Psychology and Aging Copyright 2008 by the American Psychological Association

2008, Vol. 23, No. 3, 558 –566 0882-7974/08/$12.00 DOI: 10.1037/a0012897

Age Differences in the Big Five Across the Life Span: Evidence From
Two National Samples

M. Brent Donnellan Richard E. Lucas

Michigan State University Michigan State University and German Institute for
Economic Research

Cross-sectional age differences in the Big Five personality traits were investigated using 2 large datasets
from Great Britain and Germany: the British Household Panel Study (BHPS; N ⱖ 14,039) and the
German Socio-Economic Panel Study (GSEOP; N ⱖ 20,852). Participants, who ranged in age from 16
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

to the mid-80s, completed a 15-item version of the Big Five Inventory (e.g., John & Srivastava, 1999)
This document is copyrighted by the American Psychological Association or one of its allied publishers.

in either 2005 or 2006. The observed age trends were generally consistent across both datasets.
Extraversion and Openness were negatively associated with age, whereas Agreeableness was positively
associated with age. Average levels of Conscientiousness were highest for participants in middle age. The
only exception was Neuroticism, which was slightly negatively associated with age in the BHPS and
slightly positively associated with age in the GSEOP. Neither gender nor education level were consistent
moderators of age differences in the Big Five.

Keywords: personality assessment, Big Five, personality development, British Household Panel Study,
German Socio-Economic Panel Study

Age-related differences in personality have captured human tribute to this literature by evaluating cross-sectional age differences
attention for centuries. For instance, Aristotle devoted three chap- in the Big Five personality traits in two large national datasets.
ters of Book II of his Rhetoric to the description of the character-
istics of individuals at different phases of the life span. His Age Differences in the Big Five
prediction, translated into a testable hypothesis, is that there should
be age-related differences in personality attributes. Conversely, There is something of a consensus that five broad domains
William James (1892/1985) believed that character was fixed by capture much of the variability in personality traits (John & Sriv-
age 30, so that there was little reason to expect age-linked person- astava, 1999; but see Ashton & Lee, 2007; Block, 1995). The Big
ality differences after this point in the life span (see Kelly, 1955). Five are Extraversion (traits like energetic and sociable), Agree-
In contemporary psychology, much of the interest in this question ableness (traits like considerate and kind), Conscientiousness
has centered around Costa and McCrae’s work on personality in (traits like hard-working and orderly), Neuroticism (traits like
adulthood (Helson, Kwan, John, & Jones, 2002; McCrae & Costa, nervous and tense), and Openness (traits like artistic and creative).
2003; Roberts, Walton, & Viechtbauer, 2006a; Roberts, Wood, & Mean-level differences in the Big Five across the life span were
Smith, 2005; Srivastava, John, Gosling, & Potter, 2003; Terrac- summarized by Roberts et al. (2006a), who compiled the results of
ciano, McCrae, Brant, & Costa, 2005). The present analyses con- 113 longitudinal samples involving 50,120 participants. Different
longitudinal studies examined different personality traits for vary-
ing periods of the life span, so the degree of comprehensiveness
varied for each of the traits they examined. Moreover, Roberts et
M. Brent Donnellan, Department of Psychology, Michigan State Uni-
al. drew on a distinction between two aspects of Extraversion:
versity; Richard E. Lucas, Department of Psychology, Michigan State
University, and the German Institute for Economic Research, Berlin,
traits related to independence and dominance (labeled social dom-
Germany. inance) versus traits related to positive affect, activity level, and
The data used in this article were made available by the German Institute sociability (labeled social vitality), following Helson and Kwan
for Economic Research and the UK Data Archive. The British Household (2000).
Panel Study data were originally collected by the Economic and Social Roberts et al. (2006a) found that average levels of social vitality
Research Council Research Centre on Micro-social Change at the Univer- tended to be fairly stable across the life span, although there was
sity of Essex, now incorporated within the Institute for Social and Eco- a slight spike from adolescence to the early 20s followed by
nomic Research. Neither the original collectors of the data nor the UK Data mean-level consistency from the mid-20s until the mid-50s, when
Archive bear any responsibility for the analyses or interpretations pre- there was a slight decline. In contrast, social dominance showed a
sented here. Preparation of the article was supported by National Institute
more pronounced and consistent increase from adolescence to the
on Aging Grants 1R03AG026028-01 and 1R03AG028744-01.
We thank Samuel Gosling and Sanjay Srivastava for providing us with
mid-30s, when mean levels remained consistent until the mid-50s.
the correlations between our scales and the full Big Five Inventory scales. Data on average levels of this trait were not available beyond this
Correspondence concerning this article should be addressed to M. Brent point in the life span because only seven studies examined changes
Donnellan, Department of Psychology, Michigan State University, East in social dominance for participants in their 50s and older. Agree-
Lansing, MI 48823. E-mail: donnel59@msu.edu ableness and Conscientiousness showed relatively gradual increases

