Word Search


Patterns of Continuity: A Dynamic Model for Conceptualizing the Stability of Individual Differences in Psychological Constructs Across the Life Course

In contemporary psychology there is debate over whether individual differences in psychological constructs are stable over extended periods of time. The authors argue that it is impossible to resolve such debates unless researchers focus on patterns
of 15
All materials on our website are shared by users. If you have any questions about copyright issues, please report us to resolve them. We are always happy to assist you.
  Patterns of Continuity: A Dynamic Model for Conceptualizing the Stabilityof Individual Differences in Psychological ConstructsAcross the Life Course R. Chris Fraley University of Illinois at Chicago Brent W. Roberts University of Illinois at Urbana–Champaign In contemporary psychology there is debate over whether individual differences in psychologicalconstructs are stable over extended periods of time. The authors argue that it is impossible to resolve suchdebates unless researchers focus on patterns of stability and the developmental mechanisms that may giverise to them. To facilitate this shift in emphasis, they describe a formal model that integrates 3developmental processes: stochastic-contextual processes, person–environment transactions, and devel-opmental constancies. The theoretical and mathematical analyses indicate that this model makes novelpredictions about the way in which test–retest correlations are structured across a wide range of ages andtest–retest intervals. The authors illustrate the utility of the model by comparing its predictions againstmeta-analytic data on Neuroticism. The discussion emphasizes the value of focusing on patterns of continuity, not only as phenomena to be explained but as data capable of clarifying the developmentalprocesses underlying stability and change for a variety of psychological constructs. In 1963 Director Michael Apted and his colleagues interviewed14 British 7-year-olds about their dreams, fears, and aspirations.The documentary that resulted,  7 Up , was a critically acclaimedfilm about the lives of a diverse group of children who wouldultimately become Britain’s future (Almond & Apted, 1963). Inthe years that have followed, Apted has kept in touch with theseindividuals, interviewing them every 7 years about their relation-ships, accomplishments, and disappointments. The most recentupdate,  42 Up , was released in 1999.The 7 Up series is remarkable to watch because it allows theviewer to observe the unfolding of lives—from childhood tomiddle age—over the span of a few short hours. When watchingthis series, one cannot help but be struck by the degree of conti-nuity that characterizes some of the children. The child interestedin astronomy grows up to become a tenured professor of physics,and the timid, introspective child spends decades trying to discoverhis place in society. In contrast, other children exhibit markeddiscontinuities, coming across as arrogant and rebellious at age 21,for example, and humble and conventional 7 years later. Thediversity of developmental trajectories captured by the seriesprompts the viewer to ask, “How stable are individual differencesfrom infancy to adulthood?” Indeed, it is precisely this kind of question that Apted hoped to answer by working on the 7 Upseries. Inspired by the Jesuit maxim “Give me the child until he is7, and I will show you the man,” Apted sought to determine towhat extent the personality of the child foreshadows that of theadult.Apted has not been alone in his search for an answer to thisquestion. Psychologists have spent years mapping psychologicaldevelopment and tracing the unique pathways forged by people asthey negotiate the vicissitudes of life (see Block, 1971; Bloom,1964; Caspi & Roberts, 1999; Funder, Parke, Tomlinson-Keasey,& Widaman, 1993; Roberts & DelVecchio, 2000; Schuerger,Zarrella, & Hotz, 1989). Personality psychologists, for example,have focused on the stability of individual differences in person-ality traits, as commonly quantified by test–retest correlations (seeRobins, Fraley, Roberts, & Trzesniewski, 2001). On the basis of this research we now know that certain aspects of childhoodtemperament correlate about .20–.30 with adult personality char-acteristics (e.g., Block, 1993; Kagan & Moss, 1962). Moreover,research shows that personality traits appear to become increas-ingly stable over the life course, with children and young adultsexhibiting less stability than older adults (Roberts & DelVecchio,2000). In middle to late adulthood, for example, test–retest coef-ficients for basic personality traits often average around .50–.80(Costa & McCrae, 1994), whereas stability coefficients overequivalent periods of time in adolescence tend to be .30–.50(Roberts & DelVecchio, 2000).Although previous research has been able