Content uploaded by Brent W Roberts
Author content
All content in this area was uploaded by Brent W Roberts on Jul 13, 2014
Content may be subject to copyright.
Patterns of Mean-Level Change in Personality Traits Across the Life
Course: A Meta-Analysis of Longitudinal Studies
Brent W. Roberts and Kate E. Walton
University of Illinois at Urbana–Champaign Wolfgang Viechtbauer
University of Maastricht
The present study used meta-analytic techniques (number of samples ⫽92) to determine the patterns of
mean-level change in personality traits across the life course. Results showed that people increase in
measures of social dominance (a facet of extraversion), conscientiousness, and emotional stability,
especially in young adulthood (age 20 to 40). In contrast, people increase on measures of social vitality
(a 2nd facet of extraversion) and openness in adolescence but then decrease in both of these domains in
old age. Agreeableness changed only in old age. Of the 6 trait categories, 4 demonstrated significant
change in middle and old age. Gender and attrition had minimal effects on change, whereas longer studies
and studies based on younger cohorts showed greater change.
Keywords: personality change, meta-analysis, mean-level change, personality development
How do people change in terms of personality traits across the
life course? Despite the common perspective that personality
traits—relatively enduring patterns of thoughts, feelings, and be-
havior—do not change, numerous longitudinal studies have now
reported mean-level changes in traits at various ages across the life
course (Haan, Millsap, & Hartka, 1986; Helson & Moane, 1987;
Helson & Wink, 1992; Roberts, Caspi, & Moffitt, 2001; Robins,
Fraley, Roberts, & Trzesniewski, 2001). It is assumed that mean-
level change, sometimes referred to as absolute change or norma-
tive change (Caspi & Roberts, 1999), reflects generalizable pat-
terns of personality development that apply to most people.
However, it is difficult to draw definitive conclusions regarding
the nature of mean-level change because longitudinal studies track
changes in personality traits in particular samples by using distinct
measures over specific times of the life course, which results in
somewhat idiosyncratic findings that are sometimes contradictory.
For example, Costa and McCrae (1997) argued on the basis of their
longitudinal studies that there is little meaningful mean-level
change in any personality traits past the age of 30 (Costa, Herbst,
McCrae, & Siegler, 2000). In contrast, others have argued that
personality traits continue to change in middle and old age (Hel-
son, Jones, & Kwan, 2002; Helson & Kwan, 2000; Srivastava,
John, Gosling, & Potter, 2003).
One solution to the debate over whether personality traits dem-
onstrate systematic changes in any given period of the life course
is to synthesize the existing longitudinal studies by using meta-
analytic techniques (e.g., Roberts & DelVecchio, 2000). A meta-
analytic approach affords several advantages to any single longi-
tudinal study. First, by examining change across studies, we can
effectively control for the nature of the samples used in existing
longitudinal efforts and weight findings from larger samples more
heavily. Also, we can use the Big Five taxonomy of personality
traits (John & Srivastava, 1999) to organize measures into domains
that are not specific to individual personality inventories. The Big
Five taxonomy organizes most trait terms and personality inven-
tory scales into five broad categories: extraversion, agreeableness,
conscientiousness, emotional stability, and openness to experi-
ence. By using the Big Five taxonomy to organize findings, we can
create a common conceptual framework that makes it possible to
synthesize the results across questionnaires, including those that
were not created with the Big Five in mind. Finally, we can put
together data for most of the life course, a possibility seldom
realized by individual longitudinal studies. Having data from the
majority of the life course allows difficult questions to be ad-
dressed, such as whether people change more during specific time
periods (e.g., adolescence, young adulthood, middle age) and
whether people continue to show normative changes in middle age
and beyond.
In the present study, we have assembled the largest meta-
analytic database of longitudinal studies of mean-level personality
change in existence in order to address the question of how
personality traits change across the life course. We will first review
how change is defined, then the patterns of personality change we
expect given previous research and theory on personality devel-
opment in adulthood, and finally the potential moderators of mean-
level change.
Can Personality Traits Be Both Consistent and
Changeable?
Personality traits are indisputably consistent across time and age
(Fraley & Roberts, 2005; Roberts & DelVecchio, 2000). It is
Brent W. Roberts and Kate Walton, Department of Psychology,
University of Illinois at Urbana–Champaign; Wolfgang Viechtbauer,
Department of Psychology, University of Maastricht, Maastricht, The
Netherlands.
Preparation of this article was supported by grants from the National
Institute on Aging (R03 AG19414 and R01 AG21178) and the Research
Board at the University of Illinois. We thank Tim Bogg, Robert McCrae,
Brent Donnellan, Scott Morris, Sue Duval, and Ravenna Helson for helpful
comments on drafts of this article.
Correspondence concerning this article should be addressed to Brent W.
Roberts, Department of Psychology, University of Illinois, 603 East Daniel
Street, Champaign, IL 61820. E-mail: broberts@cyrus.psych.uiuc.edu
Psychological Bulletin Copyright 2006 by the American Psychological Association
2006, Vol. 132, No. 1, 1–25 0033-2909/06/$12.00 DOI: 10.1037/0033-2909.132.1.1
1
common to believe that if a construct demonstrates temporal
consistency, it does not change. Unfortunately, this conclusion is
often premature and sometimes simply incorrect. This misunder-
standing stems from a semantic problem with the use of such terms
as “stability,” “consistency,” and “change.” These terms are often
used too broadly and interchangeably and, when used as such, fail
to take into account the fact that continuity and change come in
many forms and can also occur simultaneously—they are not the
opposites of one another (Block, 1971).
In most longitudinal studies, consistency is operationalized as
rank-order consistency, which refers to the relative placement of
individuals within a group over time. Change is most often defined
as mean-level change, which refers to whether a group of people
increases or decreases on trait dimensions over time. The existence
of consistency, at least defined in terms of rank-order consistency,
does not preclude the existence of change, especially mean-level
change over time. For example, the numbers x⫽(1, 2, 3, 4) show
perfect rank-order consistency (r⫽1) with the numbers y⫽(2, 4,
6, 8) but yet a clear change in mean level. On the other hand,
pairing the same xwith y⫽(4, 6, 6, 4) yields zero rank-order
consistency (r⫽0) and still the same amount of mean-level
change. Therefore, rank-order consistency and mean-level change
are better thought of as independent or orthogonal constructs.
Conceptually and empirically, both can exist simultaneously
(Block, 1971; Caspi & Roberts, 1999; Funder & Colvin, 1991).
Furthermore, there are now a number of clear empirical cases
where it has been demonstrated that rank-order consistency and
mean-level change coexist in the same longitudinal study (Haan et
al., 1986; Mortimer, Finch, & Kumka, 1982; Roberts et al., 2001;
Roberts, Helson, & Klohnen, 2002; Robins et al., 2001).
Mean-level change in personality traits is often equated with
normative change in personality. Normative change occurs when
most people change in the same way during a specific period
within the life course. Normative changes are thought to result
from maturational or historical processes shared by a population
(e.g., Helson & Moane, 1987; McCrae et al., 2000). These shared
processes could be biological in origin, such as the general period
when adolescence is begun or when menopause occurs in women
(e.g., Helson & Wink, 1992). The timing of these biological
phenomena is partially driven by genetic factors and tends to
happen within a specific period of the life course for most people
in the particular population of interest. It also is possible that
normative changes in personality traits arise because of engage-
ment in normative life tasks and roles, such as leaving home,
establishing a family, and starting a career, for example, which all
happen in the period of young adulthood (Helson, Kwan, John, &
Jones, 2002; Roberts, Wood, & Smith, 2005).
Needless to say, change is a multifaceted construct (Caspi &
Roberts, 1999). In addition to being tracked with rank-order con-
sistency and mean-level change, change also can be tracked in the
structure of trait covariances (e.g., the longitudinal consistency in
the factor structure; Small, Hertzog, Hultsch, & Dixon, 2003) and
in individual differences in change (Mroczek & Spiro, 2003;
Roberts, 1997). The existence of individual differences in change
qualifies the inference that change does or does not occur and that
changes are normative. For example, when there are no mean-level
changes over time, there may still be robust individual differences
in change. Subsets of individuals may be increasing and decreasing
and thus offsetting each other’s change, resulting in no mean-level
change overall. Moreover, the occurrence of normative trends does
not imply that all people change in the same direction. For exam-
ple, in one longitudinal study (Roberts et al., 2001), most individ-
uals decreased in negative emotionality in young adulthood. De-
spite this general trend to decrease, a subset of individuals actually
increased. Nonetheless, most people decreased in negative emo-
tionality, and it is this trend that is captured in the overall mean-
level change patterns that will be examined in this study.
The meta-analysis presented here will focus exclusively on
mean-level change in personality traits over time. To the extent
that these changes are consistent across studies, we can begin to
draw conclusions concerning whether the changes are normative in
that most people demonstrate a distinct pattern of change in
personality traits.
Theories of Personality Trait Development
In his review of personality and aging, Kogan (1990) high-
lighted three theoretical approaches to personality trait develop-
ment. The first model is the classical psychometric theory or trait
model of personality development (see also Conley, 1984). The
exemplar trait theory of personality development in adulthood is
the five-factor theory of personality (McCrae & Costa, 1999).
According to this perspective, traits remain so stable in adulthood
that they are essentially “temperaments” and are impervious to the
influence of the environment. In terms of personality traits, the
five-factor theory clearly states that traits develop through child-
hood and reach maturity in adulthood and are thereafter stable in
“cognitively intact individuals” (McCrae & Costa, 1999, p. 145)
and that this pattern holds across cultures (McCrae & Costa, 1994;
McCrae et al., 2000). Personality trait development is thus pre-
sumed to be governed by temperament or genetic factors rather
than environmental influences or experiences. Therefore, the pat-
terns of mean-level change demonstrated by various samples must
be attributed to genetic factors that define propensities to grow in
specific directions at specific ages during the life course.
The second theoretical approach highlighted by Kogan (1990)
emphasizes the role of the environment. These contextual models
focus on the effect of environmental contingencies often contained
within social roles and how they affect personality (e.g., Brim,
1965). The prototypical contextual approach to understanding per-
sonality development is to focus on more microanalytic social–
cognitive units of analysis (e.g., Bandura, 1999; Zelli & Dodge,
1999). Social–cognitive units of analysis are by definition context-
bound constructs, such as social skills, competencies, and personal
goals. Rather than presupposing consistency, as is done in trait
models, personality consistency is thought to emerge through
one’s transactions with the social environment (Zelli & Dodge,
1999). This model does little to inform perspectives on mean-level
changes in personality traits, as it limits its focus to phenomena
that are presumably not traitlike. Moreover, from an extreme
contextual perspective, mean-level change would result from en-
vironmental contingencies that are mostly unpredictable; therefore,
patterns of mean-level change also would be relatively unpredict-
able (Lewis, 1999).
According to Kogan (1990), the third set of developmental
models emphasizes the transactions between the traits and contexts
across the life course and is therefore interactional. The most
contextual of the interactional models is Levinson’s (1978), which
2ROBERTS, WALTON, AND VIECHTBAUER
focuses on the building of life structures in childhood, early
adulthood, middle adulthood, and late adulthood. Life structures
represent the basic pattern or design of a person’s life and largely
reflect the interplay between self-driven goals and societal and
age-graded roles. As this perspective is most relevant to the pattern
of social roles across the life course, it does not provide informa-
tion pertinent to the development of personality traits. Similarly,
Erikson’s (1950) stage theory emphasizes the change and emer-
gence of specific life tasks and associated crises at different ages.
This perspective also essentially ignores personality trait
development.
More recently, Baltes (1997; Baltes, Lindenberger, &
Staudinger, 1998) has enumerated the life span development ap-
proach, which proposes a dialectic between consistency and
change over the life course, with adaptation being the primary
focus of development. The life span perspective specifies that
people are open systems and that they exhibit both continuity and
change in personality throughout the life course. Furthermore,
according to the life span model, the effects of psychological,
social, and cultural factors diminish as people grow older often as
a result of selection, optimization, and compensation processes
(Baltes et al., 1998).
Recently, Roberts and Caspi (2003) proposed an alternative
theory of personality trait development consistent with Baltes’s
(1997) life span approach (see also Roberts & Wood, in press).
This theory proposes that identity processes can help explain the
patterns of continuity and change in personality traits across the
life course. Specifically, the development of a strong identity and
certain facets of identity structure, such as identity achievement
and certainty, are positively related to many of the mechanisms
that promote personality continuity. Furthermore, with age, a per-
son’s identity becomes clarified and strengthened, and this helps to
explain the increasing continuity in personality traits across the life
course. Finally, making normative commitments to the conven-
tional social institutions necessary to create an identity (e.g., work,
marriage, family, community) gives rise to the increases in traits
associated with psychological maturity, such as agreeableness,
conscientiousness, and emotional stability (see also Roberts,
Caspi, & Moffitt, 2003; Roberts & Wood, in press). These invest-
ments in conventional social institutions are presumed to facilitate
increases in these three core domains of the Big Five.
The theories of adult development range from static to dynamic.
If adult development conforms to the more static form of trait
models, then we should expect few if any systematic mean-level
changes across the life course. A similar conclusion can be drawn
from the contextualist approach, as the driving force behind de-
velopment will be the immediate situation of any one person or
group of persons. Moreover, because our samples are drawn from
dramatically different cohorts ranging from the generations that
experienced the Great Depression to the generations that grew up
in the 1960s, 1970s, and 1980s, one would assume little consis-
tency in the pattern of mean-level change across these studies.
Therefore, from a contextualist perspective, we would not expect
clear patterns of development across the life course because im-
mediate situations and cohort differences are relatively unpredict-
able as are their consequences (Lewis, 1999). The interactional
models all emphasize the continued dynamic nature of the life
course beyond adolescence and therefore imply that development
would occur in adulthood, even in terms of personality traits. One
consistent theme across these interactional models is the critical
nature of young adulthood. This is a period in which identities are
resolved and people make commitments to life paths (Arnett,
2000). Consequently, substantial changes in personality traits are
expected to occur during this critical period. Finally, Baltes (1997)
emphasized the plastic nature of psychological functioning across
the life course, highlighting the possibility for changes to happen
even in middle and old age. We will compare our findings with
these broad generalizations about personality development in
adulthood to see which perspective comes closest to capturing the
empirical patterns of mean-level change across the life course.
Is There a Pattern to Mean-Level Changes in Personality
Traits?
Despite the publication of numerous longitudinal studies in the
last few decades, there was, until recently, a surprising paucity of
statements or claims concerning normative trends in personality
traits over the life course. In part, this lack of a coherent descrip-
tive picture was a result of an emphasis on the more general
question of whether personality changes occur at all during the life
course. Until recently, the dominant perspective was that once
adulthood is reached, which happens around age 30, there is no
subsequent change in personality traits (McCrae & Costa, 1994).
Specifically, on the basis of their interpretation of existing longi-
tudinal studies, McCrae and Costa concluded that there is no
relationship between mean levels of personality traits and age in
cross-sectional studies of personality, nor are there age-related
trends in longitudinal data (Costa & McCrae, 1994, 1997).
A more differentiated perspective has emerged recently from
both cross-sectional studies and two new narrative overviews of
the evidence for personality trait change in adulthood. First, Mc-
Crae et al. (1999) reported cross-sectional mean-level differences
across the Big Five in five different cultures. In contrast to earlier
reports, McCrae et al. reported mean-level differences between
individuals over the age of 30 and younger people in several
cultures for neuroticism, extraversion, openness, agreeableness,
and conscientiousness. Specifically, older individuals scored
higher on agreeableness and conscientiousness and lower on ex-
traversion, neuroticism, and openness (see also Srivastava et al.,
2003).