558
 AGE DIFFERENCES IN PERSONALITY 559

in absolute scores across the life span, whereas Neuroticism examined age differences in all of the Big Five using nationally
showed relatively gradual decreases. Lastly, Openness showed a representative samples. To be sure, Costa et al. (1986) examined
mean-level increase from adolescence to the early 20s, and then cross-sectional age differences in Neuroticism, Extraversion, and
mean levels remained fairly consistent until the mid-50s, when Openness in a representative sample from the United States, and
average levels started to decline. Steunenberg, Twisk, Beekman, Deeg, and Kerkhof (2005) exam-
Terracciano et al. (2005) conducted cross-sectional and longi- ined changes in Neuroticism in a representative sample from the
tudinal analyses examining links between age and mean levels of Netherlands; however, neither study examined all five traits simul-
the Big Five. This study was completed too recently to be included taneously. Indeed, the absence of data from nationally representa-
in the Roberts et al. (2006a) meta-analysis. Terracciano et al. found tive samples is perhaps the most serious limitation of this entire
that scores on Extraversion generally declined from age 30 to 90, literature (see Roberts et al., 2006a, p. 20). For example, most of
although the drop in Extraversion was more pronounced after the the participants in Terracciano et al. (2005) were described as
mid-50s or so. Agreeableness demonstrated a fairly linear increase “generally healthy and highly educated” (p. 494). Also, few studies
with age, whereas the pattern for Conscientiousness was curvilin- included participants past the age of 70 (Terracciano et al., 2005).
ear: Scores increased up to a peak somewhere between the ages of The goal of this article is to extend research on age differences
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
This document is copyrighted by the American Psychological Association or one of its allied publishers.

50 and 70 and then declined. Average levels of Neuroticism in the Big Five using data from two large national panel studies
generally declined with age but increased slightly starting around that include participants past the age of 70. In addition, we explore
age 80. Finally, Openness showed a negative and linear association whether gender and education moderate cross-sectional age differ-
with age. In general, Terracciano et al. found similar cross- ences. Although there is persistent interest in gender differences in
sectional and longitudinal results with the exception that the cross- mean levels of personality (e.g., Chapman, Duberstein, Sörensen,
sectional zenith for Conscientiousness was around age 50 com- & Lyness, 2007; Costa, Terracciano, & McCrae, 2001; Feingold,
pared to around age 70 for the longitudinal analyses. 1994; Goldberg, Sweeney, Merenda, & Hughes, 1998), evidence
Age differences in the Big Five have also been identified in that gender moderates age differences in the Big Five is not
cross-cultural research. McCrae et al. (1999) used convenience generally robust (see Roberts et al., 2006a). Likewise, the existing
samples from Germany, Italy, Portugal, Croatia, and Korea and literature does not suggest that education effects will be large (e.g.,
found that self-reports of Extraversion and Openness were lower in Costa et al., 1986; Goldberg et al., 1998; Löckenhoff et al., in
older participants than younger participants, whereas Conscien- press). Thus, we view these analyses for gender and education as
tiousness and Agreeableness showed the reverse pattern. Results exploratory given the lack of clear trends in the existing literature.
were mixed for Neuroticism, as it was found to be lower in older
participants versus younger participants in Germany, Portugal, and
Method
Korea, whereas age differences were not statistically detectable in
Italy and Croatia. McCrae, Terracciano, and the Personality Pro- Samples
files of Cultures Project (2005) found that observers rated adults
(ages 40 –98) higher on measures of Conscientiousness but lower The British Household Panel Study (BHPS) is an ongoing panel
on measures of Extraversion and Openness when contrasted with study of British households that began in 1991. Households were
college students in research that included participants from 50 selected using a multistage probability design with systematic
countries. Age differences for Neuroticism and Agreeableness sampling, and all members of the household age 16 and older were
were moderated by gender: The negative association between age asked to participate. The sample was initially representative of the
and observer reports of Neuroticism was more evident for men population of the United Kingdom, though as would be expected,
than women, whereas the positive association between age and some attrition has occurred. Average annual attrition is relatively
observer reports of Agreeableness was more evident for women low, with about 5% of initial participants lost to attrition each year.
than men. Additional subsamples were recruited after 1991; the cumulative
In sum, the existing evidence broadly suggests that levels of attrition at the time of the Big Five assessment for these samples
Agreeableness and Conscientiousness are positively associated is thus lower than for the original sample.
with age, whereas levels of Extraversion and Openness are nega- The age range of the most recent wave of BHPS was 16 to 99
tively associated with age (see also Costa et al., 1986; Helson et years, but we trimmed the sample to participants who were less
al., 2002; Mroczek, Spiro, & Griffin, 2006; Srivastava et al., than 86 years old because of the sparse number of participants
2003). Average levels of Neuroticism are generally negatively (defined as n ⬍ 40) at older ages. The BHPS sample was approx-
associated with age, although there are exceptions in the literature imately 54% women, and the sample sizes used in these analyses
such that consistent age differences have not emerged in all coun- ranged from 14,039 to 14,055 depending on the Big Five trait. The
tries, and there are hints that this trait may increase around age 80. average age of the sample with complete Big Five data was 45.29
All in all, these broad trends translate to age-linked increases on years (SD ⫽ 18.04). Big Five measures were administered to the
Digman’s (1997) alpha factor and age-linked decreases on Dig- panel at the most recent wave (Wave 15), which was predomi-
man’s beta factor. In other words, as individuals grow older, they nately collected in September through December 2005 (approxi-
seem to increase on traits related to social interest and communion mately 94% of all Wave 15 participants). The rest of the data were
and decrease on traits related to agency and a zestful approach to collected before May 2006. Participants completed portions of the
life. survey in face-to-face interviews, though the Big Five measures
Despite the emergence of some relatively consistent age differ- were administered in a self-completion format.
ences in the Big Five, there are a couple of limitations of the The German Socio-Economic Panel Study (GSOEP) is an on-
existing literature that are worth noting. First, no single study has going study of German households that began in 1984 (see
 560 DONNELLAN AND LUCAS

