 104(4), 467–483. https://doi.org/10.1080/00223891.2021.1984246
Bem, D. J., & Allen, A. (1974). On predicting some of the people some of thetime: The search for cross-situational consistencies in behavior. Psycho-logical Review, 81(6), 506–520. https://doi.org/10.1037/h0037130
Bleidorn, W., & Hopwood, C. J. (2019). Stability and change in personalitytraits over the lifespan. In D. P. McAdams, R. L. Shiner, & J. L. Tackett(Eds.), Handbook of personality development (pp. 237–252). GuilfordPress.
Bleidorn, W., Hopwood, C. J., & Lucas, R. E. (2018). Life events andpersonality trait change. Journal of Personality, 86(1), 83–96. https://doi.org/10.1111/jopy.12286
Bleidorn, W., Klimstra, T. A., Denissen, J. J., Rentfrow, P. J., Potter, J., &Gosling, S. D. (2013). Personality maturation around the world: A cross-cultural examination of social-investment theory. Psychological Science,24(12), 2530–2540. https://doi.org/10.1177/0956797613498396
Block, J. (1971). Lives through time. Bancroft Books.Blonigen, D. M., Carlson, M. D., Hicks, B. M., Krueger, R. F., & Iacono,W. G. (2008). Stability and change in personality traits from late adoles-cence to early adulthood: A longitudinal twin study. Journal of Personal-ity, 76(2), 229–266. https://doi.org/10.1111/j.1467-6494.2007.00485.x
Bowlby, J. (1973). Attachment and loss: Volume II: Separation, anxietyand anger (pp. 1–429). Basic Books.
Branje, S. J. T., van Lieshout, C. F. M., & Gerris, J. R. M. (2007). Big Fivepersonality development in adolescence and adulthood. European Journalof Personality, 21(1), 45–62. https://doi.org/10.1002/per.596
Briley, D. A., & Tucker-Drob, E. M. (2013). Explaining the increasingheritability of cognitive ability across development: A meta-analysis oflongitudinal twin and adoption studies. Psychological Science, 24(9),1704–1713. https://doi.org/10.1177/0956797613478618
Briley, D. A., & Tucker-Drob, E. M. (2014). Genetic and environmentalcontinuity in personality development: A meta-analysis. PsychologicalBulletin, 140(5), 1303–1331. https://doi.org/10.1037/a0037091
Bürkner, P. (2017). brms: An R package for Bayesian multilevel modelsusing Stan. Journal of Statistical Software, 80(1), 1–28. https://doi.org/10.18637/jss.v080.i01
Caspi, A., & Bem, D. J. (1990). Personality continuity and change across thelife course. In L. A. Pervin (Ed.), Handbook of personality: Theory andresearch (pp. 549–575). Guilford Press.
Caspi, A., &Moffitt, T. E. (1993). When do individual differences matter? Aparadoxical theory of personality coherence. Psychological Inquiry, 4(4),247–271. https://doi.org/10.1207/s15327965pli0404_1
Thisdocumentiscopyrightedby
theAmerican
Psychological
Associatio
nor
oneof
itsalliedpublishers.
Thisarticle
isintended
solely
forthepersonal
useof
theindividual
user
andisnotto
bedissem
inated
broadly.
1334 WRIGHT AND JACKSON
Caspi, A., & Roberts, B.W. (1999). Personality continuity and change acrossthe life course. In L. A. Pervin & O. P. John (Eds.), Handbook ofpersonality: Theory and research (Vol. 2, pp. 300–326). Guilford Press.