Helson and Kwan (2000) reviewed data from three cross-
sectional studies and three longitudinal studies that covered the
majority of the life course from age 20 to age 80 and found
evidence for significant mean-level change in select personality
traits across the life course. For example, they reported increases
of a little under half a standard deviation in norm orientation,
which could be construed as a facet of conscientiousness. They
argued that previous research on extraversion has overlooked the
fact that there are two components to extraversion—namely, social
dominance and social vitality—which demonstrate distinctly dif-
ferent maturational patterns. Specifically, social dominance re-
flects such traits as dominance, independence, and self-confidence,
especially in social contexts. On the other hand, social vitality
corresponds more closely to traits like sociability, positive affect,
gregariousness, and energy level. According to Helson and
Kwan’s review, people increase in measures of social dominance
and decrease in measures of social vitality with age. Finally,
Helson and Kwan presented data indicating small increases in
3
PATTERNS OF MEAN-LEVEL CHANGE
openness, with some contradictory evidence across cross-sectional
and longitudinal studies.
Finally, Roberts, Robins, Caspi, and Trzesniewski (2003) re-
viewed a more comprehensive list of cross-sectional and longitu-
dinal studies than previously captured in narrative reviews. Ac-
cording to their interpretation of the data, there is evidence for
increases in social dominance and decreases in social vitality, as
Helson and Kwan (2000) proposed. They also argued that mea-
sures of agreeableness and conscientiousness increase across the
life course from age 18 to over age 60 and that neuroticism tends
to decrease with age, possibly reaching a plateau in old age.
Finally, they found a complex pattern of change for measures of
openness, with some evidence for increases in early young adult-
hood and contradictory evidence for later portions of the life
course.
The primary goal of this meta-analysis was to test whether the
evidence in the longitudinal database is consistent with theoretical
and narrative descriptions of personality trait development in
adulthood as described above. On the basis of a modified Big Five
framework (with extraversion further divided into social vitality
and social dominance), we expected to find that traits from the
domain of social vitality decrease with age while traits from social
dominance increase. We also expected to find traits from the
domains of agreeableness, conscientiousness, and emotional sta-
bility to increase with age, and traits from the domain of openness
to experience to show little or no change or to decrease slightly.
We also attempted to identify when and whether personality traits
stop changing during the life course, at least in terms of mean-level
change. A final goal was to identify the critical periods during the
life course when the largest changes occur.
Moderators of Mean-Level Change in Personality Traits
One distinct advantage of a meta-analytic approach is the op-
portunity to test potential moderators of changes in personality
traits across the life course. We considered four potential moder-
ators. First, we examined the effect of sex on the patterns of
change in personality traits. Although claims for and against sex
differences in personality development have been made in the past
(Guttman, 1987; Helson, Pals, & Solomon, 1997; Roberts et al.,
2001), comparisons have yet to be made across different longitu-
dinal studies. The most concrete hypothesis about sex differences
in adult personality trait development is the crossover hypothesis
(Guttman, 1987). According to this perspective, men and women
change differently in middle age because of shifting demands of
gender-related sex roles. Presumably, men become more nurturing
and emotional with age, especially during the transition from
young adulthood to middle age, because of the shift from the
self-focused drive of occupational success in young adulthood to
the more generative role of family patriarch in midlife. In contrast,
women are expected to become less nurturing and emotional and
more confident and dominant, because of the shifting focus away
from the family to a potential career as children grow older and
leave home (Guttman, 1987; Roberts & Helson, 1997; Wink &
Helson, 1993). Accordingly, we should observe larger increases on
measures of agreeableness for men between the age of 40 and 60,
whereas women should increase more so than men on measures of
social dominance in the same age period. No previous research has
led us to believe that men and women should differ in their
development on other traits at any age period.
A second moderator considered in this meta-analysis is the
effect of time on the patterns and magnitude of change in person-
ality traits. The length of time between assessments has a known
negative effect on rank-order consistency, implying that larger
changes occur as more time passes between assessments. Whether
this effect holds for a different metric of change, such as mean-
level change, has never been examined.
The third moderator we tested—namely, attrition—is one of the
most problematic study features in longitudinal research. Attrition
apparently has little effect on estimates of rank-order consistency
(Roberts & DelVecchio, 2000). Nonetheless, given the typical
findings that individuals who remain in longitudinal studies are
more conscientious (e.g., Robins et al., 2001), we suspect that
attrition may have an effect on the magnitude and types of changes
demonstrated by longitudinal studies, especially in the domain of
conscientiousness.
Finally, we tested the effect of cohort standing, meaning the
year when the sample of any given longitudinal study was born. It
has been shown in several cases that cohort standing is related to
level of psychological functioning on such traits as neuroticism
and extraversion. For example, Twenge (2000, 2001, 2002) has
reported that younger cohorts are more neurotic and extraverted
than older cohorts. Differences in mean levels across cohorts are
thought to reflect the effect of shared values that differ from
decade to decade. So, for example, individuals growing up just
after World War II may have been less narcissistic than individuals
growing up in the 1960s and 1970s because the value system
within the United States in this later period became more self-
focused (Roberts & Helson, 1997). One possibility is that people
living through periods of cultural change will show analogous
psychological change that reflects the changing value system in the
culture. For example, a sample of women followed from their early
20s to their early 40s during the 1960s and 1970s showed a pattern
of increased narcissism and decreased norm adherence, which was
consistent with the changes in values in the social climate of the
United States during that same period of history (Roberts &
Helson, 1997). Therefore, we might expect to see patterns of
change in the longitudinal data across different cohorts that are
similar to the patterns of differences found across cross-sectional
studies of different cohorts. In this case, we would expect longi-
tudinal studies tracking change in younger cohorts to show de-
creases in emotional stability and increases in either social vitality
or social dominance (i.e., extraversion).
Method
Literature Searches
We used six methods to locate studies. First, we reviewed the reference
list from an earlier meta-analysis of rank-order consistency for longitudinal
studies (Roberts & DelVecchio, 2000). Many of these studies reported
information on rank-order consistency and mean-level change, although
Roberts and DelVecchio did not examine the latter in this previous study.
Second, we reviewed additional databases developed by Brent W. Roberts
on personality development (Roberts, Robins, et al., 2003). Third, we
searched the PsychLIT and Dissertation Abstracts databases by using the
following keywords: normative personality change, mean-level personality
change, and longitudinal personality change. Fourth, we reviewed current
4ROBERTS, WALTON, AND VIECHTBAUER
issues of relevant journals (e.g., Journal of Personality and Social Psy-
chology, Journal of Personality, Developmental Psychology). Fifth, after
developing a preliminary list of studies, we reviewed the references cited
in each article for additional studies. Sixth, after developing a relatively
comprehensive list of studies, we asked knowledgeable colleagues to
review the list and alert us to any studies that were overlooked.
Criteria for Study Inclusion
We included studies if they fulfilled five criteria. First, the study had to
include dispositional or trait variables (i.e., enduring, cross-situationally
consistent). Measures of attitudes, values, self-esteem, affect, mood, intel-
ligence, cognitive functioning, sex role (e.g., femininity), temperament,
and validity scales were not included. Second, to emphasize the longitu-
dinal change of traits and to diminish potential carry-over effects that may
affect estimates of change, we included only those studies with test–retest
intervals greater than or equal to 1 year. Third, at a minimum, each study
needed to contain information on mean-level change, sample size, and age
of the sample. Fourth, the sample needed to be nonclinical. Fifth, we
excluded studies that examined change over time for samples heteroge-
neous in terms of age (e.g., a 3-year longitudinal study of individuals
ranging in age from 18 to 80), as these studies precluded the possibility of
examining the effect of age on development.
Ninety-two studies satisfied the inclusion criteria. Because many of
these studies reported data from several samples, the number of samples,
113, was greater than the total number of studies. The total number of
participants for the 113 samples was 50,120. A total of 1,682 estimates of
change were compiled.
Study Variables
Age. We recorded the age at the inception and conclusion of each wave
of assessment in each longitudinal study. Results in a few studies were
reported for a range of ages (e.g., 20 to 30, 30 to 40). For these studies, the
midpoints of the reported age ranges were used as estimates of age.
To test the effect of age on mean-level personality trait change, we
created age categories across the life course. We focused only on studies of
individuals older than age 10, as this is the age at which children begin to
describe themselves with trait terms and are administered standard person-
ality measures (Roberts & DelVecchio, 2000). We created age categories
that corresponded to adolescence (ages 10 to 17.9), the college years (ages
18 to 21.9), and the subsequent decades through age 101.
1
A study was
categorized into an age period by taking the midpoint between the age at
the first assessment and the age at follow-up. For example, if a study first
assessed individuals at age 20 and then again at age 30, the corresponding
“age” for the study was 25 and the data were categorized into the age 22
to 30 period.
Trait categories. A modified version of the Big Five taxonomy of
personality traits used in Roberts and DelVecchio’s (2000) previous meta-
analysis was used to categorize the standardized mean differences of
personality trait change. This system was similar to Roberts and DelVec-
chio’s previous system, with the exception of a division of the domain of
extraversion into two subcategories of social vitality and social dominance.
Social vitality corresponds more closely to traits like sociability, positive
affect, and gregariousness. Representative scales would be the Gregarious-
ness and Activity Scales from the Revised NEO Personality Inventory
(Costa & McCrae, 1994) and the Sociability Scale from the California
Psychological Inventory (Gough & Bradley, 1996). Social dominance
reflects such traits as dominance, independence, and self-confidence, es-
pecially in social contexts. Representative scales would be the Assertive-
ness Scale from the Revised NEO Personality Inventory, the Dominance
and Capacity for Status Scales from the California Psychological Inven-
tory, and the Dominance Scale from the 16 Personality Factor Question-
naire (Conn & Reike, 1994).
Two judges categorized the scales by using the modified Big Five
taxonomy. Interjudge reliability was quite high across the whole database
(
⫽.93). Discrepancies were subsequently discussed until a consensus
was reached on the categorization of all scales into the six personality
domains.
Moderators
Sex of sample. Sex of the sample was coded as follows: 0 ⫽men, 1 ⫽
both men and women, and 2 ⫽women.
Time interval. Time interval was coded in number of years. If the time
interval was reported in months, then the appropriate fraction of a year was
included in the coding.
Attrition. Attrition was computed by subtracting the number of partic-
ipants at the end of each stage of a longitudinal study from the number of
participants at Time 1 and converting this figure to a percentage. Two
judges computed attrition ratings. Interjudge reliability, indexed by the
intraclass correlation, was quite high across the whole database (r⫽.93).
Most discrepancies were small in magnitude, some resulting from rounding
errors. These were subsequently discussed until a consensus was reached
on the level of attrition for each study.
Cohort. Cohort was coded by subtracting the age at the time of the first
assessment from the year that the first assessment was conducted in each
longitudinal study. When the years of assessment were not available, we
used estimates of the year of assessment on the basis of information in the
article, such as year of publication.
Demographic variables. We also attempted to extract information
about ethnicity, education, and socioeconomic status (SES) to see whether
the studies were comparable on these statistics across the life course. The
information provided by researchers on these three demographic variables
was limited and often missing. For ethnicity, we coded studies as “diverse”
if they included ethnic minorities in their sample and “not diverse” if the
sample did not include ethnic minorities. For education, we coded studies
as follows: 1 ⫽most of the sample was high school educated and some of
the sample had college education, 2 ⫽most of the sample had some college
education, and 3 ⫽most of the sample had finished college or had
advanced degrees. For SES, we coded studies as follows: 1 ⫽representa-
tive (included full range of SES), 2 ⫽predominantly middle class, and 3 ⫽
predominantly upper class. There were no studies focusing exclusively on
poor or working-class citizens. The most salient feature of these three
demographic domains was that the information was missing from 40% to
55% of the studies. Therefore, we chose not to examine these variables as
moderators of personality change.
Computation and Analysis of Effect Sizes
The effect size measure used in this meta-analysis was a standardized
mean difference and was calculated by subtracting the mean of the per-
sonality trait scores at Time 2 from the mean at Time 1 and dividing this
raw mean difference by the standard deviation of the raw scores at the first
time point. This is also known in short as the single-group, pretest–posttest
raw score effect size (Morris & DeShon, 2002). When dealing with
repeated measures data as in the present case, one must choose between
two effect size metrics—namely, the change score and raw score metric.
For the change score metric, one divides the observed mean difference by
the standard deviation of the changes scores. The change score metric
incorporates the test–retest correlation into the estimate of the standardized
mean difference. Therefore, difference scores based on data with high
test–retest correlations (i.e., above .50) are increased because of the strong
1
In describing the analyses, we use the convention of describing the
changes as change from “20 to 30,” from “30 to 40,” and so on. In each
case, the actual age ranges were from 20 to 29.9, 30 to 39.9, and so on.
5
PATTERNS OF MEAN-LEVEL CHANGE
correlation over time. Difference scores based on data with low test–retest
correlations (i.e., below .50) are decreased because of the weak correlation
over time (see Morris & Deshon, 2002).
Because the test–retest correlation is a direct function of the rank-order
consistency of the data, using the change score metric provides an effect
size measure that does not differentiate between (or confounds) rank-order
consistency and mean-level change. As we were interested only in the
latter, we used the raw score metric, which is not affected by changes in
rank-order consistency. Also, for the raw score metric, the mean difference
is standardized in the units of the original scale and therefore provides an
effect size measure that is directly comparable to standardized mean
differences obtained from independent samples.
It is important to recognize that the raw score metric also uses the
test–retest information, but only to adjust the standard errors of the effect
size estimates instead of the actual standardized mean difference scores
themselves. Specifically, all else being equal, standardized mean difference
estimates with higher test–retest stabilities have smaller standard errors.
Therefore, mean-level changes with high rank-order consistency (i.e.,
mean-level changes that are more consistent on an individual-level basis)
provide more efficient effect size estimates. Finally, instead of using the
pooled standard deviation of the raw scores at Time 1 and Time 2, one must
use the standard deviation from Time 1 alone, because the repeated
measures structure of the data makes it impossible to estimate the correct
degrees of freedom for the pooled standard deviation (see Morris &
Deshon, 2002).
We used various statistics from the studies to estimate the standardized
mean difference scores. The computation was based on reported means and
standard deviations for 93% of the studies and inferred from Fand tvalues
in 7% of the studies. Because the latter studies used repeated measures F
and ttest results, we transformed these scores into the raw score metric by
using formulas provided by Morris and Deshon (2002).
The structure of the data necessitated several levels of aggregation to
ensure that the effect size estimates were independent. To retain the
developmental trend information across the life course, we aggregated
effect sizes within the age categories rather than collapsing multiple waves
of data across the life course into one effect size. For example, if a
longitudinal study reported results from age 20 to 30 and from age 30 to 40,
then these two estimates were aggregated into the age 22 to 30 and age 30
to 40 periods, respectively. If a longitudinal study reported multiple waves
of data that fell within an age category (e.g., age 22 to 23, and 23 to 24),
then these standardized mean differences and the relevant study moderator
variables were averaged within that age category (e.g., age 22 to 30).
After the aggregation within age period, the next level of aggregation
was within each Big Five category. If a study contributed multiple effect
sizes that were coded as measures of agreeableness, then these effect sizes
were aggregated into one estimate of change in agreeableness. If a study
contributed multiple effect sizes that were coded as measures of agreeable-
ness and openness, then these effect sizes were aggregated into one
estimate of change for agreeableness and one estimate of change for
openness. Finally, we further organized the data around the sex of the
sample studied, aggregating data within the categories of male, female, and
data from combined samples. This technique for aggregation meant that
each longitudinal sample could contribute an averaged standardized mean
difference score for each of the Big Five separately for men and women,
to each of the separate age categories. Therefore, each age category within
each of the Big Five essentially constitutes a separate meta-analysis.
When possible, we tested the likelihood of publication bias by using a
trim and fill procedure (Duval & Tweedie, 2000). Publication bias reflects
the possibility that the studies retrieved for the meta-analysis may not
include all studies actually conducted, with the most likely omissions being
studies that failed to find statistically significant results. The trim and fill
procedure is a nonparametric statistical technique that examines the sym-
metry and distribution of effect sizes plotted against the inverse of the
variance or standard error. This technique first estimates the number of
studies that may be missing as a result of publication bias, with publication
bias meaning studies with effect sizes that are low or near zero relative to
the average effect. Then, the trim and fill procedure calculates hypothetical
effects for potentially omitted studies and then reestimates the average
effect size and confidence intervals on the basis of the influence of studies
that would have been included in the analyses if they had been published.