Haisken-DeNew & Frick, 2005, for details). Households were of longer Big Five assessments, see Donnellan, Oswald, Baird, &
selected using a multistage random sampling technique, and all Lucas, 2006, and Rammstedt & John, 2007.)
members of the household age 16 and older were asked to partic- Mean differences between the samples might stem from artifacts
ipate. Like the BHPS, multiple samples were recruited in the years (e.g., differences in the response scales and item translations) as
since the study began, and each sample was initially representative well as real cross-cultural differences in personality. It is difficult
of the population from which it was drawn. Again, however, some to disentangle these issues with the present data, and these con-
attrition has occurred, which raises questions about the represen- cerns are not the focus of the present research. Accordingly, we
tativeness of the remaining sample. Attrition rates for the earliest calculated T scores within each sample to control for any mean-
sample are similar to those from the BHPS, with average yearly level differences across the two samples, following the logic used
attrition at around 6%. Again, it is important to note that although by McCrae et al. (1999). Scores were norm referenced to the group
the cumulative attrition for the original sample is substantial of participants age 30 to 34 within each sample, which facilitates
(which might affect its representativeness), a number of sub- an intuitive within-sample comparison: Scores below 50 indicate
samples have been added to the study much more recently. For trait levels that are lower than the reference group, whereas scores
above 50 indicate trait levels that are higher than the reference
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

instance, 44% of the sample that was used in this analysis was
This document is copyrighted by the American Psychological Association or one of its allied publishers.

recruited in 2000 or 2002. These subsamples have much lower group. Such differences can be interpreted against the overall T
rates of cumulative attrition because they have been in the study score SD of 10 for rough effect size calculations. For example, the
for much less time than the original sample. average score on Conscientiousness for individuals age 16 to 19
Big Five measures were administered to the panel at the most was 42.76 (SD ⫽ 10.97) in the BHPS and 41.49 (SD ⫽ 12.27) in
recent wave, which was collected in 2005. The age range of the the GSOEP (see Table 1). Both of these values indicate that late
most recent wave of the GSOEP was 16 to 95, but we trimmed adolescents scored more than seven tenths of a standard deviation
the sample to those participants who were less than 85, given the lower than individuals in their early 30s, a pattern that is consistent
sparse number of participants past this cutoff (i.e., n ⬍ 40). The with longitudinal research that has found increases on traits linked
GSOEP sample was approximately 52% women, and the sample with Conscientiousness during the transition from adolescence to
sizes used in these analyses ranged from 20,852 to 20,876, de- adulthood (e.g., Caspi, Roberts, & Shiner, 2005; Donnellan,
pending on the Big Five trait. The average age of the sample with Conger, & Burzette, 2007).
complete Big Five data was 46.03 years (SD ⫽ 17.23). Participants
completed the measure through an oral interview (roughly 26%),
Education
by a written questionnaire (roughly 50%), or by computer assisted
personal interviewing (CAPI; roughly 25%).1 We coded whether participants in each study had completed
basic education in their countries of origin. Following Hu, Stewart-
Brown, Twigg, and Weich (2007), we used whether or not partic-
Personality ipants had educational qualifications in the BHPS for this purpose.
Participants completed a 15-item version of the Big Five Inven- Individuals without qualifications left compulsory education in the
tory (BFI; John & Srivastava, 1999) using a 7-point scale (BHPS:
1 ⫽ does not apply, 7 ⫽ applies perfectly; GSOEP: 1 ⫽ does not 1
We examined whether format type was associated with scores on the
apply, 7 ⫽ does apply). Three items were used to measure each Big Five. It appeared that scores derived from both oral interviews and
dimension.2 Internal consistencies in the BHPS were as follows: CAPI formats were higher than written formats for some traits. Accord-
Extraversion (␣ ⫽ .54, average interitem r ⫽ .28), Agreeableness ingly, we collapsed interviews and CAPI administrations to contrast with
(␣ ⫽ .53, average interitem r ⫽ .28), Conscientiousness (␣ ⫽ .52, written administrations. The main effects of format type in the d metric
average interitem r ⫽ .28), Neuroticism (␣ ⫽ .68, average inter- were as follows: Extraversion: d ⫽ 0.03; Agreeableness: d ⫽ 0.20; Con-
item r ⫽ .41), and Openness (␣ ⫽ .67, average interitem r ⫽ .41). scientiousness: d ⫽ 0.24; Neuroticism: d ⫽ – 0.14; Openness: d ⫽ – 0.01
(positive values indicated interview formats were higher). It was also the
Internal consistencies in the GSOEP were as follows: Extraversion
case that older participants were more likely to use interview-based formats
(␣ ⫽ .66, average interitem r ⫽ .41), Agreeableness (␣ ⫽ .51,
versus written formats when compared to younger participants. The real
average interitem r ⫽ .28), Conscientiousness (␣ ⫽ .62, average question, however, was whether format type moderated age trends. We first
interitem r ⫽ .39), Neuroticism (␣ ⫽ .60, average interitem r ⫽ used an analysis of covariance model to control for format type (1 ⫽
.33), and Openness (␣ ⫽ .63, average interitem r ⫽ .37). interview or CAPI, 0 ⫽ written) for the means reported in Table 1. The
To establish the utility of these short scales, we correlated them predicted means were quite similar to those reported in Table 1, which
with the full BFI scales using data from the Gosling-Potter Internet suggests that differences in format type did not create major confounds
Personality Project (N ⫽ 628,640; see Srivastava et al., 2003). Our (table available upon request). We also conducted tests of interactions
3-item scales were strongly correlated with the full versions for all using the same regression strategy that we used for gender and education
five BFI scales (Extraversion: .90; Agreeableness: .88; Conscien- (see Results). Only one instance met our threshold for a meaningful
interaction—Conscientiousness. It appeared that age differences in Con-
tiousness: .88; Neuroticism: .89; Openness: .86), and they were
scientiousness after midlife were more pronounced for written administra-
strongly correlated with the remaining 5 to 7 items in each BFI tions than for interview-based administrations. Nonetheless, the shape of
scale that were not included in our short measure (Extraversion: the predicted age curves was quite similar.
.73; Agreeableness: .71; Conscientiousness: .73; Neuroticism: .70; 2
Extraversion was assessed with BFI Items 1, 6, and 36. Agreeableness
Openness: .70). On the basis of these results, we had confidence was assessed with Items 17, 32, and 37. Conscientiousness was assessed
that these brief measures were reasonable substitutes for the longer with Items 3, 23, and 33. Neuroticism was assessed with Items 9, 19, and
scales. (For additional evidence about the utility of shorter forms 39. Openness was assessed with Items 5, 20, and 30.
 AGE DIFFERENCES IN PERSONALITY 561

