What refer to the relatively stable properties that describe elements of personality?

Rank-order stability of SWB

Rank-order stability refers to the extent to which the rank order of individuals with respect to their level of SWB is stable over time. The rank-order stability of SWB can be estimated by the retest correlation of SWB measured at two separate time points. Meta-analyses found that for both life satisfaction and affect, the retest correlation decreases with increasing time intervals between the two time points until it reaches an asymptote of about r = .30 [Anusic & Schimmack, 2016; Schimmack & Oishi, 2005]. This asymptote can be interpreted as a lower-bound estimate of the long-term rank-order stability of SWB [Fraley & Roberts, 2005]. Hence, these estimates suggest that SWB is partly characterized by a stable trait component. Compared to the asymptote for the Big Five of about r = .60 [Anusic & Schimmack, 2016], however, SWB is less stable than personality traits.

Differential Stability

This type of stability, also called rank-order stability, refers to the degree to which the relative ordering of individuals on a given variable is maintained over time. It explicitly requires longitudinal research and is typically assessed through test–retest correlations or stability coefficients of measurement occasions separated by a specified time interval. This type of stability can address the question of whether people retain their standing on a trait dimension relative to others over time. A high test–retest correlation would implicate either that individuals are stable in a given trait over time or are changing, but in more or less the same way. This situation can occur when a normative developmental event such as retirement impacts all individuals in the same way [eg, if retirement causes everyone to decrease or increase in a personality trait by the same amount]. By contrast, a low test–retest correlation indicates that individuals are changing over time and there are individual differences in the direction of change, implying that some individuals are increasing in a personality trait whereas others are decreasing. This can occur when non-normative events impact personality traits [eg, if some individuals get divorced and decline or incline in a particular personality trait whereas others do not experience this life event and maintain the same personality trait level]. In addition, a low test–retest correlation can also occur when the factors that influence the personality trait are normative but individuals have unique reactions to these events [eg, if retirement causes some individuals to increase in a personality trait but causes others to decrease in the same trait]. Finally, from a methodological point of view, a low test–retest correlation could also simply reflect measurement error or less reliable measurements.

Longitudinal studies have been conducted to investigate differential stability of the Big Five personality traits. In order to test whether trait stability maximizes and stabilizes at a specific period in the lifespan, Roberts and DelVecchio conducted an extensive meta-analysis and included 152 longitudinal studies. Estimates of mean population test–retest correlation coefficients showed that the overall trait stability increased from 0.31 in childhood to 0.54 during the college years, to 0.64 at age 30, and then reached a plateau around 0.74 between ages 50 and 70. Their findings suggest that there is a tendency for increasing relative stability of personality traits from childhood to old adulthood, a pattern that has been termed the “cumulative stability principle.” Moreover, differential stability did not vary markedly across the Big Five traits, nor across assessment method [eg, self-reports, observer-ratings, and projective tests], or by gender. A slightly different pattern of trait stability is found for very old adults [ie, 80–100 years]. Results from previous studies indicate a decrease in differential stability in personality traits like emotional stability [or inversely, neuroticism], imagination, sensitivity, and dominance in very old age. Hence, personality seems to become more plastic and prone to change again in very old age. To conclude, although a relatively high differential stability coefficient around 0.70 is found in late middle and older adulthood, it is not perfect [1.0] and leaves room for individual change. As mentioned above, individual change can be due to unexpected life events or off-time lifespan transitions. The decrease in trait stability in very old adults might point to an increase in vulnerability as social and psychological resources diminish with age. For example, cognitive impairments, which are quite prevalent in very old individuals, can also decrease the stability of certain personality traits.

Mean-Level and Rank-Order Stability of the Big Five Across the Life Span

What are the patterns of normative personality development? What is the pattern of rank-order stability across the life span? Summarizing research on these topics has been greatly facilitated by a number of meta-analyses that were conducted in the last decade. A meta-analysis is a statistical approach for summarizing research literatures and essentially involves averaging the results from all available studies. This method permits researchers to quantify findings across the entire research literature and is perhaps optimal to narrative reviews, which simply provide qualitative impressions of previous work.

