Running head: PERSONALITY STABILITY OVER 50 YEARS 1
Sixteen Going on Sixty-Six:
A Longitudinal Study of Personality Stability and Change across 50 Years
Rodica Ioana Damian1
Marion Spengler2
Andreea Sutu1
Brent W. Roberts3
1
University of Houston; 2
University of Tübingen; 3
University of Illinois at Urbana-Champaign
Accepted Version of Article in Press at Journal of Personality and Social Psychology
© 2018, American Psychological Association. This paper is not the copy of record
and may not exactly replicate the final, authoritative version of the article. Please
do not copy or cite without authors' permission. The final article will be available,
upon publication, via its DOI: 10.1037/pspp0000210
Acknowledgements: This research uses data from Project Talent, a project directed by the
American Institutes of Research (AIR). Information on how to obtain the Project Talent data
files is available on the AIR website (http://www.air.org/). All data used in the validation study
are publicly available at the following address: https://osf.io/vxba7/. Marion Spengler’s
contribution to this paper was supported by a grant to Marion Spengler funded by the European
Social Fund and the Ministry of Science, Research and the Arts of Baden-Württemberg.
Correspondence concerning this article should be addressed to Rodica Ioana Damian,
Department of Psychology, University of Houston, 3695 Cullen Boulevard, Houston, TX 77204;
email: ridamian@uh.edu.
PERSONALITY STABILITY OVER 50 YEARS 2
Abstract
How much do people’s personalities change or remain stable from high-school to
retirement? To address these questions, we used a large US sample (N = 1,795) that assessed
people’s personality traits in adolescence and 50 years later. We also used two independent
samples, one cross-sectional and one short-term longitudinal (N = 3,934 and N = 38,
respectively), to validate the personality scales and estimate measurement error. This was the
first study to test personality stability/change over a 50-year time span in which the same data
source was tapped (i.e., self-report). This allowed us to use four different methods (rank-order
stability, mean-level change, individual-level change, and profile stability) answering different
developmental questions. We also systematically tested gender differences. We found that the
average rank-order stability was .31 (corrected for measurement error) and .23 (uncorrected).
The average mean-level change was half of a standard deviation across personality traits, and the
pattern of change showed maturation. Individual-level change also supported maturation, with
20-60% of the people showing reliable change within each trait. We tested three aspects of
personality profile stability, and found that overall personality profile stability was .37,
distinctive profile stability was .17, and profile normativeness was .51 at baseline and .62 at the
follow-up. Gender played little role in personality development across the lifespan. Our findings
suggest that personality has a stable component across the lifespan, both at the trait level and at
the profile level, and that personality is also malleable and people mature as they age.
Keywords: personality traits; rank-order stability; mean-level change; profile stability;
lifespan
PERSONALITY STABILITY OVER 50 YEARS 3
Sixteen Going on Sixty-Six:
A Longitudinal Study of Personality Stability and Change across 50 Years
In Homer’s epic poem, Odysseus, the legendary Greek king, returns home to Ithaca after
20 years of warfare and difficult journeys, only to find his wife, Penelope, faithfully waiting for
him despite her numerous suitors. The lovers are happily reunited and Odysseus reclaims his
kingdom. Perhaps even more impressive than legend is the love story of Jerzy Bielecki and Cyla
Cybulska. The two fell in love in 1943 in the Nazi concentration camp Auschwitz. After
managing to break out, they got separated, and through a series of misunderstandings they each
came to presume the other dead. Cyla moved to Brooklyn and married, while Jerzy started a
family in Poland. In 1983, Cyla told this story to her Polish house cleaner who told her that she
had seen a man tell the same story on Polish television. The two were reunited a few weeks later.
When Cyla arrived in Krakow, Jerzy gave her 39 roses, one for each year they had been apart.
They became very good friends and visited each other regularly until 2005, when Cyla died. In
2010, when Jerzy was last interviewed before passing away, he said he was “still very much in
love with Cyla” (Hevesi, 2011).
When hearing such stories, one must wonder what such reunions feel like. When
considering the personality traits (i.e., the characteristic patterns of thoughts, feelings, and
behaviors) that one exhibited in adolescence, how similar to their old self is that person likely to
be 50 years later? Are sociable teens destined to become sociable older adults? And does our
relative personality ranking with respect to other people endure over our entire lifespan? For
example, if Cyla was more sociable than Jerzy when they were 16, how likely is it that she was
still more sociable than Jerzy 50 years later? What about absolute levels of personality: Do
people change across their entire lifespan? Were both Cyla and Jerzy perhaps a bit wiser, less
PERSONALITY STABILITY OVER 50 YEARS 4
impulsive, when they met as older adults, than when they were teenagers? And do some people
change more than others across the lifespan? What about the unique constellations of traits that
people have? For example, if Cyla were more neurotic than she was sociable when she was an
adolescent, how likely is it that the same idiosyncratic pattern of personality traits would
characterize her 50 years later? Finally, are there gender differences in how people’s
personalities change as they age? Questions regarding the stability and change of personality
across the entire lifespan are some of the most interesting, because there are very few
longitudinal studies spanning over so many years. The present study seeks to address such
questions by using a large US sample that was followed over 50 years.
A major insight from recent research on personality development (e.g., Costa & McCrae,
1988, 1994; Donnellan, Conger, & Burzette, 2007; Fraley & Roberts, 2005; Hampson &
Goldberg, 2006; Roberts & Del Vecchio, 2000; Roberts, Walton, & Viechtbauer, 2006;
Terracciano, Costa & McCrae, 2006) is that personality traits are both stable and changeable.
Personality traits are defined as relatively enduring, automatic patterns of thoughts, feelings, and
behaviors that are relatively consistent across a wide variety of situations and contexts (Roberts,
2009). The most commonly used personality trait framework is the Big Five (John, Naumann, &
Soto, 2008) or Five-Factor Model (McCrae & Costa, 2008), which includes five broad traits:
openness to experience, conscientiousness, extraversion, agreeableness, and neuroticism. The
insight that personality traits are both stable and changeable reflects, in part, the systematic use
of different methodological approaches to assess stability and change in individual traits (Block
1971). The two most prominent and most widely used approaches are rank-order stability and
mean-level change. Rank-order stability (or differential stability) refers to the relative placement
of a person (on a specific trait) within a group over time. A typical research question would be:
PERSONALITY STABILITY OVER 50 YEARS 5
Does a 16-year-old who is conscientious relative to her peers develop into a 66-year-old who is
also more conscientious compared to her peers? As described below, assessing rank-order
stability across the lifespan, and especially over very long periods of time, is essential for
understanding personality development and whether personality traits may be caused, in part, by
continuous factors, such as (a) certain components of the genetic system (Roberts, in press) and
(b) individuals seeking consistent roles and/or environments across time (Roberts & Damian, in
press; Roberts & Robins, 2004; Harms, Roberts, & Winter, 2006). Mean-level change refers to
how the average level of a trait across all individuals changes over time (see Caspi, Roberts, &
Shiner, 2005). Mean-level change can be studied in cross-sectional studies with different cohorts
(e.g., 16 and 66-year-olds assessed at the same time) or longitudinally, with the same cohort
(e.g., assess people when they are 16 and then again 50 years later). Typical research questions
would be: Are 16-year-olds less responsible than 66-year-olds? Or are people less responsible
when they are 16 than 50 years later?
Although rank-order stability and mean-level change are essential in understanding
personality development at the population level, they limit the understanding of development at
the individual level (e.g., Roberts, Caspi, & Moffitt, 2001; Robins et al., 2001). Thus, it is
important to investigate individual differences in change, that is, the magnitude of increase or
decrease within each person over the course of the study. Typical research questions would be:
Across the entire sample, what percentage of people showed reliable increases or decreases in
responsibility going from 16- to 66-years-old?
Rank-order stability, mean-level change, and individual-level change are all “variablecentered”
approaches (Block, 1971), meaning that they focus on stability and change in single
personality traits. As a result, these approaches cannot account for the fact that personality is a
PERSONALITY STABILITY OVER 50 YEARS 6
‘‘peculiar patterning of attributes within the single person’’ (Allport, 1954, p. 9), as opposed to a
set of disconnected traits. Thus, a fourth way to conceptualize change and stability in personality
is to take a “person-centered approach” and focus on the stability of the pattern of personality
traits within a person across time, that is, personality profile stability (or ipsative stability).
Assessing personality profile stability requires measurements of multiple personality traits that
are ranked with respect to each other and that are collected across a minimum of two time-points
(see Bleidorn et al., 2012; Furr, 2008; Klimstra et al., 2010). A typical research question would
be: Does a 16-year-old, whose neuroticism is higher than her conscientiousness, develop into a
66-year-old whose neuroticism is still higher than her conscientiousness?
Rank-order Stability in Personality over Extended Periods of Time
Longitudinal studies that investigated the rank-order stability of personality traits across
extremely long time-spans are rare and often plagued by methodological drawbacks (e.g., having
obtained informant-reports at baseline and self-reports at the follow-up). Thus, it is unclear what
level of rank-order stability we should expect across long periods of time (e.g., 50 years).
For example, several previous studies (Shiner et al., 2017; Edmonds, Goldberg,
Hampson, & Barckley, 2013; Hampson & Goldberg, 2006) that assessed rank-order stability
across long time spans, that is, across 20 years (N = 205; ages 10-30) and 40 years (N = 799,
starting the assessment in elementary school with teacher reports and following-up with selfreports),
found that the average rank-order stability of personality traits was about .20. Note that,
although still substantial, this estimate is much lower than the rank-order stability evident over
shorter time-spans (see Roberts & DelVecchio, 2000). In other words, the stability of traits over
time should decrease as the time-span increases, but it should not asymptote to zero (Fraley &
Roberts, 2005).
PERSONALITY STABILITY OVER 50 YEARS 7
Contradicting previous research, a recent study suggested there may be little to no rankorder
stability in personality traits when very long time-intervals are considered. Specifically,
Harris and colleagues (2016) used a sample of 174 people to test the rank-order stability of
personality over 63 years. At baseline (age 14), teacher reports were collected using only one
item for each of six personality characteristics (Self-Confidence, Perseverance, Stability of
Moods, Conscientiousness, Originality, and Desire to Excel). At the follow-up (age 77), self- and
informant-reports (by close others) were collected on the same six personality items. The authors
found no statistically significant rank-order stability. The test-retest coefficients found across 63
years when correlating teacher ratings at age 14 with self-ratings at age 77 ranged from -.05
(Perseverance) to .12 (Stability of Moods). When correlating teacher ratings at age 14 with
ratings by close others at age 77, the test-retest coefficients ranged from -.14 (Desire to Excel) to
.12 (Conscientiousness). The authors’ conclusion that personality traits show little to no stability
over extended periods of time is at odds with the previous research cited above, with two metaanalyses
(Roberts & DelVecchio, 2000; Anusic & Schimack, 2016), and with genetic models
which would predict some personality stability, even over long time-spans (e.g., Bleidorn,
Kandler, Riemann, Angleitner, & Spinath, 2009). One possible explanation, which was put
forward by the authors, is that the unusually long time-span (63 years) is responsible for the lack
of observed stability. However, there are two alternative explanations, due to methodological
issues: (a) it is possible the six items did not constitute a comprehensive measure of personality
and (b) it is possible that the source of the lack of observed stability was, not the long timeinterval,
but the use of measurements from different informants at the two time-points (teacher
vs. self; teacher vs. close other). Thus, the status of personality stability over extended time
periods (i.e., over 50 years) remains unclear.
PERSONALITY STABILITY OVER 50 YEARS 8
What level of stability in personality should we expect over a time span of five decades?
Given the lack of long-term longitudinal data, it would be prudent to rely on estimates drawn
from the corpus of prior longitudinal research, as an aggregate estimate would be an ideal
benchmark upon which to base expectations. Accordingly, Fraley and Roberts (2005), used an
aggregate of stability coefficients assessed at varying time intervals (including very long
intervals) and with different starting ages to estimate the expected levels of stability over time
and age. Using a variety of models to better incorporate developmental processes, such as
random life events, person-environment transactions, and developmental constancies (e.g.,
genetic factors), Fraley and Roberts (2005) estimated that personality stability coefficients over
long time spans (including over 50 years), when the first measurement was in adolescence,
should asymptote at a value of about .20 on a correlational metric (not corrected for error),
though estimates varied across different personality traits, with asymptotes ranging from .18 for
Openness to .36 for Conscientiousness. However, their data did not include studies that covered
a stability time span longer than 30 years. Thus, the current study provides the first opportunity
to empirically test the stability of personality traits over a 50-year time span in which the same
data source was tapped (i.e., self-report).
Mean-level Changes in Personality over Extended Periods of Time
Understanding mean-level change across the lifespan is important because it can further
inform our knowledge of developmental processes. How do people change as they age? Like
with studies on stability, longitudinal studies assessing mean-level change over extended periods
of time are lacking. Such studies are important because without them we cannot discern whether
the changes seen over shorter time-spans dissipate or cumulate with time.
PERSONALITY STABILITY OVER 50 YEARS 9
Despite the lack of longitudinal studies investigating mean-level change over very long
time-spans, research has made extensive progress in understanding principles of personality
change. For example, a principle of personality change that has received extensive attention is
the maturity principle, which can explain mean-level changes in personality over time.
According to this principle, people become more psychologically mature with age, if maturity is
defined as becoming more socially adapted, and, specifically, if being socially adapted is
reflected in changes that increase a person’s ability to negotiate social relationships and
challenges more effectively (Roberts & Damian, in press). A meta-analysis of the mean-level
changes in personality traits over time solidified empirical support for the maturity principle
(Roberts, Walton, & Viechtbauer, 2006). This study found that most people become more
agreeable, conscientious, and emotionally stable over their lifespan. Interestingly, the metaanalysis
also showed robust increases in a facet of extraversion, described as social dominance,
which reflects higher levels of assertiveness, self-confidence, and dominance.
The interesting question concerning long-term personality trait change is whether the
changes seen over shorter time spans cumulate or dissipate. To the extent that personality traits
are governed by “set points” that anchor the range of potential change that a person can realize in
their life, then the longer the time span, the more likely a person will return to their set point
(Fraley, 2002). On the other hand, if changes in personality cumulate over the life course, the
longer the time period examined, the greater the amount of change that should occur. Using
meta-analytic data, Roberts and colleagues (2006) found that changes in personality traits within
different decades of life (e.g., from 20 to 30 or from 30 to 40), were each about a quarter to a
third of standard deviation in the direction of maturation and that change was consistently
positive across different age cohorts (note that, in this meta-analysis, longitudinal data spanned
PERSONALITY STABILITY OVER 50 YEARS 10
across an average of 7 years, but used different age cohorts). Given these meta-analytic data,
they estimated that changes in specific personality traits, like conscientiousness and emotional
stability, should be around a full standard deviation between ages 20 and 70 if changes seen over
shorter time spans (10 years) cumulated. While consistent with cross-sectional estimates of age
differences in personality traits (Costa & McCrae, 1988), these estimates were extrapolations
from the data set rather than reflections of actual tests of how personality traits should behave
across the life course (because most longitudinal studies included in the meta-analysis did not
span for longer than 10 years). If the magnitude of personality trait change were lower than the
estimates from the meta-analysis when examined for the first time across 50 years (i.e., lower
than one standard deviation for traits like conscientiousness and emotional stability), this would
be consistent with a set point model that would argue for a braking system on change. On the
other hand, if personality trait change continued to accumulate, we would expect estimates closer
to half to one standard deviation (i.e., estimates that are higher than the quarter to a third of a
standard deviation change estimated over shorter time spans of 10 years). The latter potential
finding would be more consistent with a plasticity model of personality traits, and it would
contradict a strong set point model, indicating that once positive gains are made, they are likely
to continue in a form of a virtuous cycle.
Individual-Level Change in Personality
Although mean-level change can help us understand personality development at the
population level, it overlooks potential individual differences in change. The existence of
individual differences is pertinent to personality development in two ways. First, if they did not
exist, this would bolster the argument that normative changes in personality traits are universal
and uniform. If people demonstrated normative increases in traits without individual differences
PERSONALITY STABILITY OVER 50 YEARS 11
in change, then one could argue that a universal genetic factor might be the cause of personality
development (McCrae & Costa, 2008). On the other hand, if individual differences in personality
trait change did exist, this would bolster the argument that personality trait change is contingent
on each person’s particular experiences. A recent review of prospective research (Bleidorn,
Hopwood, & Lucas, 2018) showed that life experiences are associated with change in personality
traits, and that different life experiences are differentially related to personality trait domains.
Specifically, the most robust findings across the review were that transitioning to the first
romantic relationship increased Extraversion and decreased Neuroticism (e.g., Neyer & Lehnart,
2007; Wagner, Becker, Lüdtke, & Trautwein, 2015), and that transitioning from high school to
college/work increased Agreeableness, Conscientiousness, Openness, and decreased Neuroticism
(e.g., Bleidorn, 2012; Lüdtke, Roberts, Trautwein, & Nagy, 2011). Furthermore, studies have
found that life experiences are associated with personality change in middle- (van Aken,
Denissen, Branje, Dubas, & Goossens, 2006) and old-adulthood (Mottus, Johnson, & Deary,
2012). Nevertheless, these effects of life experiences may be relatively modest and the evidence
is still preliminary (Costa et al., 2000; Bleidorn et al., 2018). Thus, the existence of individual
differences in personality trait change is key for understanding which model of personality
development we should hold as an assumption—one that does not propose the influence of
environmental experiences (McCrae & Costa, 2008) or one that does (Roberts, Wood, & Caspi,
2008).
One way to measure individual-level change in each personality trait, when only two
waves of data are available, is to calculate the Reliable Change Index (RCI; Christensen &
Mendoza, 1986; Jacobson & Truax, 1991), and classify people into three groups: decreased,
increased, or stayed the same on each trait level. The Reliable Change Index is a widely used and
PERSONALITY STABILITY OVER 50 YEARS 12
very conservative index of change that was primarily developed to assess whether the changes
resulted from therapeutic interventions were larger than chance (Jacobson & Truax, 1991).