Conley, J. J. (1984a). Longitudinal consistency of adult personality: Self-reported psychological characteristics across 45 years. Journal of Person-ality and Social Psychology, 47(6), 1325–1333. https://doi.org/10.1037/0022-3514.47.6.1325
Conley, J. J. (1984b). The hierarchy of consistency: A review and model oflongitudinal findings on adult individual differences in intelligence,personality and self-opinion. Personality and Individual Differences,5(1), 11–25. https://doi.org/10.1016/0191-8869(84)90133-8
Costa, P. T., Jr., Bagby, R. M., Herbst, J. H., & McCrae, R. R. (2005).Personality self-reports are concurrently reliable and valid during acutedepressive episodes. Journal of Affective Disorders, 89(1–3), 45–55.https://doi.org/10.1016/j.jad.2005.06.010
Cronbach, L. J., & Gleser, G. C. (1953). Assessing similarity between profiles.Psychological Bulletin, 50(6), 456–473. https://doi.org/10.1037/h0057173
De Fruyt, F., Bartels, M., Van Leeuwen, K. G., De Clercq, B., Decuyper, M.,& Mervielde, I. (2006). Five types of personality continuity in childhoodand adolescence. Journal of Personality and Social Psychology, 91(3),538–552. https://doi.org/10.1037/0022-3514.91.3.538
Denissen, J. J. A., Luhmann, M., Chung, J. M., & Bleidorn, W. (2019).Transactions between life events and personality traits across the adultlifespan. Journal of Personality and Social Psychology, 116(4), 612–633.https://doi.org/10.1037/pspp0000196
Donnellan, M. B., Conger, R. D., & Burzette, R. G. (2007). Personalitydevelopment from late adolescence to young adulthood: Differentialstability, normative maturity, and evidence for the maturity–stabilityhypothesis. Journal of Personality, 75(2), 237–263. https://doi.org/10.1111/j.1467-6494.2007.00438.x
Epstein, S. (1979). The stability of behavior: I. On predicting most of thepeople much of the time. Journal of Personality and Social Psychology,37(7), 1097–1126. https://doi.org/10.1037/0022-3514.37.7.1097
Fraley, R. C. (2002). Attachment stability from infancy to adulthood: Meta-analysis and dynamic modeling of developmental mechanisms. Person-ality and Social Psychology Review, 6(2), 123–151. https://doi.org/10.1207/S15327957PSPR0602_03
Fraley, R. C., & Brumbaugh, C. C. (2004). A dynamical systems approach toconceptualizing and studying stability and change in attachment security.In W. S. Rholes & J. A. Simpson (Eds.), Adult attachment: Theory,research, and clinical implications (pp. 86–132). Guilford Publications.
Fraley, R. C., Gillath, O., & Deboeck, P. R. (2021). Do life events lead toenduring changes in adult attachment styles? A naturalistic longitudinalinvestigation. Journal of Personality and Social Psychology, 120(6),1567–1606. https://doi.org/10.1037/pspi0000326
Fraley, R. C., & Roberts, B. W. (2005). Patterns of continuity: A dynamicmodel for conceptualizing the stability of individual differences in psy-chological constructs across the life course. Psychological Review, 112(1),60–74. https://doi.org/10.1037/0033-295X.112.1.60
Fraley, R. C., Roisman, G. I., Booth-LaForce, C., Owen, M. T., & Holland,A. S. (2013). Interpersonal and genetic origins of adult attachment styles:A longitudinal study from infancy to early adulthood. Journal of Person-ality and Social Psychology, 104(5), 817–838. https://doi.org/10.1037/a0031435
Gilbert, K., Whalen, D. J., Jackson, J. J., Tillman, R., Barch, D. M., & Luby,J. L. (2021). Thin slice derived personality types predict longitudinalsymptom trajectories. Personality Disorders, 12(3), 275–285. https://doi.org/10.1037/per0000455
Goldberg, L. R. (1990). An alternative “description of personality”: The BigFive factor structure. Journal of Personality and Social Psychology, 59(6),1216–1229. https://doi.org/10.1037/0022-3514.59.6.1216
Headey, B. (2008). Life goals matter to happiness: A revision of set-pointtheory. Social Indicators Research, 86(2), 213–231. https://doi.org/10.1007/s11205-007-9138-y
Headey, B., & Wearing, A. (1989). Personality, life events, and subjectivewell-being: Toward a dynamic equilibrium model. Journal of Personalityand Social Psychology, 57(4), 731–739. https://doi.org/10.1037/0022-3514.57.4.731