A meta-analysis by Roberts, Walton, and Viechtbauer provided a summary of average levels of the Big Five traits using data from 113 longitudinal samples involving 50 120 participants ranging in age from adolescence through old age. These researchers divided the extraversion domain into two facets: social dominance [traits related to independence and dominance] and social vitality [traits related to positive affect, activity level, and sociability]. Average levels of social vitality tended to be fairly flat across the life span, although there was a slight spike upward from adolescence to young adulthood followed by a plateau until the mid-50s when there was a slight decline. Social dominance, on the other hand, showed a more pronounced trend such that there was a consistent absolute increase from adolescence to the early 30s where mean levels remained consistent until the mid-50s, after which the lack of studies precluded further analyses. Agreeableness and conscientiousness showed gradual increases in absolute scores across the life span whereas neuroticism showed gradual decreases. Finally, openness showed a mean-level increase from adolescence to young adulthood and then mean levels remained constant until the mid-50s when it started to show a slight decline in average levels.

One of the more interesting results that emerged from this meta-analysis concerned the adolescent period. Contrary to the popular belief that adolescence involves tumultuous changes in personality, the Roberts et al. meta-analysis demonstrated that the largest mean-level changes in personality occur during the young adult years [i.e., the 20s]. This is the phase in the life span when individuals assume the roles of worker, committed romantic partner, and in many cases, parent and caregiver. Furthermore, it is easy to see how increasing levels of agreeableness and conscientiousness and decreasing levels of neuroticism facilitate the successful enactment of the roles of worker, parent, and committed romantic partner. Thus, average levels of traits change in ways that coincide with the time in the life span during which individuals assume mature social roles, a pattern referred to as the maturity principle of adult personality development.

In summary, the available data indicate that average levels of agreeableness and conscientiousness increase with age whereas average levels of extraversion [in the aggregate], neuroticism, and openness decline with age. There are two dominant explanations for these mean-level differences in the Big Five domains across the life span. The intrinsic maturational position holds that normative age-related changes in personality are driven by unfolding biological processes related to aging whereas the life course position posits that changes stem from investment in particular social roles and the life experiences that accompany these roles. Researchers are currently debating which perspective has the most empirical support. One of the complicating factors is that critical tests of these two explanations are nearly impossible because experimental manipulations of either biological factors or important social roles are neither ethical nor feasible.

At least two different meta-analyses have investigated rank-order stability across the life span. Roberts and DelVecchio examined test–retest correlations from 152 longitudinal studies and found that the rank-order stability of personality increases across the life span, ranging from a low around 0.30 in childhood to a high of 0.70 in late adulthood. This pattern generally held for men and women and for all five of the Big Five traits. A more recent meta-analysis by Ferguson reached similar conclusions, although in his analysis rank-order stability reached a plateau earlier in development than in the Roberts and DelVecchio analysis [perhaps because he corrected the rank-order stability estimates for measurement error].

The finding that the rank-order stability of personality increases from childhood to adulthood is known as the cumulative continuity principle of personality development – that is, personality becomes increasingly stable with age [when viewed through the lens of rank-order stability]. This naturally raises questions as to why rank-order stability increases with age. Lower stability is expected when individuals respond to experiences differently or experience personality-altering environments at different times. The transition from childhood to adolescence involves rapid maturational changes, shifting societal demands, exploration of new identities and roles, and initiation of new peer and romantic relationships. These changes may impact individuals in relatively unique ways, thus shifting their relative ordering on a trait and thereby reducing stability coefficients. In contrast, the transition to adulthood is accompanied by fewer maturational changes and social transitions that begin to stabilize. Likewise, a hallmark of adulthood is the increased ability to select environments consistent with individual dispositions. These broad developmental considerations may explain the cumulative continuity principle. Indeed, researchers are now moving beyond simply documenting patterns of mean-level and rank-order stability to testing hypotheses about the underlying mechanisms that produce personality consistency and change.