We are not aware of any previous studies that used the RCI to assess individual-level
change in personality traits across very long periods of time (e.g., 50 years), but two previous
longitudinal studies of personality traits used the RCI to assess individual-level change across
four years (Robins et al., 2001) and across eight years (Roberts et al., 2001), in samples that had
starting ages similar to ours. Across both of these studies, the observed distributions of changers
and non-changers was significantly different from chance across all personality traits examined.
However, in both studies, the percentage of people found in the group who “stayed the same”
was quite large. Specifically, across four years, 73% (Neuroticism) to 91% (Openness) of the
people in the sample stayed the same (Robins et al., 2001). Across eight years, 72.2% (Negative
emotionality) to 84.4% (Constraint) of the people in the sample stayed the same (Roberts et al.,
2001). Notably, the percentages of people increasing or decreasing on various traits were
consistent with previously found mean-level change patterns, that is, they showed a maturation
trajectory, where a higher percentage of people increased (vs. decreased) in conscientiousness,
emotional stability, agreeableness, and dominance aspects of extraversion (Roberts et al., 2001).
Personality Profile Stability
Personality profile stability, also known as ipsative stability (Caspi & Herbener, 1990) or
within-person coherence (Biesanz & West, 2000), represents the degree to which a person’s
unique pattern of traits remains stable across time (Ozer & Gjerde, 1989) and it is usually
assessed with q-correlations, that is correlations, within-individuals, between ranked sets of traits
across time.1
PERSONALITY STABILITY OVER 50 YEARS 13
Previous studies that assessed personality with self-reports found average profile stability
coefficients ranging from .37 (over a 15-year time span) to .85 (over a 3-year time span), with
the distributions of these profile stability coefficients often ranging from -.95 to 1.00 (Block,
1971; Caspi & Herbener, 1990; De Fruyt et al., 2006; Donnellan et al., 2007; Klimstra et al.,
2009; Lönnqvist et al., 2008; Ozer & Gjerde, 1989; Roberts et al., 2001; Robins et al., 2001).
Notably, based on our review, the study with the longest time-span covered (i.e., 15 years, ages
20-35, N = 74) revealed an average personality profile stability of .37 (SD = .32), with
individuals ranging from -.54 to .90 (Lönnqvist et al., 2008).
Extensive previous research has measured personality profile stability as a unitary
construct. However, Furr (2008) argued that personality profile stability is not a unitary construct
because of the so-called “normativeness problem.” Specifically, Furr (2008) proposed that (a)
most people’s personality profiles will be similar to the normative profile (i.e., the average
profile of the sample), as the degree of normativeness might reflect psychological adjustment and
adaptation; and (b) normative profiles are likely stable across time, to the extent to which similar
norms are relevant across different developmental periods. Thus, overall profile stability (i.e.,
“classic” profile stability, which was previously measured as a unitary construct) might reflect
two separate processes: (a) the tendency to retain idiosyncratic (non-normative) personality
profiles or (b) the tendency to be consistently normative across time.
To address the normativeness issue, Furr (2008) suggested decomposing overall profile
stability into two additional components: distinctive profile stability (i.e., the degree to which a
person’s personality profile consistently diverges from the normative profile within the sample)
and within-time normativeness (a.k.a., profile normativeness, or the degree to which a person’s
personality profile is similar to the average personality profile in a sample at the respective
PERSONALITY STABILITY OVER 50 YEARS 14
measurement point). Notably, overall profile stability and distinctive profile stability require two
measurement time points, whereas within-time profile stability can be computed at each
measurement point.
Following Furr’s (2008) recommendations, Klimstra and colleagues (2010) found, in a
longitudinal study of 565 college students, that the average profile stability of self-reported
personality traits was .74 (SD = .35) and the average distinctive profile stability was .61 (SD =
.43), across four consecutive 1-year intervals. The average profile normativeness across the four
time points was .58 (SD = .48), and seemed to increase with age (going from .54 at time 1 to .68
at time 4). In another study, Bleidorn and colleagues (2012) found, in a sample of 805 twin pairs
(ages 34-46) who self-reported on their personalities, that, across two 5-year intervals, the
average overall profile stability was .86 (SD = .23), whereas the average distinctive profile
stability was .76 (SD = .29). Average profile normativeness ranged from .53 at Time 1 to .71 at
Time 3, again showing an increasing trend with age. Across all stability coefficients, the range of
estimates across participants was very wide, as previously reported going from -.96 to 1.00.
Across past research, we observed several trends. First, overall personality profile
stability tended to be higher when the time span between assessments was shorter and to increase
from childhood to late adolescence (see Klimstra et al., 2009; Ozer & Gjerde, 1989). Second,
distinctive personality profile stability tended to be lower than overall profile stability (which
should happen by definition, given Furr’s theorizing on the normativeness problem; Furr, 2008).
Third, profile normativeness was about .50 to .70 within time points, and it increased with age,
where older people had more normative profiles (e.g., Bleidorn et al., 2012). Fourth, very few
studies investigated profile stability over a timespan longer than eight years (e.g., Bleidorn et al.,
2012; Caspi & Herbener, 1990; Roberts et al., 2001), and the longest time span covered was 15
PERSONALITY STABILITY OVER 50 YEARS 15
years (Lönnqvist et al., 2008). And fifth, there were only a handful of studies that assessed
personality profile stability following Furr’s (2008) recommendations, and thus, addressing the
normativeness issue.
Gender Differences in Personality Stability and Change
Previous studies have shown that men and women differ in their personality trait levels at
any given point in time, such that, women are higher in conscientiousness, agreeableness, the
sociability facet of extraversion, and lower in emotional stability and agency (e.g., Schmitt et al.,
2008; Costa et al., 2001; Roberts et al., 2001). However, cross-sectional gender differences may
not necessarily translate into gender differences in stability and change across the lifespan.
We are not aware of any previous studies on gender differences in personality stability
and change across very long time spans (e.g., 50 years), however, previous studies over shorter
time spans (e.g., 8-10 years) provide some clues. For example, a meta-analysis of 152
longitudinal studies that examined rank-order stability across the lifespan found no statistically
significant gender differences, that is, personality traits were similarly trait-like for both men and
women (Roberts & DelVecchio, 2000). Furthermore, a meta-analysis of 92 longitudinal studies
that examined mean-level change across the lifespan also found no statistically significant gender
differences, that is, men and women changed at similar rates across the lifespan (Roberts,
Walton, & Viechtbauer, 2006). Regarding gender differences in individual-level change and in
profile stability, one study, across 8 years (Roberts et al., 2001), found that men showed slightly
more reliable change, and that women showed slightly more personality profile stability,
however, it is worth noting that only overall profile stability was computed in this study (without
addressing the normativeness issue).
PERSONALITY STABILITY OVER 50 YEARS 16
In sum, extensive previous research has shown that (a) personality traits show both rankorder
stability and mean-level change over time, (b) there is reliable change at the individuallevel,
and (c) personality profiles show stability across time. Furthermore, scientists have started
uncovering the developmental antecedents and processes underlying stability and change.
Despite these advances, most previous research, including the meta-analyses described above,
has focused on relatively short time-spans, up to 10 years, in different age cohorts. Thus, very
little research has examined personality stability and change over the entire life-span (i.e., long
periods of time, over 50 years). The reason for this oversight is a lack of longitudinal samples
that included personality traits over the entire life-span. Nevertheless, the question of how
personality develops over the entire lifespan is an important one. The critical research question
becomes “If we are presented with an adolescent, can we reliably predict their personality when
they are in their 60s?” And, do personality traits continue to change in a positive direction, such
that personality trait change is cumulative across the life course? In the present investigation, we
had access to self-reported personality traits recorded in adolescence, as well as self-reported
personality traits recorded 50 years later, which enabled us to assess long-term stability and
change both at the trait level and at the person level.
The Present Investigation
In the “Main Study” of the present investigation, we used a subsample of the Project
Talent data set to test personality stability and change over a 50-year time span. Specifically, we
used a large US sample of high-school students who had their personality assessed in 1960 and
50 years later (N = 1,795), using the Project Talent Personality Inventory (PTPI). Using these
data, we tested (a) to what extent people maintained their relative standing on personality trait
dimensions relative to others over time (i.e., rank-order stability), (b) to what extent people’s
PERSONALITY STABILITY OVER 50 YEARS 17
personality traits changed across time (i.e., mean-level changes), (c) what percentage of people
had reliable change in their personality traits (i.e., individual-level change), (d) to what extent
people’s personality profiles remained stable across time (to address this question, we evaluated
three aspects of profile stability, that is, overall personality profile stability, distinctive profile
stability, and profile normativeness), and (e) gender differences in each of the four types of
continuity and change.
Based on previous research and theory (Fraley & Roberts, 2005; Roberts & DelVecchio,
2000; Roberts et al., 2006; Caspi et al., 2005), we had several predictions. First, due to the long
time-span and the relatively young age of the participants (~16) at baseline, personality should
show lower levels of rank-order stability than it did in samples tested over shorter time spans and
where testing started at older ages (see Roberts & DelVecchio, 2000). Specifically, rank-order
stability over 50 years (with the first assessment in adolescence) should be about .20 on average,
and it should not differ markedly across different traits. Second, assuming a plasticity (vs. a set
point) model of personality change across the lifespan, mean-level changes in personality traits
across 50 years should be similar to the estimated cumulative mean-level changes estimated from
the meta-analysis by Roberts and colleagues (2006), with an average effect of half to one
standard deviation, and changes should be in the direction of maturation, such that people should
become more agreeable, conscientious, and emotionally stable over their lifespan, as well as
higher in dominance-related facets of extraversion. Regarding individual-level changes in
personality traits and personality profile stability, we did not have clear point-estimate
predictions, because, to our knowledge, no previous studies have examined individual-level
changes using the Reliable Change Index or profile stability across a time span longer than eight
and fifteen years, respectively. Nevertheless, regarding individual-level changes in personality
PERSONALITY STABILITY OVER 50 YEARS 18
traits, and consistent with the idea that change across the lifespan is cumulative and depends on
life experiences as opposed to being universal and uniform (Roberts et al., 2006; Roberts, Wood,
& Caspi, 2008), we expected more than 20% of the people to show reliable changes on each
personality trait, because 20% was the average percentage of “changers” found across shorter
time spans within each trait (see Robins et al., 2001; Roberts et al., 2001). We also expected the
percentages of people increasing or decreasing on various traits to correspond to the expected
mean-level change patterns, that is, to show a maturation trajectory, where a higher percentage of
people should increase (as opposed to decrease) in conscientiousness, emotional stability,
agreeableness, and dominance aspects of extraversion (see Roberts et al., 2001). Regarding
personality profile stability, we expected our estimate to be below .37, which was the overall
profile stability observed across a 15-year time span (Lönnqvist et al., 2008). Furthermore, due to
the normativeness issue (see Furr, 2008), we expected distinctive profile stability to show a much
smaller effect than overall profile stability. We also expected profile normativeness within each
time point to range between .50 and .70, and to increase slightly with age, which is what other
studies have shown (e.g., Bleidorn et al., 2012). Finally, regarding gender differences, we
expected to replicate previously found cross-sectional differences, whereby women should be
higher in conscientiousness, agreeableness, the sociability facet of extraversion, and lower in
emotional stability and agency at any given point in time (e.g., Schmitt et al., 2008; Costa et al.,
2001; Roberts et al., 2001), but, given previous longitudinal research on shorter time spans
(Roberts & DelVecchio, 2000; Roberts et al., 2006), which found no statistically significant
gender differences, we did not expect gender differences in the patterns of personality stability
and change across 50 years.
PERSONALITY STABILITY OVER 50 YEARS 19
The inventory used at 50th
year follow-up was an abbreviated form of the original scale,
which necessitated preliminary work to be validated.2
Therefore, we conducted additional
analyses on two independent data sets (N = 3,934 and N = 38), which were previously collected
by Pozzebon and colleagues (2013). Using the first sample (N = 3,934), which was crosssectional,
we obtained descriptive statistics and correlations between long- and short-forms of
the PTPI. We also tested mean-level personality differences between two different age cohorts
(20s vs. 60s) that were included in this sample and that were similar in age to the participants
from the two waves of the main study. This allowed us to test whether short- vs. long-forms of
the PTPI showed the same patterns of cross-sectional mean-level change across different cohorts.
Furthermore, this allowed us to compare cross-sectional mean-level change across different
cohorts (computed from the validation study) with longitudinal mean-level change within the
same cohort (computed from the main study), using the same personality measures. Using this
validation sample, we also computed mean-level differences between the long- and short-forms
of the PTPI, which allowed us to conduct additional robustness checks of the longitudinal meanlevel
changes observed in the main study, because we could correct for potential mean
differences due to the type of scale used. Finally, using the cross-sectional validation sample, we
tested measurement invariance across relevant age cohorts (20- vs. 60-year-olds, which we could
not do in the main study due to the lack of item-level data). Using the second sample (N = 38),
which was longitudinal, we obtained the short-term test-retest reliabilities of the long- and shortform
PTPI scales over a two-week time span. We used the test-retest estimates obtained from
this independent validation study to help us better estimate the rank-order stability of personality
across the lifespan in the Project Talent sample (i.e., in the main study). Specifically, in addition
to raw correlation coefficients (unadjusted for error) found in the Project Talent sample, we also
PERSONALITY STABILITY OVER 50 YEARS 20
obtained correlation coefficients in a latent framework, which are likely to be more accurate, as
we accounted for measurement error (Little, 2013). To that effect, we used single-indicator latent
constructs with the marker variable convention, and used the observed variance and estimated
test-retest reliability of the long- vs. short-form PTPI scales obtained from the validation study.
Validation Study
This validation study included two samples, one cross-sectional and one short-term
longitudinal. Pozzebon and colleagues (2013) previously used these data to publish results on the
long-forms of the Project Talent Personality Inventory (PTPI). We extend those results to
encompass the short-forms of the PTPI and validate them against the long-forms.
Method
Because the data were publicly available and de-identified, this study was deemed
exempt by the University of Houston, Division of Research Institutional Review Boards (IRB
ID: STUDY00000793).
Participants (cross-sectional sample). As reported by Pozzebon and colleagues (2013),
3,934 participants (65% females; 88% European American) were available for data analysis. The
mean age was 50 (SD = 19.32) and the sampling was largely focused on young (20s) and older
(60s) adults. The goal was to test the validity of the PTPI in samples close to the age of the
Project Talent sample at baseline (when the average age was 16) and 50 years later (when the
average age was 67).
Participants (short-term longitudinal sample). As reported by Pozzebon and
colleagues (2013), a sample of 38 participants (47% females, 87% European American, Mage =
31.87, SDage = 13.58) was available for data analysis. These participants completed two
assessments collected two weeks apart.
PERSONALITY STABILITY OVER 50 YEARS 21
Measures (across both samples). Participants responded to questions about their age,
gender, and ethnicity. To measure personality traits, we used the same measure that was
administered in Project Talent, namely, the Project Talent Personality Inventory (PTPI).
Participants completed all 108 PTPI items, from which the 10 PTPI scales were scored. Item
responses were assessed using the same measurement scale used in the Project Talent surveys
and scale scores were computed using the same procedures described below in the “Main Study”
section. Specifically, item scores were dichotomized and then scale scores were computed taking
the average of the relevant items. Notably, the scoring procedures for the PTPI cannot be
changed in the Project Talent data because item-level data are not available at baseline; thus, to
validate the scales used in the main study, we followed the same scale scoring procedures in this
validation study. We computed scale scores for both long- and short-forms of the PTPI.
All the measures, data analysis scripts, and de-identified data used in this validation
study, and which are necessary to reproduce the results, are publicly available at the following
address: https://osf.io/vxba7/.
Results (cross-sectional sample)
Descriptive statistics and internal consistencies for the long- and short-forms of the PTPI
scales can be found in Tables 1 and 2, respectively. When comparing the internal consistencies
of the long- vs. short- forms, we can observe that the average decrement in reliability was only
.02 for the short-forms. Furthermore, regardless of form type (long vs. short) reliabilities did not
change markedly when examined in younger (20s) and older (60s) subgroups. Moreover, we
observe that both the long- and the short-forms showed a similar pattern of mean age differences
between the 20s and 60s groups, with the 60s age group (compared to the 20s age group) being
higher in calmness, mature personality, self-confidence, social sensitivity and tidiness. These
PERSONALITY STABILITY OVER 50 YEARS 22
findings were consistent with the maturation hypothesis and previously published age differences
in the Big Five personality traits (see Roberts, Walton, & Viechtbauer, 2006).
Table 3 shows the correlations among the 10 PTPI scales in both their long- and shortforms,
as well as across the long- and short-forms. Importantly, all correlations between longand
short-form versions of the same scale were higher than .83.
When comparing the means of long- vs. short- forms (across the entire sample), paired
samples t-tests showed some statistically significant differences (see Table 4), where in some
cases the short-form score was slightly lower (e.g., vigor, impulsiveness, self-confidence,
culture, and tidiness), whereas in other cases the short-form score was slightly higher (e.g.,
mature personality, sociability, and social sensitivity). To address this issue in the Project Talent
data, where the long-form was administered at baseline and the short-form was administered at
the 50th
year follow-up, we conducted a robustness check in the context of the longitudinal
mean-level change analysis, where we adjusted the means from the 50th
year follow-up by
adding or subtracting a constant equal to the mean-difference observed between long- and shortforms
in this validation study.
In longitudinal studies, it is important to ascertain that the changes observed in manifest
indicators (e.g., personality traits in the present case) are due to real changes in the phenomena
being studied and not due to changes in measurement properties across time or across different
age groups, that is, one needs to establish measurement invariance (Schmitt & Kuljanin, 2008;
Widaman et al., 2010). To test for measurement invariance, however, item-level data must be
available at all time points. Unfortunately, in the main longitudinal study presented in this paper
(Project Talent sample), we did not have item-level data available at baseline. To alleviate this
limitation, we used the cross-sectional data from this validation study to test for measurement
PERSONALITY STABILITY OVER 50 YEARS 23
invariance in the PTPI scales across the two relevant age groups: 20- vs. 60-year-olds. We
performed the analyses for each PTPI scale separately, both for long- and short-form versions of
each scale. Across all scales, we found evidence for configural (similar factor structure in twogroup
confirmatory factor analysis), metric (equal factor loadings), scalar (equal factor loadings
and intercepts), and strict (equal factor loadings, intercepts, and residuals) measurement
invariance across the two age groups (the only exception was for the short-form of the selfconfidence
scale, which showed evidence for metric, but not scalar invariance). The results,
along with further details on how to interpret them, can be found in the Supplemental Materials
in Tables 4S and 5S. These findings suggested that people across different age groups (age
groups that were very similar to the age groups we had in the main longitudinal study) used the
PTPI measures in similar ways. This brought some empirical evidence that measurement
invariance might also hold in the longitudinal data form the main study, which allowed us to
interpret the coefficients resulting from the longitudinal analyses described below as we did
(with the caveat that mean-level differences for self-confidence should be treated with caution
since the evidence for scalar invariance was limited in this case).