Ibáñez, M. I., Viruela, A. M., Mezquita, L., Moya, J., Villa, H., Camacho, L.,& Ortet, G. (2016). An investigation of five types of personality traitcontinuity: A two-wave longitudinal study of Spanish adolescents fromage 12 to age 15. Frontiers in Psychology, 7, Article 512. https://doi.org/10.3389/fpsyg.2016.00512
Jackson, J. J., & Beck, E. D. (2021). Personality development beyond themean: Do life events shape personality variability, structure, and ipsativecontinuity? The Journals of Gerontology: Series B, 76(1), 20–30. https://doi.org/10.1093/geronb/gbaa093
Jones, C. J., &Meredith,W. (1996). Patterns of personality change across thelife span. Psychology and Aging, 11(1), 57–65. https://doi.org/10.1037/0882-7974.11.1.57
Juster, F. T., & Suzman, R. (1995). An overview of the Health andRetirement Study. The Journal of Human Resources, 30, S7–S56.https://doi.org/10.2307/146277
Kazak, A. E. (2018). Editorial: Journal article reporting standards. AmericanPsychologist, 73(1), 1–2. https://doi.org/10.1037/amp0000263
Kendler, K. S., Gatz, M., Gardner, C. O., & Pedersen, N. L. (2006).Personality and major depression: A Swedish longitudinal, population-based twin study. Archives of General Psychiatry, 63(10), 1113–1120.https://doi.org/10.1001/archpsyc.63.10.1113
Klimstra, T. A., Hale, W. W., III, Raaijmakers, Q. A. W., Branje, S. J. T., &Meeus, W. H. J. (2009). Maturation of personality in adolescence. Journalof Personality and Social Psychology, 96(4), 898–912. https://doi.org/10.1037/a0014746
Lamiell, J. T. (1981). Toward an idiothetic psychology of personality. AmericanPsychologist, 36(3), 276–289. https://doi.org/10.1037/0003-066X.36.3.276
Lewis, M. (1997). Altering fate: Why the past does not predict the future.Guilford Press.
Lewis, M. (2001). Issues in the study of personality development. Psychologi-cal Inquiry, 12(2), 67–83. https://doi.org/10.1207/S15327965PLI1202_02
Lönnqvist, J.-E., Mäkinen, S., Paunonen, S. V., Henriksson, M., & Verka-salo, M. (2008). Psychosocial functioning in young men predicts theirpersonality stability over 15 years. Journal of Research in Personality,42(3), 599–621. https://doi.org/10.1016/j.jrp.2007.08.006
Lucas, R. E., Clark, A. E., Georgellis, Y., & Diener, E. (2003). Reexaminingadaptation and the set point model of happiness: Reactions to changes inmarital status. Journal of Personality and Social Psychology, 84(3), 527–539. https://doi.org/10.1037/0022-3514.84.3.527
Lucas, R. E., & Donnellan, M. B. (2011). Personality development across thelife span: Longitudinal analyses with a national sample from Germany.Journal of Personality and Social Psychology, 101(4), 847–861. https://doi.org/10.1037/a0024298
Lüdtke, O., Roberts, B. W., Trautwein, U., & Nagy, G. (2011). A randomwalk down university avenue: Life paths, life events, and personality traitchange at the transition to university life. Journal of Personality and SocialPsychology, 101(3), 620–637. https://doi.org/10.1037/a0023743
Magnus, K., Diener, E., Fujita, F., & Pavot, W. (1993). Extraversion andneuroticism as predictors of objective life events: A longitudinal analysis.Journal of Personality and Social Psychology, 65(5), 1046–1053. https://doi.org/10.1037/0022-3514.65.5.1046
McCrae, R. R. (1993). Moderated analyses of longitudinal personalitystability. Journal of Personality and Social Psychology, 65(3), 577–585. https://doi.org/10.1037/0022-3514.65.3.577
McCrae, R. R., & Costa, P. T., Jr. (2003). Personality in adulthood: A five-factor theory perspective (2nd ed.). Guilford Press. https://doi.org/10.4324/9780203428412
McCrae, R. R., & Costa, P. T., Jr. (2008). The five-factor theory ofpersonality. In O. P. John, R.W. Robins, & L. A. Pervin (Eds.),Handbookof personality: Theory and research (pp. 159–181). Guilford Press.
Thisdocumentiscopyrightedby
theAmerican
Psychological
Associatio
nor
oneof
itsalliedpublishers.
Thisarticle
isintended
solely
forthepersonal
useof
theindividual
user
andisnotto
bedissem
inated
broadly.