3.1 Social dominance, competition, and aggression

Despite diurnal declines, assessments of testosterone [and other hormones] show fairly high rank-order stability over repeated assessments, particularly when those assessments occur at the same time of day [e.g., Liening, Stanton, Saini, & Schultheiss, 2010]. Baseline or endogenous testosterone measures are therefore considered somewhat “trait-like” [Sellers, Mehl, & Josephs, 2007], making it possible to assess correlations between testosterone and other stable constructs, such as personality traits. In general, baseline testosterone has not been consistently associated with individual differences in broad personality traits, such as the “Big Five” [i.e., neuroticism, extraversion, openness to experience, agreeableness, and conscientiousness]: A recent meta-analysis of nearly 4,000 participants from 25 samples revealed fairly weak and mostly nonsignificant associations between baseline salivary testosterone levels and the Big Five personality traits, although there was some evidence for small negative associations with conscientiousness [Sundin et al., 2021].

Instead, higher baseline testosterone has been most consistently been linked with somewhat narrower traits that are implicated in competition and dominance, including higher self-reported competitiveness, need for power, social dominance, and sensation- or novelty-seeking [e.g., Arnocky, Albert, Carré, & Ortiz, 2018; Campbell et al., 2010; Määttänen et al., 2013; Perini, Ditzen, Hengartner, & Ehlert, 2012; Sellers et al., 2007; Stanton & Schultheiss, 2009; Turan, Guo, Boggiano, & Bedgood, 2014]. Participants with higher baseline testosterone levels, and those whose testosterone has been temporarily increased via pharmacological administration or experimental manipulation, also show more risk-taking and competitive behavior [e.g., Apicella, Dreber, & Mollerstrom, 2014; Goudriaan et al., 2010; Kordsmeyer & Penke, 2019; Stanton, Liening, & Schultheiss, 2011; Zilioli & Watson, 2014]. Importantly, although women are much less likely than men to be included in these studies, associations are often similar for men and women when both are included in the same study [e.g., Sellers et al., 2007; Stanton et al., 2011; Sundin et al., 2021].

Engaging in competitive activities, such as sports or videogames, can also increase testosterone in both men and women [Casto & Edwards, 2016; Edwards & Casto, 2013]. For instance, in a study of male and female cross-country runners, both men and women showed pre- to post-race increases in testosterone, and these increases were unrelated to the runners’ finish times [Casto, Elliott, & Edwards, 2014]. In attempt to isolate the effects of competition specifically from those involved in physical exertion more generally, the “competition effect” has also been assessed in more tightly controlled laboratory scenarios [e.g., using video games or competitive puzzles]. Findings from such studies suggest that competition that does not involve physical exertion may nevertheless increase testosterone in both men and women, regardless of the competition outcome [see Casto & Edwards, 2016, for a review]. Competition effects from laboratory studies appear to be less consistent than those conducted in the field, however, and point to considerable variability in testosterone responses.

At least one source of variability in testosterone responses is the competition outcome: In many cases there is evidence for a “winner-loser effect,” such that testosterone increases are often larger among winners compared to losers, who in some cases show testosterone decreases [e.g., Casto & Edwards, 2016; Geniole, Bird, Ruddick, & Carré, 2017; Jiménez, Aguilar, & Alvero-Cruz, 2012]. In one study, for instance, men who won a [rigged] videogame competition against another male participant showed increases in testosterone following the competition compared to men who lost the competition. Moreover, men who showed larger testosterone increases, regardless of whether they won or lost the competition, performed better in a subsequent competitive activity [Zilioli & Watson, 2014], suggesting that increases in testosterone may feed forward to influence competitive behavior.

Perceptions of ability and performance have also been associated with testosterone levels and changes in testosterone: Female flag football players showed a pre-to post-game increase in testosterone, regardless of whether they won or lost; however, both winners and losers who felt more positively about their performance showed higher testosterone across time points and larger pre- to post-game increases in testosterone [Casto, Rivell, & Edwards, 2017]. Interestingly, findings from a recent meta-analysis suggest that winner-loser effects also appear to be stronger for studies conducted in the field [e.g., sporting events] versus the lab [e.g., playing videogames; Geniole et al., 2017]. Differences between lab and field studies suggest not only that it may be more difficult to elicit winner-loser effects in the lab, but also that there are likely contextual factors influencing testosterone reactivity that are not yet fully understood. Winner-loser effects also appear to be more consistently observed among male versus female participants; however, women are much less likely to be included in studies of competition than men [Carré & Olmstead, 2015].