to establish some of the ways in which stability coefficients are patterned across the lifecourse, both for personality traits and other psychological con-structs, many psychologists continue to focus on the size of test–retest coefficients and, more specifically, on the question of whether the size of those coefficients suggests a small or largedegree of stability. One of the arguments we make in this article isthat the size of test–retest coefficients for a psychological constructhas little to say about the kinds of processes that promote stabilityand change. Our objectives in this article are to demonstrate thispoint and to propose a theoretical model that highlights patterns of continuity and the developmental processes that give rise to them.We begin by discussing some of the limitations of point estimatesof stability for addressing questions about developmental pro- R. Chris Fraley, Department of Psychology, University of Illinois atChicago; Brent W. Roberts, Department of Psychology, University of Illinois at Urbana–Champaign.Correspondence concerning this article should be addressed to R. ChrisFraley, who is now at the Department of Psychology, University of Illinoisat Urbana–Champaign, Champaign, IL 61820. E-mail: rcfraley@uiuc.edu Psychological Review Copyright 2005 by the American Psychological Association2005, Vol. 112, No. 1, 60–74 0033-295X/05/$12.00 DOI: 10.1037/0033-295X.112.1.60 60  cesses. As we show, point estimates often obscure informationabout the kinds of processes that may produce those estimates—information that is more readily revealed by focusing on patternsof coefficients across different ages and test–retest intervals. Tofacilitate this shift in emphasis, we discuss the patterns entailed bythree processes that have been emphasized in the literature onpsychological development: stochastic-contextual processes (e.g.,Lewis, 1997, 1999, 2001a, 2001b), person–environment transac-tions (e.g., Caspi & Bem, 1990; Caspi, Bem, & Elder, 1989; Neyer& Asendorpf, 2001; Sameroff, 1975), and developmental constan-cies (e.g., Bowlby, 1973; McGue, Bacon, & Lykken, 1993; Rob-erts & Capsi, 2003; Roberts & Wood, in press). We show howthese distinct processes can be conceptualized as elements of amore complete theoretical model and, via dynamic modeling andmathematical analysis (e.g., Haefner, 1996; Huckfeldt, Kohfeld, &Likens, 1982; van Geert, 1994), illustrate the novel kinds of patterns this theoretical model predicts. In doing so, we hope todemonstrate the value in focusing on patterns of continuity, notonly as phenomena to be explained but as data capable of clari-fying the developmental processes underlying stability and changefor a variety of constructs of interest to psychological scientists. Points and Patterns Researchers studying stability and change often assess variationin a psychological quality on two occasions and estimate theconstruct’s stability via a single test–retest coefficient. This ap-proach carries with it two assumptions. The first is that the endur-ing nature of a psychological variable can be revealed by themagnitude of its test–retest coefficient. This assumption is oftenexplicit. In fact, much of the theory and research concerningindividual differences is guided by the notion that stability coef-ficients should be high. The second assumption is that informationabout age or the duration of the test–retest interval is largelyirrelevant to understanding psychological dynamics. This assump-tion is largely implicit but can be inferred by noting that mostresearch tends to focus only on two assessment waves, and in casesin which more than one wave of data are available, researchersfrequently aggregate all pairwise test–retest coefficients to yield asingle, composite estimate of the stability of the construct underinvestigation.To see why both of these assumptions may be problematic,consider the histogram of test–retest coefficients diagrammed inthe left-hand panel of Figure 1. These coefficients correspond tothe stability of individual differences in the Big Five personalitytraits and are the same coefficients that Costa and McCrae (1994)analyzed when they concluded that personality traits do not changein adulthood. The average of these coefficients is approximately.65. This estimate, however, poses a number of interpretive ambi-guities. For one, it is difficult to ascertain whether a value of .65is consistent or inconsistent with the theoretical perspective underdiscussion because researchers rarely make  point predictions  (i.e.,precise quantitative predictions) about the magnitude of the aver-age test–retest coefficient that should be observed over time. Acoefficient of .65 clearly indicates some degree of stability, but itis not clear whether this particular value is more consistent with aperspective that emphasizes instability (e.g., Lewis, 1999) over Figure 1.  