PERSON-CENTERED PERSONALITY CONSISTENCY 1335
McGue, M., Bacon, S., & Lykken, D. T. (1993). Personality stability andchange in early adulthood: A behavioral genetic analysis. DevelopmentalPsychology, 29(1), 96–109. https://doi.org/10.1037/0012-1649.29.1.96
Meeus, W., van de Schoot, R., Keijsers, L., & Branje, S. (2012). Identitystatuses as developmental trajectories: A five-wave longitudinal study inearly-to-middle and middle-to-late adolescents. Journal of Youth and Ado-lescence, 41(8), 1008–1021. https://doi.org/10.1007/s10964-011-9730-y
Mischel, W., & Shoda, Y. (1998). Reconciling processing dynamics andpersonality dispositions. Annual Review of Psychology, 49(1), 229–258.https://doi.org/10.1146/annurev.psych.49.1.229
Morizot, J., & Le Blanc, M. (2005). Searching for a developmental typologyof personality and its relations to antisocial behavior: A longitudinal studyof a representative sample of men. Journal of Personality, 73(1), 139–182.https://doi.org/10.1111/j.1467-6494.2004.00307.x
Mroczek, D. K., & Spiro, A., III (2003). Modeling intraindividual change inpersonality traits: Findings from the normative aging study. The Journalsof Gerontology Series B: Psychological Sciences and Social Sciences,58(3), 153–165. https://doi.org/10.1093/geronb/58.3.P153
Neyer, F. J., & Asendorpf, J. B. (2001). Personality–relationship transactionin young adulthood. Journal of Personality and Social Psychology, 81(6),1190–1204. https://doi.org/10.1037/0022-3514.81.6.1190
Neyer, F. J., & Lehnart, J. (2007). Relationships matter in personalitydevelopment: Evidence from an 8-year longitudinal study across youngadulthood. Journal of Personality, 75(3), 535–568. https://doi.org/10.1111/j.1467-6494.2007.00448.x
Ozer, D. J., & Benet-Martínez, V. (2006). Personality and the prediction ofconsequential outcomes. Annual Review of Psychology, 57(1), 401–421.https://doi.org/10.1146/annurev.psych.57.102904.190127
Ozer, D. J., & Gjerde, P. F. (1989). Patterns of personality consistency andchange from childhood through adolescence. Journal of Personality,57(2), 483–507. https://doi.org/10.1111/j.1467-6494.1989.tb00490.x
Plomin, R. (1986). Behavioral genetic methods. Journal of Personality,54(1), 226–261. https://doi.org/10.1111/j.1467-6494.1986.tb00394.x
Plomin, R., & Spinath, F. M. (2004). Intelligence: Genetics, genes, andgenomics. Journal of Personality and Social Psychology, 86(1), 112–129.https://doi.org/10.1037/0022-3514.86.1.112
R Core Team. (2021). R: A language and environment for statisticalcomputing. R Foundation for Statistical Computing. http://www.R-project.org/
Reiss, D., Neiderhiser, J., Mavis, E., Hetherington, E. M., & Plomin, R.(2000). The relationship code: Deciphering genetic and social influenceson adolescent development. Harvard University Press.
Revelle, W. (2021). psych: Procedures for psychological, psychometric, andpersonality research (R package Version 2.1.3). Northwestern University.https://CRAN.R-project.org/package=psych
Revelle, W., & Wilt, J. (2020). The dynamics of personality. In J. F.Rauthmann (Ed.), Handbook of personality dynamics and processes(1st ed.). Elsevier.