Higher testosterone has also been linked with dominance and aggression in romantic relationships specifically: In a sample of partnered men, those with higher baseline testosterone reported more verbal and physical aggression directed toward their female partners [Soler, Vinayak, & Quadagno, 2000]. Similarly, in a study of dating couples—one of the few to include both men and women—Kaiser and Powers [2006] found that men with higher baseline testosterone were more likely to report using physically and psychologically aggressive strategies with their female partners during conflict, but only when their partners also had relatively high testosterone compared to other women in the sample. When higher testosterone men were paired with lower testosterone women, they were less likely to engage in physically aggressive behaviors. The authors argued that lower testosterone women may have been more effective in diffusing tension and conflict in the relationship, perhaps contributing to higher testosterone men's lower aggression in this particular pairing. Effects were largely nonsignificant for women's behavior, although there was a trend suggesting similar patterns for women's physical aggression. Together, these findings suggest that conflict strategies may be at least somewhat dependent on the relative hormone concentrations of each partner, and that women's aggression may be less closely tied than men's to their or their partner's testosterone levels. More generally, these findings highlight the importance of considering the hormone levels of both relationship partners when assessing interpersonal outcomes.

Continuity and stability of personality traits across the life-span

Personality stability is itself a complex notion because there are many different kinds of continuity and change [Caspi & Shiner, 2006]. First, “rank-order stability” refers to the degree to which the relative ordering of individuals on a given trait is maintained over time. Rank-order stability is high if people in a group maintain their position on a trait relative to each other over time, even if the group as a whole increases or decreases on that trait over time. It is typically indexed by correlations between scores on the same trait measured across two points in time [i.e., test–retest correlations]. Differential continuity describes the degree to which the relative differences among individuals remain invariant across time. Mean-level stability refers to the extent to which personality scores change over time. Investigations of mean-level change address the question of whether people, on average, tend to increase or decrease on particular trait or symptom measures during different life periods.

To investigate differential stability, longitudinal designs are required, whereas mean-age stability can be examined through longitudinal data [Roberts, Walton, & Viechtbauer, 2006]. In addition, mean traits scores from cross-sectional age cohorts can be employed for mean-level stability comparisons [McCrae et al., 2000]. Structural continuity refers to the invariance of the covariance structure across time and is a prerequisite for the assessment of mean-level stability [Biesanz, West, & Kwok, 2003]. Individual-level change refers to the magnitude of increase or decrease exhibited by a person on any given trait. Ipsative stability refers to the continuity of the configuration of traits within the individual and provides information on the stability of the patterning of traits within a person across time, hence facilitating a person-centered approach to personality development [Robins & Tracy, 2003].

De Fruyt et al. [2006a] examined these five types of personality stability [structural, mean-level, individual-level, differential, and ipsative] in a representative population of children and young adolescents [N = 498] and a twin and sibling sampling sample [N = 548] of children and adolescents. Parents described their children on two consecutive occasions with a 3-year interval using the Hierarchical Personality Inventory for Children [HiPIC; Mervielde & De Fruyt, 1999]. The results confirmed structural continuity in the two samples, and personality appeared to be largely differentially stable. A large percentage of children had a stable trait profile indicative of ipsative stability, and mean-level personality changes were generally small in magnitude. Continuity findings were generally attributed to genetic and nonshared environmental factors. The evidence for different types of personality continuity supports and extends previous research revealing that the level of continuity in childhood and adolescence is higher than often expected [Roberts et al., 2006]. A large number of empirical studies have examined the patterns of continuity and change in personality traits [e.g., Roberts et al., 2006] and their relation to well-being in adulthood. In recent years cross-sectional and longitudinal studies have demonstrated that agreeableness, conscientiousness, emotional stability, and social dominance [e.g., social self-confidence] increase from young adulthood to middle age [Lucas & Donnellan, 2011; McAdams & Olson, 2010].