Together, these findings suggest that the short-form versions represent good measures of
the original constructs, with the caveat that the scale means are slightly different (see Table 4),
but we addressed this issue in the main study by conducting a robustness check using data from
this validation study. Furthermore, given the findings for measurement invariance across the two
critical age groups in the cross-sectional validation study data, we believe that the results
presented in the main longitudinal study (Project Talent sample) can be taken as not being the
result of measurement artifacts.
PERSONALITY STABILITY OVER 50 YEARS 24
Results (short-term longitudinal sample)
Descriptive statistics can be found in Table 2S of the Supplementary Material. The testretest
reliabilities for the long-form PTPI scales were as follows: Vigor (.81), Calmness (.77),
Mature Personality (.79), Impulsiveness (.61), Self-Confidence (.89), Culture (.82), Sociability
(.80), Leadership (.76), Social Sensitivity (.85), Tidiness (.88).
The test-retest reliabilities for the short-form PTPI scales were as follows: Vigor (.80),
Calmness (.73), Mature Personality (.66), Impulsiveness (.68), Self-Confidence (.84), Culture
(.79), Sociability (.78), Leadership (.76), Social Sensitivity (.80), Tidiness (.80).
Thus, the average decrement in test-retest reliability resulting from using short- versus
long-form scales was .04, indicating that the short-form versions represent good measures of the
original constructs. Furthermore, these results provided us with critical information needed to
address measurement error in the rank-order stability analyses from the main study (see Table
2S).3
Conclusion
In sum, this validation study provided us with enough information to be confident that the
short-forms represent good approximations of the long-forms and likely measure the same
constructs. Furthermore, this study provided us with the necessary information to (a) account for
measurement error in a latent framework (using the observed test-retest reliabilities and variance
of the scales from the validation study to inform measurement error in the main study), and (b)
conduct a robustness check of the mean-level change analysis in the main study, by accounting
for mean differences observed in the validation study between long- and short-form versions of
the PTPI.
PERSONALITY STABILITY OVER 50 YEARS 25
Main Study
Method
Because the data were publicly available and de-identified, this study was deemed
exempt by the University of Houston, Division of Research Institutional Review Boards (IRB
ID: STUDY00000793).
Participants. The data came from Project Talent (see Wise et al., 1979), a longitudinal
study that started in 1960 with a 5% representative sample of US high-school students.4
Over
440,000 students in grades 9 through 12, coming from 1,300 schools, participated at the baseline
assessment in 1960. Personality measures were available at baseline and at the 50th
year followup.
Thus, we used these two waves of data to test the rank order stability, mean-level change,
and profile stability of personality, over the lifespan. Participants in the 50th
year follow-up were
selected using the following procedures. First, a representative subsample of 4,879 participants
was randomly selected from a 10% random subsample of the schools that were originally
surveyed in 1960. Next, using a wide variety of tracking methods (see Stone et al., 2014), the
project team managed to locate 84.8% of the random subsample: 15.5% were deceased, 50.3%
were located with an address and verified, 19% were located with an address and not verified.
Survey materials were mailed to the presumably surviving subjects whose address had been
identified (i.e., 3,462 people). Of these, about 56% responded to the survey and were included in
the final dataset of the 50th
year follow-up (N = 1,952, out of which 1,858 were coded as
“credible,” see “Data Analysis” below, and therefore used in our analyses), however, due to
missing data on the personality variables, our final longitudinal sample using listwise deletion
was N = 1,795.
PERSONALITY STABILITY OVER 50 YEARS 26
The participant demographics across the two waves used were as follows: (a) the gender
distribution was stable across the two time-points, with 52.5% females at baseline and 52%
females at the follow-up; (b) the race/ethnicity distribution was also fairly similar across waves
with 93% Whites/Caucasians at baseline and 95.3% at the follow-up; and (c) the ages were on
average 16 years old at baseline (with participants ranging from 9th
to 12th
grades) and 67 years
old at the follow-up.
Measures.
Project Talent Personality Inventory (PTPI; baseline). The PTPI included 108 items
from which ten different scale composites were scored and recorded. All the scale items are
available in Table 1S in the Supplementary Materials.
The Vigor scale (7 items) measures the physical activity level of a person. The Calmness
scale (9 items) measures the ability to react to emotional situations in an appropriate manner
without extreme emotions. The Mature Personality scale (24 items) measures the ability to get
work done efficiently and to accept assigned responsibility. The Impulsiveness scale (9 items)
measures the tendency to make quick decisions without full consideration of the outcomes. The
Self-Confidence scale (12 items) measures one’s feelings of social acceptability and the
willingness to act and think independently. The Culture scale (10 items) measures the tendency
to recognize the value of aesthetic things, and to display refinement and good taste. The
Sociability scale (12 items) measures the tendency to enjoy being with people. The Leadership
scale (5 items) measures activities such as taking charge and seeking out responsibilities. The
Social Sensitivity scale (9 items) measures the propensity to put oneself in another’s place.
Finally, the Tidiness scale (11 items) measures the desire for order and neatness in one’s
environment. For each item, participants rated how well the item described them on a 5-point
scale (“extremely well” to “not very well”). Item-level data are unfortunately not available to
PERSONALITY STABILITY OVER 50 YEARS 27
researchers today for the entire sample (only for 4% of the sample), which is why we relied on
the scale scores computed by the Project Talent staff. Furthermore, when computing the PTPI
scale scores that are currently available to researchers, the Project Talent staff did not use the
original Likert scale coding. Instead, they dichotomized the individual item scores as follows:
answers A (extremely well) and B (quite well) were coded as 1, whereas answers C (fairly well),
D (slightly), and E (not very well) were coded as 0; in the case of reverse scored items, answers
D and E were coded as 1, whereas answers A, B, and C were coded as 0. After the item-level
answers were dichotomized, the Project Talent staff summed them up, to form the 10 PTPI scale
scores that are currently available. To make the long-form scale scores used at baseline
comparable with the short-form scale scores used at the follow-up, we computed item averages,
instead of sums, for each scale.5
As mentioned earlier, in previous work on independent but comparable samples
(Pozzebon et al., 2013), researchers established the validity and reliability of the 10 (long-form)
PTPI scales (see also Table 1 in the present paper), and they identified how the 10 PTPI scales
relate to modern Big Five inventories (e.g., John, Donahue, & Kentle, 1991). Thus, SelfConfidence
and Calmness were most reflective of Emotional Stability; Sociability, Vigor, and
Leadership were most reflective of Extraversion; Culture was best reflective of Openness, Social
Sensitivity reflected Agreeableness; and Mature Personality, Impulsiveness (reverse scored), and
Tidiness reflected Conscientiousness.
Project Talent Personality Inventory (PTPI; 50th year follow-up). At the 50th
year
follow-up, the Project Talent staff administered a short-form version of the PTPI. Specifically,
the 10 PTPI scales were measured using a subset of 5 of the original items for each scale. All the
scale items are available in Table 1S in the Supplementary Materials, and detailed codebooks
PERSONALITY STABILITY OVER 50 YEARS 28
and scale construction syntax (note that the scale construction syntax used was the same as the
one used in the validation study, as we recoded the item names to match) are publicly available
at the following address: https://osf.io/vxba7/. As it was the case at baseline, participants rated
how well each item described them on a 5-point scale (“extremely well” to “not very well”).
Although item-level data are available at the 50th
year follow-up, to make the short-form versions
of the PTPI scales as comparable as possible to the long-form versions used at baseline, we
followed the same scale computation procedure that was used by the Project Talent staff in 1960.
Specifically, we dichotomized and averaged the items. Therefore, all the analyses presented in
this paper are based on the dichotomized coding of the items.
Given the similar internal consistencies, test-retest reliabilities, and high intercorrelations
between the long- and short-form versions of the PTPI, as presented in the validation
study, we concluded that the two measures were highly comparable and therefore testing the
rank-order stability, mean-level change, individual-level change of personality traits, and
personality profile stability over time using these measures was appropriate.
Data Analysis
Participants were excluded prior to all analyses based on response credibility.
Specifically, we only analyzed cases that were coded as “credible” on the original response
credibility index (see Wise et al., 1979). This credibility index was computed based on a
Screening scale, which included questions such as “How many days are in a week?” that should
have been answered easily by anyone who did not suffer from a reading problem, a clerical
problem in recording answers, general slowness, or a lack of cooperation. Because the 50th
year
follow-up was conducted on a representative sub-sample of the original baseline data (i.e., as
described earlier, at the 50th
year follow-up, the researchers did not attempt to contact everyone
who had participated at baseline, but only a representative 10% sub-sample), for the purposes of
PERSONALITY STABILITY OVER 50 YEARS 29
this study, the baseline sample consisted of 4,879 participants. Of these, 4,513 cases were coded
as credible. Furthermore, cases were excluded based on having missing data on personality
measures at baseline. Thus, we were left with a total sample of 4,510 participants at baseline.
Due to the longitudinal design of this study and the long time-span covered, we had missing data
at the 50th
year follow-up. To better understand how participants who stayed in the study at the
50th
year follow-up differed from the participants who dropped out, we conducted an attrition
analysis, which we present in Table 6S of our Supplementary Material. We present the results of
these analyses in the next section. Because participants who stayed in the study differed
systematically from participants who dropped out (see Table 6S), we dealt with missing data in
two different ways. First, in all our subsequent analyses, we analyzed the data using listwise
deletion (N = 1,795). Second, we used the full information maximum likelihood (FIML)
approach (maximum likelihood estimation with robust standard errors) to obtain parameter
estimates and standard errors that accounted for the missing data (Enders, 2010; Enders &
Bandalos, 2001). In this approach, all the model covariates were used to predict the missing data,
and the estimation was based on N = 4,510, which was the baseline sample relevant for this
study. We present FIML results in Tables 7S and 8S of our Supplementary Material. Notably,
there were no meaningful differences between the results using listwise deletion and those using
FIML estimation.
The main analysis consisted of five parts. First, we assessed the rank-order stability of the
PTPI scales across 50 years. To do so, we obtained a correlation matrix across all 10 PTPI scales
and across both time points. The correlations of interest for testing the rank-order stability of
personality over the lifespan were those between the same scales across time points (e.g., the
correlation between Vigor at baseline and Vigor at the 50th
year follow-up, represents the rank-
PERSONALITY STABILITY OVER 50 YEARS 30
order stability over 50 years of the trait Vigor). All the PTPI scales scores were based on
dichotomized item coding and averaging the relevant items. In addition to the raw rank-order
stability coefficients, we also obtained rank-order stability coefficients in a structural equation
modeling framework, which are likely to be more accurate, as we accounted for measurement
error (Little, 2013). To that effect, we used single-indicator latent constructs with the marker
variable convention, and used the observed variance and estimated test-retest reliability of the
long- and short-form PTPI scales, respectively, obtained from the validation study to account for
measurement error (Watson, 2004).
Second, we tested the mean-level change of the 10 PTPI scores across 50 years using
paired-samples t-tests. Because the validation study showed slight mean-level differences
between long- and short-form versions of the PTPI scales, and because the Project Talent data
included long-forms at baseline and short-forms at the 50th
year follow-up, we addressed this
issue by conducting a robustness check, where we adjusted the means from the 50th
year followup
by adding or subtracting a constant equal to the mean-difference observed between long- and
short-forms in the validation study. After adjusting the means, we re-computed longitudinal
mean-level change.
Third, we tested individual-level change using the Reliable Change Index (RCI;
Christensen & Mendoza, 1986; Jacobson & Truax, 1991). The RCI is calculated for each trait,
for each person, separately. RCI = (X2 – X1)/Sdiff, where X2 is a person’s score at Time 2, X1 is a
person’s score at Time 1, and Sdiff = (2(SE)2
)1/2
, that is, the standard error of difference between
the two test scores. The standard error of measurement, SE = Sdev(1-αtest-retest)1/2
, where Sdev is the
standard deviation of the measure and αtest-retest is its test-retest reliability. Because the Project
Talent data included long-forms at baseline and short-forms at the 50th
year follow-up, we
PERSONALITY STABILITY OVER 50 YEARS 31
computed two different standard errors of measurement, one for each form, using data from the
short-term longitudinal validation study.6
RCI scores larger than 1.96 or smaller than -1.96 are
considered indicative of reliable change. Based on RCI scores, people are split into increasers,
decreasers, and nonchangers and then the distribution of people across these three groups is
compared, via a chi-square test, with a distribution expected by chance (i.e., 2.5% increasers,
2.5% decreasers, and 95% nonchangers).
Fourth, following Furr (2008) and Klimstra and colleagues (2010), we used qcorrelations
to assess personality profile stability based on the PTPI self-reports. To assess
overall profile stability, we correlated the rank-ordered set of PTPI traits assessed at baseline
with the same set assessed at the 50th
year follow-up, within each person. To assess distinctive
profile stability, we first subtracted average scale scores from the corresponding raw scores for
each person at each time point. Next, we computed distinctive stability scores, for each person,
by correlating the rank-ordered set of PTPI difference scores at baseline with the rank-ordered
set of the same difference scores at the 50th
year follow-up. To assess within-time normativeness
(a.k.a. profile normativeness) we correlated each person’s rank-ordered set of PTPI traits at each
time point with the rank-ordered set of sample means on PTPI traits at that same time point, thus
rendering two sets of profile normativeness (one for baseline and one for the 50th
year followup).
To assess whether the magnitude of the within-person correlations observed across time was
meaningful, we had to evaluate them against the distribution of within-person correlations that
could be found in a sample with the same mean and standard deviation, but in which profiles had
been randomly paired across time. Thus, we conducted a simulation study following previous
recommendations (Robins et al., 2001; De Fruyt et al., 2006). We used the same original data set,
but instead of using the correct pairs across time, we randomly assigned follow-up measurement
PERSONALITY STABILITY OVER 50 YEARS 32
occasions to baseline measurement occasions. Note that the simulation benchmarks are only
relevant for overall profile stability and distinctive stability (i.e., the cross-time analyses).
Fifth, we systematically tested gender differences. To test for gender differences in rankorder
stability, we obtained stability coefficients for each trait and for each gender and then
conducted Fisher tests to compare the magnitudes of the stability coefficients. To test for gender
differences in mean-levels, we conducted independent-samples t-tests to test cross-sectional
gender mean-differences at baseline and at the 50th
year follow-up, and we used repeatedmeasures
ANOVA to test the interaction between change in personality across time and gender.
To test for gender differences in individual-level change, we dummy-coded RCI scores, such that
people who showed reliable change in either direction got a 1 and people who did not show a
reliable change got a 0. Then we cross-tabulated that with gender (0=male, 1= female) and
obtained phi correlations, which indicated whether women or men were more likely to show
reliable change on each personality trait across 50 years. To test for gender differences in
personality profile stability, we correlated gender with each of the three aspects of personality
profile stability (overall profile stability, distinctive profile stability, and profile normativeness).
All the data analysis scripts necessary to reproduce these results are publicly available at
the following address: https://osf.io/vxba7/. Furthermore, all the output files can be found at the
same address in case the reader is interested in exact p-values for each of our effects, in addition
to the effect sizes and 95% confidence intervals that we report in this paper.
PERSONALITY STABILITY OVER 50 YEARS 33
Results
Attrition Analysis
In the attrition analyses we tested mean-level differences in the PTPI scales measured at
baseline between participants who dropped out and those who stayed in the study at the 50th
year
follow-up. Results can be found in Table 6S of the Supplementary Material.
The attrition analyses showed that participants who stayed in the study at the follow-up
differed slightly on their PTPI scores. Specifically, participants who stayed (vs. those who
dropped) were higher in vigor, calmness, and mature personality (Cohen’s d = .18). All other
attrition effects were not statistically significant and smaller than a Cohen’s d of .10.
Rank-order Stability
The raw correlation coefficients for the rank-order stability of the 10 PTPI scales can be
found in Table 5. The average rank-order stability over 50 years was .23, ranging from .09 for
Impulsiveness to .34 for Culture, with most scales showing rank-order stability above .20. The
latent framework rank-order stabilities (corrected for error using test-retest reliabilities and
variances of the PTPI scales obtained from the longitudinal validation sample) were as follows:
Vigor (.22), Calmness (.32), Mature Personality (.43), Impulsiveness (.13), Self-Confidence
(.28), Culture (.41), Sociability (.29), Leadership (.32), Social Sensitivity (.35), Tidiness (.36).
Thus, when using the latent framework that accounts for measurement error (as estimated from
the validation study and as detailed in Table 2S in the Supplementary Material), the average
rank-order stability across PTPI scales over 50 years was .31, with most scales showing rankorder
stability above .25.7
The raw and error-adjusted rank-order stabilities can be found in
Figure 1 along with their 95% confidence intervals.
PERSONALITY STABILITY OVER 50 YEARS 34
Mean-level Change
Table 6 presents descriptive statistics of each of the 10 PTPI scales at baseline and at the
50th
year follow-up, as well as standardized mean-level changes across the 50-year period. For
the sake of comparison with previous research, to calculate standardized mean-level changes, we
used the same procedure used by Roberts and colleagues (2006), namely the single-group,
pretest-posttest raw score effect size (Morris & DeShon, 2002). Specifically, d = (Mfollow-up –
Mbaseline)/SDbaseline.