For effect sizes that were predominantly in the negative direction, we first
reversed the sign of the effects before running the trim and fill analyses
because the procedure assumes that the effects are generally positive. In
several instances, the data from our meta-analysis violated one of the basic
assumptions of the trim and fill procedure. Specifically, if the distribution
includes large positive and negative values, this tends to violate the
distributional assumptions of the procedure and biases the estimates. We
examined the distributions for each effect and did not run the trim and fill
procedure on data that demonstrated this type of pattern. The trim and fill
procedure was performed with the DVBID library (Biggerstaff, 2000) by
using the S-Plus statistical computing program. The procedure as imple-
mented in the DVBID library generates three estimators of missing studies,
L
0
,R
0
, and Q
0
. We used the L
0
estimator as it is the most robust estimator
in the case of problematic distributions described above (S. Duval, personal
communication, November 19, 2003).
To determine whether each set of standardized mean difference scores
shared a common effect size, we calculated the homogeneity statistic Q
within each of the age categories and for each personality factor. When the
effect sizes were found to be homogeneous, we applied the fixed-effects
model. On the other hand, a random-effects model was used for heteroge-
neous effect sizes. In either case, we obtained an estimate of the overall
effect size (denoted by d) for each age category and personality factor,
which was also tested for statistical significance by computing a 95%
confidence interval.
To examine the relationships between the moderator variables and the
effect sizes, we first created data sets for each trait category that included
as many nonoverlapping studies as possible. Testing the moderators within
each age and trait category separately would greatly reduce the power to
detect any moderating effects, because of the smaller number of studies in
each age period. Moreover, the likelihood of making a Type I error
increases quickly when conducting multiple hypothesis tests, such as when
testing the moderators within each age and trait category separately.
Finally, using as many studies as possible in each trait category allows for
more efficient tests of the moderators as we could include all the moderator
variables simultaneously in the model. Thus, we were able to test for the
independent effect of each moderator, controlling for the remaining
moderators.
We proceeded as follows to create data sets with independent effect sizes
for each of the trait categories. When a sample provided results across
multiple assessment waves, we retained the effect size with the largest
sample size. When the sample size did not vary across effect sizes, we
selected effect sizes with the longest time period, as these were rarer than
those with short time periods. If neither sample size nor time period varied,
we selected effect sizes from the oldest age period from the sample, as
these were rarer than other types of change estimates. The resulting six
databases contained between 31 and 102 independent effect sizes, depend-
ing on the personality trait.
To test the moderators, we used a random-effects model with moderators
or, in other words, a mixed-effects model (Overton, 1998; Raudenbush,
1994). Mixed-effect models provide a more stringent test of moderators
and help to diminish the possibility of Type I errors, which can become
severely inflated in the typical fixed-effects with moderators approach to
meta-analysis (Overton, 1998; Viechtbauer, 2004a, 2004b). Mixed-effects
models allow for the possibility of residual heterogeneity, meaning heter-
ogeneity in the effect sizes over and beyond that which we would expect
on the basis of the moderators alone. Therefore, after estimating the
amount of residual heterogeneity, we tested the relationships between the
moderators and the mean-level change in personality measures by fitting a
6ROBERTS, WALTON, AND VIECHTBAUER
weighted least squares regression model within each trait category that
incorporated this estimate of residual heterogeneity. We report parameter
estimates, indicating the change in the effect size for a one-unit increase in
the corresponding moderator value. We also tested whether each moderator
was significantly related to the effect sizes and whether residual heteroge-
neity was present with the Q
E
statistic. The method we used for fitting the
mixed-effects model and analyzing moderators is essentially identical to
that discussed by Raudenbush (1994), except that the amount of hetero-
geneity was estimated via restricted maximum-likelihood estimation (Rau-
denbush & Bryk, 2002). Restricted maximum-likelihood estimates are
approximately unbiased, as opposed to regular maximum-likelihood esti-
mates, which tend to be too small on average. The analysis was conducted
in S-Plus, by using a custom function written by Wolfgang Viechtbauer,
which is available upon request.
Results
Study Characteristics
Table 1 shows the author, sample size, measures, types of traits,
attrition, and ages for each study.
2
Table 2 shows the descriptive
statistics for the relevant variables in the unaggregated database.
The age at the initiation of study ranged from 10 to 101. The
average time span between assessments was 9.5 years. This esti-
mate is somewhat misleading as the median time span was 6 years
and the modal study was 2 years in length. The oldest cohort was
born in 1898 and the youngest was born in 1982, with the average
cohort being born in 1944. The average attrition rate across sam-
ples was 44%. Analysis of the quartiles of the attrition distribution
showed that 25% of the effect sizes came from studies with less
than 27% attrition, whereas another 25% came from studies with
more than 59% attrition. Twenty-six percent of the estimates came
from samples that did not report results separately for men and
women. Thirty-four percent of the estimates were derived from
samples of men and 40% from samples of women. The majority of
the effect sizes came from studies that were heterogeneous in
terms of ethnicity. Also, SES of the samples was evenly distributed
across those that included working-class, middle-class, and afflu-
ent samples. There was a distinct lack of studies of the poor,
however. Most of the samples had some college education, with a
substantial minority being highly educated. The most common trait
category studied was emotional stability (29%), followed by open-
ness (19%), conscientiousness (18%), and social vitality (18%).
The least common trait domains studied were agreeableness (12%)
and social dominance (5%).
To determine the comparability of the samples across the life
course, we broke down the study characteristics and moderator
variables by the age categories that were used to organize the data
(see Table 3). The demographic characteristics revealed several
interesting patterns. Studies of adolescents and college-age stu-
dents appear to be the most representative, in that these samples
included ethnic minorities and tended to have more of a mixed
SES pattern. The data drawn from longitudinal studies that tracked
adults in midlife tended to be more homogeneous in terms of
ethnicity, SES, and college education. The median length of the
longitudinal studies in each age period also showed a differentiated
pattern. Compared with the remaining periods of the life course,
studies in adolescence and college tended to be short-term studies.
The longest studies were derived from samples that first started in
young adulthood. This makes sense as some of the classic longi-
tudinal studies in the field have followed college samples for long
periods of time (e.g., Helson & Wink, 1992). Cohort had a strong
negative relationship with age. This reflects the fact that longitu-
dinal research is a relatively recent phenomenon, at least in terms
of cohort. For example, there has not yet been the opportunity to
study people born in the 1970s and 1980s into middle or old age.
In turn, studies of older cohorts when they were young were not
carried out because longitudinal studies were relatively rare before
the 1960s. Finally, the gender composition and attrition levels
showed the least amount of variability and no clear pattern across
the different age periods.
Mean-Level Changes in Personality Traits Across the Life
Course
We analyzed aggregate change in traits within specific age
categories across the life course. Specifically, each age category
can be thought of as its own meta-analysis, as each study could
contribute one and only one effect size per age category. We will
first present data for each trait across the life course or the portion
of the life course for which there were enough studies to make
estimates.
Table 4 shows the standardized mean-level changes in the facet
of extraversion described as social vitality. On the basis of narra-
tive reviews and the cross-sectional data, we expected these scores
to decrease with age. The standardized mean-level change data
revealed a more complex pattern. Traits that fell within the social
vitality domain showed small but statistically significant increases
in the college period (d⫽.06, p⬍.05). In contrast, social vitality
demonstrated statistically significant decreases in the age 22 to 30
period (d⫽⫺.14, p⬍.05) as well as the age 60 to 70 period (d⫽
⫺.16, p⬍.05), with no systematic change in the intervening
decades of the life course. Overall, the pattern is consistent with
the perspective that there are small decreases in social vitality as
people approach old age. None of the social vitality effects dem-
onstrated potential publication bias.
Table 5 shows the population estimates of mean-level change in
social dominance, the second facet of extraversion. The pattern of
change was very consistent with the argument that social domi-
nance increases with age (Helson & Kwan, 2000). We found
statistically significant increases in social dominance in adoles-
cence (d⫽.20, p⬍.05), the college years (d⫽.41, p⬍.05), and
the two decades of young adulthood (d⫽.28 and .18, respectively,
ps⬍.05). Beyond age 40, changes in social dominance did not
demonstrate sizable increases or decreases. It is interesting to note
that none of the statistically significant increases were heteroge-
neous. The trim and fill analyses identified one estimate that was
potentially biased. The age 18 to 22 effect size had two potential
missing studies. Nonetheless, the estimated effect of including
these potentially missing studies was small, and the reestimated
effects were still statistically significant. Thus, traits from the
domain of social dominance showed a consistent pattern of in-
crease from adolescence through age 40.
Next, we examined standardized mean-level change in measures
of agreeableness (see Table 6). We hypothesized that traits from
2
We also considered method (e.g., self-report, observer ratings, projec-
tive tests) as a moderator, but there were too few studies in the observer and
projective test categories to test the effect.
7
PATTERNS OF MEAN-LEVEL CHANGE
Table 1
Longitudinal Studies of Mean-Level Change in Personality
Authors NMeasure Traits
a
% attrition Ages and time intervals
b
Asendorpf (unpublished data) 93, 81 German CAQ E, A, C, N, O 6 12–17
Asendorpf & Wilpers (1998) 132 NEO-FFI E, A, C, N, O 44 20–22
Baltes & Nesselroade (1972) 1,249 Cattell’s High School Personality Questionnaire E, A, C, N, O 24 14–15
Barefoot et al. (2001) 151 MMPI N 76, 78 50–60, 50–80, 60–80
Berdie (1968) 148 Minnesota Counseling Inventory C, N 41 14–18
Block (1971) 159 California Q-Sort N 83 14–18, 14–30, 17–30
Branje et al. (2004) 285 Goldberg’s 30 adjectives (Dutch version) E, A, C, N, O 1 12–13, 12–14, 15–16
15–17, 16–17, 42–43, 42–44,
43–44, 44–45, 44–46, 45–46
Bronson (1966) 85 Interviewer ratings of personality traits E, A, C, N, O 31 11–14
Cairns et al. (1990) 2,429 Locus of Control N 16 17–18
Cantoni (1955) 211 Bell Adjustment Inventory N 10 14–18, 14–27, 18–27
Caputo et al. (1966) 52 Edwards Personal Preference Schedule E, A, C, N, O 34 18–20
Carmichael & McGue (1994) 121 Eysenck Personality Inventory E, N 63 16–35
Cartwright & Wink (1994) 40 California Psychological Inventory E, C 31 24–31, 24–46, 31–46
Cooper (2004) 1,363 NEO-PI-R C 44 17–22, 17–27, 21–27
Costa et al. (2000) 2,274 NEO-PI-R E, A, C, N, O 26 41–50, 44–50
Costa & McCrae (1988) 398 NEO-PI E, C, N, O 37 35–38, 35–41, 45–48, 45–51,
55–58, 55–61, 65–68, 65–71,
75–78, 75–81
Costa & McCrae (1992) 175 Guilford–Zimmerman Temperament Survey E, A, C, N, O M 53–60
Costa et al. (1980) 433 Guilford–Zimmerman Temperament Survey E, A, C, N, O M 36–42, 45–51, 54–60
64–70, 74–80
Cramer (1998) 88 Narcissistic Personality Inventory A 25 18–22
Cramer (2003) 155 California Q-Sort E, A, C, N, O 64 33–43, 33–57, 43–57
Crook (1943) 52 Thurstone Personality Schedule N 50 18–25
Davis & Franzoi (1991) 205 Interpersonal Reactivity Index A, N, O 74 16–18
Dudek & Hall (1991) 70 California Psychological Inventory E, C, N, O 44 49–74
Field & Millsap (1991) 72 Personality trait ratings E, A, N, O 83 66–80, 75–89
Gold & Henderson (1990) 74 Revised Imaginal Processes Inventory O 76 12–13
Academic Curiosity Scale
Revised Children’s Reactive Curiosity
Grigoriadis & Fekken (1992) 89 MMPI E, C, N 7 30–33
Gustavsson et al. (1997) 130 Karolinska Scales of Personality E, A, C, N 24 43–52
Haan, Millsap, & Hartka (1986) 118 California Q-Sort E, A, C, N, O 74 14–17, 14–57, 17–33, 33–43,
43–57
Hair & Graziano (2003) 147 Goldberg’s bipolar self-rating markers E, A, C, N, O 23 13–17
Hathaway & Monachesi (1963) 3,976 MMPI E, C, N 74 14–18
Helson & Kwan (2000) 40, 106, 45 California Psychological Inventory E, A, C, O 31, 24, 79 24–46, 21–60, 33–75
Helson & Moanc (1987) 81 California Psychological Inventory E, A, C, O 42 21–27, 21–43, 27–43
Helson et al. (1995) 104 California Psychological Inventory O 26 21–43, 21–52, 43–52
Helson & Wink (1992) 101 California Psychological Inventory E, C, N 28 43–52
Adjective Check List 31
Holmlund (1991) 349 Cesarec–Marke Personality Schedule E, A, C, N 31 15–25
Karney & Bradbury (1997) 106 Eysenck Personality Questionnaire N 12 24–28, 25–29
Karney & Coombs (2000)
c
91 California Psychological Inventory E, A, N 48 26–36, 26–46, 36–46
Keltikangas-Ja¨rvinen (1989) 1,737 AFMS questionnaire A 17–46 12–15, 15–18, 18–21, 21–24
Kitchener et al. (1984) 61 Sentence Completion Task O 15, 25, 30 16–18, 20–22, 28–30
8ROBERTS, WALTON, AND VIECHTBAUER
Table 1 (continued)
Authors NMeasure Traits
a
% attrition Ages and time intervals
b
Klopsch (1983) 150 MMPI E, C, N 39 25–30
Labouvie-Vief & Jain (2002) 300 California Psychological Inventory E, C, N 17, 24, 38 23–25, 23–29, 25–29, 39–41,
39–45, 41–45, 53–55, 53–59,
55–59, 72–74, 72–78, 74–78
Leadbeater et al. (1999) 460 Depressive Experiences Questionnaire for
Adolescents
N 8 12–13
Youth Self-Report A, C, N
Leon et al. (1979) 71 MMPI E, N 75 49–79
Loehlin et al. (1990) 83 Parent ratings of personality E, C, N 50 10–20
Loevinger et al. (1985) 298 Sentence Completion Task O 42–53, M 18–19, 18–20, 18–21, 19–21
J. Martin & Redmore (1978) 32 Sentence Completion Task O 70 12–18
P. Martin (2002) 179 16-PF E, A, C, N, O 45 65–70, 83–88, 101–103
McGue et al. (1993) 254 Multidimensional Personality Questionnaire E, A, C, N, O M 20–30
Mclamed et al. (1974) 62 16-PF E, A, C, N, O 35 36–39
Morizot & LeBlanc (2003) 145, 277 Jesness Personality Inventory E, A, C, N 74, 41 15–17, 15–30, 15–40
Eysenck Personality Inventory 17–30, 17–40, 30–40
Mortimer et al. (1982) 368 Semantic Differential Scale E, N, C 43 18–22, 18–32, 22–32
Mroczek et al. (2002) 205 Goldberg 100 E, A, C, N, O 29 13–14
Mroczek & Spiro (2003) 521, 742, 409 Eysenck Personality Inventory E, N 27 42–58, 59–68, 69–96
Mussen et al. (1980) 53 Observer trait ratings E, A, N, O 58 30–70
Nesselroade & Baltes (1974) 816 Cattell’s High School Personality Questionnaire E, A, C, N, O 38 14–16
Personality Research Form
Neyer & Asendorpf (2001) 489 NEO-FFI E, A, C, N 23 24–28
Nichols (1967) 636 16-PF E, A, C, N, O 46 18–22
Ogawa et al. (1997) 48 Youth Self-Report of Behavior Checklist N 37 17–20
Ormel & Rijsdijk (2000) 383 Amsterdamse Biografische Vragenlijst N 23, 33, 42 35–40, 35–41, 35–49, 35–51
40–41, 40–49, 40–51, 41–49
41–51, 49–51
Ormel & Schaufeli (1991) 226, 389 Rotter’s Locus of Control N 41, 39 39–46, 23–24
Pederson (1991) 553 General Health Questionnaire N 7 17–19
Zuckerman Sensation Seeking Scale (Form V) E, C, O
Peterson & Lane (2001) 69 Altemeyer’s Right-Wing Authoritarianism O 65 18–22
Piccione et al. (1989) 50 Stanford Hypnosis Susceptibility Scale O 92 20–30, 20–45, 30–45
Plant (1965) 1,177, 974 Modified California Ethnocentric Scale O 50, 59 18–20, 18–22
Gough Revision of the F Scale
Rokeach Dogmatism Scale
Plant & Telford (1966) 1,713
d
California Psychological Inventory E, C, O 62 18–20
Pogue-Geile & Rose (1985) 266 MMPI A, C, N 34 20–25
Popham & Holden (1991) 55 MMPI E, A, N, O 31 20–21
Roberts (1997) 81 California Psychological Inventory E 42 21–27, 21–43, 27–43
Roberts et al. (2001) 980 Multidimensional Personality Questionnaire E, A, C, N 5 18–26
Roberts & Chapman (2000) 77 California Psychological Inventory N 45 21–27, 21–43, 21–52
27–43, 27–52, 43–52
Roberts et al. (2002) 78 California Psychological Inventory E, A, C, O 44 21–27, 21–43, 21–52
27–43, 27–52, 43–52
Robins et al. (2001) 270 NEO-FFI E, A, C, N, O 45 18–22
Sanderman & Ranchor (1994) 225 Eysenck Personality Questionnaire E, A, N 74 42–48
Sanford (1962) 108, 118 Vassar Scale E, C, O M 18–19, 18–20
California Psychological Inventory
MMPI
Personality trait ratings
9
PATTERNS OF MEAN-LEVEL CHANGE
Table 1 (continued)
Authors NMeasure Traits
a
% attrition Ages and time intervals
b
Schofield (1953) 83 MMPI C, N 0 22–24
Scollon (2004) 400 Eysenck Personality Inventory E, N 57 24–32
Small et al. (2003) 223 NEO-PI E, A, C, N, O 54 68–74
Stein et al. (1986) 654 Bentler Psychological Inventory E, A, C, N, O 60 13–21
Stevens & Truss (1985) 92, 85 Edwards Personal Preference Schedule E, A, C, N, O 91, 93 18–30, 20–42, 18–42
Tenerowicz (1992) 62 Personality Research Form E, A, C, N, O 22 27–29
Townsend et al. (1989) 112 Zung Self-Rated Depression Scale N 38 47–48
Vaidya et al. (2002) 392 Big Five Inventory E, A, C, N, O 48 18–21
PANAS E, N
van der Velde et al. (1995) 314 Depression Adjective Checklist E, C, N 29 18–22, 20–24
Eysenck Personality Inventory
Zuckerman’s Sensation Seeking Scale
Feij’s Impulsiveness Scale
Heesink’s Locus of Control
Viken et al. (1994) 14,932
d
Eysenck Personality Inventory E, N 5–14 20–26, 26–32, 32–38
38–44, 44–50, 50–56
Watson & Slack (1993) 82 Multidimensional Personality Questionnaire E, N 46 45–47
Watson & Walker (1996) 237 PANAS E, N 78 18–25
Weinryb et al. (1992) 37 Karolinska Scales of Personality E, A, C, N 43 39–41
Westenberg & Gjerde (1999) 97 Sentence Completion Task O 7 14–23
Wilhelm & Parker (1990) 163 Costello–Comrey Trait Depression Scale N 4 23–33
Wink & Helson (1993) 42, 21 Adjective Checklist E, A, C, N, O 68 27–52, 31–56
Woodall & Matthews (1993) 108 Type A Structured Interview A 29 14–18
MMPI
Woodruff & Birren (1972) 54 California Test of Personality N 89 20–45
Note. Multiple numbers in the Ncolumn indicate the sample sizes for multiple sample sizes reported within the same study. Multiple numbers in the attrition column correspond to the different samples
reported in each study. Ranges indicate that more than three samples were reported on in the study. CAQ ⫽California Adult Q-Sort; E ⫽Extraversion, including measures of Social Dominance and
Social Vitality; A ⫽Agreeableness; C ⫽Conscientiousness; N ⫽Neuroticsm or Emotional Stability; O ⫽Openness to Experience; NEO-FFI ⫽NEO Five-Factor Inventory; MMPI ⫽Minnesota
Multiphasic Personality Inventory; NEO-PI-R ⫽Revised NEO Personality Inventory; NEO-PI ⫽NEO Personality Inventory; M ⫽missing data; 16-PF ⫽16 Personality Factor Questionnaire;
PANAS ⫽Positive and Negative Affect Scale for Children.