In a study, Josefsson et al. [2013] examined the developmental patterns of the Temperament and Character Inventory [TCI] traits in a large population-based longitudinal study of Finnish men and women aged 20–45 years. Mean-level changes demonstrated qualitatively distinct developmental patterns for character [self-directedness, cooperativeness, and self-transcendence] and temperament [novelty seeking, harm avoidance, reward dependence, and persistence]. Personality developed toward greater maturity, although self-transcendence decreased with age. However, self-transcendence was the strongest predictor of overall personality change. Cohort effects indicated lower level of self-transcendence and higher level of self-directedness and cooperativeness in younger birth cohorts. Regarding temperament, novelty seeking decreased and persistence increased slightly with age. Both high novelty seeking and high persistence predicted overall personality change. These findings suggest that temperament and character traits follow different developmental trajectories.

Although personality traits have traditionally been defined as enduring patterns of thinking, feeling, and behaving [Costa & McCrae, 1997], contemporary theories combine a dynamic perspective that conceptualizes traits as developmental constructs subject to change and adaptation throughout the life-span [e.g., Caspi, Roberts, & Shiner, 2005]. Efforts to evaluate these processes [1] have focused on describing group-level change, [2] have focused on higher-order traits [rather than those at the facet level], and [3] were limited in their ability to determine nonlinear change [due to their analytic framework or use of only two or three waves of assessments].

It is important to distinguish between mean-level personality change, which evaluates how individuals develop over time on average, and rank-order change [i.e., change in the relative position of individuals on a trait over time] [Caspi et al., 2005b]. Mean-level personality change combined with rank-order stability implies that the mean-level change is due to normative [i.e., norm-factoring] change in personality [Klimstra, Hale, Raaijmakers, Branje, & Meeus, 2009].

Theories of the process that could underlie personality developments could be improved by more accurate knowledge about the progress of mean-level changes with age [i.e., linear vs. nonlinear], their degree of consistency across different facets of higher-order traits, and the extent to which individuals deviate from mean-level trajectories at the population level. A meta-analysis of 14 studies with samples aged 10–20 years revealed that early adolescence was associated with decreases in conscientiousness, openness, extraversion, and emotional stability [Denissen, Aken, Penke, & Wood, 2013], supporting the findings of earlier studies [e.g., Harden & Tucker-Drob, 2011; Soto, John, Gosling, & Potter, 2011].

Second, findings regarding sex differences have been less consistent than those for general maturational trends. Some evidence suggests that the changes in higher-order traits in young adulthood are fairly uniform across genders [Blonigen, Carlson, Hicks, Krueger, & Iacono, 2008; Donnellan et al., 2007]. During late adolescence and young adulthood, traits associated with behavioral constraint [e.g., conscientiousness] have been found to increase more rapidly in females than in males [Blonigen et al., 2008; Branje, Van Lieshout, & Gerris, 2007; Donnellan et al., 2007; Klimstra et al., 2009; Soto et al., 2011].

Third, most prospective studies of personality development in either adolescence or young adulthood have focused on changes at the higher-order level of the trait hierarchy—Big Five or Big Three domains [De Fruyt et al., 2006b; Hopwood et al., 2011; McCrae et al., 2002]. However, in their cross-sectional study, Soto et al. [2011] observed different age trends for several facets from the same domain. Differences are most prevalent across facets of neuroticism/negative emotionality and conscientiousness/behavioral constraint [Jackson et al., 2009]. Similarly, in an epidemiological sample of youths aged 12–24 years, Harden and Tucker-Drob [2011] reported divergent patterns of change for facets of behavioral constraint. Mean levels of impulsivity declined, whereas levels of sensation seeking exhibited a nonlinear pattern, increasing during early adolescence then gradually declining over late adolescence and young adulthood. Overall, these findings suggest that lower-order traits may reveal a more complex picture of the rate and timing of personality maturation.

In a study, Hicks et al. [2013] used data from large community epidemiological samples to explore trajectories of personality change between 11 and 30 years of age. Data were collected using the Multidimensional Personality Questionnaire [MPQ; Tellegen & Waller, 2008] through four waves of assessment. Data were analyzed using multilevel modeling to explore for nonlinear patterns of change, quantify change parameters at both the group and individual levels, and test for differences between genders.

For 9 of the 11 MPQ scales, models including quadratic and cubic terms provided significantly better fit than those including only linear terms. This finding highlights the importance of using multiple assessments [necessary for detecting nonlinear changes] and of considering a long period of developmental time.