Table 1
Means and Standard Deviations for the Big Five T Scores by Age Categories

Minimum Extraversion Agreeableness Conscientiousness Neuroticism Openness

Age group sample size M (SD) M (SD) M (SD) M (SD) M (SD)

British Household Panel Study

16–19 1,007 53.01 (9.86) 48.61 (10.61) 42.76 (10.97) 50.47 (10.86) 50.45 (10.43)
20–29 2,216 51.58 (9.93) 50.00 (10.11) 47.88 (10.11) 50.10 (10.36) 51.08 (10.21)
30–39 2,590 49.70 (10.33) 50.43 (10.02) 50.35 (10.21) 49.92 (10.03) 49.79 (10.16)
40–49 2,625 48.54 (10.94) 50.91 (10.03) 50.82 (10.62) 49.39 (10.34) 48.64 (10.56)
50–59 2,220 47.47 (11.03) 51.32 (10.51) 50.80 (10.99) 48.99 (10.81) 48.06 (11.45)
60–69 1,697 46.98 (11.29) 50.98 (11.37) 49.24 (12.32) 47.87 (11.17) 46.28 (12.54)
70–79 1,250 45.56 (12.35) 51.43 (11.96) 47.20 (13.21) 46.25 (11.22) 44.27 (13.58)
80–85 434 45.41 (12.10) 51.44 (11.92) 46.77 (13.29) 46.52 (11.74) 42.47 (12.86)
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

German Socio-Economic Panel Study

This document is copyrighted by the American Psychological Association or one of its allied publishers.

16–19 1,344 51.17 (10.32) 49.64 (10.26) 41.49 (12.27) 48.80 (9.96) 51.75 (10.44)
20–29 2,835 50.94 (10.25) 49.65 (9.97) 47.15 (10.92) 49.99 (10.42) 51.46 (9.89)
30–39 3,745 50.12 (9.96) 49.79 (10.17) 50.22 (9.98) 50.04 (10.18) 50.23 (9.90)
40–49 4,275 49.84 (9.99) 50.31 (10.16) 51.22 (9.72) 50.36 (10.22) 50.15 (10.18)
50–59 3,271 49.08 (9.79) 50.21 (10.48) 51.16 (10.23) 51.10 (10.44) 50.43 (10.68)
60–69 3,293 48.27 (10.05) 50.56 (10.45) 50.23 (10.78) 51.51 (10.32) 49.43 (11.07)
70–79 1,683 47.54 (10.28) 52.46 (10.63) 50.46 (10.69) 51.38 (10.48) 47.66 (11.62)
80–84 403 47.57 (10.44) 54.16 (10.23) 49.84 (11.42) 50.74 (11.41) 45.56 (11.95)

Note. T scores were created by standardizing scores to the mean and standard deviation for individuals age 30 –34 within each sample.