a
The Big Five categories included in the meta-analysis do not always correspond perfectly to the Big Five categories included in the measure, such as the NEO-PI-R. The discrepancy resulted from
a variety of factors, such as the authors not reporting the statistics for all of the scales contained in the personality inventory or because of problems with the way scales were measured over time (e.g.,
the items were changed as scales were developed).
b
To determine which age categories each study contributed to, identify the age at the midpoint between the ages listed under the Ages and time intervals column. This age was used to categorize studies
into the various age categories (22–30, 40–50, etc.).
c
Data were unreported means on personality measures from the samples described in the respective studies that we acquired from the authors of the original studies.
d
These studies reported on five or more samples, so the aggregate Nis provided. Also, we used the overall attrition as attrition for each subsample was not reported.
10 ROBERTS, WALTON, AND VIECHTBAUER
the domain of agreeableness would increase with age. This hy-
pothesis was not strongly supported. Up through the age 50 to 60
period the changes were positive, but all the estimates included
zero in their confidence intervals, indicating that we cannot rule
out the null hypothesis. We should also note that all the effect sizes
through the age 50 to 60 period were heterogeneous and thus were
evaluated with the random effects model. We did find a statisti-
cally significant increase in agreeableness in the age 50 to 60
period (d⫽.30, p⬍.05). None of the estimates were biased
according to the trim and fill procedure.
Table 7 shows the pattern of standardized mean-level changes in
measures of conscientiousness. We hypothesized that traits from
the domain of conscientiousness would also increase with age and
found strong support for this hypothesis. We found little or no
change in measures of conscientiousness in adolescence and the
college period. In contrast, conscientiousness increased from age
20 to 30 (d⫽.22, p⬍.05), from age 30 to 40 (d⫽.26, p⬍.05),
and from age 40 to 50 (d⫽.10, p⬍.05). In addition, conscien-
tiousness increased during the age 60 to 70 period (d⫽.22, p⬍
.05). Three of the effects, from adolescence, the college period,
and middle age (40 to 50), were heterogeneous. None of the
statistically significant effects demonstrated evidence of publica-
tion bias.
The patterns of standardized mean-level change for measures of
emotional stability were similar to those of conscientiousness but
centered earlier in the life course (see Table 8). People showed
small, positive increases in emotional stability in their teens (d⫽
.16, p⬍.05), the college period (d⫽.12, p⬍.05), as well as in
their 20s (d⫽.23, p⬍.05) and their 30s (d⫽.26, p⬍.05). We
also found a very small but statistically significant increase in the
age 50 to 60 period (d⫽.06, p⬍.05), although the latter effect
demonstrated publication bias and should be interpreted
cautiously.
Table 4
Population Estimates of Mean-Level Change in Social Vitality
Facet of Extraversion Across the Life Course
Age KN d CI Q
h
10–18 12 6,132 .11 ⫺.06, .27 32.3*
18–22 15 3,929 .06* .01, .10 20.0
22–30 21 7,306 ⫺.16* ⫺.27, ⫺.05 56.6*
30–40 14 2,310 .04 ⫺.07, .15 58.8*
40–50 20 4,948 ⫺.02 ⫺.06, .02 38.7*
50–60
a
9 1,266 .01 ⫺.04, .04 7.5
60–70 11 1,406 ⫺.16* ⫺.27, ⫺.05 26.3*
70⫹10 1,045 ⫺.01 ⫺.05, .04 8.3
Note. K ⫽number of samples; d⫽standardized mean difference; CI ⫽
95% fixed effects confidence intervals; Q
h
⫽heterogeneity statistic.
a
Trim and fill procedure was not carried out on these data because the
distribution of effects violated assumptions of the procedure.
*p⬍.05.
Table 2
Descriptive Statistics for Characteristics Associated With Trait
Consistency
Study Characteristics MSD Range
Time interval in years 9.0 (8.7) 1 to 43 years
Age range of samples 10 to 101
Cohort 1942 (18.2) 1898 to 1982
Attrition 42% (22) 0% to 93%
Male samples 34%
Female samples 40%
Both 26%
Ethnicity
Homogeneous 30%
Heterogeneous 70%
Socioeconomic status
Includes working class 36%
Predominantly middle class 32%
Predominantly affluent 32%
Education
Mostly high school educated with
some with college education
33%
Most have some college education 54%
Most have completed college or have
advanced degrees
12%
Trait category
Social vitality 18%
Social dominance 5%
Agreeableness 12%
Conscientiousness 18%
Emotional stability 29%
Openness to experience 19%
Note. Descriptive statistics are based on overall database before aggregation.
Table 3
Characteristics of the Samples Across the Life Course
Age
Not
monoethnic
Mixed
SES
Most have
some college Time Cohort
Gender
AttritionMen Women
10–18 96% 79% 3 1960 31% 27% 35.6
18–22 100% 73% 2 1955 23% 42% 40.9
22–30 100% 37% 54% 10 1951 42% 35% 48.6
30–40 54% 38% 71% 10 1939 41% 47% 49.2
40–50 51% 39% 72% 7.6 1941 37% 39% 35.6
50–60 42% 42% 57% 7.3 1926 39% 44% 47.1
60–70 28% 6% 94% 7.3 1916 58% 22% 46.3
70⫹50% 27% 61% 6 1914 26% 21% 48.2
Note. Descriptive statistics are based on aggregated database. Time estimates are medians rather than means because of skewed distributions. Gender
percentages do not add up to 100% because the remaining samples were combined samples. SES ⫽socioeconomic status.
11
PATTERNS OF MEAN-LEVEL CHANGE
Traits from the domain of openness to experience showed a
similar pattern to social vitality (see Table 9). People showed
increases in openness in adolescence (though not statistically sig-
nificant) and statistically significant increases in the college years
(d⫽.37, p⬍.05). In the ensuing decades, openness either did not
change or declined. Like social vitality, the standardized mean-
level change in openness dropped in old age (d⫽⫺.19, p⬍.05).
Also, changes in openness to experience showed significant het-
erogeneity mostly in adolescence and young adulthood. None of
the statistically significant effects for the domain of openness to
experience demonstrated evidence of publication bias.
Three aspects of the results stand out. First, it is clear that
personality trait change occurs well past ages previously used to
demarcate when traits purportedly stop changing, such as age 30
(McCrae & Costa, 1994). All six trait domains demonstrated
statistically significant changes past the age of 30, and four of the
six trait domains demonstrated statistically significant changes in
middle or old age. Second, as is shown in Figure 1, when the
aggregate amount of change across the six trait categories is
plotted against age, we see that personality traits change most
during the period of young adulthood (age 20 to 40), rather than in
adolescence. This finding indicates that the window for personality
development is open well into adulthood. Finally, as can be seen in
the panels of Figure 2, most of the change was in the positive
direction. Each of these panels shows the cumulative dvalues
across the life course for the six trait domains. The assumption
underlying these figures is that the change in personality traits is
cumulative across the life course, which is somewhat risky as none
of the known longitudinal studies have tracked individuals from
birth to death. Thus, we would propose that the change shown here
is most likely an upper bound estimate of personality trait change.
The patterns and magnitude of change are easily discernable in
these figures. The changes in social vitality were small and in the
downward direction. The increases in social dominance were ro-
bust in young adulthood and cumulate to over one standard devi-
ation. Similar patterns are shown for conscientiousness and emo-
tional stability. Despite demonstrating few significant increases in
any given age period, agreeableness shows a clear trend toward
increasing across the life course. Finally, openness to experience
shows a clear curvilinear pattern of change.
Moderators of Mean-Level Change in Personality
We investigated four potential moderators of mean-level change
in personality traits: sex, time between assessments, attrition, and
cohort standing (see Table 10). To control for the effect of age on
the effect sizes, we also entered this variable in the model. Ac-
Table 5
Population Estimates of Mean-Level Change in Social Dominance Facet of Extraversion Across
the Life Course
Age KN d CI Q
h
d
a
CI
a
Q
ha
10–18 5 1,700 .20* .01, .39 3.9
18–22 9 1,655 .41* .13, .69 3.7 .32* (2)
a
.07, .58 6.0
22–30 14 2,445 .28* .12, .44 9.2
30–40 8 807 .18* .004, .35 2.4
40–50 7 2,009 .02 ⫺.02, .07 8.6
50⫹7 418 .01 ⫺.16, .15 22.9*
Note. K ⫽number of samples; d⫽standardized mean difference; CI ⫽95% fixed effects confidence intervals;
Q
h
⫽heterogeneity statistic; d
a
⫽fixed effects standardized mean difference adjusted for potential publication
bias; CI
a
⫽95% fixed effects confidence interval for the standardized mean difference adjusted for potential
publication bias; Q
ha
⫽heterogeneity statistic for the mean difference adjusted for potential publication bias.
a
Represents the number of studies presumed to be missing from the distribution of scores according to the trim
and fill procedure.
*p⬍.05.
Table 6
Population Estimates of Mean-Level Change in Agreeableness
Across the Life Course
Age KN d CI Q
h
10–18 20 4,378 .01 ⫺.09, .11 61.3*
18–22 11 2,239 .05 ⫺.08, .18 39.4*
22–30 17 2,962 .17 ⫺.02, .34 47.6*
30–40 11 1,046 .06 ⫺.10, .22 24.6*
40–50 14 3,079 .03 ⫺.05, .11 55.7*
50–60 5 281 .30* .003, .61 4.5
60⫹7 551 .04 ⫺.04, .13 12.0
Note. K ⫽number of samples; d⫽standardized mean difference; CI ⫽
95% fixed effects confidence intervals; Q
h
⫽heterogeneity statistic.
*p⬍.05.
Table 7
Population Estimates of Mean-Level Change in
Conscientiousness Across the Life Course
Age KN d CI Q
h
10–18 17 7,506 .03 ⫺.09, .14 170.2*
18–22 18 5,226 .04 ⫺.18, .11 89.6*
22–30 22 4,827 .22* .11, .32 28.6
30–40 12 1,079 .26* .09, .43 18.7
40–50 13 2,838 .10* .01, .19 66.8*
50–60 5 240 .06 ⫺.06, .19 6.5
60–70 7 434 .22* .01, .43 9.8
70⫹7 444 .03 ⫺.04, .11 7.2
Note. K ⫽number of samples; d⫽standardized mean difference; CI ⫽
95% fixed effects confidence intervals; Q
h
⫽heterogeneity statistic.
*p⬍.05.
12 ROBERTS, WALTON, AND VIECHTBAUER
cording to the crossover hypothesis, we expected to find evidence
for gender differences in changes in social dominance and agree-
ableness. However, given the complete lack of statistically signif-
icant relationships between gender and standardized mean-level
change, we found no support for this hypothesis.
The second moderator, time span of the longitudinal study, has
a well-known positive relationship to a lack of rank-order consis-
tency, which led us to believe that longer longitudinal studies may
report larger mean-level change. Consistent with this hypothesis,
length of time between assessments was positively related to
mean-level change in agreeableness and conscientiousness. Figure
3 shows the bivariate plot of standardized mean-level change in
conscientiousness by time span of the longitudinal study. A very
clear linear relationship can be seen in which all of the effect sizes
become positive and increase in magnitude as longitudinal inves-
tigations exceeded 10 years in length. In contrast, time had a
negative relationship with mean-level change in social vitality. On
further examination of this effect, it became apparent that the
predominantly negative changes in social vitality complicated the
interpretation of these effects. Rather than interpreting this as
showing less change for longer studies, these effects should be
interpreted as showing that time and attrition were associated with
larger decreases in social vitality, which was the most common
pattern of change on this trait.