Several developmental processes could potentially account for the patterns of deviation from maturation identified in adolescence. For example, early-adolescent personality development may be influenced by fluctuations in identity development processes of commitment versus explorations of different decisions and roles [Klimstra et al., 2010], or the experience of normative and nonnormative life events [Lüdtke, Roberts, Trautwein, & Nagy, 2011].

Questions to be addressed

Several questions surrounding the development of implicit personality seem ripe to be addressed. For example, Does the development of implicit personality show parallels to the developmental course of self-reports regarding mean-level and rank-order stability and change? How are implicit and explicit measures of traits correlated with each other as well as other data sources measuring the same construct across the lifespan [age-dependent convergent construct validity]? Do they influence each other in cross-lagged ways, such that when one changes the other also changes [temporo-sequential and causal contiguities]? How are they [differentially] related to different life outcomes across the lifespan [age-dependent criterion validity and nomological networks]? Which biological mechanisms and environmental conditions as well as their interactions operate for stability and change in implicit traits?

In addition to these basic research questions revolving around “normative” personality change that takes place automatically and naturally across life, more applied questions targeting the voluntary and purposefully induced change of traits may be addressed also. For example, do interventions targeted at personality [Magidson et al. 2014; Hudson & Fraley, 2015] and people’s explicit goals to change their traits [Hudson & Roberts, 2014] result in changes of implicit traits also? If yes, why is this the case, how sustainable and long lasting is implicit trait change, and which consequences and trajectories does it entail? According to Magidson et al. [2014, p. 1443], automatic aspects of traits may be the key to lasting personality change as the challenge for any intervention to changing personality traits is not only to overcome the nonconscious nature of personality traits but also to inculcate a level of change that is so complete it is automatic and instantiated over time in an enduring way.

All of the preceding questions can be addressed in a general sense, targeting population estimates for groups of people, as well as in a differential sense, targeting individual deviations from normative group estimates. For example, are there interindividual differences in intraindividual change of implicit traits? If so, which underlying biological, psychological, and environmental mechanisms and processes drive such differential change within persons? And which short-, middle-, and long-term consequences do such differences have?

Set like plaster

An earlier version of the FFT suggested that almost all of the intrinsic personality development happened before age 30, after which personality would become “set like plaster” [Costa & McCrae, 1994], whereas this claim seems to be somewhat softened in the later descriptions of the theory [Terracciano, Costa, & McCrae, 2006]. Indeed, although mean-level personality change continues into very old age [Specht et al., 2014], there is remarkable evidence for very high rank-order stability even in the latest decades of life, regardless of often notable changes and individual differences in people’s health, cognitive functioning, and abilities to independently cope with life [Mõttus, Johnson, & Deary, 2012; Mõttus et al., in press; but see evidence for decreasing rank-order stability in older age in Ardelt, 2000; Specht et al., 2011]. Furthermore, mean-level changes may represent intrinsic maturation throughout life: for example, declining Conscientiousness in later life may reflect some sort of general functional decline that is characteristic of most people at that age. Likewise, individual differences in the rates of change may reflect intrinsic differences throughout life. On the other hand, high rank-order stability may also result from increasingly stable environmental influences [Briley & Tucker-Drob, 2014]. Thus although the idea of personality be set like plaster after age 30 does not seem to be entirely correct—people, on average, do keep changing and individual differences are never perfectly stable—this does not seem to have major implications for the principal suggestion of the FFT—the development of the basic tendencies reflects intrinsic maturation rather than external influences.

Abstract

In this chapter, I discuss the main methods that are currently used to describe, predict, and explain long-term development, with a particular focus on between-person differences in individual trajectories of change. First, I distinguish between three concepts of change [individual, mean, and differential change] and associated concepts of stability [mean-level and positional/rank-order stability], and highlight the equivalence of between-person differences in change and change in between-person differences in the case of linear change. Subsequently, I contrast three variable-centered models for describing differential change with each other [multilevel, latent growth curve, and autoregressive models] and outline two person-centered approaches to the description of change [development of personality types and types of personality development]. Finally, I discuss four approaches to explaining personality development [statistically controlled prediction, intervention studies, natural experiments, and crosslagged analysis, including longitudinal mediation] with an eye on differences between within-person and between-person mechanisms of change.