Alternative explanations for stability and change. Left: A histogram of test–retest correlations thathave been observed in adulthood for the Big Five personality traits. Right: Curves that fit these data but havedramatically different implications for the test–retest correlations over increasing test–retest intervals (in years).The solid curve illustrates a function that approaches zero in the limit; the dashed curve, in contrast, approachesa value of approximately .55. 61 MODELING DEVELOPMENTAL MECHANISMS  stability (e.g., Costa & McCrae, 1994; McCrae et al., 2000).Although values of 0.00 or 1.00 have clear interpretations, valuesbetween these extremes comprise a vast gray area in which theimplications of the data for theory become elusive.It is important to note, however, that there is additional infor-mation available to us that has the potential to inform our under-standing of continuity and change. For example, if we disaggregatethe coefficients in this example, it is possible to discern a patternin which the magnitude of the coefficients decreases as the lengthof the test–retest interval increases (see the right-hand panel of Figure 1). This pattern is noteworthy because it suggests that theway in which personality traits change is systematic and, therefore,begs for a theoretical explanation. Moreover, with some imagina-tive extrapolation, one can envision alternative curves that not onlycapture these data but have different implications for data yet to beobserved. For example, if these coefficients were to decrease tozero as the test–retest interval increases, as illustrated by the solidline in the right-hand panel of Figure 1, the implications for thestudy of personality development would be radically different thanif the coefficients approached a nonzero asymptote, as illustratedby the dashed line. The former curve suggests that even if we canpredict a person’s trait level with a high degree of accuracy overthe span of a few months, we will not be able to do so over longerperiods of time. Specifically, the predicted test–retest correlationapproaches zero as the test–retest interval gets larger. The lattercurve, in contrast, indicates that it might be possible to predictindividual differences even over quite extended periods of time.Most important, this curve suggests that there is an enduringquality to the personality trait, although the overall magnitude of the stability coefficient is not very high.The observation that the stability coefficients may asymptote atdifferent values suggests that the common assumption that thestability of a psychological variable is reflected in the size of anyone test–retest correlation is incomplete and potentially mislead-ing. Instead, we propose that the stability of a construct is reflectedin the way in which its test–retest coefficients decay across in-creasingly long intervals or, more specifically, the way in whichcoefficients are patterned across a range of ages and test–retestintervals. It is possible for measurements to exhibit high test–retestcorrelations across two time points, but if these correlations ap-proach zero as the test–retest interval increases, the psychologicalentity is clearly not an enduring one.It is important to note that this distinction can be captured onlyby attending to the pattern of test–retest coefficients; the magni-tude of any one coefficient does not reveal the way in which thecurves decay across time. As a consequence, a single coefficienthas little to say about the dynamic processes that may underliestability and change. We believe that psychological science wouldbenefit enormously by treating such patterns as a valuable sourceof data—as not only phenomena that need explanation but also asphenomena that have the potential to provide insight into the kindsof developmental processes that give rise to continuity and change. Empirical Patterns of Continuity: An Illustration The observation that different theoretical functions can explaina single test–retest coefficient implies that it is impossible tounderstand the processes underlying psychological developmentwithout examining patterns of stability and change. To advancecurrent knowledge about psychological continuity, researchers inthe field must move beyond the current focus on the magnitude of test–retest correlations and begin taking into account the way inwhich those correlations are patterned across different ages andtemporal intervals. In the following sections we take a first step inthis direction by using empirical data on the stability of variouspersonality traits to illustrate the kinds of patterns that exist in onearea of psychological inquiry. Although we focus on the continuityof personality traits, it is important to emphasize that empiricalpatterns could be reconstructed for any kind of psychologicalconstruct for which people vary (e.g., anxiety, attachment security,attitudes, depression, intelligence, resilience, self-esteem, subjec-tive well being, working memory capacity). We highlight basicpersonality traits here because of the extensive amount of longi-tudinal research that has accumulated in the study of personalitytrait development and our expertise in this specific area of research. 