On the basis of cross-sectional cohort differences, we expected
to find increases in social vitality or social dominance and de-
creases in emotional stability in younger cohorts (Twenge, 2000,
2001, 2002). We found no statistically significant relationship
between cohort standing and either social vitality or emotional
stability. In contrast, and consistent with Twenge’s findings, we
did find a positive relationship between cohort standing and
changes in social dominance. This indicates that younger cohorts
were showing larger increases in social dominance than older
cohorts. Figure 4, which shows the relationship between standard-
ized mean-level changes in social dominance and cohort, demon-
strates a clear linear relationship in which cohorts born in 1940 and
after demonstrated increases in social dominance. Unexpectedly,
we found cohort differences for agreeableness and conscientious-
ness. In both cases, bivariate plots revealed a more curvilinear
relationship between change and cohort, such that cohorts born
before 1930 and after 1960 showed increases on both dimensions.
We confirmed this finding by testing the quadratic relationship
between cohort and the effect sizes and found it to be statistically
significant for both personality traits.
In examining the effect of cohort, it became clear that the effect
was indistinguishable from age. That is, in the case of social
dominance, most of the change occurred in young adulthood, and
these were also the cohorts from the last half of the 20th century.
Similarly, for conscientiousness, there were increases in young
adulthood and old age, which also happened to correspond quite
strongly with the cohorts from after 1960 and before 1930. In fact,
the average linear correlation between age and cohort was .69
across all six trait domains. Even though we controlled for age in
the estimate of the effect of cohort, the almost perfect overlap
renders the interpretation difficult if not impossible to make.
Therefore, we tested the effect of cohort within the age periods that
demonstrated the largest change (college age, 20 to 30, and 30 to
40) and provided the largest number of change estimates. This also
provides a more direct replication of Twenge’s (2001) cross-
sectional research, as it focused almost exclusively on adolescents
and young adults. In the case of social dominance, cohort remained
Table 8
Population Estimates of Mean-Level Change in Emotional Stability Across the Life Course
Age KN d CI Q
h
d
a
CI
a
Q
ha
10–18 23 10,557 .16* .09, .23 32.6
18–22
a
15 3,621 .12* .004, .24 38.5*
22–30 31 10,480 .23* .14, .32 26.9
30–40 20 4,025 .26* .05, .47 84.9*
40–50 27 7,153 .06 ⫺.03, .15 158.9*
50–60 12 2,002 .06* .004, .12 9.0 .03 (2)
b
⫺.02, .08 15.4
60–70 13 1,560 .01 ⫺.15, .17 68.3*
70⫹10 1,227 ⫺.05 ⫺.22, .12 66.2*
Note. K ⫽number of samples; d⫽standardized mean difference; CI ⫽95% fixed effects confidence intervals; Q
h
⫽heterogeneity statistic; d
a
⫽fixed
effects standardized mean difference adjusted for potential publication bias; CI
a
⫽95% fixed effects confidence interval for the standardized mean
difference adjusted for potential publication bias; Q
ha
⫽heterogeneity statistic for the mean difference adjusted for potential publication bias.
a
Trim and fill procedure was not carried out on these data because the distribution of effects violated assumptions of the procedure.
b
Represents the number of studies presumed to be missing from the distribution of scores according to the trim and fill procedure.
*p⬍.05.
Table 9
Population Estimates of Mean-Level Change in Openness to
Experience Across the Life Course
Age KN d CI Q
h
10–18 13 2,911 .23 .00, .48 37.2*
18–22 37 3,998 .37* .18, .56 76.6*
22–30 12 702 ⫺.01 ⫺.14, .12 27.8*
30–40 11 541 .07 ⫺.06, .21 6.8
40–50 12 2,660 ⫺.01 ⫺.07, .04 10.6
50–60 6 317 .11 ⫺.16, .39 21.1*
60–70 8 507 ⫺.19* ⫺.37, ⫺.02 11.3
70⫹8 458 ⫺.08 ⫺.16, .01 5.1
Note. K ⫽number of samples; d⫽standardized mean difference; CI ⫽
95% fixed effects confidence intervals; Q
h
⫽heterogeneity statistic.
*p⬍.05.
13
PATTERNS OF MEAN-LEVEL CHANGE
significantly related to standardized mean-level change when the
analyses were limited to the college age, 20 to 30, and 30 to 40 age
periods (number of samples [K]⫽18,

⫽.12, p⬍.05). The
effect of cohort on change in conscientiousness was still statisti-
cally significant when the analyses were restricted to the period of
late adolescence and young adulthood (K⫽30,

⫽.09, p⬍.05),
as was the effect for agreeableness (K⫽27,

⫽.11, p⬍.01).
Thus, we can say with confidence that younger cohorts increased
more in terms of social dominance, agreeableness, and conscien-
tiousness than did older cohorts. Moreover, we did not find the
expected relationships with changes in social vitality or emotional
stability.
Finally, attrition had no relationship with standardized mean-
level changes in any domain of personality. The moderators failed
to account for all of the residual heterogeneity in the effect sizes,
with the exception of social dominance.
Discussion
This study demonstrates that personality traits show a clear
pattern of normative change across the life course. People become
more socially dominant, conscientious, and emotionally stable
mostly in young adulthood, but in several cases also in middle and
old age. We found that individuals demonstrated gains in social
vitality and openness to experience early in life and then decreases
in these two trait domains in old age. These findings are directly
relevant to the theoretical models that guide research on person-
ality trait development. Many trait models emphasize stability and
go so far as to hypothesize that specific ages, such as age 30,
constitute significant developmental markers that provide adequate
information to infer that personality traits stop changing. This
meta-analysis clearly contradicts the notion that there is a specific
age at which personality traits stop changing, as we found evidence
for change in middle and old age for four of the six trait categories
studied. In general, the findings were most consistent with inter-
actional models of personality development (Baltes, 1997; Roberts
& Caspi, 2003).
An argument could be made that the changes were all small in
magnitude, and this would be a reasonable conclusion. But this
argument merits further consideration. First, what is a small effect
size in terms of changing personality traits? Traits are presumed to
be some of the least changeable factors studied in personality
psychology, if not psychology in general (Conley, 1984). Given
their known effect on such significant life outcomes as occupa-
tional success, longevity, and health (Bogg & Roberts, 2004;
Friedman, 2000; Judge, Higgins, Thoreson, & Barrick, 1999; Rob-
erts & Bogg, 2004), any change in these attributes may reap
unknown benefits and pitfalls for those individuals who do change.
It is quite possible that even small changes in personality traits
may have profound effects on successful development across the
life course. A second reason to reconsider the magnitude of the
effects is that they were constrained to specific periods of the life
course and did not reflect the potential accumulation of change
over the life course. If we assume that the changes are independent
across different age periods, we can sum the absolute values of
change in order to gain an upper bound estimate of change across
the life course (see Figure 2). When we do this, the lower bound
amount of change over the life course is over one half of a standard
deviation (social vitality), and the upper bound is over one full
standard deviation (social dominance), with most domains show-
ing changes close to one standard deviation across the life course.
Whole populations changing one full standard deviation consti-
tutes what is typically considered a large effect in psychology
(Cohen, 1992). To change this much in terms of personality traits
over the entire life course is clearly more than a trivial amount of
change.
One of the most noteworthy findings was that personality traits
changed more often in young adulthood than any other period of
the life course, including adolescence. Stereotypically, personality
development is thought to be a phenomenon of childhood and
adolescence. Moreover, personality is thought to stop developing
once adulthood is reached, with chronological age markers for
adulthood ranging from 18 to 30 (Caspi & Roberts, 1999; Roberts
Figure 1. Aggregate mean-level changes in personality traits across the life course.
14 ROBERTS, WALTON, AND VIECHTBAUER
& DelVecchio, 2000). The picture derived from theory and prac-
tice in developmental psychology is that the crucible of personality
development is centered early in life with subtle refinements made
as people approach adulthood. Our data directly contradict this
implicit developmental worldview. Moreover, adolescence appar-
ently is not the crucial period during which personality is matured.
Rather, young adulthood, the period of life in which people tran-
sition from their family of origin to their family of destination,
from compulsory education to a career and to being active mem-
bers of their community, is the time during which we see the most
personality trait change and a uniformly positive pattern of change
at that.
Moderators of Mean-Level Change in Personality Traits
We considered the potential effect of several moderators of
personality change in adulthood. We found less than definitive
support for the crossover hypothesis that women should increase in
terms of social dominance and men in agreeableness during middle
age. Gender had no relationship to changes in social dominance
and agreeableness. We also found no gender differences in esti-
mates of standardized mean change in the domains of conscien-
tiousness, emotional stability, and openness to experience. So in
general, we conclude that there is very little support for the idea
that men and women change in distinct ways or that they change
Figure 2. Cumulative dscores for each trait domain across the life course.
15
PATTERNS OF MEAN-LEVEL CHANGE
in ways that are related to their sex roles across the life course
(Guttman, 1987).
Several factors may have contributed to the patterns contradict-
ing Guttman’s (1987) crossover hypothesis and the fact that there
were little or no systematic differences between men and women
in terms of mean-level personality change. For example, the ag-
gregation of the personality measures into the broad categories of
the Big Five may have effectively washed out possible gender
differences, as the measures specific to Guttman’s ideas may have
been merged with too many other dispositions. Nonetheless, given
Guttman’s position, we expected women to increase faster on
measures of social dominance in middle age and men to increase
faster on measures of agreeableness in middle age, and we did not
find these patterns. Overall, the inconsistent patterns of sex differ-
ences in personality trait development would seem to indicate that
the effect of sex on personality development is neither large nor
widespread across the life course.
Previous research has never addressed whether time has a no-
ticeable effect on patterns of mean-level change. We know from
studies of rank-order consistency that the longer the time between
assessments, the lower the levels of rank-order consistency. What
we did not know was whether longer time periods would also
result in larger mean-level changes. On the basis of the findings of
this study, there appears to be a consistent effect for longitudinal
studies that track change over longer time periods to show larger
mean-level changes. We found that time was associated with larger
increases for agreeableness and conscientiousness and with larger
decreases for social vitality. This may have resulted from the
largely linear change in these trait domains, especially in young
adulthood and midlife. With a longer longitudinal study, it was
more likely that changes would cumulate in one direction and thus
be captured more readily in longer longitudinal studies. These
findings imply that longer longitudinal studies will provide larger
estimates of change, a result that needs to be taken into consider-
ation when planning new studies and evaluating the results of
existing longitudinal research.
We also tested for cohort effects on personality change. Previ-
ous research has shown that older cohorts tended to score lower on
measures of extraversion and neuroticism than did younger co-
horts. The developmental inference is complicated, but it is pre-
sumed that the differences reflected the effects of changing values
and social context within a given culture (Roberts & Helson, 1997;
Twenge, 2001). These changing values affect child-rearing prac-
tices and cohort experiences resulting in shifts in mean levels of
certain traits across cohorts.
The interesting developmental question is whether these
changes in culture are robust enough to change individuals living
through those shifts in cultural values. For example, Roberts and
Figure 3. Bivariate plot of standardized mean-level change in conscientiousness and time.
Table 10
Moderators of Personality Trait Development Across the Life Course
Trait KGender Time Cohort Attrition Q
E
Social vitality 76 .0308 ⫺.0135** ⫺.0066 .0000 198.7**
Social dominance 31 .0470 ⫺.0027 .1183** .0043 22.1
Agreeableness 62 ⫺.0736 .0204** .0676** .0002 127.3**
Conscientiousness 65 ⫺.0147 .0183** .0420** .0028 169.5**
Emotional stability 102 ⫺.0025 .0037 .0164 .0010 174.4**
Openness to experience 66 .0506 .0070 .0527 ⫺.0021 371.7**
Note. Coefficients are unstandardized beta weights in the metric of the standardized mean-level difference
scores. K⫽number of samples; Q
E
⫽test for residual heterogeneity after accounting the effect for moderators.
** p⬍.01.
16 ROBERTS, WALTON, AND VIECHTBAUER
Helson (1997) showed that women living through the 1960s and
1970s became more individualistic, just as separate cohorts of
students did during that same period of history. In the present
study, we found that younger cohorts had larger standardized
mean-level changes in terms of social dominance. The changes in
social dominance were consistent with the cross-sectional patterns
identified in previous research that indicate that younger cohorts
are more assertive (e.g., Twenge, 2000). In addition, these effects
held even when we examined a narrower range of ages more
consistent with the cross-sectional studies reported by Twenge.
Therefore, we have a partial replication of the cross-sectional
studies, indicating that the increase in self-assertion culturally over
the last 50 years may have impacted the individuals living through
those changes. We did not find a relationship between cohort and
either social vitality or emotional stability, which was expected
from the cross-sectional differences in cohorts. The lack of repli-
cation for the latter may have resulted because the cross-sectional
findings have higher fidelity as they focused on narrower facets of
emotional stability, such as anxiety (Twenge, 2000). Nonetheless,
it is clear that cross-sectional cohort differences do not automati-
cally translate into longitudinal changes in personality.
We found unexpected cohort effects for agreeableness and con-
scientiousness. It appears that younger and older cohorts increased
on both of these trait domains, with a tendency for less of an
increase in cohorts born in the middle of the 20th century. This
pattern could be attributed to the general trend toward questioning
norms and traditions in cohorts that came of age in the 1960s and
1970s, which might have been manifest in less of an increase on
conscientiousness (Helson, Jones, & Kwan, 2002). Also, this same
period has been described as the age of narcissism (Lasch, 1979),
which may have attenuated increases in agreeableness. For both
traits, it appears that the general trend across time is for increases
and that this period in Western culture depressed this tendency.
Overall, it appears that cohort does influence some estimates of
longitudinal change and that some of these influences translate
from cross-sectional to longitudinal studies.
Finally, we tested the effect of attrition across each trait cate-
gory. Most longitudinal methodologists recommend that attrition
should be avoided at all costs. Consistent with our previous meta-
analysis (Roberts & DelVecchio, 2000), we found that attrition had
no discernable effect on estimates of mean-level change over time.
We would caution that these findings should not be taken to
indicate that attrition is not a problem for longitudinal studies.
Many studies have reported that individuals who remain in longi-
tudinal studies tend to be female, more conscientious, and higher
on measures of cognitive ability (e.g., Robins et al., 2001). From
a sampling perspective, then, it is still ideal to retain as many
participants as possible. Nonetheless, it appears that attrition was
not a biasing factor in estimates of mean-level personality change
over time.
Can Personality Change? A Meta-Analytic Resolution of
a Controversy
Historically, there has been an ongoing controversy over the
existence of personality change in adulthood. One position, hark-
ening back to Mischel (1968), was that personality traits did not
exist and therefore could not develop (Lewis, 2001). With the
accumulation of knowledge of both the long-term continuity of
personality (Fraley & Roberts, 2005) and the increasing levels of
consistency with age (Roberts & DelVecchio, 2000), coupled with
the burgeoning evidence for the predictive validity of personality
traits in important life domains, such as work (Judge et al., 1999),
marriage (Robins, Caspi, & Moffitt, 2002), and health (Bogg &
Roberts, 2004), we assume that this radical position can be con-
sidered refuted.
On the other hand, there has been an ongoing debate on the other
side of the spectrum, with some arguing for the immutability of
personality, especially in adulthood (McCrae & Costa, 1999), and
others arguing that personality traits continue to develop, some-
times even in midlife and old age (Field & Millsap, 1991; Helson,
Jones, & Kwan, 2002; Helson & Stewart, 1994; Roberts, 1997).
We see the meta-analytic approach as a viable way of resolving
this debate. It is interesting to note that our meta-analysis includes
the data presented by all of these authors and, of course, dozens of
others who were not invested in this debate. The impartial aggre-
Figure 4. Bivariate plot of standardized mean-level change in social dominance and cohort.
17
PATTERNS OF MEAN-LEVEL CHANGE
gation of studies provides the fairest way of determining the nature
of personality development and whether there are normative trends
in personality. In this sense, the data speak for themselves. Per-
sonality traits show clear patterns of normative change, continue to
change after age 30, and in several cases change in old age.