The means for all the scales were reasonably close to the theoretical midpoints (i.e., .50)
and the standard deviations were reasonably wide. We can observe that across the board, people
increased over time in all the traits, except for Impulsiveness on which they decreased. The
average change (in absolute value) was slightly above one half of a standard deviation (d = .63),
the standardized mean-level change ranging between -.17 for Impulsiveness and 1.64 for Mature
Personality. The average mean-level change as well as the direction of change was consistent
with previous research on the Big Five personality traits (Roberts et al., 2005). Thus, when
people were in their 60s (as opposed to when they were in their teens), they were higher in
Calmness and Self-confidence (indicative of higher Emotional Stability), higher in Mature
Personality and Tidiness (indicative of higher Conscientiousness), higher in Leadership
(indicative of the dominance facet of Extraversion), and higher in Social Sensitivity (indicative
of Agreeableness). These standardized mean-level changes found in the Project Talent
longitudinal sample are also consistent (albeit larger) with the standardized mean-level changes
we observed in our cross-sectional validation study (see Tables 1 and 2), when comparing
different age cohorts. The effect directions were very similar, but the average effect size was
smaller in the cross-sectional analyses, only about a quarter of a standard deviation (d = .25 when
PERSONALITY STABILITY OVER 50 YEARS 35
using the long-forms of the PTPI and d = .21 when using the short-forms), as opposed to the half
standard deviation observed in the longitudinal sample.
One possibility is that the observed longitudinal mean-level change observed in Project
Talent was inflated by using different scale versions (long-form at baseline and short-form at the
follow-up). To address this issue, we conducted a robustness check. Specifically, we adjusted the
means from the 50th
year follow-up by adding or subtracting a constant equal to the meandifference
observed between long- and short-forms in the validation study (for mean-differences
observed in the validation study, see Table 4). After adjusting the means, we re-computed
longitudinal mean-level change, using the same formula used before, namely the single-group,
pretest-posttest raw score effect size (Morris & DeShon, 2002). As we can see in Table 6, the
mean-level changes that included the robustness check were very similar to those without the
robustness check. Some effects increased slightly while others decreased slightly, but the average
change observed (in absolute value) was exactly the same, that is, about half a standard deviation
(d = .63). The pattern of change observed was consistent, with two exceptions. Specifically, the
Impulsiveness effect changed from a small negative to a non-statistically significant effect, and
the Sociability effect decreased from about a third of a standard deviation to close to zero. The
new Sociability effect is more consistent with past research, which does not predict that people
should increase in Sociability as they grow older. In sum, the robustness-check suggests our
results hold up and are consistent with past research, that is, personality changes across the
lifespan in the maturational direction. Furthermore, this is the first study to test mean level
changes in personality in a longitudinal setting across 50 years.
PERSONALITY STABILITY OVER 50 YEARS 36
Individual-level Change
Table 7 shows the percentages of people, for each trait, who either increased, decreased,
or stayed the same according to the Reliable Change Index across the 50-years between the two
assessments. Of the ten personality traits assessed, Mature Personality exhibited the highest level
of reliable change, with 60.7% of the people in the sample showing change, mostly increases
(58.7%). The lowest level of reliable change was found for Leadership, where 21.4% of the
people in the sample showed reliable change, mostly increases again (17.3%). On average, on
any given trait, about 40% of the people in the sample showed reliable change, whereas 60%
showed no reliable change.
As expected, the percentages of people increasing or decreasing on any given trait
corresponded to the mean-level changes presented in Table 6 and followed a maturation pattern,
with highest percentages of “increasers” (vs. “decreasers”) being found for traits indicative of
Conscientiousness (Mature Personality, Tidiness), Emotional Stability (Calmness, Selfconfidence),
Agreeableness (Social Sensitivity), and dominance facets of Extraversion
(Leadership). Furthermore, as seen in Table 7, chi-square tests showed that, for each of the ten
traits, the pattern of increasers, nonchangers, and decreasers differed significantly from an
expected random-change pattern where 95% of people would show no change, 2.5% would show
increases, and 2.5% would show decreases. Thus, there appears to be reliable change in each of
the ten personality traits. Looking across traits, according to the RCI, 97.9% of the people
showed reliable change on at least one of the ten personality traits assessed across the 50-year
period, 58.9% of the people showed reliable change on four or more traits, and only .2% of
people showed reliable change on all ten traits.
PERSONALITY STABILITY OVER 50 YEARS 37
In sum, across 50 years, there was evidence of reliable change for every single trait
assessed, patterns of individual-level change followed mean-level maturational patterns, almost
everyone showed reliable change on at least one trait, more than half of the people showed
reliable change on four or more traits, but very few people changed reliable on all ten traits.
Personality Profile Stability
Overall profile stability, which represents the similarity between a person’s trait profile at
one time and their trait profile at a later time (e.g., Ozer & Gjerde, 1989), over a 50-year time
span, ranged from -.69 to .98, with a mean of .37 (SD = .31) and a median of .40. Interestingly,
this is the same overall profile stability estimate that Lönnqvist and colleagues (2008) found
across 15 years (going from age 20 to age 35). Furthermore, our overall profile stability estimate
was well above the corresponding estimate produced in our simulation study, where profiles
were matched randomly across time (M = .25, SD = .33, Mdn =.28), suggesting that overall
profile stability across 50-years was higher than chance (Mdiff = .12, t(3588) = 11.23, p < .001,
95% CI Mdiff [.10; .14], Cohen’s d = .38). Notably, the overall profile stability estimates
produced in our simulation were very similar to estimates produced in previous simulations; for
example, Robins and colleagues (2001) found an average value of .20 in their simulation; the
somewhat large overall profile stability estimates in simulations where data were randomly
matched across time are likely due to what Furr (2008) called the normativeness problem, which
is why we assessed, in addition to overall profile stability, two additional estimates of profile
stability.
Distinctive profile stability, which represents the similarity between a person’s distinctive
trait profile at one time and their distinctive trait profile at a later time (Furr, 2008), over a 50year
time span, ranged from -.80 to .97, with a mean of .17 (SD = .35) and a median of .20.
PERSONALITY STABILITY OVER 50 YEARS 38
These estimates were well above the corresponding estimates produced in the simulation (M =
.01, SD = .35, Mdn =.01), suggesting that distinctive profile stability across 50-years was higher
than chance (Mdiff = .16, t(3588) = 13.7, p < .001, 95% CI Mdiff [.14; .18], Cohen’s d = .46).
Regarding within-time normativeness or profile normativeness, which represents the
degree to which a person’s profile is similar to the average profile within each developmental
period (Furr, 2008), we had two sets of estimates. During adolescence (at baseline), within-time
normativeness ranged from -.76 to .98, with a mean of .51 (SD = .28) and a median of .56.
Within-time normativeness during older adulthood (at the 50th
year follow-up) ranged from -.71
to .97, with a mean of .62 (SD = .24) and a median of .69. These estimates are in line with
previous findings (e.g., Bleidorn et al., 2012; Klimstra et al., 2010) and suggest that most people
have personality profiles that are normative, regardless of age. To better understand what profile
normativeness might mean across time, we also investigated generalized normative stability,
which reflects the degree to which two age periods (adolescence and older adulthood in our case)
have similar normative trait profiles. Figure 2 shows the normative trait profiles that we found in
1960 adolescents (average age 16) versus older adults (50 years later). Based on these profiles,
generalized normative stability was .77 (note that this is a single estimate, not the mean of a
distribution).
Gender Differences in Personality Stability and Change
We tested gender differences across all four kinds of stability and change. We started
with cross-sectional gender differences, for each time point (baseline and 50th
year follow-up).
As seen in Table 8, at baseline (average age 16), women scored higher than men in Mature
Personality and Tidiness (indicative of Conscientiousness), Sociability (Extraversion), Culture
(Openness), and Social Sensitivity (Agreeableness). These same effects replicated at the follow-
PERSONALITY STABILITY OVER 50 YEARS 39
up, and additionally, women scored lower than men in Self-confidence (indicative of lower
Emotional Stability).
Regarding the rank-order stability of personality traits across 50 years, we found that the
average stability was .21 for women and .20 for men, and none of the stability coefficients
differed by gender at p < .001.
Regarding mean-level change, men and women showed similar patterns of change, and
only two gender-by-time interactions were statistically significant at p < .001, specifically, men
(vs. women) increased more in self-confidence across 50 years (d = .86 vs. d = .50, respectively),
whereas women (vs. men) increased more in social sensitivity as they aged (d = 1.24 vs. d =
1.04, respectively).
Regarding individual-level change, we found no statistically significant gender
differences in reliable change.
The correlation between gender and personality profile stability was .16 (p < .001),
indicating that women showed higher levels of overall profile stability. However, this was likely
because women showed higher levels of profile normativeness within each time point (the
correlation between gender and profile normativeness was .12, p < .001, at baseline; and .19, p <
.001 at the 50th
year follow-up). Indeed, there were no gender differences in distinctive profile
stability across the 50 years (r = .04, p = .087).
Across the board, these results are largely consistent with past research, suggesting that,
although there are cross-sectional gender differences in personality, men and women do not
differ much in their patterns of personality stability and change across the lifespan.
PERSONALITY STABILITY OVER 50 YEARS 40
Discussion
In the present paper, we used a large US sample that assessed people’s personality during
adolescence, as well as 50 years later. We tested rank-order stability, mean-level change,
individual-level change, personality profile stability, gender differences in stability and change,
and we also used two independent samples to validate the short-forms of the PTPI against the
long-forms.
We found that the average rank-order stability across the 10 personality traits and across
50 years was .31 (when accounting for measurement error in a latent framework) and .23 (when
estimated without correcting for unreliability). The only previous study inconsistent with our
finding is the recent study by Harris and colleagues (2016), but this study may have
underestimated the stability of personality traits over the lifespan due to their use of one itemmeasures
and different data sources at the different time points (teacher at time 1 vs. self or close
other at time 2). Furthermore, our present results, where the average raw stability coefficient was
.23, with an average 95% confidence interval of [.19; .27] (see Fig. 1) are consistent with the full
developmental model put forward by Fraley and Roberts (2005), where all three developmental
processes (stochastic-contextual, person–environment transactions, and developmental
constancies) are present.
Why would personality traits be consistent from age 16 through age 66? Broadly
speaking authors have argued for environmental factors, genetics, or both when attempting to
explain why individuals would remain consistent over long periods of time (Fraley & Roberts,
2005). What environments or environmental factors could promote consistency over a 50-year
period stretching from adolescence to old age? It is difficult to imagine that many people in their
60s would find themselves in similar environments to those occupied in their teens. Thus, except
PERSONALITY STABILITY OVER 50 YEARS 41
for people who remained in their parents’ home or nearby in the same community, strictly
environmental consistency would be an unlikely explanation. Alternatively, one could imagine
that people could play similar roles in adolescence and old age and that the continuity in the role
may help explain the continuity in personality (Roberts, 2007). If a person played the clown in
high school, or was the leader or nurturer, it would not be out of the question that they could play
the same role in their 60s. The other factor thought to contribute to personality continuity is
genetics. One argument would be that some temperamental factor that resulted from genetic
differences at conception would play out as a permanent signal in one’s personality over time.
The extreme version of this argument, that people are “hard wired from birth” to possess a
specific personality, is difficult to support given the incredibly small relation between early
childhood temperament and personality traits in adulthood (Caspi & Silva, 1995). It would be
impossible for a signal of such small magnitude (correlations at or below .10) to result in the .20
correlation that we find on average from adolescence to old age. However, newer perspectives on
genetics that focus on developmental genetics may provide a potential answer. Specifically, a
sociogenomic perspective on personality trait development (Roberts, in press) argues that
experience acts on gene systems during development and that this process results in fixed
phenotypes that emerge through development (referred to as “pliable” systems). Thus,
experience modifies the genome through epigenetic mechanisms to help create a phenotype that
reflects both fixed and dynamic genetic processes that may then create a consistent signal in
personality from adolescence to adulthood. Of course, both environmental and genetic
explanations for continuity in personality across the life course are necessarily speculative
because no longitudinal study has tracked yet, genes, epigenetic systems, nor experience over the
time-line of this study.
PERSONALITY STABILITY OVER 50 YEARS 42
We found that the average mean-level change was about half of a standard deviation
across the 10 personality traits, which is also consistent with past research. The pattern of change
was consistent with the maturity principle, where most traits are assumed to increase in an
adaptive manner (e.g., higher conscientiousness, agreeableness, emotional stability, and the
dominance facet of extraversion) (Roberts et al., 2006). The standardized mean-level changes
found in our Project Talent longitudinal sample were also consistent with the standardized meanlevel
changes we observed in our cross-sectional validation study, though the effects were twice
the size. In sum, mean-level change in personality traits over 50 years was, on average, slightly
over a half of a standard deviation. For several scales, such as the maturity scale, the amount of
change was over one standard deviation, which is substantial. In fact, the changes in the scales
that reflect the largest changes estimated from prior work (e.g., agreeableness, conscientiousness,
and emotional stability) all showed changes around 1 to 1.5 standard deviations which are quite
large by psychological standards. The present findings that personality trait change across 50
years was (a) larger than change across shorter periods of time (e.g., 10 years, where effects were
about a quarter to a third of a standard deviation), and (b) consistent with estimates of cumulative
change across the lifespan extrapolated from meta-analytic findings (Roberts et al., 2006),
suggest that the plasticity model, where change continues and cumulates across the lifespan,
might be more fitting than a “set point” model that would argue for a braking system on change.
The mean-level changes also beg the question of why individuals would show such
dramatic shifts with age. Like explanations for rank-order stability, the explanations for meanlevel
change have focused on genetic and environmental factors. While longitudinal twin studies
point to the typical finding that both are important for development (Bleidorn et al., 2009), as of
yet, there are no insights into how genetics would contribute to these changes. It is a possibility
PERSONALITY STABILITY OVER 50 YEARS 43
that personality trait development works like puberty in that it is pre-programed and universal,
but to date no molecular genetic evidence supports that claim, and the present findings on
individual-level change bring further evidence against this claim (that is, change is unlikely to be
universal if there are individual-differences in change). Some of the environmental factors
thought to contribute to personality development include normative transitions to adulthood
(Roberts & Damian, in press). These experiences include applying oneself to achievement
situations, such as school and work (Bleidorn et al., 2013; Hudson, Roberts, & Lodi-Smith,
2012) and some aspects of relationships, such as relationship duration and satisfaction (Lehnart,
Neyer, & Eccles, 2010). Unfortunately, because there was such a long-time span between the
assessments of the Project Talent sample, we do not have the experiential data to test these ideas.
The maturational patterns we found are also consistent with the classic account of Erik Erikson
(1963), who postulated that people mature as they age, and change continuously throughout the
lifespan, pressed to adapt by the ever-increasing social demands and developmental tasks
required by each life stage.
Regarding individual-level change, we found that, on average, on any given trait, about
40% of the people in the sample showed reliable change, whereas 60% did not. This is higher
than the 30-10% estimates of people who showed reliable change across shorter time spans (e.g.,
Robins et al., 2001; Roberts et al., 2001), which suggests that greater change accumulates with
time. Looking across traits, we found that 97.9% of the people showed reliable change on one or
more of the 10 personality traits assessed across the 50-year period (compared to 84% over 8
years in the study by Roberts and colleagues, 2001), again suggesting that more personality
change may accumulate with time, thus causing more individuals to show reliable change on
PERSONALITY STABILITY OVER 50 YEARS 44
more traits. Furthermore, like mean-level changes, individual-level change patterns supported the
maturation hypothesis.
The fact that most people showed reliable change in one or more personality traits
supports the perspective that individual differences in change are an important developmental
phenomenon. Not everyone changes in the same way despite normative trends. Some people
change less than their peers, while others change more than the norm. Our study is consistent
with past research identifying the existence of individual differences in change in shorter
longitudinal studies (e.g., De Fruyt et al., 2006; Mõttus, Johnson, & Deary, 2012; Robins, et al,
2001; Roberts, Caspi, & Moffitt, 2001; Schwaba & Bleidorn, 2017). It makes sense that with a
longer time span, more people showed unique patterns of change. These non-normative patterns
of change beg the question of why? In many other studies, this question has been answered by
showing that life experiences correlate with individual differences in change (Bleidorn, 2012;
Göllner et al., 2017; Lehnart, Neyer, & Eccles, 2010; Takahashi, et al., 2013). Unfortunately, the
unique feature of the present study—the 50-year time lag, also prevented the tracking of life
experiences that could have shown relations to these individual differences in change in the
present case. Future research would, optimally, track personality and life experiences over a 50year
period, with multiple assessments, to further establish evidence for the links between life
experiences and personality trait development in adulthood.
Regarding personality profile stability, we found that overall profile stability across 50
years was .37, which was consistent with previous findings of profile stability over 15 years
(e.g., Lönnqvist et al., 2008). As expected, we found distinctive profile stability to be lower, with
a mean of .17. Importantly, both these estimates were well above estimates of stability from the
simulation study that we ran, indicating that profile stability across 50 years (overall and
PERSONALITY STABILITY OVER 50 YEARS 45
distinctive) was higher than chance. Profile normativeness was high at each age period and it
increased as people aged (.51 and .62, during adolescence and older age, respectively), which
was also consistent with previous findings that looked at profile normativeness across much
shorter time spans (e.g., Bleidorn et al., 2012; Klimstra et al., 2010).8
Interestingly, normative
personality profiles were very similar between 1960 adolescents and the same sample assessed
50 years later correlating .77 (see Fig. 2). To better understand how normative personality
profiles might have changed (or not changed) across historical cohorts, we conducted additional
exploratory analyses using the cross-sectional data used from the validation study, for 20- and
60-year-olds separately (see these additional normative trait profiles in Figure 2). Notably, the
correlation between the normative trait profile of 16-year-olds in 1960 and 20-year-olds in 2013
was .67. This estimate is comparable to previously reported generalized normative stability
estimates (.64-.78), where 12- and 20-year-olds’ normative profiles were compared (Klimstra et
al., 2012; note that in this latter study, youth cohorts were assessed within a 5-year period of each
other, unlike across our two studies, where the gap was 53 years). All in all, the relatively high
degree of consistency in personality norms, both across US history and across time, suggests that
certain personality profiles may be normative across time because they might facilitate
adjustment (e.g., it might always be advantageous to be more sociable than you are impulsive,
regardless of age and historical time).