British system without formal certification of their skills or voca- models) when the higher order term improved overall model fit at
tional training. We used whether or not participants had completed F ⬎ 25.00. We followed similar strategies for addressing questions
the equivalent of high school in the GSOEP, using data from the about the potential moderating effects of gender and education.
Cross-National Equivalent File, a dataset constructed from the
GSOEP responses to facilitate cross-national comparisons across
five panel studies conducted in the United States, United King-
Age Differences in the Big Five From Age 16
dom, Canada, Germany, and Australia. We restricted the exami- to the Mid-80s
nation of education-related differences to those participants who Table 1 displays average levels of personality traits for the eight
were 30 or older to avoid confusing education-linked differences age groups. To facilitate an intuitive understanding of the direction
with age-linked differences in personality. Approximately 23% of and magnitude of the age differences, we identified the age group
those 30 and older did not have any qualifications in the BHPS, associated with the maximum score for each trait and the age
whereas approximately 14% of those 30 and older did not com- group associated with the minimum score for each trait. We then
plete the equivalent of high school in the GSOEP. computed the simple difference between the two groups, which
yields a difference score that is expressed in age 30 –34 T units. We
Results considered differences of around 2 points as small differences,
differences of around 5 points as medium differences, and differ-
Overview of Analytic Strategy
ences above 8 points as large— designations that parallel the
We first created eight age groups (16- to 19-year-olds, 20- to conventions used for interpreting standardized mean differences
29-year-olds, 30- to 39-year-olds, 40- to 49-year-olds, 50- to (e.g., McCartney & Rosenthal, 2000).
59-year-olds, 60- to 69-year-olds, 70- to 79-year-olds, and indi- We first examined the pattern of age differences in the BHPS.
viduals over 80) to examine broad age trends in the Big Five. We Extraversion showed a linear pattern of age differences such that the
then used a hierarchical regression approach to formally model the youngest group scored the highest and the oldest group scored the
association between age and personality traits. Given the large lowest on this measure. This difference was nearly large (7.60 T
sample sizes, we placed a higher premium on the size of the effects units). Agreeableness showed the opposite linear pattern such that
and on obtaining replicable models than on statistical significance. the oldest group scored highest and the youngest group scored
This follows in the tradition of the “less is more” perspective used lowest; however, the difference was fairly small (2.83 T units).
by McCrae et al. (1999) and Srivastava et al. (2003). The most Conscientiousness showed a curvilinear pattern of age differences
complicated models that we evaluated involved cubic terms for such that there was a large difference (8.06 T units) between the
age, a decision motivated by an inspection of the age curves (see youngest age group and the 40- to 49-year-olds (the highest
Figure 1) and the fact that few other studies have used anything group), whereas there was a small to medium difference between
higher than cubic terms to model age–personality curves (e.g., the 40- to 49-year-olds and the oldest age group (4.05 T units).
Srivastava et al., 2003; Terracciano et al., 2005). Following Sriv- Neuroticism demonstrated a linear pattern of age differences
astava et al. (2003), we selected more complicated models (e.g., whereby the youngest group scored the highest and the 70- to
models with quadratic terms for age as opposed to simpler linear 79-year-olds scored the lowest (4.22 T units). Finally, Openness
 562 DONNELLAN AND LUCAS

Extraversion Agreeableness

60
60

55 55

T Score
T Score

50 50

45 45

40 40
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

16
21
26
31
36
41
46
51
56
61
66
71
76
81
16
21
26
31
36
41
46
51
56
61
66
71
76
81
This document is copyrighted by the American Psychological Association or one of its allied publishers.

Age Age

Conscientiousness Neuroticism

60 60

55 55
T Score
T Score

50 50

45 45

40 40
16
21
26
31
36
41
46
51
56
61
66
71
76
81
16
20
24
28
32
36
40
44
48
52
56
60
64
68
72
76
80

Age Age

Openness

60

55

GSOEP
T Score

50
BHPS
45

40
16
21
26
31
36
41
46
51
56
61
66
71
76
81

Age

Figure 1. Age differences in the Big Five.

showed a fairly linear pattern such that the 20- to 29-year-olds 79-year-old group scored the lowest for Extraversion; the differ-
scored the highest and the oldest group scored the lowest; the ence was small to medium (3.63 T units). The oldest group scored
difference was large by our conventions (8.61 T units). highest and the youngest group scored lowest for Agreeableness (a
Roughly similar patterns of age differences were observed in the difference of 4.52 T units). There was a large difference (9.73 T
GSOEP. The youngest group scored the highest and the 70- to units) between the youngest age group and the 40- to 49-year-olds
 AGE DIFFERENCES IN PERSONALITY 563

for Conscientiousness, whereas there was a slight difference be- Table 2

tween the 40- to 49-year-old group and the oldest group (1.38 T Regression Models Linking Age to the Big Five
units). Neuroticism showed a fairly linear increase in Germany
such that the youngest group scored the lowest and the 60- to Variable BHPS GSOEP
69-year-old group scored the highest. This difference was small Extraversion
(2.71 T units). Last, Openness showed a linear pattern of age Intercept 48.813 (.091) 49.526 (.070)
differences whereby the 16- to 19-year-olds scored the highest and Age ⫺.118 (.005) ⫺.063 (.004)
the oldest group scored the lowest; the difference was medium by R .19 .11
F for Linear 546.253 247.101
our conventions (6.19 T units). All in all, we concluded that the
⌬F for Quadratic 14.754 1.161
most notable difference between the BHPS and GSOEP was that ⌬F for Cubic 2.267 0.128
that Neuroticism trends were in the opposite directions. Even so, Agreeableness
there was a considerable degree of overlap in the broad patterns of Intercept 50.655 (.089) 49.946 (.099)
age differences across the two datasets. To be sure, Conscientious- Age .036 (.005) .041 (.004)
Age2 — .001 (.000)
ness appeared to be the personality domain with the largest repli-
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

R .06 .08
This document is copyrighted by the American Psychological Association or one of its allied publishers.