The second advantage that a meta-analysis affords is the oppor-
tunity to examine the history of the field and how the persons on
either side of this debate came to their positions. The position that
there are no normative trends in personality traits derived almost
exclusively from the arguments of Costa and McCrae (1988) and
their ongoing Baltimore Longitudinal Study of Aging (BLSA).
The position that personality can develop derives most strongly
from the work of Helson (Helson, Jones, & Kwan, 2002; Helson &
Moane, 1987; Helson & Wink, 1992) and her ongoing study of
Mills graduates and from the team of researchers involved with the
Institute of Human Development studies (Field & Millsap, 1991;
Haan et al., 1986). The positions of each of these teams were
eminently reasonable given the nature of their studies and the data
that they have collected. The BLSA is made primarily of men and
women over the age of 40. Furthermore, the early BLSA data
focused most on the trait domains of social vitality, neuroticism,
and openness. As we have seen in this meta-analysis, these three
trait domains demonstrate the least amount of change after the age
of 40. Moreover, many of the studies reporting on BLSA data have
not broken change out for specific decades of the life course (e.g.,
Costa & McCrae, 1988). Therefore, small yet meaningful changes
during specific age periods were not examined. Finally, their
analyses tended to be over shorter time spans of 3 and 6 years,
which also attenuates the amount of change that can be found. The
Mills and Institute of Human Development studies, in contrast,
concentrated more on change in young adulthood than in midlife,
at least initially (cf. Field & Millsap, 1991). Each also incorporated
a much greater focus on traits from the social dominance and
conscientiousness domains, which we see from the meta-analysis
are more prone to change, especially in young adulthood. More-
over, each of these studies has been focused on a single cohort
moving through specific periods of the life course over quite long
periods of time. All of these factors combined enhanced their
ability to find statistically significant changes in personality traits
over time.
So it is clear how this debate could have developed, given that
each author tends to weigh his or her own data and ongoing study
more heavily when drawing conclusions about the existence of
normative changes in personality. The meta-analytic approach, in
contrast, provides a more comprehensive, inclusive, and less arbi-
trary test of these ideas and therefore has the potential to resolve
these debates.
Why Does Personality Change in Adulthood?
The clear question that emerges from this consistent pattern of
mean-level changes across the life course is why people change in
this way. According to five-factor theory (McCrae, 2004; McCrae
& Costa, 1999; McCrae et al., 2000; ), mean-level changes arise
because of genetic predispositions to change in particular ways.
According to the five-factor theory, traits are “endogenous dispo-
sitions that follow intrinsic paths of development essentially inde-
pendent of environmental influences” (McCrae et al., 2000, p.
173). This position paints a very elegant if extreme portrait of
personality development. Life experiences, random life events,
shifts in cultural values, and simple lessons learned from living life
have no effect on the development of personality traits. If person-
ality traits do change, it is because human beings have a species-
wide genetic predisposition to develop in certain directions. Hu-
man beings are, within this perspective, hard-wired to become
more socially dominant, conscientious, and emotionally stable and
less open to experience with age.
As has been noted elsewhere, there are very few data to support
this position (Roberts et al., 2005). The few longitudinal studies of
twins have shown that childhood personality change appears to be
largely genetic, whereas in adulthood genetics has only a small
influence over personality change (Plomin & Nesselroade, 1990).
In adulthood, the largest estimate of the heritability of personality
trait change is around 30%, with the average being much lower
(McGue, Bacon, & Lykken, 1993). This indicates that environ-
mental factors play a larger role in personality trait change in
adulthood than do genetic factors. Moreover, the findings from the
present study showing that patterns of change were associated with
cohort would also contradict the notion that personality trait de-
velopment is independent of environmental influences. Thus, there
is little evidence to support the idea that development of person-
ality over time is independent of environmental influences.
On the basis of our analysis of the evidence to date, we believe
that life experiences and life lessons centered in young adulthood
are the most likely reason for the patterns of development we see
in this meta-analysis, especially the increases in social dominance,
conscientiousness, and emotional stability (Roberts et al., 2005).
Specifically, the universal tasks of social living in young adult-
hood, such as finding a marital partner, starting a family, and
establishing one’s career, appear to be candidate experiences
through which people also experience increases in such traits as
conscientiousness and emotional stability. As all dominant cultures
support if not promote these activities, they may be the catalysts
for the widespread pattern of personality trait development found
in adulthood and across cultures (Helson, Kwan, et al., 2002;
Roberts et al., 2005).
If this position holds, then most personality change occurs
through the press of contingencies found in age-graded social
roles. Specifically, these contingencies come in the form of ex-
pectations for how one should behave if he or she occupies a
specific role (Sarbin, 1964). For example, a person’s first job may
bring expectations to show up on time, work hard while at work,
spend extra hours of the day at work, and interact in an agreeable
manner with coworkers. These role expectations can affect change
through either punishing inappropriate behavior or rewarding ap-
propriate behavior. Violation of these expectations can lead to
withdrawal of social approval and even tangible negative out-
comes, such as losing one’s job. In turn, meeting expectations
should lead to greater levels of acceptance and social reinforce-
ment. Therefore, role expectations can facilitate personality
change by serving as guides for how one should act and possibly
how one should change.
To some extent then, if we know the life course patterning of
roles and the expectations that derive from these roles, then we
should get a clearer picture of how personality should develop. The
life course has been characterized as a series of interdependent
trajectories of work life, marriage, and parenthood (Elder, 1985).
We would expand this limited list to childhood roles, which
18 ROBERTS, WALTON, AND VIECHTBAUER
include being a child, teenager, friend, and student, and the roles of
late life, which include being a friend, retired person, and grand-
parent. Clearly, the initial stages of life are dominated by the child,
friend, and student roles. Young adulthood and middle age are
marked by a clear emphasis on work, marriage, and family roles
(Modell, 1989). Old age is a period dominated by disengagement
with the roles of middle adulthood and the transition out of the
labor force to become a retired person, grandparent, and possibly
a widower.
Several longitudinal studies have examined the relationship
between role experiences and personality trait change in young
adulthood and have found clear relationships to changes in traits
associated with social dominance, conscientiousness, and emo-
tional stability. These studies lend further support to the idea that
age-graded role experiences organized around universal tasks of
human living are in part responsible for the changes reported in the
present study. For example, compared with men who achieved the
same or less than their fathers, upwardly mobile men became more
dependable and responsible, independent, and motivated for suc-
cess (Elder, 1969). Women who had higher labor force participa-
tion showed increases in self-confidence (Clausen & Gilens, 1990)
and social dominance (Roberts, 1997). Furthermore, work satis-
faction was associated with decreases in measures of neuroticism
in women (Roberts & Chapman, 2000). Many of these effects were
replicated in a longitudinal study of men and women, in which, for
example, it was found that occupational success was related to
increases in dominance, whereas satisfaction was associated with
decreases in negative emotionality (Roberts, Caspi, & Moffitt,
2003).
Marital and family experiences also are associated with changes
in personality traits. Helson and Picano (1990) tracked personality
change from age 20 to age 43 in women who occupied either
traditional or neo- or nontraditional role configurations. For ex-
ample, women who occupied a traditional role configuration (e.g.,
homemaker) in young adulthood demonstrated fewer positive de-
velopmental gains in personality traits when compared with
women who occupied neo-traditional (e.g., some involvement in
the paid labor force) or nontraditional (e.g., working full time for
whole career) role configurations (Helson & Picano, 1990). More-
over, changes in motherhood status and the experience of divorce
were associated with changes in femininity and dominance, re-
spectively (Roberts, Helson, & Klohnen, 2002). Finally, experi-
encing more satisfying relationships is associated with increases in
emotional stability (Roberts & Chapman, 2000; Robins et al.,
2002).
These findings are much in line with life span developmental
theory that personality development occurs largely as a conse-
quence of the expectations and experiences that come with age-
graded roles (Roberts & Caspi, 2003; Roberts & Wood, in press).
Clearly, there may be additional mechanisms and experiences that
facilitate change, but when considering normative changes, espe-
cially those that appear to generalize across cultures (e.g., McCrae
et al., 1999), then the dominant, universal tasks of social living in
young adulthood appear to be the most likely factors contributing
to this pattern of change.
One of the primary implications of the fact that change in
personality traits comes about, in part, through social role experi-
ences is that chronological age is a less than ideal marker of
development. If people do change in response to life experiences
that can vary quite significantly, then normative patterns of per-
sonality trait development that result from these life experiences
can vary also. For example, the age at which individuals embark
on their careers has changed markedly over the last two centuries
(Modell, 1989). With the decrease in agriculture and manufactur-
ing and increase in technological and service jobs, people in
Western countries have extended their educational experiences and
delayed their careers from teens now well into their 20s and 30s.
This means that if changes within a culture also change the age of
onset of these major life transitions, then we might expect the
normative age at which personality traits change to shift also.
Many developmental psychologists have called for alternatives to
chronological age, often referring to “psychological age” as a more
appropriate depiction of development. One of the primary chal-
lenges of defining psychological age is knowing what factors
would need to be accounted for in the conceptualization of the
construct. One possibility, highlighted here, would be an index that
accounts for when adult social roles are engaged and committed to.
Integrating Patterns of Rank-Order Consistency and
Mean-Level Changes Across the Life Course
The pattern of mean-level change across the life course appears
to be more complex than that found for rank-order consistency.
The latter showed a progressive increase with age (Roberts &
Delvecchio, 2000), whereas the patterns of mean-level change,
though centered primarily in young adulthood, were found to exist
in almost all age periods in the life course. The distinct patterns of
rank-order consistency and mean-level change demonstrated
across these two meta-analyses provide definitive support for
Block’s (1971) original argument that different indexes of
continuity–change are relatively independent of one another.
Thus, populations can demonstrate high rank-order consistency
while simultaneously demonstrating significant mean-level
change. The distinctiveness of these different indexes of change is
often overlooked in characterizations of personality continuity and
change, especially when broad descriptions of the changeability of
personality are attempted.
In combination, the results of these two meta-analyses provide
a unique and compelling picture of adult personality trait devel-
opment. With age, people become increasingly consistent at least
in relationship to one another (rank-order consistency). The in-
crease in consistency is largely linear, with increasing increments
occurring through at least age 50 when there is an apparent plateau
in consistency. In contrast, mean-level changes appear to be cen-
tered primarily in young adulthood, with some continuing patterns
of change occurring throughout the remainder of the life course.
The combined pattern has relatively strong implications for our
understanding of personality development in the transition from
adolescence into young adulthood. Adolescence is marked by less
distinct patterns of mean-level change and a lower level of rank-
order consistency. Thus, rather than being a critical period in
which dispositions are formalized and crystallized, it may be that
adolescence is akin to a period of personality trait moratorium.
Definitive gains and losses may not be centered on this period of
the life course, as traits change in similar or greater magnitudes
later in the life course. In combination with the lower levels of
rank-order consistency, this implies that adolescence is a time not
only of exploration in terms of identity but also of flux and
19
PATTERNS OF MEAN-LEVEL CHANGE
exploration in terms of dispositional qualities. It is during young
adulthood when people begin to confront the realities of becoming
an adult and when we find significant gains in personality traits.
This fact alone argues for opening the developmental window to
ages quite a bit older than typically considered in most depictions
of personality trait development.
One distinct possibility is that the two patterns are linked.
Specifically, the gains in terms of personality traits may, in turn,
facilitate increased continuity. For the most part, people increased
on desirable traits during this period, such as conscientiousness
and emotional stability. It is interesting to note that a number of
studies have demonstrated a positive association between these
same trait domains and higher levels of continuity over time. For
example, Asendorpf and van Aken (1991) found that ego resil-
iency, which is, in part, related to emotional stability (Klohnen,
1996), predicted personality consistency over time in a longitudi-
nal sample of children. More specifically, children who were more
resilient tended to be more consistent over time. Similarly,
Schuerger, Zarrella, and Hotz (1989) found that clinical samples,
which we can assume are less emotionally stable, were less con-
sistent than nonclinical samples. In an 8-year longitudinal study,
men and women who were more controlled, less neurotic, and
more pro-socially oriented demonstrated less change in personality
traits and greater profile consistency across personality traits (Rob-
erts et al., 2001). If we can extrapolate from these longitudinal
studies, the peak of consistency found in middle age is in part the
result of increases in conscientiousness and emotional stability that
occur in young adulthood.
It should be noted that neither this meta-analysis nor the previ-
ous one on rank-order consistency provided any information on the
existence of individual differences in change in personality traits
(Nesselroade, 1991). There is now accumulating evidence for the
existence of individual differences in personality trait change in
young adulthood (Robins et al., 2001), middle age (Roberts et al.,
2002), and old age (Mroczek & Spiro, 2003; Small et al., 2003).
Individual differences in change speak to the unique patterns of
development particular to individual lives. For the most part,
individual differences in change are linked more strongly to mean-
level changes than to rank-order consistency. For example, if there
are strong normative trends to increase on a trait, such as consci-
entiousness, then more people in that sample show patterns of
increase than of decrease (Roberts et al., 2001; Robins et al.,
2001). Nonetheless, there remains a sizable minority that does not
follow the normative trend and demonstrates, for example, a
decrease in conscientiousness. Moreover, these nonnormative pat-
terns of change can be predicted from life experiences, such as
having an unstable marriage or participating in unconventional
activities, such as smoking marijuana (Roberts & Bogg, 2004;
Roberts et al., 2002).
One inference is that the existence of individual differences in
change calls into question the existence of normative trends in
personality trait development. Given the fact that these two in-
dexes of change are positively associated, we think that this
conclusion would be too extreme. However, it would be appropri-
ate to qualify normative trends by saying that a normative trend to
increase in a personality trait means that there is a higher proba-
bility for individuals to increase on this trait during a given period
of the life course, but there is not a guarantee that this change will
occur for all people. Moreover, the inclusion of individual differ-
ences in change is strongly relevant to any inference that person-
ality traits can and do change. Thus, in the absence of mean-level
changes, there may still be quite significant levels of individual
differences in change. Before broad or sweeping descriptions of
the changeability of personality traits can be made, a systematic
investigation of the existence of individual differences in change
across the life course needs to be made. Unfortunately, there are
very few data to date on this topic as it has only recently become
a focus of longitudinal researchers (Mroczek & Spiro, 2003; Rob-
erts, 1997; Small et al., 2003).
Limitations, Future Directions, and Conclusions
Despite the comprehensive nature of this review, there are
several glaring omissions in the longitudinal database and numer-
ous questions that remain unanswered. It is clear from our review
that many more studies performed on a wider variety of samples
are needed before definitive statements can be made concerning
the patterns of change for specific traits, such as social dominance
and agreeableness. Also, a disproportionate number of longitudinal
studies of personality have been based on highly educated, middle-
class or affluent samples. Studies of ethnic minorities, the poor,
and the working class are still a rarity in the field of personality
development. Moreover, more studies of middle-aged and older
individuals would help clarify some of the patterns of personality
development. We hope that the results of the present study will
motivate researchers to include personality measures in the longi-
tudinal studies of older people, in part because personality traits
apparently continue to change in old age. Moreover, it is clear
from the distribution of cohorts across the life course that the
enterprise of studying individuals longitudinally is a relatively
recent phenomenon. Only with the ongoing collection of data in
the coming decades will we be able to tease apart the effects of
cohort and age on personality development.
A second limitation to the present study was the necessity of
categorizing various personality measures into the Big Five do-
mains. Although this approach has greatly enhanced our ability to
synthesize research findings in numerous domains, including job
performance (Hogan & Holland, 2003) and creativity (Feist, 1998)
in addition to personality development, the act of categorizing
personality measures into broad domains inherently leads to some
loss of information. For example, much information is lost about
particular measures that represent subcomponents of each domain.
The breakdown of the domain of extraversion into the facets of
social vitality and social dominance serves as an excellent exam-
ple. Without differentiating the domain of extraversion, the pat-
terns of development would have been unclear. Of course, before
a more fine-grained analysis of each Big Five domain can be
performed, we need replicable lower order models of each of the
Big Five domains. Unfortunately, this more fine-grained under-
standing of the Big Five has yet to emerge.