The stability of individual differences in self-esteem

Although the average level of self-esteem changes in systematic ways across the life course, research suggests—as will be reviewed in this section—that individual differences in self-esteem are relatively stable across long periods. As early as at the end of the 19th century, William James pointed to the stability of self-esteem by observing that “there is a certain average tone of self-feeling which each one of us carries about with him” [James, 1890, p. 306]. Later, empirical studies have tested for the rank-order stability of self-esteem. Typically, estimates of stability were based on test–retest correlations between two assessments of the same sample, where the second assessment is conducted some time—e.g., 1 or 2 years—after the first assessment [note that a test–retest correlation of 1 indicates perfect stability and a correlation of 0 indicates complete absence of stability]. Generally, these studies suggested that the stability of self-esteem is high [Alsaker & Olweus, 1992; Block & Robins, 1993; Marsh, Craven, & Debus, 1998; O’Malley & Bachman, 1983]. Trzesniewski, Donnellan, and Robins [2003] meta-analyzed the findings of 50 published studies and, in addition, examined data from 4 large nationally representative samples. Their findings suggested that stability is low during childhood, increases in adolescence, is highest in young and middle adulthood, and decreases during old age. For intervals of 3 years [i.e., the average observed time interval across studies], the rank-order stability of self-esteem was estimated as 0.64 when corrected for the effect of measurement error. Moreover, Trzesniewski et al. [2003] found that the pattern of findings replicated across gender, ethnicity, measure of self-esteem, and year of publication.

However, although estimates of rank-order stability provide some information about the stability of constructs such as self-esteem, a complete understanding requires information about the pattern of stability estimates across intervals of different length [Fraley & Roberts, 2005]. Clearly, rank-order stability decreases, as the interval between assessments increases. But as Fraley and Roberts [2005] have demonstrated, the crucial question is whether the stability of a construct asymptotically approaches zero or a nonzero, positive value when intervals become very long. A nonzero asymptote has important theoretical implications, because it indicates that constant factors—such as genetic influences, formative experiences in early childhood, or stable environmental conditions—contribute to the maintenance of individual differences in the construct. In contrast, a zero asymptote suggests that only transient factors shape the individuals’ standing on the construct.

Therefore, in a study with a large sample that was assessed multiple times across 29 years, Kuster and Orth [2013] examined the time-dependent decline of stability in self-esteem and tested alternative functions that might explain the pattern of stability across time. The results showed that the decline in stability followed an exponential decay function with a nonzero asymptote at about 0.40. Thus, as the time interval increased, stability first quickly declined but in the long run leveled off at a medium-sized value. Moreover, the pattern of results held across gender and across age groups from adolescence to old age. The findings suggest that individual differences in self-esteem are relatively stable across very long periods and that constant factors that account for the long-term stability of self-esteem must be present. The time-dependent pattern of stability in self-esteem was similar to findings on the Big Five personality traits [Fraley & Roberts, 2005], although the asymptotic value might be somewhat smaller for self-esteem than for the Big Five [Anusic & Schimmack, 2016; Kandler, Zimmermann, & McAdams, 2014]. Nevertheless, the results overall suggest that self-esteem exhibits trait-like stability. Thus individuals who have relatively high [or low] self-esteem at one developmental stage are likely to have high [or low] self-esteem 10, 20, or even 30 years later.

Another approach to gain information about the stability of a construct is to test latent trait-state models, using structural equation modeling [Cole, 2012; Kenny & Zautra, 1995, 2001]. These models allow disentangling stable and unstable variance components [i.e., trait and state factors] of a construct over time. Three recent longitudinal studies have used this approach to examine the stability of individual differences in self-esteem across long periods [Donnellan, Kenny, Trzesniewski, Lucas, & Conger, 2012; Kuster & Orth, 2013; Wagner, Lüdtke, & Trautwein, 2016]. The findings of these studies showed that a stable trait factor is needed to explain the patterns of change and stability in the data. Across the three studies, about 70% to 85% of the variance in self-esteem was accounted for by trait factors, whereas only 15% to 30% was state variance or measurement error. A short-term longitudinal study that used data from four assessments across 18 months yielded similar estimates of trait and state components of self-esteem, which strengthens the generalizability of the conclusions from the long-term studies [Orth & Luciano, 2016].