Why would personality profiles be stable across 50 years and why would profile
normativeness increase with age? As with rank-order stability, the answer to the first question
probably lies with a combination of genetic and environmental factors, whereby profile stability
reflects fixed phenotypes that emerge through development (see Bleidorn et al., 2012; Roberts, in
press). Regarding the second question, researchers have suggested that profile normativeness
PERSONALITY STABILITY OVER 50 YEARS 46
might increase across time because it reflects psychological adjustment (Bleidorn et al., 2012;
Klimstra et al., 2010). Presumably, a higher level of psychological adjustment, might allow
people to select environments that fit them better, thus exposing themselves to less
environmental pressure to change (see also Donnellan et al., 2007), which would in turn lead to
increasing stability and normativeness across time.
Although our study has many advantages, including its large sample and the inclusion of
personality measures across a 50-year time span, our study also has several limitations. First,
item-level data were not available at baseline, which means that we were not able to conduct
more complex statistical analyses, such as structural equation modeling, which would have
allowed us to test for invariance over time in the measurement model and use latent constructs.
To alleviate this limitation, we used cross-sectional data from the validation study to test for
measurement invariance in the PTPI scales across the two relevant age groups: 20- vs. 60-yearolds.
Across all scales (except for the short-form of self-confidence), we found evidence for
configural, metric, scalar, and strict invariance, which gave us more confidence that we may
interpret the longitudinal findings as we did (though the mean-level changes in self-confidence
should be interpreted with caution since we only had evidence for metric but not scalar
invariance for its short-form version). Second, as the attrition analyses presented in Tables 6S of
the supplementary materials showed, the sample available at the follow-up differed slightly from
the sample that dropped out. Specifically, the people who stayed in the study at the 50th
year
follow-up (as opposed to those who dropped out) were higher in vigor, calmness, and mature
personality at baseline (i.e., during adolescence), which means our sample was no longer
representative of the US high-school population in 1960. It is possible that this impacted our
estimates of personality stability and change, though it is difficult to assess in what way. Another
PERSONALITY STABILITY OVER 50 YEARS 47
limitation is that the Project Talent staff used long forms of the personality measures at baseline,
but short forms at the follow-up. We tried to alleviate this shortcoming by conducting a
validation study on two independent but comparable samples to obtain internal and test-retest
reliabilities for both the long and the short forms (see validation study), as well as investigate
their mean-level differences (which we adjusted for in the robustness check) and cross-sectional
correlations (which were in the order of .90 or higher, indicating that the short- and long-forms
likely measure the same constructs). Although we did everything we could to address the issue of
long- versus short-forms at the two time-points, it is still possible that our estimates of change
are biased. Furthermore, it is possible that the robustness check employed did not correct for this
issue because our robustness check made one crucial assumption, namely that differences
between long- and short-forms of the PTPI would be constant across samples and across
historical time. This is a rather strict assumption that is likely to be violated, so our mean-level
change estimates should be treated with caution. However, the fact that the effects found are
consistent with past research and with the cross-sectional results gives us some confidence.
Finally, another limitation is that the present study only assessed personality at two time-points,
which prevented us from delving more deeply into developmental patterns and processes. For
instance, by having such a long time-span in between the two assessments with no other
assessments in between, we were not able to capture when changes occurred or whether some
changes occurred that were later reversed. We were also limited in the kinds of models we could
fit to the data. With three or more assessments we might have been able to fit growth curve
models to better understand developmental trajectories.
Despite these limitations, the present study advances our understanding of personality
stability and change, because it examines rank-order stability, mean-level change, individual-
PERSONALITY STABILITY OVER 50 YEARS 48
level change, and profile stability in a large sample and over a very long timespan (50 years). We
found evidence for continuity in the way individuals ranked relative to each other, but there was
also a great deal of change in mean-levels of specific traits, a large percentage of individuals
showed reliable change, and their individual patterns of change varied, though overall there was
evidence for maturation. Furthermore, patterns of personality also showed some degree of
stability, within individuals across the lifespan, and this stability was explained by both
distinctive stability and within-time normativeness. Together, these results suggest that, although
individuals maintain some of their core personality across the lifespan, they also change.
Moreover, that change in personality across the lifespan is likely cumulative, following a
maturational adaptive pattern, and that change is not uniform or universal, meaning that
environmental factors, such as life experiences, are likely to contribute to change, thus, resulting
in individual differences in change. These results are in line with previous research, supporting
developmental models that include a developmental constancy (e.g., genetic component) and
personality stability over the lifespan (Fraley & Roberts, 2005; Roberts, in press). These results
are also in line with the maturity principle, whereby people adapt and mature as they grow older
so they can successfully engage in life’s responsibilities, and they support a plasticity (vs. a “set
point” model) of personality change (Roberts et al., 2006; Fraley, 2002). The profile stability
results are also in line with genetic models, with the cumulative continuity principle, and with
the social investment principle, whereby people’s personality profiles may increase in
normativeness across the lifespan as they invest in normative social roles (see Bleidorn et al.,
2012; Klimstra et al., 2010).
PERSONALITY STABILITY OVER 50 YEARS 49
Future research should include more long-term studies, over long periods of time, that
assess personality at multiple waves, to enable scientists to investigate more deeply
developmental process as well as look at inter-individual variation in stability and change.
In sum, there is evidence for stability in personality traits across the lifespan, while at the
same time there is evidence of change presumably resulting from life’s trials and tribulations.
Going back to the love story we presented at the beginning, an interesting question raised by
these findings is how likely are people to stay within the same personality “zones” (e.g., above or
below the median) after a 50-year long hiatus? Looking at the variable-centered approach, where
raw rank-order stability was .23 on average, and using the Binomial Effect Size Display (BESD;
Rosenthal & Rubin, 1982) to transform it, we could say that Jerzy and Cyla would each have
about 60:40 odds of having stayed in the same personality “zone,” on each trait, during their long
separation. But people are not characterized by single traits. Thus, looking at the whole person,
where overall profile stability was .37 on average, and using BESD to transform it, we could say
that Jerzy and Cyla would each have about 70:30 odds of having stayed in the same personality
“zone” after their long separation. Given our findings in mean-level changes, individual-level
changes, and profile normativeness, we can also assume that both Jerzy and Cyla would have
been likely to be more mature and more normative in their personality profiles when they met
later in life. These being said, perhaps it makes more sense now how Jerzy and Cyla could fall in
love all over again after 40 years of being apart. Although life’s experiences had undoubtedly
changed them (likely for the better), somewhere deep down they each managed to find in the
other a glimmer of the person they had fallen in love with many years before.
PERSONALITY STABILITY OVER 50 YEARS 50
References
Allemand, M., Zimprich, D., & Hertzog, C. (2007). Cross-sectional age differences and
longitudinal age changes of personality in middle adulthood and old age. Journal of
Personality, 75, 323-358.
Allport, G. W. (1954). Personality: A psychological interpretation. New York: Henry Holt.
Anusic, I., & Schimmack, U. (2016). Stability and change of personality traits, self-esteem, and
well-being: Introducing the meta-analytic stability and change model of retest
correlations. Journal of Personality and Social Psychology, 110, 766-81.
Biesanz, J. C., & West, S. G. (2000). Personality coherence: Moderating self-other profile
agreement and profile consensus. Journal of Personality and Social Psychology, 79, 425–
437.
Bleidorn, W. (2012). Hitting the road to adulthood: Short-term personality development during a
major life transition. Personality and social psychology bulletin, 38(12), 1594-1608.
Bleidorn, W., Hopwood, C. J., & Lucas, R. E. (2018). Life events and personality trait change.
Journal of Personality, 86, 83-96.
Bleidorn, W., Kandler, C., Riemann, R., Angleitner, A., & Spinath, F. M. (2009). Patterns and
sources of adult personality development: Growth curve analyses of the NEO PI-R scales
in a longitudinal twin study. Journal of personality and social psychology, 97(1), 142.
Bleidorn, W., Kandler, C., Riemann, R., Angleitner, A., & Spinath, F. M. (2012). Genetic and
environmental influences on personality profile stability: Unraveling the normativeness
problem. Journal of personality, 80(4), 1029-1060.
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-
PERSONALITY STABILITY OVER 50 YEARS 51
investment theory. Psychological Science, 24(12), 2530-2540.
Block, J. (1971). Lives through time. Berkeley, CA: Bancroft Books.
Caspi, A., & Herbener, E. S. (1990). Continuity and change: Assortative marriage and the
consistency of personality in adulthood. Journal of Personality and Social Psychology,
58, 250–258.
Caspi, A., Roberts, B. W., & Shiner, R. L. (2005). Personality development: Stability and
change. Annu. Rev. Psychol., 56, 453-484.
Caspi, A., & Silva, P. A. (1995). Temperamental qualities at age three predict personality traits
in young adulthood: Longitudinal evidence from a birth cohort. Child
development, 66(2), 486-498.
Cattell, R.B. (1981). Sixteen Personality Factor Questionnaire (16PF). Psykologien Kustannus
OY, Helsinki.
Cheung, G. W., & Rensvold, R. B. (2002). Evaluating goodness-of-fit indexes for testing
measurement invariance. Structural Equation Modeling, 9, 233-255.
Costa Jr, P. T., Herbst, J. H., McCrae, R. R., & Siegler, I. C. (2000). Personality at midlife:
Stability, intrinsic maturation, and response to life events. Assessment, 7(4), 365-378.
Costa, P. T., & McCrae, R. R. (1988). Personality in adulthood: a six-year longitudinal study of
self-reports and spouse ratings on the NEO Personality Inventory. Journal of personality
and social psychology, 54(5), 853.
Costa, P. T., Jr., & McCrae, R. R. (1992). Revised NEO Personality Inventory (NEO-PI-R) and
NEO Five-Factor Inventory: Professional manual. Odessa, FL: Psychological
Assessment Resources.
Costa, P. T., & McCrae, R. R. (2008). The revised neo personality inventory (neo-pi-r). The
PERSONALITY STABILITY OVER 50 YEARS 52
SAGE handbook of personality theory and assessment, 2, 179-198.
Damian, R. I., & Roberts, B. W. (2015). The associations of birth order with personality and
intelligence in a representative sample of U.S. high school students. Journal of Research
in Personality, 58, 96-105.
Damian, R. I., Spengler, M., & Roberts, B. W. (2017). Whose job will be taken over by a
computer? The role of personality in predicting job computerizability over the lifespan.
European Journal of Personality, 31, 291–310.
Damian, R. I., Su, R., Shanahan, M., Trautwein, U., Roberts, B. (2015). Can personality traits
and intelligence compensate for background disadvantage? Predicting status attainment
in adulthood. Journal of Personality and Social Psychology, 109, 473-89.
De Fruyt, F., Bartels, M., van Leeuwen, K. G., de Clercq, B., Decuyper, M., & Mervielde, I.
(2006). Five types of personality continuity in childhood and adolescence. Journal of
Personality and Social Psychology, 91, 538–552.
Donnellan, M. B., Conger, R. D., & Burzette, R. G. (2007). Personality development from late
adolescence to young adulthood: Differential stability, normative maturity, and evidence
for the maturity‐stability hypothesis. Journal of personality, 75(2), 237-264.
Edmonds, G. W., Goldberg, L. R., Hampson, S. E., & Barckley, M. (2013). Personality stability
from childhood to midlife: relating teachers’ assessments in elementary school to
observer-and self-ratings 40years later. Journal of research in personality, 47(5), 505-
513.
Ferguson, C. J. (2010). A meta-analysis of normal and disordered personality across the life
span. Journal of Personality and Social Psychology, 98, 659-67.
Fraley, R. C. (2002). Attachment stability from infancy to adulthood: Meta-analysis and dynamic
PERSONALITY STABILITY OVER 50 YEARS 53
modeling of developmental mechanisms. Personality and Social Psychology Review, 6,
123-151.
Fraley, R. C., & Roberts, B. W. (2005). Patterns of continuity: a dynamic model for
conceptualizing the stability of individual differences in psychological constructs across
the life course. Psychological review, 112(1), 60.
Furr, M. R. (2008). A framework for profile similarity: Integrating similarity, normativeness, and
distinctiveness. Journal of Personality, 76, 1267–1316.
Göllner, R., Damian, R. I., Rose, N., Spengler, M., Trautwein, U., Nagengast, B., & Roberts, B.
W. (2017). Is doing your homework associated with becoming more conscientious?
Journal of Research in Personality, 71, 1-12.
Hampson, S. E., & Goldberg, L. R. (2006). A first large cohort study of personality trait stability
over the 40 years between elementary school and midlife. Journal of personality and
social psychology, 91(4), 763.
Harms, P.D., Roberts, B.W., & Winter, D. (2006). Becoming the Harvard man: Personenvironment
fit, personality development, and academic success. Personality and Social
Psychology Bulletin, 32, 851-865.
Harris, M. A., Brett, C. E., Johnson, W., & Deary, I. J. (2016). Personality stability from age 14
to age 77 years. Psychology and aging, 31(8), 862.
Hevesi, D. (2011). Jerzy Bielecki dies at 90; Fell in love in a Nazi camp. The New York Times.
Retrieved from http://www.nytimes.com/2011/10/24/world/europe/jerzy-bielecki-dies-at-
90-fell-in-love-in-a-nazi-camp.html?_r=0
Hoekstra, H. A., Ormel, J., & De Fruyt, F. (1996). NEO persoonlijkheids vragenlijsten: NEO-PIR,
NEO-FFI. Handleiding [NEO personality questionnaires: NEO-PI-R and NEO-FFI-
PERSONALITY STABILITY OVER 50 YEARS 54
manual]. Lisse, The Netherlands: Swets & Zeitlinger.
Hudson, N.W., Roberts, B.W., & Lodi-Smith, J. (2012). Personality trait development and
social investment at work. Journal of Research in Personality, 46, 334-344.
John, O. P., Donahue, E. M., & Kentle, R. L. (1991). The big five inventory—versions 4a and
54.
John, O. P., Naumann, L. P., & Soto, C. J. (2008). Paradigm shift to the integrative big five trait
taxonomy. Handbook of personality: Theory and research, 3, 114-158
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. Journal of Personality and Social
Psychology, 96, 898–912.
Klimstra, T. A., Luyckx, K., Hale, W. W., III, Goossens, L., & Meeus, W. H. J. (2010).
Longitudinal associations between personality profile stability and adjustment in college
students: Distinguishing among overall stability, distinctive stability, and within-time
normativeness. Journal of Personality, 78, 1163–1184.
Lehnart, J., Neyer, F. J., & Eccles, J. (2010). Long‐term effects of social investment: The case of
partnering in young adulthood. Journal of personality, 78(2), 639-670.
Little, T. D. (2013). Longitudinal structural equation modeling. Guilford Press.
Lodi-Smith, J., & Roberts, B. W. (2007). Social investment and personality: A meta-analysis of
the relationship of personality traits to investment in work, family, religion, and
volunteerism. Personality and Social Psychology Review, 11(1), 68-86
Lönnqvist, J.-E., Mäkinen, S., Paunonen, S. V., Henriksson, M., & Verkasalo, M. (2008).
Psychosocial functioning in young men predicts their personality stability over 15 years.
Journal of Research in Personality, 42, 599–621.
PERSONALITY STABILITY OVER 50 YEARS 55
Lüdtke, O., Roberts, B. W., Trautwein, U., & Nagy, G. (2011). A random walk down university
avenue: life paths, life events, and personality trait change at the transition to university
life. Journal of personality and social psychology, 101(3), 620.
MacCallum, R. C., Browne, M. W., & Sugawara, H. M. (1996). Power analysis and
determination of sample size for covariance structure modeling. Psychological Methods,
1, 130.
Major, J. T., Johnson, W., & Deary, I. J. (2014). Linear and nonlinear associations between
general intelligence and personality in Project Talent. Journal of personality and social
psychology, 106, 638.
McCrae, R. R. (2008). A note on some measures of profile agreement. Journal of Personality
Assessment, 90, 105–109.
McCrae, R. R., & Costa, P. T. (1994). The stability of personality: Observations and
evaluations. Current directions in psychological science, 3(6), 173-175.
McCrae, R. R., & Costa, P. T. (2008). Empirical and theoretical status of the five-factor model of
personality traits. The SAGE handbook of personality theory and assessment, 1, 273-294.
Morris, S. B., & DeShon, R. P. (2002). Combining effect size estimates in meta-analysis with
repeated measures and independent-groups designs. Psychological methods, 7(1), 105.
Mõttus, R., Johnson, W., & Deary, I. J. (2012). Personality traits in old age: measurement and
rank-order stability and some mean-level change. Psychology and aging, 27(1), 243.
Neyer, F. J., & Lehnart, J. (2007). Relationships matter in personality development: Evidence
from an 8‐year longitudinal study across young adulthood. Journal of personality, 75(3),
535-568.
Ostendorf, F., & Angleitner, A. (2004). NEO Persönlichkeitsinventar nach Costa und McCrae:
PERSONALITY STABILITY OVER 50 YEARS 56
Revidierte Fassung [NEO Personality Inventory by Costa and McCrae: Revised edition].
Göttingen, Germany: Hogrefe.
Ozer, D. J., & Gjerde, P. F. (1989). Patterns of personality consistency and change from
childhood through adolescence. Journal of Personality, 57, 483–507.
Pozzebon, J., Damian, R. I., Hill, P. L., Lin, Y., Lapham, S., & Roberts, B. W. (2013).
Establishing the validity and reliability of the Project Talent Personality Inventory.
Frontiers in Psychology—Personality Science and Individual Differences, 4, 968.
Roberts, B. W. (2007). Contextualizing personality psychology. Journal of Personality, 75,
1071-1081.
Roberts, B. W. (2009). Back to the future: Personality and assessment and personality
development. Journal of research in personality, 43(2), 137-145.
Roberts, B.W. (in press). A revised sociogenomic model of personality traits. Journal of
Personality.
Roberts, B. W., Caspi, A., & Moffitt, T. E. (2001). The kids are alright: Growth and stability in
personality development from adolescence to adulthood. Journal of Personality and
Social Psychology, 81, 670–683.
Roberts, B. W., & Damian, R. I. (in press). The principles of personality trait development and
their relation to psychopathology. In D. Lynam & D. Samuel (Eds.), Using basic
personality research to inform the personality disorders. Oxford, UK: Oxford University
Press.