Roberts, B. W. (2018). A revised sociogenomic model of personality traits.Journal of Personality, 86(1), 23–35. https://doi.org/10.1111/jopy.12323
Roberts, B. W., & Caspi, A. (2003). The cumulative continuity model ofpersonality development: Striking a balance between continuity andchange in personality traits across the life course. In R. M. Staudinger& U. Lindenberger (Eds.), Understanding human development: Lifespanpsychology in exchange with other disciplines (pp. 183–214). KluwerAcademic. https://doi.org/10.1007/978-1-4615-0357-6_9
Roberts, B. W., Caspi, A., & Moffitt, T. E. (2001). The kids are alright:Growth and stability in personality development from adolescence toadulthood. Journal of Personality and Social Psychology, 81(4), 670–683.https://doi.org/10.1037/0022-3514.81.4.670
Roberts, B. W., & DelVecchio, W. F. (2000). The rank-order consistency ofpersonality traits from childhood to old age: A quantitative review oflongitudinal studies. Psychological Bulletin, 126(1), 3–25. https://doi.org/10.1037/0033-2909.126.1.3
Roberts, B. W., & Jackson, J. J. (2008). Sociogenomic personality psychol-ogy. Journal of Personality, 76(6), 1523–1544. https://doi.org/10.1111/j.1467-6494.2008.00530.x
Roberts, B. W., &Mroczek, D. (2008). Personality trait change in adulthood.Current Directions in Psychological Science, 17(1), 31–35. https://doi.org/10.1111/j.1467-8721.2008.00543.x
Roberts, B. W., Robins, R. W., Caspi, A., & Trzesniewski, K. (2003).Personality trait development in adulthood. In J. Mortimer &M. Shanahan(Eds.), Handbook of the life course (pp. 579–598). Kluwer Academic.https://doi.org/10.1007/978-0-306-48247-2_26
Roberts, B. W., Walton, K. E., & Viechtbauer, W. (2006). Patterns of mean-level change in personality traits across the life course: A meta-analysis oflongitudinal studies. Psychological Bulletin, 132(1), 1–25. https://doi.org/10.1037/0033-2909.132.1.1
Roberts, B.W., &Wood, D. (2006). Personality development in the context of theneo-socioanalytic model of personality. In D. K.Mroczek& T. D. Little (Eds.),Handbook of personality development (pp. 11–39). Lawrence Erlbaum.
Roberts, B. W., Wood, D., & Caspi, A. (2008). The development ofpersonality traits in adulthood. In O. P. John, R. W. Robins, & L. A.Pervin (Eds.), Handbook of personality: Theory and research (pp. 375–398). Guilford Press.
Robins, R. W., Fraley, R. C., Roberts, B. W., & Trzesniewski, K. H. (2001).A longitudinal study of personality change in young adulthood. Journal ofPersonality, 69(4), 617–640. https://doi.org/10.1111/1467-6494.694157
Robins, R.W., & Tracy, J. L. (2003). Setting an agenda for a person-centeredapproach to personality development. Monographs of the Society forResearch in Child Development, 68(1), 110–122. https://doi.org/10.1111/j.1540-5834.2003.06801012.x
Roisman, G. I., Collins, W. A., Sroufe, L. A., & Egeland, B. (2005).Predictors of young adults’ representations of and behavior in their currentromantic relationship: Prospective tests of the prototype hypothesis.Attachment and Human Development, 7(2), 105–121. https://doi.org/10.1080/14616730500134928
Scherpenzeel, A. C., & Das, M. (2010). “True” longitudinal and probability-based internet panels: Evidence from the Netherlands. In M. Das, P. Ester,& L. Kaczmirek (Eds.), Social and behavioral research and the internet:Advances in applied methods and research strategies (pp. 77–104). Taylorand Francis.
Schmidt, F. L., & Hunter, J. E. (1996). Measurement error in psychologicalresearch: Lessons from 26 research scenarios. Psychological Methods,1(2), 199–223. https://doi.org/10.1037/1082-989X.1.2.199
Schuerger, J. M., Zarrella, K. L., & Hotz, A. S. (1989). Factors that influencethe temporal stability of personality by questionnaire. Journal of Person-ality and Social Psychology, 56(5), 777–783. https://doi.org/10.1037/0022-3514.56.5.777
Sherman, R. A., & Serfass, D. G. (2015). The comprehensive approach toanalyzing multivariate constructs. Journal of Research in Personality, 54,40–50. https://doi.org/10.1016/j.jrp.2014.05.002
Socio-Economic Panel. (2018). Socio-Economic Panel (SOEP) data foryears 1984–2018 (Version 35).
Specht, J., Bleidorn, W., Denissen, J. J., Hennecke, M., Hutteman, R.,Kandler, C., Luhmann, M., Orth, U., Reitz, A. K., & Zimmermann, J.(2014). What drives adult personality development? A comparison oftheoretical perspectives and empirical evidence. European Journal ofPersonality, 28(3), 216–230. https://doi.org/10.1002/per.1966
Specht, J., Egloff, B., & Schmukle, S. C. (2011). Stability and change ofpersonality across the life course: The impact of age and major life eventson mean-level and rank-order stability of the Big Five. Journal ofPersonality and Social Psychology, 101(4), 862–882. https://doi.org/10.1037/a0024950
Specht, J., Luhmann, M., & Geiser, C. (2014). On the consistency ofpersonality types across adulthood: Latent profile analyses in twolarge-scale panel studies. Journal of Personality and Social Psychology,107(3), 540–556. https://doi.org/10.1037/a0036863
Thisdocumentiscopyrightedby
theAmerican
Psychological
Associatio
nor
oneof
itsalliedpublishers.