cable effect size when considering the difference between adoles- F for Linear 54.214 109.273
cent and middle-aged participants. ⌬F for Quadratic 12.481 36.543
We then modeled age differences using regression analyses; ⌬F for Cubic 4.168 19.039
Conscientiousness
results are reported in Table 2 and displayed in Figure 1. Age was
Intercept 51.252 (.133) 51.460 (.101)
mean-centered within each dataset before higher order terms were Age ⫺.019 (.012) ⫺.001 (.009)
calculated. The regression results mapped closely to the age trends Age2 ⫺.007 (.000) ⫺.007 (.000)
depicted in Table 1. That is, Extraversion and Openness scores Age3 .0001 (.000) .0002 (.000)
were predicted to be lower in older individuals compared to R .20 .24
F for Linear 31.280 449.358
younger individuals, whereas scores on Agreeableness were pre- ⌬F for Quadratic 483.736 596.991
dicted to be higher in older individuals compared to younger ⌬F for Cubic 52.636 156.908
individuals. Age differences in Conscientiousness had a predicted Neuroticism
curvilinear pattern such that average levels of this trait were Intercept 49.051 (.090) 50.541 (.071)
Age ⫺.067 (.005) .039 (.004)
highest in middle adulthood. Last, age difference in Neuroticism
R .11 .07
diverged in the two samples along the previously described lines: F for Linear 179.200 88.137
Older individuals were predicted to score lower on Neuroticism ⌬F for Quadratic 13.095 4.887
than younger individuals in the BHPS, whereas this pattern was ⌬F for Cubic 0.441 2.360
reversed in the GSOEP. Openness
Intercept 48.888 (.132) 50.284 (.101)
Age ⫺.114 (.005) ⫺.006 (.009)
Gender and Education Effects Age2 ⫺.002 (.000) ⫺.001 (.000)
Age3 — ⫺.0001 (.000)
We first examined overall gender differences in the Big Five R .20 .11
using d metric effect sizes that were scored so that positive scores F for Linear 525.857 199.951
indicated that women scored higher than men. These are displayed ⌬F for Quadratic 29.079 15.202
in Table 3. The effect sizes were generally similar in the BHPS and ⌬F for Cubic 0.323 40.586
the GSOEP. The one caveat was that the direction of the gender
Note. BHPS ⫽ British Household Panel Study; GSEOP ⫽ German
difference for Openness was reversed across the two datasets. In Socio-Economic Panel Study. Unstandardized coefficients reported. Stan-
addition, the overall gender differences were more or less consis- dard errors reported in parentheses. Age was mean-centered within each
tent with the effect sizes reported by Costa et al. (2001), given that sample (age in BHPS ⫽ 45.50; age in GSOEP ⫽ 46.09). More complicated
we found the biggest differences for Neuroticism, Agreeableness, models involving age polynomials were selected only when the inclusion
of the higher order term improved overall model fit at F ⬎ 25.00. Coef-
and Extraversion and relatively smaller effect sizes for Conscien-
ficients for age were reported from the final selected model within each
tiousness and Openness. We also calculated d metric effect sizes sample. See text for complete details and justification of the model selec-
within each of our eight age groups, and these differences are also tion strategy.
reported in Table 3. As seen in Table 3, there did not seem to be
striking evidence that effect sizes varied systematically across the
age groups in either dataset. 3. a model with gender, linear age, gender by linear age, and
We then formally evaluated whether gender moderated the quadratic age terms;
association between age and personality. For these analyses, we
followed a hierarchical strategy similar to the strategy we used to 4. a model with gender, linear age, gender by linear age,
model higher order terms involving nonlinear age effects. We quadratic age, and gender by quadratic age terms;
tested for interactions involving gender by modeling effects in six
steps: 5. a model with gender, linear age, gender by linear age,
quadratic age, gender by quadratic age, and cubic age
1. a model with gender and linear age terms; terms; and finally

2. a model with gender, linear age, and gender by linear age 6. a model that added the gender by cubic age term to the
terms; previous model.
 564 DONNELLAN AND LUCAS

Table 3
Effect Sizes for Gender and Education Differences by Age Categories

Extraversion Agreeableness Conscientiousness Neuroticism Openness

Age group BHPS GSOEP BHPS GSOEP BHPS GSOEP BHPS GSOEP BHPS GSOEP

Gender differences
16–19 .38 .24 .29 .31 .13 .34 .57 .47 ⫺.10 .36
20–29 .26 .13 .23 .30 .30 .13 .56 .47 ⫺.18 .17
30–39 .27 .18 .39 .26 .16 .12 .45 .45 ⫺.16 .12
40–49 .22 .23 .33 .37 .16 .13 .47 .35 ⫺.21 .19
50–59 .16 .17 .38 .39 .08 .06 .64 .30 ⫺.08 .09
60–69 .13 .14 .35 .42 .07 .09 .47 .37 ⫺.14 .08
70–79 .08 ⫺.01 .19 .38 ⫺.08 .03 .51 .42 ⫺.15 ⫺.09
80–85/84 .06 .18 .22 .23 ⫺.17 ⫺.04 .39 .37 ⫺.12 ⫺.03
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

Overall .20 .16 .31 .35 .11 .11 .51 .39 ⫺.15 .12
This document is copyrighted by the American Psychological Association or one of its allied publishers.