A final limitation is the generalizability of the findings, as there
are to date no longitudinal studies tracking continuity and change
in personality traits from many parts of the world, such as Africa
or Asia. Therefore, the results are particularly germane to Western
cultures, and their generalizability to non-Western countries is still
unknown and has yet to be established. This omission in the
longitudinal database is more conspicuous when one considers the
wealth of studies arguing that culture, especially in the particular
20 ROBERTS, WALTON, AND VIECHTBAUER
forms found in Asian countries, has a profound effect on psycho-
logical functioning (e.g., Markus & Kityama, 1991). In the absence
of longitudinal studies of Asian populations, it is impossible to
know whether people in these cultures demonstrate different de-
velopmental patterns in terms of personality traits than in Western
cultures. For example, as a result of different socialization prac-
tices in many Asian cultures, children experience more shame and
social comparison at an earlier age than in the West (Fung, 1999;
Miller, Wiley, Fung, & Liang, 1997). One would assume that
Asian parenting practices that center more on shame would result
in higher levels of conscientiousness at an earlier age, but there are
no longitudinal studies to test this hypothesis at this point.
In conclusion, the results from our meta-analysis demonstrate
that the patterns of personality trait change are intrinsically posi-
tive. People tend to become more socially dominant, conscien-
tious, and emotionally stable through midlife. Moreover, the pe-
riod of young adulthood rather than adolescence is the primary
period of mean-level personality trait development. We also pro-
vided definitive evidence for the continued plasticity of personality
traits beyond age 30 and well into old age in the case of specific
traits, such as social vitality, agreeableness, conscientiousness, and
openness to experience. It appears that personality trait develop-
ment is not just a phenomenon of childhood but also of all
adulthood.
References
References marked with an asterisk indicate studies included in the
meta-analysis.
Arnett, J. J. (2000). Emerging adulthood: A theory of development from
the late teens through the twenties. American Psychologist, 55, 469–
480.
*Asendorpf, J. B. (2003). [Personality trait change]. Unpublished raw data.
Asendorpf, J. B., & van Aken, M. A. G. (1991). Correlates of the temporal
consistency of personality patterns in childhood. Journal of Personality,
59, 689–703.
*Asendorpf, J. B., & Wilpers, S. (1998). Personality effects on social
relationships. Journal of Personality and Social Psychology, 74, 1531–
1544.
Baltes, P. B. (1997). On the incomplete architecture of human ontogeny.
American Psychologist, 52, 366–380.
Baltes, P. B., Lindenberger, U., & Staudinger, U. M. (1998). Life-span
theory in developmental psychology. In W. Damon & R. M. Lerner
(Eds.), Handbook of child psychology (5th ed., Vol. 1, pp. 1029–1143).
New York: Wiley.
*Baltes, P. B., & Nesselroade, J. R. (1972). Cultural change and adolescent
personality development: An application of longitudinal sequences. De-
velopmental Psychology, 7, 244–256.
Bandura, A. (1999). Social cognitive theory of personality. In L. A. Pervin
& O. P. John (Eds.), Handbook of personality: Theory and research (2nd
ed., pp. 154–196). New York: Guilford Press.
*Barefoot, J. C., Mortensen, E. L., Helms, M. J., Avlund, K., & Schroll, M.
(2001). A longitudinal study of gender differences in depressive symp-
toms from age 50 to 80. Psychology and Aging, 16, 342–345.
*Berdie, R. F. (1968). Personality changes from high school entrance to
college matriculation. Journal of Counseling Psychology, 15, 376–380.
Biggerstaff, B. (2000). S-Plus library DVBID (dvbidlib.exe) [Computer
software]. Retrieved August 29, 2003, from http://www.stat.colostate
.edu/⬃bradb/files/
*Block, J. (1971). Lives through time. Berkeley, CA: Bancroft.
Bogg, T., & Roberts, B. W. (2004). Conscientiousness and health-related
behaviors: A meta-analysis of the leading behavioral contributors to
mortality. Psychological Bulletin, 130, 887–919.
*Branje, S. J. T., van Lieshout, C. F. M., & van Aken, M. A. G. (2004).
Relations between Big Five personality characteristics and perceived
support in adolescents’ families. Journal of Personality and Social
Psychology, 86, 615–628.
Brim, O. C. (1965). Adult socialization. In J. A. Clausen (Ed.), Socializa-
tion and society (pp. 182–226). Boston: Little, Brown.
*Bronson, W. C. (1966). Central orientations: A study of behavior orga-
nization from childhood to adolescence. Child Development, 37, 125–
155.
*Cairns, E., McWhirter, L., Duffy, U., & Barry, R. (1990). The stability of
self-concept in late adolescence: Gender and situational effects. Person-
ality and Individual Differences, 11, 937–944.
*Cantoni, L. J. (1955). A study of emotional adjustment: The correlation of
student and adult forms of the Bell Adjustment Inventory over a period
of thirteen years. Educational and Psychological Measurement, 15,
137–143.
*Caputo, D. V., Psathas, G., & Plapp, J. M. (1966). Test–retest reliability
of the EPPS. Educational and Psychological Measurement, 26, 883–
886.
*Carmichael, C. M., & McGue, M. (1994). A longitudinal family study of
personality change and stability. Journal of Personality, 62, 1–20.
*Cartwright, L. K., & Wink, P. (1994). Personality change in women
physicians from medical student years to mid-40s. Psychology of Women
Quarterly, 18, 291–308.
Caspi, A., & Roberts, B. W. (1999). Personality change and continuity
across the life course. In L. A. Pervin & O. P. John (Eds.), Handbook of
personality theory and research (pp. 300–326). New York: Guilford
Press.
Clausen, J. A., & Gilens, M. (1990). Personality and labor force partici-
pation across the life course: A longitudinal study of women’s careers.
Sociological Forum, 5, 595–618.
Cohen, J. (1992). A power primer. Psychological Bulletin, 112, 155–159.
Conley, J. J. (1984). The hierarchy of consistency: A review and model of
longitudinal findings on adult individual differences in intelligence,
personality, and self-opinion. Personality and Individual Differences, 5,
11–26.
Conn, S., & Reike, M. L. (Eds.). (1994). The 16PF fifth edition technical
manual. Champaign, IL: Institute for Personality and Ability Testing.
*Cooper, M. L. (2004). [Changes in impulse control].Unpublished raw
data.
*Costa, P. T., Jr., Herbst, J. H., McCrae, R. R., & Siegler, I. C. (2000).
Personality at midlife: Stability, intrinsic maturation, and response to life
events. Assessment, 7, 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, 853–863.
*Costa, P. T., & McCrae, R. R. (1992). Multiple uses for longitudinal
personality data. European Journal of Personality, 6, 85–102.
Costa, P. T., Jr., & McCrae, R. R. (1994). Revised NEO Personality
Inventory (NEO-PI-R) and NEO-Five-Factor Inventory (NEO-FFI) pro-
fessional manual. Odessa, FL: Psychological Assessment Resources.
Costa, P. T., & McCrae, R. R. (1997). Longitudinal stability of adult
personality. In R. Hogan, J. Johnson, & S. Briggs (Eds.), Handbook of
personality psychology (pp. 269–292). San Diego: Academic Press.
*Costa, P. T., McCrae, R. R., & Arenberg, D. (1980). Enduring disposi-
tions in adult males. Journal of Personality and Social Psychology, 38,
793–800.
*Cramer, P. (1998). Freshman to senior year: A follow-up study of iden-
tity, narcissism, and defense mechanisms. Journal of Research in Per-
sonality, 32, 156–172.
*Cramer, P. (2003). Personality change in later adulthood is predicted by
21
PATTERNS OF MEAN-LEVEL CHANGE
defense mechanism use in early adulthood. Journal of Research in
Personality, 37, 76–104.
*Crook, M. N. (1943). A retest with the Thurstone Personality Schedule
after six and one-half years. Journal of General Psychology, 28, 111–
120.
*Davis, M. H., & Franzoi, S. L. (1991). Stability and change in adolescent
self-consciousness and empathy. Journal of Research in Personality, 25,
70–87.
*Dudek, S. Z., & Hall, W. B. (1991). Personality consistency: Eminent
architects 25 years later. Creativity Research Journal, 4, 213–231.
Duval, S., & Tweedie, R. (2000). A nonparametric “trim and fill” method
of accounting for publication bias in meta-analysis. Journal of the
American Statistical Association, 95, 89–98.
Elder, G. H. (1969). Occupational mobility. Life patterns, and personality.
Journal of Health and Social Behavior, 10, 308–323.
Elder, G. H., Jr. (1985). Household, kinship, and the life course: Perspec-
tives on Black families and children. In M. B. Spencer, G. K. Brookins,
& W. R. Allen (Eds.), Beginnings: The social and affective development
of black children (pp. 29–43). Hillsdale, NJ: Erlbaum.
Erikson, E. H. (1950). Childhood and society. New York: Norton.
Feist, G. J. (1998). A meta-analysis of personality in scientific and artistic
creativity. Personality and Social Psychology Review, 2, 290–309.
*Field, D., & Millsap, R. E. (1991). Personality in advanced old age:
Continuity or change? Journal of Gerontology: Series B: Psychological
Sciences and Social Sciences, 46, P299–P308.
Fraley, 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, 60–74.
Friedman, H. S. (2000). Long-term relations of personality and health:
Dynamism, mechanisms, tropisms. Journal of Personality, 68, 1089–
1107.
Funder, D. C., & Colvin, C. R. (1991). Explorations in behavioral consis-
tency: Properties of persons, situations, and behaviors. Journal of Per-
sonality and Social Psychology, 60, 773–794.
Fung, H. (1999). Becoming a moral child: The socialization of shame
among young Chinese children. Ethos, 27, 180–209.
*Gold, S. R., & Henderson, B. B. (1990). Daydreaming and curiosity:
Stability and change in gifted children and adolescents. Adolescence, 25,
701–708.
Gough, H. G., & Bradley, P. (1996). California Psychological Inventory
manual (3rd ed.). Palo Alto, CA: Consulting Psychologists Press.
*Grigoriadis, S., & Fekken, G. C. (1992). Person reliability on the Min-
nesota Multiphasic Personality Inventory. Personality and Individual
Differences, 13, 491–500.
*Gustavsson, J. P., Weinryb, R. M., Go¨ransson, S., Pedersen, N. L., &
Åsberg, M. (1997). Stability and predictive ability of personality traits
across 9 years. Personality and Individual Differences, 22, 783–791.
Guttman, D. L. (1987). Reclaimed powers: Toward a new psychology of
men and women in later life. New York: Basic Books.
*Haan, N., Millsap, R., & Hartka, E. (1986). As time goes by: Change and
stability in personality over fifty years. Psychology and Aging, 1, 220–
232.
*Hair, E. C., & Graziano, W. G. (2003). Self-esteem, personality and
achievement in high school: A prospective longitudinal study in Texas.
Journal of Personality, 71, 971–994.
*Hathaway, S. R., & Monachesi, E. D. (1963). Adolescent personality and
behavior: MMPI patterns of normal, delinquent, dropout, and other
outcomes. Minneapolis, MN: University of Minnesota Press.
Helson, R., Jones, C., & Kwan, V. S. Y. (2002). Personality change over
40 years of adulthood: Hierarchical linear modeling analyses of two
longitudinal samples. Journal of Personality and Social Psychology, 83,
752–766.
*Helson, R., & Kwan, V. S. Y. (2000). Personality development in adult-
hood: The broad picture and processes in one longitudinal sample. In S.
Hampson (Ed.), Advances in personality psychology (Vol. 1, pp. 77–
106). London: Routledge.
Helson, R., Kwan, V. S. Y., John, O. P., & Jones, C. (2002). The growing
evidence for personality change in adulthood: Findings from research
with personality inventories. Journal of Research in Personality, 36,
287–306.
*Helson, R., & Moane, G. (1987). Personality change in women from
college to midlife. Journal of Personality and Social Psychology, 53,
176–186.
Helson, R., Pals, J., & Solomon, M. (1997). Is there adult development
distinctive to women? In R. Hogan, J. Johnson, & S. Briggs (Eds.),
Handbook of personality psychology (pp. 293–317). San Diego, CA:
Academic Press.
Helson, R., & Picano, J. (1990). Is the traditional role bad for women?
Journal of Personality and Social Psychology, 59, 311–320.
*Helson, R., Roberts, B., & Agronick, G. (1995). Enduringness and change
in creative personality and the prediction of occupational creativity.
Journal of Personality and Social Psychology, 69, 1173–1183.
Helson, R., & Stewart, A. (1994). Personality change in adulthood. In T. F.
Heatherton & J. L. Weinberger (Eds.), Can personality change? (pp.
201–225). Washington, DC: American Psychological Association.
*Helson, R., & Wink, P. (1992). Personality change in women from the
early 40s to the early 50s. Psychology and Aging, 7, 46–55.
Hogan, J., & Holland, B. (2003). Using theory to evaluate personality and
job-performance relations: A socioanalytic perspective. Journal of Ap-
plied Psychology, 88, 100–112.
*Holmlund, U. (1991). Change and stability of needs from middle adoles-
cence to young adulthood in Swedish females. European Journal of
Personality, 5, 379–385.
John, O. P., & Srivastava, S. (1999). The Big Five trait taxonomy: History,
measurement, and theoretical perspectives. In L. A. Pervin & O. P. John
(Eds.), Handbook of personality theory and research (Vol. 2, pp. 102–
138). New York: Guilford Press.
Judge, T. A., Higgins, C. A., Thoreson, C. J., & Barrick, M. R. (1999). The
Big Five personality traits, general mental ability, and career success
across the life span. Personnel Psychology, 52, 621–652.
*Karney, B. R., & Bradbury, T. N. (1997). Neuroticism, marital interac-
tion, and the trajectory of marital satisfaction. Journal of Personality and
Social Psychology, 72, 1075–1092.
*Karney, B. R., & Coombs, R. H. (2000). Memory bias in long-term close
relationships: Consistency or improvement? Personality and Social Psy-
chology Bulletin, 26, 959–970.
*Keltikangas-Ja¨rvinen, L. (1989). Stability of Type A behavior during
adolescence, young adulthood, and adulthood. Journal of Behavioral
Medicine, 12, 387–396.
*Kitchener, K. S., King, P. M., Davison, M. L., Parker, C. A., & Wood,
P. K. (1984). A longitudinal study of moral and ego development in
young adults. Journal of Youth and Adolescence, 13, 197–211.
Klohnen, E. C. (1996). Conceptual analysis and measurement of the
construct of ego-resiliency. Journal of Personality and Social Psychol-
ogy, 70, 1067–1079.
*Klopsch, J. W. (1983). Police personality change as measured by the
MMPI—A five-year longitudinal study. Unpublished doctoral disserta-
tion, Fuller Theological Seminary, Pasadena, CA.
Kogan, N. (1990). Personality and aging. In J. E. Birren & S. W. Schaie
(Eds.), Handbook of the psychology of aging (pp. 330–346). San Diego:
Academic Press.
*Labouvie-Vief, G., & Jain, E. (2002). Personality stability and change
across the lifespan: A six-year longitudinal sequential analysis. Unpub-
lished manuscript, Wayne State University, Detroit, MI.
Lasch, C. (l979). The culture of narcissism: American life in an age of
diminishing expectations. New York: Norton.
*Leadbeater, B. J., Kuperminc, G. P., Hertzog, C., & Blatt, S. J. (1999). A
22 ROBERTS, WALTON, AND VIECHTBAUER
multivariate model of gender differences in adolescents’ internalizing
and externalizing problems. Developmental Psychology, 35, 1268–1282.
*Leon, G. R., Gillum, B., Gillum, R., & Gouze, M. (1979). Personality
stability and change over a 30 year period—Middle age to old age.
Journal of Consulting and Clinical Psychology, 47, 517–524.
Levinson, D. J. (1978). The seasons of a man’s life. New York: Ballantine
Books.
Lewis, M. (1999). On the development of personality. In L. A. Pervin &
O. P. John (Eds.), Handbook of personality theory and research (pp.
327–346). New York: Guilford Press.
Lewis, M. (2001). Issues in the study of personality development. Psycho-
logical Inquiry, 12, 67–83.
*Loehlin, J. C., Horn, J. M., & Willerman, L. (1990). Heredity, environ-
ment, and personality change: Evidence from the Texas Adoption
Project. Journal of Personality, 58, 221–243.