Thisarticle
isintended
solely
forthepersonal
useof
theindividual
user
andisnotto
bedissem
inated
broadly.
1336 WRIGHT AND JACKSON
Steca, P., Alessandri, G., & Caprara, G. V. (2010). The utility of a well-knownpersonality typology in studying successful aging: Resilients, undercontrol-lers, and overcontrollers in old age. Personality and Individual Differences,48(4), 442–446. https://doi.org/10.1016/j.paid.2009.11.016
Terracciano, A., Costa, P. T., & McCrae, R. R. (2006). Personality plasticityafter age 30. Personality and Social Psychology Bulletin, 32(8), 999–1009. https://doi.org/10.1177/0146167206288599
Terracciano, A., McCrae, R. R., Brant, L. J., & Costa, P. T., Jr. (2005).Hierarchical linear modeling analyses of the NEO-PI-R scales in theBaltimore Longitudinal Study of Aging. Psychology and Aging, 20(3),493–506. https://doi.org/10.1037/0882-7974.20.3.493
Terracciano, A., McCrae, R. R., & Costa, P. T., Jr. (2010). Intra-individualchange in personality stability and age. Journal of Research in Personal-ity, 44(1), 31–37. https://doi.org/10.1016/j.jrp.2009.09.006
Terracciano, A., Stephan, Y., Luchetti, M., & Sutin, A. R. (2018).Cognitive impairment, dementia, and personality stability among olderadults. Assessment, 25(3), 336–347. https://doi.org/10.1177/1073191117691844
Viken, R. J., Rose, R. J., Kaprio, J., & Koskenvuo, M. (1994). A develop-mental genetic analysis of adult personality: Extraversion and neuroticismfrom 18 to 59 years of age. Journal of Personality and Social Psychology,66(4), 722–730. https://doi.org/10.1037/0022-3514.66.4.722
Watson, N., &Wooden, M. (2012). The HILDA Survey: A case study in thedesign and development of a successful Household Panel Study. Longi-tudinal and Life Course Studies, 3(3), 369–381. https://doi.org/10.14301/llcs.v3i3.208
Wortman, J., Lucas, R. E., & Donnellan, M. B. (2012). Stability and changein the Big Five personality domains: Evidence from a longitudinal study ofAustralians. Psychology and Aging, 27(4), 867–874. https://doi.org/10.1037/a0029322
Received January 30, 2022Revision received May 31, 2022
Accepted June 2, 2022 ▪
Thisdocumentiscopyrightedby
theAmerican
Psychological
Associatio
nor
oneof
itsalliedpublishers.
Thisarticle
isintended
solely
forthepersonal
useof
theindividual
user
andisnotto
bedissem
inated
broadly.
PERSON-CENTERED PERSONALITY CONSISTENCY 1337
- Are Some People More Consistent? Examining the Stability and Underlying Processes of Personality Profile Consistency
- Outline placeholder
- Stability and Consistency of Personality
- Indices of Personality Profile Similarity and Consistency
- Processes Underlying Patterns of Personality Consistency
- Potential Factors Impacting Person-Centered Profile Consistency
- The Present Study
- Method
- Participants
- German Socioeconomic Panel Study
- Household Income and Labour Dynamics in Australia Study
- Health and Retirement Study
- Longitudinal Studies for the Social Sciences
- Measures
- Big Five
- GSOEP
- HILDA
- HRS
- LISS
- Moderators
- Transparency and Openness
- Analytic Plan
- Intraindividual Indices of Personality Profile Similarity and Consistency
- D Metrics
- Overall Profile Correlations
- Distinct Profile Correlations
- Interindividual Differences in Indices of Personality Profile Similarity and Consistency
- D Metrics
- Profile Correlations
- Results
- The D Metrics of Profile Similarity
- What Are Typical Levels of Within-Person Profile Consistency?
- Are Some People More Consistent Than Others in Their Personality Profiles?
- Which Processes Contribute to Personality Consistency?
- Discussion
- People Differ in Dispositional Personality Profile Consistency
- There Is Variability in the Processes Underlying Personality Consistency
- Limitations
- Conclusion
- References