1 For our illustration we reexamined the meta-analytic data src-inally compiled by Roberts and DelVecchio (2000). The details of their data set are reported in depth in their srcinal article; there-fore, we focus here on the novel ways in which we reconfiguredand examined that data set. Roberts and DelVecchio were inter-ested in whether personality trait stability tends to be higher inadulthood than in childhood. Accordingly, they meta-analyzedtest–retest coefficients from a variety of longitudinal studies andregressed the magnitude of those coefficients on age, while hold-ing constant the length of the test–retest interval. Although thisstrategy allowed Roberts and DelVecchio to address the waystability varies as a function of age, it does not allow one toconsider the ways in which stability coefficients might be pat-terned over varying temporal intervals. To investigate such pat-terns, one must assemble a test–retest correlation matrix thatcaptures the meta-analytic stability coefficients observed acrossnumerous pairwise ages (e.g., age 1 to age 2, age 1 to age 10, age23 to age 30).One advantage of reexamining this meta-analytic database inparticular is that it contains information on a broad spectrum of personality traits—traits that have been organized according to theBig Five personality trait taxonomy (i.e., Extraversion, Agreeable-ness, Conscientiousness, Neuroticism, and Openness; John, 1990).Although some of the traits were not srcinally studied within theframework of the Big Five, it is useful to classify each trait asfalling within one of the five categories to aid the synthesis andanalysis of the patterns of continuity. In this sense, we are notexamining the Big Five as objective entities but are using thetaxonic qualities of the Big Five framework to facilitate the com-munication of findings across numerous personality trait measures.To construct a meta-analytic correlation matrix to characterizethe stability of each personality trait across all available pairwiseages, we first rounded the ages for each assessment across studiesto the nearest integer (e.g., age 3.7 became age 4). Next, for eachof the five major trait domains (i.e., Extraversion, Agreeableness,Conscientiousness, Neuroticism, and Openness), we filled in as 1 We should note that even within the science of personality, traitapproaches are only one of several ways of studying personality. Somealternative approaches are discussed by Cervone and Shoda (1999),Mischel and Shoda (1995), and McAdams (1995). 62  FRALEY AND ROBERTS  many cells of the corresponding correlation matrix as we couldwith studies reporting test–retest coefficients corresponding to thatparticular cell. A total of 3,218 rank-order stability coefficientswere examined based on 50,207 participants across 124 longitu-dinal samples. For cells in which multiple studies existed, theempirical coefficients were aggregated using Fisher  r  -to-  z -to- r  methods, with each sample correlation weighted by its sample size(Hedges & Olkin, 1985). 2 It should be noted that the averageinternal consistency estimate across these studies was .72 and thatthis value was invariant across ages. As such, there is no reason toassume that cross-age variations in trait stability are due to cross-age variations in measurement precision.Because the five correlation matrices were remarkably similar,we have presented the matrix for Neuroticism for illustrativepurposes in Table 1. The first thing to note is that certain cells of this (and other) meta-analytic correlation matrices were easier tofill than others. The bulk of empirical longitudinal data exists forearly childhood and the 20s rather than other parts of the life span.Furthermore, although empirical data exist for lengthy test–retestintervals (e.g., periods spanning 10 or more years), such data existprimarily for time spans covering later adulthood and early child-hood. There were no studies, for example, that investigated thestability of personality traits from age 8 to age 30. Because of therelative lack of data for ages 30 and beyond, we have onlypresented the data for ages 1–30 in Table 1.There are several noteworthy patterns that can be discerned inTable 1. Notice first that the meta-analytic correlations betweenage 1 and all subsequent ages do not approach zero. In otherwords, it is not the case that the stability coefficients get smallerand smaller as the length of the test–retest interval increases.Although the coefficients decay quickly over brief intervals, theirdecay is not continuous and appears to plateau at modest values.Thus, individual