Roberts, B. W., & DelVecchio, W. F. (2000). The rank-order consistency of personality traits
from childhood to old age: a quantitative review of longitudinal studies. Psychological
bulletin, 126(1), 3.
PERSONALITY STABILITY OVER 50 YEARS 57
Roberts, B.W., & Robins, R. W. (2004). A longitudinal study of person-environment fit and
personality development. Journal of Personality, 72, 89-110.
Roberts, B. W., Walton, K. E., & Viechtbauer, W. (2006). Patterns of mean-level change in
personality traits across the life course: a meta-analysis of longitudinal
studies. Psychological bulletin, 132(1), 1.
Roberts, B.W., Wood, D, & Caspi, A. (2008). The development of personality traits in
adulthood. In O.P. John, R.W. Robins, & L. A. Pervin (Eds.), Handbook of personality:
theory and research (3rd edition, Ch 14, pp. 375-398). New York, NY: Guilford.
Roberts, B. W., Wood, D., & Smith, J. L. (2005). Evaluating five factor theory and social
investment perspectives on personality trait development. Journal of Research in
Personality, 39(1), 166-184.
Robins, R. W., Fraley, R. C., Roberts, B. W., & Trzesniewski, K. H. (2001). A longitudinal study
of personality change in young adulthood. Journal of Personality, 69, 617–640.
Shiner, R. L., Allen, T. A., & Masten, A. S. (2017). Adversity in adolescence predicts
personality trait change from childhood to adulthood. Journal of Research in
Personality, 67, 171-182.
Stone, C., Scott, L., Battle, D., & Maher, P. (2014). Locating longitudinal respondents after a
50-Year hiatus. Journal of Official Statistics, 30, 311-334.
Takahashi, Y., Edmonds, G.E., Jackson, J.J., & Roberts, B.W. (2013). Longitudinal correlated
changes in conscientiousness, preventative health-related behaviors, and self-perceived
physical health. Journal of Personality, 81, 417-427.
Terracciano, A., Costa Jr, P. T., & McCrae, R. R. (2006). Personality plasticity after age
30. Personality and Social Psychology Bulletin, 32(8), 999-1009.
PERSONALITY STABILITY OVER 50 YEARS 58
Van Aken, M. A., Denissen, J. J., Branje, S. J., Dubas, J. S., & Goossens, L. (2006). Midlife
concerns and short‐term personality change in middle adulthood.European Journal of
Personality, 20(6), 497-513.
Wise, L., McLaughlin, D., & Steel, L. (1979). The Project TALENT Data Bank Handbook,
Revised. Palo Alto, CA: American Institutes for Research.
Wagner, J., Becker, M., Lüdtke, O., & Trautwein, U. (2015). The first partnership experience
and personality development: A propensity score matching study in young adulthood.
Social Psychological and Personality Science, 6, 455-463.
Watson, D. (2004). Stability versus change, dependability versus error: Issues in the assessment
of personality over time. Journal of Research in Personality, 38, 319–350.
PERSONALITY STABILITY OVER 50 YEARS 59
Footnotes
1
Notably, q-correlations reflect only the degree of similarity in the shape of two
personality profiles across time, leaving out information regarding the elevation (i.e., the mean of
the different profile elements) and scatter (i.e., the variability around the profile’s elevation) of
the profiles (see Biesanz & West, 2000). Some studies (e.g., Donnelan et al., 2007; Robins et al.,
2001) assessed profile stability with a related index (D2
), in addition to q-correlations, but these
studies suggested that both indices led to the same conclusions. Furthermore, McCrae (2008)
showed that q-correlations are reliable measures of profile stability.
2
The items of the original scales at baseline were not recorded so we were unable to
create duplicate scales at both time points.
3
Using the short-term longitudinal validation study data, another way to compute the
residual errors used to account for measurement error in the latent models from the “Main
Study,” is to use correlations between the long-forms at Time 1 and the short-forms at Time 2
instead of test-retest reliabilities for each form, in order to simultaneously take into account
unreliability due to (a) re-testing and (b) switching from long- to short-forms. More details on
how these residuals were computed can be found in Table 3S in the Supplementary Materials.
Corrected stability coefficients using these alternative residual errors can be found in Footnote 7
and all syntax and output can be found on our OSF page in the file titled “Main Study Syntax.”
4
Although several previous papers have been published using the personality data
available at baseline in Project Talent (Damian & Roberts, 2015; Damian et al., 2015; 2017;
Major et al., 2015; Spengler et al., 2018), to our knowledge, no previous papers have used the
personality data available at the 50th year follow-up, and thus, no previous papers have analyzed
stability and change in personality across 50 years, using this data set, which is the topic of our
PERSONALITY STABILITY OVER 50 YEARS 60
current submission. A comprehensive list of papers published using other variables from the
Project Talent dataset can be found at the following link:
http://www.projecttalent.org/about/biblio.
5
To test for mean-level changes in personality across time, we could not use the summed
scale scores because the PTPI scales had different number of items at baseline vs. the follow-up.
Therefore, we used individual mean scale scores. At the follow-up, where we had item-level data
available we simply averaged the item scores for each scale. At baseline, where we did not have
item-level data available (we only had the summed scale scores available), we simply divided the
available summed score by the known number of items present in each scale (reported in the
methods section). This method is somewhat problematic because we have no way of knowing
whether there were missing items for any of the participants in the scales measured at baseline.
Specifically, because all the items were dichotomized and summed, a score of 0 going into the
sum could either mean that an item had a score of 0 or it could be a missing item. Thus, without
this knowledge, dividing the total sum score available by the total number of known items might
underestimate the mean scores at baseline. Because this problem is not present at the follow-up
(where we had item-level data and could therefore distinguish between items with a zero score
and missing items), it follows that the standardized mean-level change reported in Table 6 could
be overestimated.
6
An alternative way to compute RCI scores was to use only one standard error of
measurement (as opposed to a different one for each kind of form) computed using the
correlations between long-forms of the PTPI at Time 1 and short-forms of the PTPI at Time 2
(from the short-term longitudinal validation study), to simultaneously correct for measurement
error resulted from two different sources (a) re-testing and (b) switching from long- to short-
PERSONALITY STABILITY OVER 50 YEARS 61
forms. The results using this alternative RCI computation can be found in Table 9S (in the
Supplementary Materials) and they did not differ in any meaningful way from the results
presented in Table 7 (which used the RCI computation presented in the paper). Specifically, the
average percentage of people who stayed the same was 59.67% in Table 7 vs. 60% in Table 9S.
7
The latent framework rank-order stabilities (corrected for error as estimated in Table 3S
in the Supplementary Material, see also Footnote 3 for more details) were as follows: Vigor
(.22), Calmness (.32), Mature Personality (.31), Impulsiveness (.13), Self-Confidence (.31),
Culture (.41), Sociability (.29), Leadership (.32), Social Sensitivity (.35), Tidiness (.37). Thus,
using this alternative way to account for measurement error (as opposed to the way proposed in
Table 2S in the Supplementary Material), the average rank-order stability across PTPI scales
over 50 years was .30 (compared to .31 when we used errors from Table 2S). Thus, the particular
way of computing the errors used in the latent models (Table 2S vs. 3S) did not make a
meaningful difference.
8
We also conducted one further analysis, where we looked at within-time normativeness
in our own cross-sectional data from the validation study and we looked at 20-year-olds and 60year-olds
separately. We found that, at age 20, the average profile normativeness was .57 (SD =
.34), and that, at age 60, the average profile normativeness was .62 (SD = .25). This replicated
our longitudinal findings, thus, bringing further evidence that normativeness might increase with
age.
Running head: PERSONALITY STABILITY OVER 50 YEARS 62
Table 1. Descriptive statistics of the long forms of Project Talent Personality Inventory Scales and mean-level age differences (cross-sectional validation study).
Total 20s 60s
Scale α M SD α M SD α M SD d
Vigor .82 .33 .32 .79 .34 .31 .84 .32 .33 -.06
Calmness .85 .68 .30 .84 .60 .32 .84 .73 .29 .41
Mature Personality .91 .69 .25 .92 .62 .27 .89 .73 .22 .41
Impulsiveness .56 .19 .17 .62 .19 .19 .53 .20 .16 .05
Self Confidence .78 .52 .25 .75 .40 .24 .74 .58 .23 .75
Culture .79 .53 .27 .78 .52 .27 .80 .55 .27 .11
Sociability .79 .42 .25 .80 .39 .25 .78 .44 .24 .20
Leadership .76 .27 .30 .74 .26 .30 .77 .28 .30 .07
Social Sensitivity .82 .68 .29 .83 .65 .31 .80 .70 .28 .16
Tidiness .85 .55 .30 .84 .51 .30 .85 .58 .29 .23
Note. Ns = 3,907-3,926 for total sample, N for the 20s = 1,243-1,246, N for the 60s = 2,405-2,420. The remaining participants were not in their 20s or 60s. Minor
differences in Ns across scales were due to different numbers of missing items on each scale, which sometimes prevented scale computation. d = standardized
mean-level change between 60 and 20 year olds. For the sake of comparison with previous research, to calculate d, we used the same procedure used by Roberts
et al., 2006, namely the single-group, pretest-posttest raw score effect size (Morris & DeShon, 2002). Specifically, d = (Mage60 – Mage20)/SDage20. Bold font
indicates effect was statistically significant at p < .001.
PERSONALITY STABILITY OVER 50 YEARS 63
Table 2. Descriptive statistics of the short forms of Project Talent Personality Inventory Scales and mean-level age differences (cross-sectional validation study).
Total 20s 60s
Scale α M SD α M SD α M SD d
Vigor .83 .32 .35 .80 .30 .34 .84 .33 .36 .09
Calmness .83 .68 .35 .82 .61 .37 .82 .73 .33 .32
Mature Personality .79 .75 .31 .82 .69 .34 .75 .79 .28 .29
Impulsiveness .69 .15 .24 .74 .18 .26 .66 .14 .22 -.15
Self Confidence .68 .45 .31 .69 .33 .30 .64 .52 .29 .63
Culture .75 .45 .33 .73 .45 .33 .76 .45 .34 .00
Sociability .77 .51 .34 .76 .48 .34 .77 .53 .34 .15
Leadership .76 .27 .30 .74 .26 .30 .77 .28 .30 .07
Social Sensitivity .84 .73 .34 .83 .69 .36 .84 .77 .33 .22
Tidiness .83 .50 .38 .80 .47 .37 .85 .52 .39 .14
Note. Ns = 3,920-3,927 for total sample, N for the 20s = 1,244-1,246, N for the 60s = 2,414-2,420. The remaining participants were not in their 20s or 60s. Minor
differences in Ns across scales were due to different numbers of missing items on each scale, which sometimes prevented scale computation. d = standardized
mean-level change between 60 and 20 year olds. For the sake of comparison with previous research, to calculate d, we used the same procedure used by Roberts
et al., 2006, namely the single-group, pretest-posttest raw score effect size (Morris & DeShon, 2002). Specifically, d = (Mage60 – Mage20)/SDage20. Bold font
indicates effect was statistically significant at p < .001
PERSONALITY STABILITY OVER 50 YEARS 64
Table 3. Inter-correlations among the Project Talent Personality Inventory scales (cross-sectional validation study).
Scale 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
1. Vigor_L 1 .33 .47 .20 .33 .40 .45 .45 .30 .39 .95 .36 .42 .20 .11 .41 .46 .45 .27 .38
2. Calmness_ L 1 .65 .00 .52 .53 .42 .30 .60 .49 .36 .93 .59 -.05 .31 .43 .47 .30 .56 .43
3. Mature Personality_ L 1 .10 .52 .57 .47 .45 .62 .63 .49 .67 .89 .04 .24 .49 .51 .45 .57 .53
4. Impulsiveness_ L 1 .17 .07 .21 .28 .04 .03 .22 .02 .09 .83 .09 .11 .19 .28 .05 .04
5. Self-Confidence_ L 1 .30 .43 .40 .28 .29 .37 .44 .42 -.05 .83 .26 .38 .41 .22 .23
6. Culture_ L 1 .50 .38 .62 .57 .43 .54 .54 .11 .04 .92 .57 .38 .56 .53
7. Sociability_ L 1 .40 .51 .37 .50 .42 .45 .19 .14 .47 .91 .40 .47 .35
8. Leadership_ L 1 .29 .29 .48 .33 .38 .23 .14 .41 .40 1.00 .25 .25
9. Social Sensitivity_ L 1 .46 .33 .59 .59 .07 .02 .50 .58 .29 .93 .41
10. Tidiness_ L 1 .41 .52 .58 .04 .09 .53 .41 .29 .41 .93
11. Vigor_S 1 .38 .44 .20 .14 .43 .52 .48 .29 .40
12. Calmness_ S 1 .63 .00 .21 .46 .48 .33 .56 .47
13. Mature Personality_ S 1 .07 .15 .46 .50 .38 .56 .51
14. Impulsiveness_ S 1 -.16 .16 .19 .23 .08 .06
15. Self-Confidence_ S 1 .00 .06 .14 -.02 .04
16. Culture_ S 1 .53 .41 .44 .50
17. Sociability_ S 1 .40 .53 .39
18. Leadership_ S 1 .25 .25
19. Social Sensitivity_ S 1 .38
20. Tidiness_ S 1
Note. Ns = 3,891-3,927. All values in bold font were statistically significant at p < .001. Values show correlations for the total sample. _L = long-form PTPI
scales, _S = short-form PTPI scales.
PERSONALITY STABILITY OVER 50 YEARS 65
Table 4. Mean-level differences between long- and short-forms of each PTPI scale in cross-sectional validation study (from paired samples t-tests).
Long Form Short form Mean Difference (Paired samples t-test)
PTPI Scale M SD M SD M-Diff SD 95% CI
Vigor .33 .32 .32 .35 .01 .11 [.01;.01]
Calmness .68 .3 .68 .35 .00 .13 [-.01;.00]
Mature Personality .69 .25 .75 .31 -.06 .14 [-.07;-.06]
Impulsiveness .19 .17 .15 .24 .04 .13 [.04;.05]
Self Confidence .52 .25 .45 .31 .06 .17 [.06;.07]
Culture .53 .27 .45 .33 .09 .14 [.09;.09]
Sociability .42 .25 .51 .34 -.09 .15 [-.10;-.09]
Leadership .27 .3 .27 .3 .00 .13 N/A*
Social Sensitivity .68 .29 .73 .34 -.05 .15 [-.06;-.05]
Tidiness .55 .3 .5 .38 .05 .11 [.05;.06]
Note: All mean differences in bold font were statistically significant at p < .001. CI= confidence interval. *95% CI was not available for the leadership scale
because the test could not be performed as the long- and short- forms of the leadership scale were exactly the same.
PERSONALITY STABILITY OVER 50 YEARS 66
Table 5. Rank-order stabilities and inter-correlations among the Project Talent Personality Inventory scales at baseline (T1) and at the 50th
year follow-up (T2).
Scale
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
1. VigorT1 .81
2. CalmnessT1 .40 .77
3. Mature
PersonalityT1 .48 .57 .79
4. ImpulsivenessT1 .23 .12 .20 .61
5. SelfConfidenceT1
.30 .44 .39 .11 .89
6. CultureT1 .39 .50 .55 .17 .28 .82
7. SociabilityT1 .48 .38 .33 .24 .35 .39 .80
8. LeadershipT1 .41 .38 .47 .23 .32 .40 .37 .76
9. Social
SensitivityT1 .39 .53 .52 .20 .25 .56 .48 .36 .85
10. TidinessT1 .37 .50 .59 .09 .24 .57 .35 .33 .46 .88
11. VigorT2 .20 .09 .17 .08 .08 .13 .15 .16 .11 .14 .80
12. CalmnessT2 .13 .22 .19 .01 .11 .19 .17 .13 .19 .20 .41 .73
13. Mature
PersonalityT2 .11 .06 .18 .06 .08 .13 .15 .14 .15 .15 .54 .55 .66
14. ImpulsivenessT2 .04 -.04 -.01 .09 -.03 .01 .07 .04 .02 .04 .24 .09 .17 .68
15. SelfConfidenceT2
.08 .12 .10 .01 .23 .04 .04 .04 .00 -.01 .07 .14 .05 -.24 .84
16. CultureT2 .09 .18 .22 .06 .10 .34 .17 .18 .27 .25 .40 .45 .43 .18 -.03 .79
17. SociabilityT2 .13 .10 .14 .06 .07 .19 .24 .12 .17 .16 .43 .43 .40 .25 -.03 .51 .78
18. LeadershipT2 .14 .13 .19 .07 .16 .16 .14 .25 .13 .14 .44 .39 .47 .25 .11 .44 .41 .76
19. Social
SensitivityT2 .07 .13 .17 .03 .04 .24 .17 .10 .26 .22 .29 .50 .40 .15 -.12 .50 .56 .30 .80
20. TidinessT2 .08 .12 .16 .00 .03 .18 .17 .09 .18 .30 .40 .48 .48 .13 -.08 .47 .39 .28 .42 .80
Note. N = 1,795 (we used listwise deletion to deal with missing data). On the main diagonal are the test-retest reliabilities from the validation study. On the grey
diagonal are the rank-order stabilities of each personality trait across 50 years. All values in Bold font were statistically significant at p < .001.
PERSONALITY STABILITY OVER 50 YEARS 67
Table 6. Descriptive statistics and mean-level changes in Project Talent Personality Inventory scales at baseline (T1) and at the 50th
year follow-up (T2)
PTPI Scale (Big Five Corresponding Scale)
Time 1 (Baseline) Time 2 (50th
year follow-up)
Standardized
Mean-level
change
Standardized
Mean-level
change with
robustness check
M SD M SD d d-adjusted
Vigor (Extraversion) .56 .31 .56 .40 .00 .03
Calmness (Emotional Stability) .50 .28 .80 .28 1.07 1.07
Mature Personality (Conscientiousness) .49 .22 .85 .23 1.64 1.34
Impulsiveness (Low Conscientiousness) .22 .18 .19 .24 -.17 .07
Self-Confidence (Emotional Stability) .44 .21 .58 .26 .67 .97
Culture (Openness/Intellect) .53 .24 .55 .33 .08 .46
Sociability (Extraversion) .57 .24 .65 .33 .33 -.04
Leadership (Extraversion) .27 .27 .40 .33 .48 .48
Social Sensitivity (Agreeableness) .52 .26 .81 .30 1.12 .91
Tidiness (Conscientiousness) .52 .26 .71 .36 .73 .93
Note. N = 1,795. d = standardized mean-level change between baseline and 50th
year follow-up. For the sake of comparison with previous research, to calculate d,
we used the same procedure used by Roberts et al., 2006, namely the single-group, pretest-posttest raw score effect size (Morris & DeShon, 2002). Specifically, d
= (Mfollow-up – Mbaseline)/SDbaseline. Bold font indicates effect was statistically significant at p < .001.