Taken together, the studies reviewed in this section suggest that self-esteem shows trait-like stability, even across very long periods. Put differently, the findings suggest that self-esteem is a relatively enduring personality characteristic rather than a state-like construct such as mood.

1 Cognitive control ability—An early warning system?

A substantial body of work has focused on establishing the relative contribution of genetic and environmental factors on cognitive control ability across different stages of the lifespan. Inter-individual variability on latent measures of both general and specific cognitive control abilities is highly heritable, especially in young adults [Friedman, Nessler, Cycowicz, & Horton, 2009]. To a lesser degree, both shared and non-shared environmental factors also influence individual differences in cognitive control ability [Boogert, Madden, Morand-Ferron, & Thornton, 2018]. Indeed, across the lifespan, any factor that impacts quality of life [e.g., stress, loss of income, interpersonal relationships, poor physical or mental health, mental overload] can have a transient or protracted impact on prefrontal cortical function and associated cognitive control abilities [Diamond, 2013; Friedman et al., 2008].

Importantly, the contribution of both genetic and environmental factors on cognition shows high longitudinal rank-order stability—the pattern of individual differences remains highly stable over time [Friedman et al., 2016; Gustavson et al., 2018; Tucker-Drob & Briley, 2014]. The level of rank-order stability is lower in childhood and strengthens across the lifespan. Moreover, the relative contribution of genetic and environmental influences on cognition rank-order stability varies across childhood and different stages of adulthood. For example, shared environmental factors are the primary determinants of cognitive phenotypic stability in early childhood, whereas genetic factors are the primary mediators of stability throughout adulthood. The role of genetic factors reduces again by mid-late life, when non-shared environmental factors acquire a substantial influence on cognitive stability. Latent measures targeting both specific and general cognitive control processes show a similar pattern; they remain stable over time [at least over relatively brief 4–6 year intervals] in both young and older adults, and this stability is influenced largely by genetic, but also environmental, factors [Gustavson et al., 2018; Wray-Lake, Syvertsen, & Flanagan, 2016].

Hence, cognitive control abilities are critical for purposeful, goal-directed behavior, are sensitive to both genetic and environmental influences across the developmental lifespan, and, on average, show high inter-individual variability and intra-individual stability. Moreover, high cognitive control ability [relative to an appropriate normative group] as well as sustained or improved cognitive control ability [relative to one's past level] are associated with better age-appropriate real-world outcomes. This pattern is consistent with Diamond's conceptualization of cognitive control ability as the “canary in the coalmine”—an early warning signal of physical or mental ill health that indicates the need for primordial or primary intervention to prevent or reduce the risk of negative future outcomes [Diamond, 2013].

However, assessing and monitoring cognitive control ability over the lifespan is a mammoth task, especially with current approaches. Neuropsychological assessments are time-, labor- and cost-intensive; experimental tasks that assess cognitive control mostly target one specific process; and self-report questionnaires confound cognitive and affective state [Merema, Speelman, Foster, & Kaczmarek, 2013]. Instead, we need to learn from other screening tools, like the Framingham Risk Score for estimating stroke risk [Lloyd-Jones et al., 2004], the Kessler-10 for identifying risk of psychological distress [Kessler et al., 2002], and the Mini-Mental State Examination for dementia screening [Folstein, Folstein, & McHugh, 1975]. We need to develop a brief, accessible, and reliable approach to assess core cognitive control processes so as to characterize subtle but meaningful deviation from population-based norms and monitor intra-individual variability over time in healthy and clinical populations. Assessment will need to be scalable and repeatable at a population level, so as to allow early detection of poor level and/or a relative decline in cognitive control ability that may be indicative of underlying neuropathology and signal the need for further evaluation and potential intervention. It will also need to be adaptive to avoid ceiling effects and flexible to accommodate varying profiles of cognitive control across the lifespan. Such a tool would contribute to the epidemiologist's and general practitioners' toolkit, being used to screen groups and individuals to identify cases that show cause for concern. As done with other screening tools, these cases may then be assessed more regularly and potentially be referred on for more comprehensive investigation to characterize specific areas of concern, identify underlying causes, and suggest alternatives for intervention at a societal or personal level.