differences in Neuroticism tend to be highlystable across brief periods of time but become less stable as thetime interval increases. Instead of becoming increasingly unstable, 2 We acknowledge that aggregating data across a variety of differentforms of assessment has the potential to obscure, rather than clarify, thepatterns of interest. We should note, however, that analyses reported byRoberts and DelVecchio (2000) suggest that stability coefficients based onself-report instruments were no stronger or weaker than those based onother kinds of assessments (e.g., observer ratings). Moreover, we havetaken steps to only aggregate coefficients corresponding to measures thatare believed, on the basis of theory or evidence, to reflect similar individualdifference constructs (Angleitner & Ostendorf, 1994; Shiner, 1998, 2000).In short, although aggregation has the potential to introduce noise to thepatterns we are attempting to understand, aggregation is the only means bywhich we can assemble patterns of this breadth and scope. As is discussedin subsequent sections, the patterns that we obtained are remarkably clear,despite the noise inherent in the process. Table 1  Meta-Analytic Test–Retest Correlations for the Trait of Neuroticism for Ages 1 Through 30 AgeAge1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 301 —2 .30 —3 .22 .27 —4 .22 .34 .39 —5 .13 .06 .18 .60 —6 .14 .28 — .56 .48 —7 — — .34 .54 .54 .54 —8 — .34 .26 .45 .59 .54 — —9 — — — .45 — .51 .63 .47 —10 — .21 — .54 .53 .34 .39 — .66 —11 — — .22 .46 — — .41 .13 — — —12 — .29 .37 .26 .42 .20 — .62 .50 .63 .41 —13 — — — — — — — — .56 .49 .37 .54 —14 — — — — — — — — — .33 .47 .55 .58 —15 .12 — — — — .22 — — .38 — — .37 .46 .61 —16 — — — .31 — — .31 .40 .39 .63 — .61 — .54 .55 —17 — — — — — — — — — — — — — .42 .13 .49 —18 — —   .06   .23 — — — — — — — — — .40 .49 .31 .67 —19 — — — — — — — — — — — — — — — — .57 .63 —20 .18 — — .36 .15 — — — — — — — — — — .55 .33 .59 .53 —21 — — — — — — — — — — — — — — — — .42 .56 .58 .83 —22 — — — — — — — — — — — — — .29 — — — .55 — — — —23 — — .15 — — — .07 — — — — — — — — — — .74 — .35 — — —24 — — — — — — — — — — — — — — — — .22 — — — — — — —25 — — — — — — — — — — — — — — — — — — — — — .67 — — —26 — — — — — — — — — — — — — — — — — — — .59 — — — — — —27 — — — — — — — — — — — — — .35 — — — .41 — — .54 .36 — — — — —28 — — — — — — — — — — — — — — — — — — — .52 — .32 — — — — — —29 — — — — — — — — — — — — — — — — — — — — — — — — — — — — —30 — — — — — — — — — — — — — — — — — .24 .45 .54 — — — — — — — — — —  Note.  Dashes appear in cells for which no data existed. 63 MODELING DEVELOPMENTAL MECHANISMS  however, the degree of stability plateaus at a value of about.20–.30. Second, notice that the test–retest correlations tend to behigher later in the life span than early on. This suggests that theway in which personality dynamics play out in adulthood may bedifferent than the way they play out in childhood. In summary,although the longitudinal data on trait stability are far from beingcomprehensive, it is possible to configure those data in a mannerthat allows specific patterns to be delineated.One of the challenges for psychological science is to developtheoretical models that are capable of making predictions aboutpatterns of coefficients and to determine whether those models canaccount for the kinds of patterns we have highlighted. In thesections that follow we attempt to facilitate this endeavor byreviewing three developmental processes; integrating those pro-cesses into a formal, mathematical model; and, via simulation andmathematical analysis, illustrating the kinds of patterns this modelpredicts. After we clarify the predictions of this theoretical model,we highlight some of the ways in which the model is able toexplain the data summarized in Table 1 as well as some ways inwhich the model may be elaborated in the future. Developmental Processes Giving Rise to Stability andChange Over the last few decades, researchers have discussed a varietyof mechanisms that might give rise to stability and change in avariety of psychological constructs. Although these differentmechanisms are sometimes treated as if they are mutually exclu-sive or in competition with one another, we believe that they canbe integrated to provide a more complete and constructive frame-work for conceptualizing the developmental processes that affectrank-order stability. In the sections that follow, we review threemechanisms of stability and change that have been discussed in thecontemporary literature: stochastic-contextual processes, person–environment transactions, and developmental constancy factors.We focus on these dynamic mechanisms in particular because theyare widely used to explain stability and change for a variety of constructs of interest to psychologists. Transactional processes, forexample, have been proposed as one set of mechanisms underlyingstability