PERSONALITY STABILITY OVER 50 YEARS 68
Table 7. Individual-Level Change in Personality Traits from baseline (T1) to the 50th
year follow-up (T2)
PTPI Scale (Big Five Corresponding Scale) Decreased (%) Stayed the same (%) Increased (%) χ2
(2, N = 1,795)
Vigor (Extraversion) 17.4 67.2 15.4 2,915.5
Calmness (Emotional Stability) 3.4 54.7 41.9 11,452
Mature Personality (Conscientiousness) 2 39.3 58.7 23,212
Impulsiveness (Low Conscientiousness) 16.4 72.7 10.9 1,989.9
Self-Confidence (Emotional Stability) 10.2 52.5 37.3 9,416.6
Culture (Openness/Intellect) 12.6 69.9 17.5 2,475.9
Sociability (Extraversion) 12.3 66.3 21.4 3,405
Leadership (Extraversion) 4.1 78.6 17.3 1,642.1
Social Sensitivity (Agreeableness) 5.6 43.7 50.7 17,191
Tidiness (Conscientiousness) 9.9 51.8 38.3 9,927.2
Note. N = 1,795. Percentages for decrease, staying the same, and increase were based on the Reliable Change Index (RCI), where change less than -1.96 or
greater than 1.96 is considered reliable change. The RCIs used for this table were computed using two different standard errors of measurement derived from two
separate sets of test-retest reliabilities (from long and short scales, respectively). The chi-square tested whether the observed distribution of non-changers and
changers differed from the expected distribution if change were random (i.e., 95% stayed the same, 2.5% each increased and decreased). Bold font indicates χ2
test was statistically significant at p < .001
PERSONALITY STABILITY OVER 50 YEARS 69
Table 8. Gender Differences in Personality Stability and Change from baseline (T1) to the 50th
year follow-up (T2).
PTPI Scale (Big Five Corresponding
Scale)
Cross-sectional differences Longitudinal differences in change
Baseline
(T1)
50th
year
(T2)
Gender d score Change
over time
d
Gender X
Time
Partial eta
squared
Gender
correlation
with reliable
change
Rank-order
stability
M SD M SD Baseline
(T1)
50th
year
(T2)
Vigor (Extraversion) -.06 -.03 .00 -.01
Women .55 .31 .56 .39 .03 .19
Men .57 .31 .57 .40 .00 .20
Calmness (Emotional Stability) .11 .14 .00 -.01
Women .51 .29 .82 .26 1.07 .20
Men .48 .28 .78 .30 1.07 .22
Mature Personality (Conscientiousness) .14 .13 .00 -.05
Women .50 .22 .86 .21 1.64 .18
Men .47 .22 .83 .24 1.64 .17
Impulsiveness (Low Conscientiousness) .05 .08 .00 .00
Women .23 .19 .20 .24 -.16 .09
Men .22 .18 .18 .24 -.22 .08
Self-Confidence (Emotional Stability) .05 -.24 .01 -.06
Women .44 .22 .55 .26 .50 .28
Men .43 .21 .61 .24 .86 .18
Culture (Openness/Intellect) .53 .49 .00 -.06
Women .58 .22 .63 .31 .23 .32
Men .46 .23 .47 .34 .04 .27
Sociability (Extraversion) .34 .36 .00 -.05
Women .61 .23 .71 .32 .43 .22
Men .53 .24 .59 .34 .25 .21
Leadership (Extraversion) .00 -.12 .00 -.03
Women .27 .28 .38 .33 .39 .25
Men .27 .27 .42 .33 .56 .25
Social Sensitivity (Agreeableness) .52 .64 .01 -.07
Women .58 .25 .89 .23 1.24 .22
Men .45 .25 .71 .33 1.04 .19
Tidiness (Conscientiousness) .40 .43 .01 -.05
Women .57 .25 .78 .32 .84 .26
Men .47 .25 .63 .38 .64 .29
PERSONALITY STABILITY OVER 50 YEARS 70
Note. Nmales = 860; Nfemales = 935; Gender d score = differences between women and men divided by the pooled standard deviation (for the sake of
comparison with Roberts et al., 2001). Change over time d = standardized mean-level change between baseline and 50th
year follow-up; for the sake of
comparison with previous research, to calculate the change over time d, we used the same procedure used by Roberts et al., 2006, namely the single-group,
pretest-posttest raw score effect size (Morris & DeShon, 2002), that is, d = (Mfollow-up – Mbaseline)/SDbaseline. The Gender X Time Partial Eta Squared
represents the magnitude of the effect size of the interaction between gender and time on personality change. Reliable change was recoded so that 0 = no change
and 1 = reliable change; gender was coded 0=male and 1=female; the numbers represent phi correlations. Bold font indicates effect was statistically significant at
p < .001.
PERSONALITY STABILITY OVER 50 YEARS 71
Figure 1. Rank-order stability in Project Talent over 50 years (raw coefficients and coefficients adjusted for measurement error).
Note. Light gray bars represent raw stability coefficients, whereas dark gray bars represent stability coefficients obtained from using single-item latent models,
corrected for measurement error. Bars represent 95% confidence intervals.
.20 .22
.18
.09
.23
.34
.24 .25 .26
.30
.22
.32
.43
.13
.28
.41
.29 .32 .35 .36
0
0.1
0.2
0.3
0.4
0.5
0.6
Rank-order
stability
Personality Scales
PERSONALITY STABILITY OVER 50 YEARS 72
Figure 2. Normative Personality Trait Profiles from the main longitudinal study (baseline = adolescent norms in 1960; 50th
year follow-up = older adult norms
2010) and from the cross-sectional validation study (20-year-olds vs. 60-year-olds norms in 2013).
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
Average
Personality
Score
Personality Scales
Adolescent norms (1960) Older adult norms (2010) 20s norms (2013) 60s norms (2013)
PERSONALITY STABILITY OVER 50 YEARS 73
Online Supplementary Material for:
Sixteen Going on Sixty-Six:
A Longitudinal Study of Personality Stability and Change Across 50 Years
PERSONALITY STABILITY OVER 50 YEARS 74
Table 1S. Items included in Long and Short Forms of the Project Talent Personality Inventory (PTPI). Long Forms Include ALL the items below, whereas shortforms
include only items followed by an asterisk.
PTPI Items Organized by Scale Coding
Vigor
I can work or play outdoors for hours without getting tired.
I am a fast walker.
I am full of pep and energy.*
People seem to think I lead a vigorous life.*
I am active.*
I am vigorous.*
I am energetic.*
Calmness
I often lose my temper. R
I can usually keep my wits about me even in difficult situations.
People seem to think I get angry easily. R
People seem to think I have good self-control.*
People consider me level-headed.*
I am even-tempered.
I am calm.*
I am stable.*
I am usually self-controlled.*
Mature Personality
I make good use of all my time.
I never seem to get things done on time. R
I work fast and get a lot done.
PERSONALITY STABILITY OVER 50 YEARS 75
When I say I’ll do something I get it done.
It bothers me to leave a task half done.
I can turn out a lot more work than average.
I am hard-working.*
People consider me an efficient worker.
I do my job, even when I don’t like it.
I find it hard to keep working toward long-range goals. R
I am productive.*
As soon as I finish one project or assignment, I always have something else I want to
begin.
I never volunteer for a tough job. R
I think that if something is worth starting its worth finishing.
I do things the best I know how, even if no one checks up on me.
I lose interest in most projects before I get them done. R
People seem to think they can count on me.
People consider me persistent.
I am dependable.*
People have criticized me for leaving things undone. R
I am conscientious.
I am persistent.
I am reliable.*
People consider me determined.*
Impulsiveness
I like to do things on the spur of the moment.*
I usually act on the first plan that comes to mind.*
I feel that I'm impulsive.*
People seem to think I sometimes make decisions too quickly.*
I am impulsive.*
PERSONALITY STABILITY OVER 50 YEARS 76
I don’t believe in rushing into things. R
I am cautious. R
When I have a problem, I make up my mind and don't worry about it.
It takes me quite a while to come to a decision. R
Self-Confidence
I am confident.
I'd enjoy speaking to a club group on a subject I know
Being around strangers makes me ill-at-ease. R
I'm troubled by people making fun of me. R
People seem to think my feelings are hurt too easily.* R
I am usually at ease.
People seem to think I am easily discouraged when criticized.* R
I am often self-conscious.* R
People consider me shy. R
I am sensitive.* R
I am often worried.* R
People seem to think I usually do a good job on whatever I'm doing.
Culture
I enjoy beautiful things.
I feel that good manners are very necessary for everyone.
I think culture is more important than wealth.
I enjoy cultural things.*
I am a cultured person.*
People seem to think I have good taste.*
I take part in the cultural activities in my community.*
I tend to have good taste.*
PERSONALITY STABILITY OVER 50 YEARS 77
I am refined.
I am sometimes crude. R
Sociability
I like to spend a good deal of time by myself. R
I’d rather be with a group of friends than at home by myself.
People consider me the quiet type. R
People seem to think I make new friends more quickly than most people do.
I couldn't get along without having people around me most of the time.
I enjoy getting to know people.*
I like to be with people most of the time.*
I go out of my way to be with friends.*
I prefer reading a good book to going out with friends. R
People consider me good-natured.
People consider me sociable.*
I am friendly.*
Leadership
I am the leader in my group.*
I am influential*
I have held a lot of elected offices.*
People naturally follow my lead.*
I like to make decisions.*
Social Sensitivity
I like to tease people. R
I never hurt another person’s feelings if I can avoid it.*
I seem to know how other people will feel about things.
PERSONALITY STABILITY OVER 50 YEARS 78
I sympathize with my friends and encourage them when they have problems.*
People consider me a sympathetic listener.*
People consider me very helpful in dealing with other people.
I am sympathetic*
I am considerate.*
People consider me understanding.*
Tidiness
I am never sloppy in my personal appearance.*
I have a definite place for all of my things.
Before I start a task, I spend some time getting it organized.
It bothers me to be with someone who dresses carelessly.
I like to do things systematically.
My work suffers from lack of neatness. R
People consider me very careful about my personal appearance.*
I am tidy.*
I am neat.*
I am orderly.*
I tend to be untidy. R
Note: PTPI Instructions read as follows: “For each statement below mark which the one of the five choices which best describes how the statement applies to
you. Regarding the things I do and the way I do them, this statement describes me: 1 (not very well), 2 (slightly), 3 (fairly well), 4 (quite well), 5 (extremely
well).” “R” means the item was reverse-coded. *item included in the short-form version of the scale.
PERSONALITY STABILITY OVER 50 YEARS 79
Validation Study
Table 2S. Descriptive statistics for short-term longitudinal validation study and residual (error) used in the latent-model (see “Main Study”).
Scale
Type Variable Name
Mean
Time 1
Variance
Time1
Mean
Time 2
Variance
Time 2
Sample Size
T1 and T2
Pooled
Variance
Testretest
Error
Long Vigor 0.33 0.06 0.33 0.08 38.00 0.07 0.81 0.01
Long Calmness 0.70 0.08 0.65 0.10 38.00 0.09 0.77 0.02
Long Mature Personality 0.64 0.07 0.59 0.07 38.00 0.07 0.79 0.01
Long Impulsiveness 0.19 0.03 0.19 0.03 38.00 0.03 0.61 0.01
Long Self Confidence 0.48 0.05 0.48 0.07 38.00 0.06 0.89 0.01
Long Culture 0.49 0.07 0.47 0.06 38.00 0.07 0.82 0.01
Long Sociability 0.31 0.05 0.31 0.05 38.00 0.05 0.80 0.01
Long Leadership 0.27 0.08 0.23 0.08 38.00 0.08 0.76 0.02
Long Social Sensitivity 0.57 0.10 0.60 0.10 38.00 0.10 0.85 0.01
Long Tidiness 0.46 0.08 0.44 0.07 38.00 0.08 0.88 0.01
Short Vigor 0.27 0.07 0.26 0.10 38.00 0.09 0.80 0.02
Short Calmness 0.67 0.11 0.63 0.12 38.00 0.12 0.73 0.03
Short Mature Personality 0.74 0.11 0.65 0.15 38.00 0.13 0.66 0.04
Short Impulsiveness 0.16 0.06 0.13 0.05 38.00 0.05 0.68 0.02
Short Self Confidence 0.50 0.07 0.52 0.10 38.00 0.09 0.84 0.01
Short Culture 0.39 0.11 0.39 0.10 38.00 0.11 0.79 0.02
Short Sociability 0.38 0.10 0.36 0.08 38.00 0.09 0.78 0.02
Short Leadership 0.27 0.08 0.23 0.08 38.00 0.08 0.76 0.02
Short Social Sensitivity 0.66 0.15 0.66 0.15 38.00 0.15 0.80 0.03
Short Tidiness 0.41 0.11 0.37 0.13 38.00 0.12 0.80 0.02
Note: Formula used for pooled variance: [Var_t1*(N_t1-1)+Var_t2(N_t2 -1)]/[(N_t1-1)+(N_t2 -1)]. Formula used for residual (error) that we fixed in the latent
framework: Pooled_var*(1-Test-retest).
PERSONALITY STABILITY OVER 50 YEARS 80
Table 3S. Descriptive statistics for short-term longitudinal validation study and residual (error) used in the second set of latent-models (see Footnotes 2 and 7).
Scale Type Variable Name
Mean
Time 1
Variance
Time1
Mean
Time 2
Variance
Time 2
Sample Size
T1 and T2
Pooled
Variance
Testretest
Error
Long Vigor 0.33 0.06 0.33 0.08 38.00 0.07 0.81 0.01
Long Calmness 0.70 0.08 0.65 0.10 38.00 0.09 0.77 0.02
Long Mature Personality 0.64 0.07 0.59 0.07 38.00 0.07 0.79 0.01
Long Impulsiveness 0.19 0.03 0.19 0.03 38.00 0.03 0.61 0.01
Long Self Confidence 0.48 0.05 0.48 0.07 38.00 0.06 0.89 0.01
Long Culture 0.49 0.07 0.47 0.06 38.00 0.07 0.82 0.01
Long Sociability 0.31 0.05 0.31 0.05 38.00 0.05 0.80 0.01
Long Leadership 0.27 0.08 0.23 0.08 38.00 0.08 0.76 0.02
Long Social Sensitivity 0.57 0.10 0.60 0.10 38.00 0.10 0.85 0.01
Long Tidiness 0.46 0.08 0.44 0.07 38.00 0.08 0.88 0.01
Long T1 to Short T2 Vigor 0.33 0.06 0.26 0.10 38.00 0.08 0.76 0.02
Long T1 to Short T2 Calmness 0.70 0.08 0.63 0.12 38.00 0.10 0.68 0.03
Long T1 to Short T2 Mature Personality 0.64 0.07 0.65 0.15 38.00 0.11 0.72 0.03
Long T1 to Short T2 Impulsiveness 0.19 0.03 0.13 0.05 38.00 0.04 0.42 0.02
Long T1 to Short T2 Self Confidence 0.48 0.05 0.52 0.10 38.00 0.08 0.75 0.02
Long T1 to Short T2 Culture 0.49 0.07 0.39 0.10 38.00 0.08 0.73 0.02
Long T1 to Short T2 Sociability 0.31 0.05 0.36 0.08 38.00 0.07 0.74 0.02
Long T1 to Short T2 Leadership 0.27 0.08 0.23 0.08 38.00 0.08 0.76 0.02
Long T1 to Short T2 Social Sensitivity 0.57 0.10 0.66 0.15 38.00 0.12 0.77 0.03
Long T1 to Short T2 Tidiness 0.46 0.08 0.37 0.13 38.00 0.10 0.74 0.03
Note: Formula used for pooled variance: [Var_t1*(N_t1-1)+Var_t2(N_t2 -1)]/[(N_t1-1)+(N_t2 -1)]. Formula used for residual (error) that we fixed in the latent
framework: Pooled_var*(1-Test-retest). In the “Main Study,” when using this alternative set of residuals to correct for measurement error, for Time 1 (baseline),
we used the same set of residuals as before (like in Table 2S, which is why the Long-form section of Table 3S is identical to the Long-form section of Table 2S),
however, for Time 2 (50th
year follow-up), instead of using means, standard deviations, and test-retest reliabilities obtained from the short-term longitudinal
validation study for the short-forms of the PTPI, we used data from both long- and short-forms of the PTPI and correlations between long-forms of the PTPI at
Time 1 and short-forms of the PTPI at Time 2 (from the short-term longitudinal validation study), to simultaneously correct for unreliability resulted from two
different sources (a) re-testing and (b) switching from long- to short-forms.
PERSONALITY STABILITY OVER 50 YEARS 81
Notes regarding measurement invariance testing (see Tables 4S ad 5S below):
We ran 20 sets of invariance models in a multi-group framework to test for measurement invariance across the 20-year-olds and the 60-year-olds in the
cross-sectional validation study. For individual models, model fit is considered adequate if CFI is > 0.95 and RMSEA is < .08. Most individual models, across
scales, showed reasonably good fit, though short-forms showed better fit than long-forms, and some scales had larger RMSEA values. We did not consider this to
be a major problem, because we are not using these latent factors in our main analyses (we only used manifest variables in the main analyses, and this was
mentioned as a limitation in the discussion section on p. 46 of the manuscript). Furthermore, the focus when testing measurement invariance is on relative fit.