*Loevinger, J., Cohn, L. D., Bonneville, L. P., Redmore, C. D., Streich,
D. D., & Sargent, M. (1985). Ego development in college. Journal of
Personality and Social Psychology, 48, 947–962.
Markus, H. R., & Kitayama, S. (1991). Culture and the self: Implications
for cognition, emotion, and motivation. Psychological Review, 98, 224–
253.
*Martin, J., & Redmore, C. (1978). A longitudinal study of ego develop-
ment. Developmental Psychology, 14, 189–190.
*Martin, P., Long, M. V., & Poon, L. W. (2002). Age changes and
differences in personality traits and states of the old and very old.
Journals of Gerontology: Series B: Psychological Sciences and Social
Sciences, 57, P144–P152.
McCrae, R. R. (2004). Human nature and culture: A trait perspective.
Journal of Research in Personality, 38, 3–14.
McCrae, R. R., & Costa, P. T. (1994). The stability of personality: Obser-
vation and evaluations. Current Directions in Psychological Science, 3,
173–175.
McCrae, R. R., & Costa, P. T., Jr. (1999). A five-factor theory of person-
ality. In L. A. Pervin & O. P. John (Eds.), Handbook of personality
theory and research (Vol. 2, pp. 139–153). New York: Guilford Press.
McCrae, R. R., Costa, P. T., Jr., Ostendorf, F., Angleitner, A., Hrebickova,
M., Avia, M. D., et al. (2000). Nature over nurture: Temperament,
personality, and life span development. Journal of Personality and
Social Psychology, 78, 173–186.
McCrae, R. R., Costa, P. T., Jr., Pedroso de Lima, M., Simoes, A.,
Ostendorf, F., Angleitner, A., et al. (1999). Age differences in person-
ality across the adult life span: Parallels in five cultures. Developmental
Psychology, 35, 466–477.
*McGue, M., Bacon, S., & Lykken, D. T. (1993). Personality stability and
change in early adulthood: A behavioral genetic analysis. Developmental
Psychology, 29, 96–109.
*Melamed, A. R., Silverman, M. S., & Lewis, G. J. (1974). Three year
follow-up of women religious on the 16 Personality Factor Question-
naire. Review of Religious Research, 15, 64–70.
Miller, P. J., Wiley, A. R., Fung, H., & Liang, C. (1997). Personal
storytelling as a medium of socialization in Chinese and American
families. Child Development, 68, 557–568.
Mischel, W. (1968). Personality and assessment. New York: Wiley.
Modell, J. (1989). Into one’s own: From youth to adulthood in the United
States, 1920 –1975. Berkeley: University of California Press.
*Morizot, J., & LeBlanc, M. (2003). Continuity and change in personality
traits from adolescence to midlife: A 25-year longitudinal study com-
paring representative and adjudicated men. Journal of Personality, 71,
705–755.
Morris, S. B., & Deshon, R. P. (2002). Combining effect size estimates in
meta-analysis with repeated measures and independent-groups designs.
Psychological Methods, 7, 105–125.
*Mortimer, J. T., Finch, M. D., & Kumka, D. (1982). Persistence and
change in development: The multidimensional self-concept. In P. B.
Baltes & O. G. Brim (Eds.), Life-span development and behavior (Vol.
4, pp. 263–313). Hillsdale, NJ: Erlbaum.
*Mroczek, D., McLaughlin, D., Finkelson, J., & Cirstofaro, A. (2002,
January/February). Intraindividual personality trait trajectories in ado-
lescents. Paper presented at the meeting of the Society for Personality
and Social Psychology, Savannah, GA.
*Mroczek, D. K., & Spiro, A. (2003). Modeling intraindividual change in
personality traits: Findings from the Normative Aging Study. Journals of
Gerontology: Series B: Psychological Sciences and Social Sciences, 58,
P305–P306.
*Mussen, P., Eichorn, D. H., Honzik, M. P., Bieber, S. L., & Meredith,
W. M. (1980). Continuity and change in women’s characteristics over
four decades. International Journal of Behavioral Development, 3, 333–
347.
Nesselroade, J. R. (1991). Interindividual differences in intraindividual
change. In L. M. Collins & J. L. Horn (Eds.), Best methods for the
analysis of change: Recent advances, unanswered questions, future
directions (pp. 92–105). Washington, DC: American Psychological As-
sociation.
*Nesselroade, J. R., & Baltes, P. B. (1974). Adolescent personality devel-
opment and historical change: 1970–1972. Monographs of the Society
for Research in Child Development, 39, 1–79.
*Neyer, F. J., & Asendorpf, J. B. (2001). Personality–relationship trans-
action in young adulthood. Journal of Personality and Social Psychol-
ogy, 81, 1190–1204.
*Nichols, R. C. (1967). Personality change and the college. American
Educational Research Journal, 4, 173–190.
*Ogawa, J. R., Sroufe, L. A., Weinfield, N. S., Carlson, E. A., & Egeland,
B. (1997). Development and the fragmented self: Longitudinal study of
dissociative symptomatology in a nonclinical sample. Development and
Psychopathology, 9, 855–879.
*Ormel, J., & Rijsdijk, F. V. (2000). Continuing change in neuroticism
during adulthood—Structural modeling of a 16-year, 5-wave community
study. Personality and Individual Differences, 28, 461–478.
*Ormel, J., & Schaufeli, W. B. (1991). Stability and change in psycholog-
ical distress and their relationship with self-esteem and locus of control:
A dynamic equilibrium model. Journal of Personality and Social Psy-
chology, 60, 288–299.
Overton, R. C. (1998). A comparison of fixed-effects and mixed (random-
effects) models for meta-analysis tests of moderator variable effects.
Psychological Methods, 3, 354–379.
*Pedersen, W. (1991). Mental health, sensation seeking and drug use
patterns: A longitudinal study. British Journal of Addiction, 86, 195–
204.
*Peterson, B. E., & Lane, M. D. (2001). Implications of authoritarianism
for young adulthood: Longitudinal analysis of college experiences and
future goals. Personality and Social Psychology Bulletin, 27, 678–690.
*Piccione, C., Hilgard, E. R., & Zimbardo, P. G. (1989). On the degree of
stability of measured hypnotizability over a 25-year period. Journal of
Personality and Social Psychology, 56, 289–295.
*Plant, W. T. (1965). Longitudinal changes in intolerance and authoritari-
anism for subjects differing in amount of college education over four
years. Genetic Psychology Monographs, 72, 247–287.
*Plant, W. T., & Telford, C. W. (1966). Changes in personality for groups
completing different amounts of college over two years. Genetic Psy-
chology Monographs, 74, 3–36.
Plomin, R., & Nesselroade, J. R. (1990). Behavioral genetics and person-
ality change. Journal of Personality, 58, 191–220.
*Pogue-Geile, M. F., & Rose, R. J. (1985). Developmental genetic studies
of adult personality. Developmental Psychology, 21, 547–557.
*Popham, S. M., & Holden, R. R. (1991). Psychometric properties of
MMPI factor scales. Personality and Individual Differences, 12, 513–
517.
Raudenbush, S. W. (1994). Random effects models. In H. M. Cooper &
23
PATTERNS OF MEAN-LEVEL CHANGE
L. V. Hedges (Eds.), The handbook of research synthesis (pp. 301–321).
New York: Sage.
Raudenbush, S. W., & Bryk, A. S. (2002). Hierarchical linear models:
Applications and data analysis methods. Thousand Oaks, CA, Sage.
*Roberts, B. W. (1997). Plaster or plasticity: Are adult work experiences
associated with personality changes in women? Journal of Personality,
65, 205–232.
Roberts, B. W., & Bogg, T. (2004). A longitudinal study of the relation-
ships between conscientiousness and the social-environmental factors
and substance-use behaviors that influence health. Journal of Personal-
ity, 72, 325–353.
Roberts, B. W., & Caspi, A. (2003). The cumulative continuity model of
personality development: Striking a balance between continuity and
change in personality traits across the life course. In U. Staudinger & U.
Lindenberger (Eds.), Understanding human development: Lifespan psy-
chology in exchange with other disciplines (pp. 183–214). Dordrecht,
The Netherlands: Kluwer Academic Publishers.
*Roberts, B. W., Caspi, A., & Moffitt, T. (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., Caspi, A., & Moffitt, T. (2003). Work experiences and
personality development in young adulthood. Journal of Personality and
Social Psychology, 84, 582–593.
*Roberts, B. W., & Chapman, C. N. (2000). Change in dispositional
well-being and its relation to role quality: A 30-year longitudinal study.
Journal of Research in Personality, 34, 26–41.
Roberts, B. W., & DelVecchio, W. F. (2000). The rank-order consistency
of personality from childhood to old age: A quantitative review of
longitudinal studies. Psychological Bulletin, 126, 3–25.
Roberts, B. W., & Helson, R. (1997). Changes in culture, changes in
personality: The influence of individualism in a longitudinal study of
women. Journal of Personality and Social Psychology, 72, 641–651.
*Roberts, B. W., Helson, R., & Klohnen, E. C. (2002). Personality devel-
opment and growth in women across 30 years: Three perspectives.
Journal of Personality, 70, 79–102.
Roberts, B. W., Robins, R. W., Caspi, A., & Trzesniewski, K. (2003).
Personality trait development in adulthood. In J. Mortimer & M. Shana-
han (Eds.), Handbook of the life course (pp. 579–598). New York:
Kluwer Academic Publishers/Plenum Press.
Roberts, B. W., & Wood, D. (in press). The cumulative continuity model
of personality trait development. In D. Mroczek & T. Little (Eds.),
Handbook of personality development. Mahwah, NJ: Erlbaum.
Roberts, B. W., Wood, D., & Smith, J. L. (2005). Evaluating five factor
theory and social investment perspectives on personality trait develop-
ment. Journal of Research in Personality, 39, 166–184.
Robins, R. W., Caspi, A., & Moffitt, T. E. (2002). It’s not just who you’re
with, it’s who you are: Personality and relationship experiences across
multiple relationships. Journal of Personality, 70, 925–964.
*Robins, R. W., Fraley, R. C., Roberts, B. W., & Trzesniewski, K. (2001).
A longitudinal study of personality change in young adulthood. Journal
of Personality, 69, 617–640.
*Sanderman, R., & Ranchor, A. V. (1994). Stability of personality traits
and psychological distress over six years. Perceptual and Motor Skills,
78, 89–90.
*Sanford, N. (1962). The American college. New York: Wiley.
Sarbin, T. R. (1964). Role theoretical interpretation of psychological
change. In P. Worchel & D. Byrne (Eds.), Personality change (pp.
176–219). New York: Wiley.
*Schofield, W. (1953). A study of medical students with the MMPI: II.
Group and individual changes after two years. Journal of Psychology,
36, 137–141.
Schuerger, J. M., Zarrella, K. L., & Hotz, A. S. (1989). Factors that
influence the temporal stability of personality by questionnaire. Journal
of Personality and Social Psychology, 56, 777–783.
*Scollon, C. N. (2004). Predictors of intraindividual change in personality
and well-being. Unpublished doctoral dissertation, University of Illinois
at Urbana–Champaign.
*Small, B. J., Hertzog, C., Hultsch, D. F., & Dixon, R. A. (2003). Stability
and change in adult personality over 6 years: Findings from the Victoria
Longitudinal Study. Journal of Gerontology: Series B: Psychological
Sciences and Social Sciences, 58, P166–P176.
Srivastava, S., John, O. P., Gosling, S. D., & Potter, J. (2003). Develop-
ment of personality in early and middle adulthood: Set like plaster or
persistent change? Journal of Personality and Social Psychology, 84,
1041–1053.
*Stein, J. A., Newcomb, M. D., & Bentler, P. M. (1986). Stability and
change in personality: A longitudinal study from early adolescence to
young adulthood. Journal of Research in Personality, 20, 276–291.
*Stevens, D. P., & Truss, C. V. (1985). Stability and change in adult
personality over 12 and 20 years. Developmental Psychology, 21, 568–
584.
*Tenerowicz, C. M. (1992). A longitudinal study of personality change in
urban police officers: Educational implications. Unpublished doctoral
dissertation, Cleveland State University, Cleveland, OH.
*Townsend, A., Noelker, L., Deimling, G., & Bass, D. (1989). Longitu-
dinal impact of interhousehold caregiving on adult children’s mental
health. Psychology and Aging, 4, 393–401.
Twenge, J. M. (2000). The age of anxiety? Birth cohort change in anxiety
and neuroticism, 1952–1993. Journal of Personality and Social Psychol-
ogy, 79, 1007–1021.
Twenge, J. M. (2001). Changes in women’s assertiveness in response to
status and roles: A cross-temporal meta-analysis, 1931–1993. Journal of
Personality and Social Psychology, 81, 133–145.
Twenge, J. M. (2002). Birth cohort, social change, and personality: The
interplay of dysphoria and individualism in the 20th century. In D.
Cervone & W. Mischel (Eds.), Advances in personality science (pp.
196–218). New York: Guilford Press.
*Vaidya, J. G., Gray, E. K., Haig, J., & Watson, D. (2002). On the temporal
stability of personality: Evidence for differential stability and the role of
life experiences. Journal of Personality and Social Psychology, 83,
1469–1484.
*van der Velde, M. E. G., Feij, J. A., & Taris, T. W. (1995). Stability and
change of person characteristics among young adults: The effect of the
transition from school to work. Personality and Individual Differences,
18, 89–99.
Viechtbauer, W. (2004a). Model selection strategies in meta-analysis:
Choosing between the fixed-, random-, and mixed-effects model. Manu-
script submitted for publication.
Viechtbauer, W. (2004b). Moderator analysis and model selection via the
mixed-effects model in meta-analysis. Manuscript submitted for publication.
*Viken, R. J., Rose, R. J., Kaprio, J., & Koskenvuo, M. (1994). A
developmental genetic analysis of adult personality: Extraversion and
neuroticism from 18 to 59 years of age. Journal of Personality and
Social Psychology, 4, 722–730.
*Watson, D., & Slack, K. (1993). General factors of affective temperament
and their relation to job satisfaction over time. Organizational Behavior
and Human Decision Processes, 54, 181–202.
*Watson, D., & Walker, L. M. (1996). The long-term stability and predic-
tive validity of trait measures of affect. Journal of Personality and Social
Psychology, 70, 567–577.
*Weinryb, R. M., Gustavsson, J. P., Åsberg, M., & Ro¨ssel, R. J. (1992).
Stability over time of character assessment using a psychodynamic
instrument and personality inventories. Acta Psychiatrica Scandinavica,
86, 179–184.
*Westenberg, P. M., & Gjerde, P. F. (1999). Ego development during the
transition from adolescence to young adulthood: A 9-year longitudinal
study. Journal of Research in Personality, 33, 233–252.
*Wilhelm, K., & Parker, G. (1990). Reliability of the parental bonding
24 ROBERTS, WALTON, AND VIECHTBAUER
instrument and intimate bond measure scales. Australian and New Zea-
land Journal of Psychiatry, 24, 199–202.
*Wink, P., & Helson, R. (1993). Personality change in women and their
partners. Journal of Personality and Social Psychology, 65, 597–605.
*Woodall, K. L., & Matthews, K. A. (1993). Changes in and stability of
hostile characteristics: Results from a 4-year longitudinal study of chil-
dren. Journal of Personality and Social Psychology, 64, 491–499.
*Woodruff, D. S., & Birren, J. E. (1972). Age changes and cohort differ-
ences in personality. Developmental Psychology, 6, 252–259.
Zelli, A., & Dodge, K. A. (1999). Personality development from the bottom
up. In D. Cervone & Y. Shoda (Eds.), The coherence of personality:
Social– cognitive bases of consistency, variability, and organization (pp.
94–126). New York: Guilford Press.
Received May 14, 2003
Revision received January 31, 2005
Accepted February 28, 2005 䡲
25
PATTERNS OF MEAN-LEVEL CHANGE
A preview of this full-text is provided by American Psychological Association.
Content available from Psychological Bulletin
This content is subject to copyright. Terms and conditions apply.