and change in domains as diverse as attachment theory(Bowlby, 1973), information-processing approaches to under-standing child aggression (Dodge, 1986), and extraversion (e.g.,Neyer & Asendorpf, 2001). As we review each set of processes,we highlight the broader thrust of the theoretical perspectives inwhich they are embedded. It is not our intention to review all of theprocesses and nuances that are embodied by different perspectives;instead, our goal is to distill the central ideas that are discussed inthe contemporary literature. After doing this, we show how theseprocesses can be mathematically formalized within a more com-prehensive developmental model—a model that has the potentialto clarify the patterns of stability and change that may exist indifferent psychological constructs. Stochastic Mechanisms In 1997, Michael Lewis published a book titled  Altering Fate:Why the Past Does not Predict the Future . In this influential andcontroversial volume, Lewis made the argument that an important,and often overlooked, mechanism in psychological development ischance. At various points in life we might relocate to a new town,change classrooms, lose a loved one, or discover a new talent.Each of these events has the potential to influence psychologicalfunctioning and, in some cases, can be conceptualized as varyingacross people in a random manner. To the extent that these randomor stochastic factors affect the ebb and flow of psychologicaldevelopment, we might expect instability in individual differencesover time.The stochastic perspective offers an important anchor for dis-cussions of development because it implies that people are likelyto change substantially over time or, more precisely, that thedegree of change depends on the stability of context. Becausedevelopment is contextual, the behaviors people exhibit in onesituation may or may not carry over to other situations (see alsoHarris, 1998). Similarly, because the situations that people facemay be distributed randomly across time, people are expected todevelop in a manner that is difficult to anticipate without fullknowledge of their present circumstances.The ideas proposed by Lewis (1997) are very similar to thosediscussed by other developmental psychologists. Kagan (1996),for example, recently argued that psychologists tend to labor underthe belief in three “pleasing,” but false, ideas—one of these beingthat early experiences and behavioral patterns play an importantrole in foreshadowing the adult personality. Kagan’s position, likeLewis’s, recognizes that there are a number of psychologicallyinfluential events that can intervene in development, thereby de-creasing the likelihood that shy children, for example, will grow upto become shy adults. Harris (1998) has made a similar argument,suggesting that competencies acquired in one situation (e.g., theearly family environment) are unlikely to carry over or transfer tonew situations (e.g., peer relations) because those situations willdemand their own set of skills and adaptations. As a consequence,there is no reason to expect people to grow up to be the same kindsof people they were as children. Person–Environment Transactional Mechanisms Many writers, while recognizing that there are aspects of lifethat are beyond our control, have argued that people play an activerole in shaping their social, emotional, and intellectual environ-ments. As a consequence, the environmental events that come toinfluence the person are caused, in part, by the person. Thetransactional dynamics that take place between persons and theirenvironments have been emphasized by a number of personality,developmental, and social–cognitive psychologists (e.g., Bowlby,1973; Caspi & Bem, 1990; Dodge, 1986; Magnusson, 1990; Neyer& Asendorph, 2001; Sameroff, 1975) and are considered to becritical for promoting the persistence of people’s attitudes, behav-iors, and feelings.In an influential review of transactional mechanisms, Caspi andBem (1990) summarized at least three ways in which systematictransactions between people and their environments can promotestability (see also Caspi & Roberts, 1999). One way in whichtransactional processes can do so is through  proactive  means, suchas when people select social environments that are consistent withtheir existing dispositions. For example, a sociable individual maychoose to affiliate with outgoing people, thereby increasing his orher tendency to see himself or herself as a sociable person. Anotherway in which transactional processes can promote stability is 64  FRALEY AND ROBERTS
Related Documents
View more
Related Search
We Need Your Support
Thank you for visiting our website and your interest in our free products and services. We are nonprofit website to share and download documents. To the running of this website, we need your help to support us.

Thanks to everyone for your continued support.

No, Thanks