Based on the measurement invariance literature, we determined the level of invariance with ΔCFI (Cheung & Rensvold, 2002) of two consecutive
models. For a model to be accepted, it had to yield acceptable relative fit compared with the preceding invariance model (ΔCFI > -.01, that is, if the model had a
worse fit, it was not supposed to deviate more than -.01, but to be clear, if the CFI increased in a subsequent, more constrained, model then the rule did not need
to be applied, because in those cases the model fit better, so there was no reason to reject it). As can be seen in Table 4S (long-forms), following the ΔCFI > -.01
rule, all scales showed evidence for measurement invariance (configural, metric, scalar, and strict) (the Sociability scale’s CFI deviated by -.011 when going
from metric to scalar, but in that case, the RMSEA 90% confidence intervals overlapped, so we still had some evidence for invariance). As can be seen in Table
5S (short-forms), following the ΔCFI > -.01 rule, 9 out of 10 scales showed evidence for measurement invariance (configural, metric, scalar, and strict), the only
exception being self-confidence, which showed evidence for metric, but not scalar invariance. As alluded to earlier, a second guideline for testing relative fit and
assessing measurement invariance is to check whether the 90% confidence intervals for the RMSEA are overlapping across consecutive models (Allemand,
Zimprich, & Hertzog, 2007; MacCallum, Browne, & Sugawara, 1996). Following, this guideline, not all scales showed measurement invariance, but according to
Little (2013), these are just general guidelines, cutoffs being somewhat arbitrary, and a ΔCFI > -.01 might bring sufficient evidence for measurement invariance,
especially in cases where the RMSEA fit itself does not change or even becomes better with subsequent models (which was the case for several of our scales).
To further facilitate a detailed interpretation of the results presented in Tables 4S and 5S, we give one specific example. The model with the
unconstrained factor loadings and intercepts can be found in the first row labeled “configural.” A configural invariance model was initially specified in which a
PERSONALITY STABILITY OVER 50 YEARS 82
single factor was estimated simultaneously in both age groups. All item factor loadings (one per item) and thresholds (one per item given two response options)
were estimated. The residual variances are not uniquely identified in the configural invariance model (when applying it to dichotomous/categorical items) and as
such were all constrained to 1 in both groups. As shown in Table 5S, the configural invariance model for the Social Sensitivity scale had a good fit. The analysis
proceeded by applying parameter constraints in successive models to examine potential decreases in fit resulting from measurement non-invariance between 20year-olds
and 60-year-olds. In the metric invariance model, we constrained the respective factor loadings to be equal across age-groups. The metric invariance
model did not fit significantly worse than the configural invariance model. In a next step, the scalar invariance model was fitted with the item thresholds
constrained to be equal across age-groups. The scalar invariance model did not fit significantly worse than the metric invariance model. In a last step we tested
for strict measurement invariance. The model comparison at this step proceeded backwards, such that a model with all residual variances freely estimated in the
60-year-olds was fitted first, and then compared with a model in which all residual variances were fixed to 1 in both groups. Model “strict b” did not fit
significantly worse than model “strict a”. Therefore, strict measurement invariance in Social Sensitivity can be assumed across the 20- and 60-year-olds.
PERSONALITY STABILITY OVER 50 YEARS 83
Table 4S. Measurement Invariance (20- vs. 60-year-olds) Long Forms of the PTPI
χ² df p CFI RMSEA 90% CI
Calmness
Configural 2,354.26 54 .00 .930 .152 [.147; .158]
Metric 1,931.04 62 .00 .943 .128 [.123; .133]
Scalar 2,175.54 70 .00 .936 .128 [.123; .133]
Strict a 2,418.93 61 .00 .929 .145 [.140 ; .150]
Strict b 2,175.54 70 .00 .936 .128 [123; .133]
Culture
Configural 1,492.36 70 .00 .936 .105 [.101; .110]
Metric 1,223.88 79 .00 .948 .089 [.084; .093]
Scalar 1,412.24 88 .00 .940 .091 [.086; .095]
Strict a 1,422.21 78 .00 .939 .097 [.093; .101]
Strict b 1,412.24 88 .00 .940 .091 [.086; .095]
Impulsiveness
Configural 493.27 54 .00 .990 .067 [.061; .072]
Metric 430.29 62 .00 .992 .057 [.052; .062]
Scalar 684.76 70 .00 .986 .069 [.065; .074]
Strict a 644.35 64 .00 .990 .070 [.065; .075]
Strict b 684.76 70 .00 .986 .069 [.065; .074]
Leadership
Configural 36.50 10 .00 .999 .038 [.025; .052]
Metric 35.62 14 .00 .999 .029 [.017; .041]
Scalar 115.96 18 .00 .995 .054 [.045; .064]
Strict a 74.88 13 .00 .997 .051 [.040; .062]
Strict b 115.96 18 .00 .995 .054 [.045; .064]
PERSONALITY STABILITY OVER 50 YEARS 84
Mature Personality
Configural 8,332.79 504 .00 .903 .092 [.090; .094]
Metric 6,533.02 527 .00 .925 .079 [.077; .081]
Scalar 7,064.26 550 .00 .919 .080 [.079; .082]
Strict a 8,443.72 526 .00 .902 .091 [.089; .092]
Strict b 7,064.26 550 .00 .919 .080 [.079; .082]
Self Confidence
Configural 2,461.33 108 .00 .796 .109 [.105; .113]
Metric 2,261.96 119 .00 .815 .099 [.096; .103]
Scalar 2,389.24 130 .00 .805 .097 [.094; .101]
Strict a 2,508.78 118 .00 .793 .105 [.102; .109]
Strict b 2,389.24 130 .00 .805 .097 [.094; .101]
Sociability
Configural 2,103.48 108 .00 .902 .100 [.097; .104]
Metric 1,723.052 119 .00 .921 .086 [.082; .089]
Scalar 1,945.30 130 .00 .910 .087 [.084; .091]
Strict a 2,116.30 118 .00 .901 .096 [.092; .100]
Strict b 1,945.30 130 .00 .910 .087 [.084; .091]
Social Sensitivity
Configural 528.64 54 .00 .982 .069 [.064; .075]
Metric 452.81 62 .00 .985 .059 [.054; .064]
Scalar 624.28 70 .00 .979 .066 [.061; .070]
Strict a 517.23 61 .00 .983 .064 [.059; .069]
Strict b 624.28 70 .00 .979 .066 [.061; .070]
Tidiness
PERSONALITY STABILITY OVER 50 YEARS 85
Configural 2,007.03 88 .00 .956 .109 [.105; .113]
Metric 1,490.30 98 .00 .968 .088 [.084; .092]
Scalar 1,773.30 108 .00 .962 .092 [.088; .095]
Strict a 1,982.09 97 .00 .957 .103 [.099; .107]
Strict b 1,773.30 108 .00 .962 .092 [.088; .095]
Vigor
Configural 283.45 28 .00 .991 .071 [.063; .078]
Metric 235.94 34 .00 .993 .057 [.050; .064]
Scalar 487.96 40 .00 .984 .078 [.072; .084]
Strict a 504.86 33 .00 .983 .088 [.082; .095]
Strict b 487.96 40 .00 .984 .078 [.072; .084]
PERSONALITY STABILITY OVER 50 YEARS 86
Table 5S. Measurement Invariance (20- vs. 60-year-olds) Short Forms of the PTPI
χ² df p CFI RMSEA 90% CI
Calmness
Configural 53.51 10 .00 .998 .049 [.036; .062]
Metric 58.99 14 .00 .997 .042 [.031; .053]
Scalar 197.59 18 .00 .990 .074 [.065; .083]
Strict a 97.76 13 .00 .995 .060 [.049; .071]
Strict b 197.59 18 .00 .990 .074 [.065; .083]
Culture
Configural 733.41 10 .00 .945 .199 [.187; .211]
Metric 681.83 14 .00 .949 .161 [.151; .172]
Scalar 719,109 18 .00 .946 .146 [.137; .155]
Strict a 769,347 13 .00 .942 .178 [.167; .189]
Strict b 719,109 18 .00 .946 .146 [.137; .155]
Impulsiveness
Configural 68.45 10 .00 .999 .056 [.044; .069]
Metric 76,671 14 .00 .999 .049 [.039; .060]
Scalar 95,918 18 .00 .998 .049 [.039; .058]
Strict a 69,263 13 .00 .999 .049 [.038; .060]
Strict b 95,918 18 .00 .998 .049 [.039; .058]
Leadership
Configural 36.50 10 .00 .999 .038 [.025; .052]
Metric 35.62 14 .00 .999 .029 [.017; .041]
Scalar 115.97 18 .00 .995 .054 [.045; .064]
Strict a 74.88 13 .00 .997 .051 [.040; .062]
Strict b 115.97 18 .00 .995 .054 [.045; .064]
PERSONALITY STABILITY OVER 50 YEARS 87
Mature Personality
Configural 327.49 10 .00 .991 .132 [.120; .144]
Metric 348.75 14 .00 .991 .114 [.104; .125]
Scalar 423.44 18 .00 .989 .111 [.102; .120]
Strict a 371.63 13 .00 .990 .123 [.112; .134]
Strict b 423.44 18 .00 .989 .111 [.102; .120]
Self Confidence
Configural 46.96 10 .00 .990 .045 [.032; .058]
Metric 59.71 14 .00 .987 .042 [.031; .053]
Scalar 158.24 18 .00 .961 .065 [.056; .075]
Strict a 126.54 13 .00 .969 .069 [.058; .080]
Strict b 158.24 18 .00 .961 .065 [.056; .075]
Sociability
Configural 144.26 10 .00 .988 .086 [.073; .098]
Metric 140.30 14 .00 .988 .070 [.060; .081]
Scalar 258.22 18 .00 .978 .085 [.076; .095]
Strict a 231.50 13 .00 .980 .096 [.085; .107]
Strict b 258.22 18 .00 .978 .085 [.076; .095]
Social Sensitivity
Configural 174.39 10 .00 .992 .095 [.083; .107]
Metric 160.37 14 .00 .992 .075 [.065; .086]
Scalar 192,597 18 .00 .991 .073 [.064; .082]
Strict a 181.75 13 .00 .991 .084 [.074; .095]
Strict b 192,597 18 .00 .991 .073 [.064; .082]
Tidiness
PERSONALITY STABILITY OVER 50 YEARS 88
Configural 566.46 10 .00 .983 .174 [.162; .187]
Metric 471.89 14 .00 .986 .134 [.123; .144]
Scalar 508.70 18 .00 .985 .122 [.113; .131]
Strict a 621.54 13 .00 .982 .160 [.149; .171]
Strict b 508.70 18 .00 .985 .122 [.113; .131]
Vigor
Configural 76.74 10 .00 .997 .060 [.048; .073]
Metric 89.58 14 .00 .997 .054 [.044; .065]
Scalar 141.51 18 .00 .995 .061 [.052; .071]
Strict a 103.21 13 .00 .996 .061 [.051; .073]
Strict b 141.51 18 .00 .995 .061 [.052; .071]
PERSONALITY STABILITY OVER 50 YEARS 89
Main Study
Table 6S. Attrition analyses based on participation in personality assessments at baseline and at the 50th
year follow-up.
Baseline only vs. Year 50 Frequency distributions
Variable Mean Difference 95% CI d Baseline only Year 50
Gender (Female) N/A N/A N/A 52.8% 52%
Race (White) N/A N/A N/A 90% 95.3%
Vigor -.06 [-.07; -.04] .18 N/A N/A
Calmness -.05 [-.07; -.03] .18 N/A N/A
Mature Personality -.04 [-.05; -.02] .18 N/A N/A
Impulsiveness -.01 [-.02; .00] .04 N/A N/A
Self-Confidence -.02 [-.03; .00] .08 N/A N/A
Culture -.02 [-.03; .00] .07 N/A N/A
Sociability -.02 [-.04; -.01] .09 N/A N/A
Leadership -.02 [-.04; .00] .08 N/A N/A
Social Sensitivity -.02 [-.04; -.01] .09 N/A N/A
Tidiness -.02 [-.03; .00] .06 N/A N/A
Notes. For the continuous variables, we provide mean differences, 95% confidence intervals, and absolute effect sizes (Cohen’s d), where negative mean
differences indicate higher scores for the people who stayed in the study at year 50 versus those who dropped out. For the dichotomous variables, we provide
frequency distributions for the people who dropped (baseline only) and the people who stayed in the study at year 50. N/A indicates that the respective statistics
are not available because the variable is either dichotomous or continuous, respectively. Total sample size of people who dropped out was 2,655. Total sample
size of people who participated at year 50 was 1,858.
PERSONALITY STABILITY OVER 50 YEARS 90
Table 7S (equivalent of Table 5, but with FIML). Rank-order stabilities and inter-correlations among the Project Talent Personality Inventory scales at baseline
(T1) and at the 50th
year follow-up (T2).
Scale
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
21. VigorT1 .81
22. CalmnessT1 .42 .77
23. Mature
PersonalityT1 .51 .60 .79
24. ImpulsivenessT1 .24 .13 .19 .61
25. SelfConfidenceT1
.29 .44 .39 .09 .89
26. CultureT1 .42 .53 .58 .20 .29 .82
27. SociabilityT1 .50 .42 .40 .25 .34 .44 .80
28. LeadershipT1 .40 .39 .48 .26 .29 .43 .36 .76
29. Social
SensitivityT1 .43 .57 .57 .23 .27 .61 .51 .41 .85
30. TidinessT1 .40 .53 .62 .12 .26 .59 .42 .35 .53 .88
31. VigorT2 .20 .10 .17 .08 .08 .13 .16 .16 .12 .14 .80
32. CalmnessT2 .14 .22 .21 .02 .11 .20 .18 .14 .20 .21 .41 .73
33. Mature
PersonalityT2 .12 .08 .19 .06 .08 .13 .15 .15 .16 .16 .54 .56 .66
34. ImpulsivenessT2 .05 -.02 -.00 .09 -.03 .02 .07 .05 .03 .05 .24 .10 .17 .68
35. SelfConfidenceT2
.07 .10 .09 .00 .22 .03 .03 .04 .00 -.01 .07 .14 .05 -.24 .84
36. CultureT2 .11 .19 .24 .07 .11 .34 .18 .20 .28 .26 .40 .46 .44 .19 -.03 .79
37. SociabilityT2 .14 .12 .16 .07 .08 .20 .24 .12 .18 .18 .43 .45 .40 .25 -.03 .52 .78
38. LeadershipT2 .14 .14 .20 .08 .15 .16 .14 .24 .15 .15 .45 .40 .47 .25 .10 .44 .41 .76
39. Social
SensitivityT2 .10 .15 .19 .04 .05 .25 .19 .12 .27 .25 .30 .51 .41 .15 -.11 .50 .57 .31 .80
40. TidinessT2 .08 .14 .18 .01 .04 .19 .19 .10 .19 .31 .40 .49 .49 .13 -.08 .47 .40 .28 .43 .80
Note. These analyses used a FIML estimation based on N = 4,510, which was the baseline sample. On the main diagonal are the test-retest reliabilities from the
validation study. On the grey diagonal are the rank-order stabilities of each personality trait across 50 years. All values in Bold font are statistically significant at
p < .001.
PERSONALITY STABILITY OVER 50 YEARS 91
Table 8S (equivalent of Table 6, but with FIML). Descriptive statistics and mean-level changes.
PTPI Scale (Big Five Corresponding Scale)
Time 1 (Baseline) Time 2 (50th
year follow-up)
Standardized
Mean-level
change
Standardized
Mean-level
change with
robustness check
M SD M SD d d-adjusted
Vigor (Extraversion) .53 .30 .56 .40 .12 .15
Calmness (Emotional Stability) .46 .28 .80 .28 1.20 1.20
Mature Personality (Conscientiousness) .46 .21 .85 .23 1.79 1.49
Impulsiveness (Low Conscientiousness) .22 .18 .19 .24 -.15 .08
Self-Confidence (Emotional Stability) .43 .20 .57 .25 .72 1.03
Culture (Openness/Intellect) .51 .23 .56 .33 .18 .56
Sociability (Extraversion) .56 .24 .65 .33 .40 .03
Leadership (Extraversion) .26 .27 .40 .33 .50 .50
Social Sensitivity (Agreeableness) .51 .26 .80 .30 1.14 .94
Tidiness (Conscientiousness) .52 .25 .71 .36 .76 .97
Note. These analyses used a FIML estimation based on N = 4,510, which was the baseline sample. All mean differences in bold font were statistically significant
at p < .001. d = standardized mean-level change between baseline and the 50th
year follow-up. For the sake of comparison with previous research, to calculate d,
we used the same procedure used by Roberts et al., 2006, namely the single-group, pretest-posttest raw score effect size (Morris & DeShon, 2002). Specifically, d
= (Mfollow-up – Mbaseline)/SDbaseline. Bold font indicates effect was statistically significant at p < .001
PERSONALITY STABILITY OVER 50 YEARS 92
Table 9S. Individual-Level Change in Personality Traits from baseline (T1) to the 50th
year follow-up (T2)
PTPI Scale (Big Five Corresponding Scale) Decreased (%) Stayed the same (%) Increased (%) χ2
(2, N = 1,795)
Vigor (Extraversion) 19.2 61.7 19.1 4,183.2
Calmness (Emotional Stability) 3.2 55.7 41.1 10,968
Mature Personality (Conscientiousness) 2 39.3 58.7 23,212
Impulsiveness (Low Conscientiousness) 8.2 81.3 10.5 727.49
Self-Confidence (Emotional Stability) 7.6 60.1 32.3 6,774.4
Culture (Openness/Intellect) 12.6 70 17.5 2,475.9
Sociability (Extraversion) 14.9 58 27.1 5,695.2
Leadership (Extraversion) 11.3 58.6 30.1 6,264.9
Social Sensitivity (Agreeableness) 4.8 53.5 41.7 11,377
Tidiness (Conscientiousness) 8 61.8 30.2 5,910.2
Note. N = 1,795. Percentages for decrease, staying the same, and increase were based on the Reliable Change Index (RCI), where change less than -1.96 or
greater than 1.96 is considered reliable change. The RCIs used for this table were computed using only one standard error of measurement derived from the
correlation between the long-form at Time 1 and the short-form at Time 2 in the validation short-term longitudinal study. The chi-square tested whether the
observed distribution of non-changers and changers differed from the expected distribution if change were random (i.e., 95% stayed the same, 2.5% each
increased and decreased). Bold font indicates χ2
test was statistically significant at p < .001