Education differences
30–39 .20 .18 ⫺.03 ⫺.11 .17 .07 ⫺.16 ⫺.27 .51 .19
40–49 .27 .18 .04 .06 .34 .21 ⫺.36 ⫺.20 .62 .42
50–59 .11 .26 ⫺.05 ⫺.07 .14 .11 ⫺.20 ⫺.25 .52 .41
60–69 .10 .23 ⫺.16 ⫺.12 .05 .18 ⫺.24 ⫺.23 .54 .46
70–79 .07 .20 ⫺.14 ⫺.01 .07 .16 ⫺.19 ⫺.27 .37 .47
80–85/84 .03 .09 ⫺.09 .08 .07 .22 ⫺.20 ⫺.23 .54 .30
Overall .22 .16 ⫺.10 ⫺.06 .20 .22 ⫺.10 ⫺.17 .60 .32

Note. BHPS ⫽ British Household Panel Study; GSEOP ⫽ German Socio-Economic Panel Study. Effect sizes for gender were calculated so that positive
scores indicated that women scored higher than men. Effect sizes for education were calculated so that positive scores indicated that more-educated
individuals scored higher than less-educated individuals.

We were interested in whether the gender by age interaction terms same general strategy outlined for testing for gender effects (i.e.,
improved overall model fit at F ⬎ 25.00 for Models 2, 4, and 6 we replaced gender and gender by age interaction terms with
compared with Models 1, 3, and 5, respectively. None of the F education and education by age interaction terms). None of the
change statistics for the relevant comparisons suggested that gen- relevant model comparisons suggested that education acted as a
der acted as a moderator of age effects according to our standards. compelling moderator of age differences. This was the case in both
This was the case for both the BHPS and the GSEOP datasets. the BHPS and the GSEOP datasets.
These results seem consistent with the conclusion drawn by Rob-
erts et al. (2006a) that “there is very little support for the idea that General Discussion
men and women change in distinct ways” (p. 15).
We followed a similar approach to examine questions about To our knowledge, this is the first study to examine age differ-
education. Recall that these analyses were restricted to individuals ences in all of the Big Five from age 16 to the mid-80s using two
who were 30 or older to cover the era in the life span when most large national datasets from Britain and Germany. In both coun-
people in both countries have completed formal schooling. Effect tries, there is evidence that Extraversion and Openness are nega-
sizes were calculated so that positive scores indicated that indi- tively associated with age, whereas Agreeableness is positively
viduals with more education scored higher than individuals with associated with age. Likewise, mean levels of Conscientiousness
less education. These effect sizes were roughly similar in the two were highest for middle-aged participants in both Britain and
datasets, as seen in Table 3, and the largest overall effect sizes Germany. In general, cross-sectional age differences in the Big
were for Openness. As a point of comparison, we calculated d Five were detectable past age 30, which seems inconsistent with
metric effect sizes from Table 1 in Löckenhoff et al. (in press), the idea that personality traits are completely fixed at some point
who compared individuals who completed more than 12 years of in the life span (i.e., the “hard plaster” hypothesis; see also Sriv-
education to those who did not complete at least 12 years of astava et al., 2003). In addition, there did not appear to be consis-
education in a sample from the United States. The d metric effect tent evidence that gender or education level moderated cross-
sizes in that report were 0.09, ⫺0.13, 0.20, ⫺0.28, and 0.57 for sectional age differences. We now comment on several of the more
Extraversion, Agreeableness, Conscientiousness, Neuroticism, and notable results.
Openness, respectively. Thus, Löckenhoff et al. (in press) found One of our more interesting findings concerned the curvilinear
the most pronounced difference for Openness, which was consis- association between age and Conscientiousness. The association
tent with our results. between age and Conscientiousness is often broadly characterized
We also calculated d metric effect sizes within each of our age as linear in the existing literature. However, Terracciano et al.
groups (see Table 3), and the education effects looked generally (2005) also reported a curvilinear association between age and this
similar across age groups in both datasets. Last, we formally trait. We also found that scores on Conscientiousness had a cur-
evaluated whether education status moderated the association be- vilinear association with age such that the biggest age differences
tween age and personality. For these analyses, we followed the were found when comparing average levels for late adolescents
 AGE DIFFERENCES IN PERSONALITY 565

with average levels of middle-aged participants. It even appeared oldest group was very small when considered in T units. On the
that average levels were lower in the oldest adults when compared other hand, the general trend fails to replicate the broad trends in
to middle-aged participants. One explanation for any apparent the existing literature. Future cross-sectional work in Germany
discrepancies between these results for Conscientiousness and the may benefit from using a longer measure of Neuroticism that
existing literature is that few studies have actually examined this assesses its lower order facets to provide a more nuanced under-
association past age 60 using analytic strategies that are capable of standing of this issue. Moreover, future work using samples from
detecting nonlinear effects. That is, Srivastava et al. (2003) did not other nations is needed to examine other potential cross-national
include adults older than 60 in their study, and the strategy of differences in the association between age and Neuroticism.
comparing groups of older and younger adults (e.g., McCrae et al., In addition to this puzzle, there are at least three important
1999) or older and very old adults (e.g., Weiss et al., 2005) does limitations of the present work. The first has to do with the
not permit an evaluation of nonlinear associations. Thus, our representativeness of the oldest members of the samples. The issue
results fill an important gap in the existing literature and point to is that the oldest individuals in the BHPS and GSOEP are actually
a more nuanced association between age and Conscientiousness. select members of their respective birth cohorts given that they
Future work is needed to resolve the discrepancies between the have lived longer than what would be considered typical (see
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
This document is copyrighted by the American Psychological Association or one of its allied publishers.

cross-sectional results and the longitudinal results of Terracciano Hofer & Sliwinski, 2006). Such a select subsample may not be
et al. (2005) showing that the peak average level of Conscientious- ideal for drawing inferences about normative levels of traits for
ness is near age 70. older individuals. This fact is one of the major conceptual and
Given that we found replicable age-linked differences for four methodological issues facing researchers interested in aging and
out of the five Big Five traits, a natural question is why such age personality—the oldest participants (e.g., those in their 70s) in
differences exist. There are currently two dominant explanations current samples are somewhat exceptional in terms of their lon-
for age differences in personality traits—the intrinsic maturation gevity, and it is possible that they may be exceptional in terms of
perspective and the life course perspective. The intrinsic matura- their personalities.
tional argument holds that normative age-related changes in per- The second limitation is the relatively low internal consistency
sonality adulthood are driven by preprogrammed biological pro- of our measures. In our favor, the items on these short scales were
cesses (e.g., Costa & McCrae, 2006), whereas the life course drawn from a reliable and well-validated parent instrument. Like-
argument posits that the major roles of adult life involving occu- wise, it is useful to bear in mind that there are constraints on the
pational pursuits, romantic relations, and parenthood drive adult length of assessments in these large panel studies, and extensive
personality development (e.g., Helson et al., 2002; Roberts et al., personality assessments are not always possible. However, it is
2005). As we have argued elsewhere (Donnellan et al., 2007), it is important to be precise about the consequences of measurement
ultimately difficult to conduct crucial tests of these two explana- error: Measurement error attenuates the ability to find differences,
tions because true experiments are neither feasible nor ethical in which would limit our ability to find systematic age trends. The
the realm of adult personality development. fact that many of our results replicated previous research perhaps
Fortunately, there are alternatives to true experiments for par- mitigates concerns over this issue.
tially resolving this debate. Evidence of replicable associations The more serious final limitation is the cross-sectional design of
between role changes and personality changes would favor the life this study. We readily acknowledge that cross-sectional studies are
course perspective over a strict intrinsic maturation explanation. unable to disentangle age effects from cohort effects (e.g., Costa &
However, there is much controversy over the conclusiveness of the McCrae, 1982). Future longitudinal and cohort-sequential studies
existing evidence for associations between personality changes are clearly needed to address these issues. On the other hand, the
and adult life experiences or role transitions (see, e.g., the ex- current evidence in favor of strong cohort effects on the Big Five
change between Costa & McCrae, 2006, and Roberts et al., is generally mixed and inconclusive (see McCrae & Costa, 2003,
2006b). Thus, more work is needed, and the general interest in p. 80). For example, meta-analytic findings by Twenge (2000)
resolving this debate may be helpful for convincing those in charge suggest that there are cohort-linked increases in Neuroticism, but
of the BHPS and GSOEP to administer measures of the Big Five these findings are inconclusive because they are based on conve-
in future assessments. As it stands, the important contribution of nience samples (i.e., nonprobability samples; see Donnellan &
the present study is that we have presented clear evidence that (a) Trzesniewski, 2008), and these effects have not always replicated
age differences in the Big Five are detectable in large national (e.g., McCrae & Costa, 2003; Roberts et al., 2006a; Terracciano et
datasets; (b) the pattern of age differences is similar in Britain and al., 2005). Nonetheless, it is possible that different sociohistorical
Germany, with the noteworthy exception of Neuroticism; and (c) factors in Britain and Germany might explain the divergent pat-
neither gender nor education seem to consistently moderate cross- terns for Neuroticism.
sectional age differences across the two datasets. In closing, we believe that the present findings are noteworthy
because the age trends were derived from two large panel studies
Puzzles, Limitations, and Conclusions and were generally consistent across the two countries. At present,
we think it is safe to conclude that there are real age differences in
This investigation yielded at least one puzzling finding— personality and many of these differences generalize to broad
namely the inconsistent results for Neuroticism in Britain and in populations of individuals in Western countries. At this point,
Germany. In particular, the results for the German sample were Aristotle appears to have been right and William James appears to
somewhat exceptional given that older individuals were found to have been wrong: Normative personality differences exist after age
score higher on this trait than younger individuals. On the one 30, and there is good reason to think that detectable absolute
hand, the absolute difference between the youngest group and the changes in personality occur across the life span.
 566 DONNELLAN AND LUCAS

References James, W. (1892/1985). Psychology: The briefer course. Notre Dame, IN:
University of Notre Dame.
Ashton, M. C., & Lee, K. (2007). Empirical, theoretical, and practical John, O. P., & Srivastava, S. (1999). The Big Five trait taxonomy: History,
advantages of the HEXACO model of personality structure. Personality measurement, and theoretical perspectives. In L. A. Pervin & O. P. John
and Social Psychology Review, 11, 150 –166. (Eds.), Handbook of personality: Theory and research (2nd ed., pp.
Block, J. (1995). A contrarian view of the Five-Factor approach to per- 102–138). New York: Guilford Press.
sonality description. Psychological Bulletin, 117, 187–215. Kelly, E. L. (1955). Consistency of the adult personality. American Psy-
Caspi, A., Roberts, B. W., & Shiner, R. L. (2005). Personality develop- chologist, 10, 659 – 681.
ment: Stability and change. Annual Review of Psychology, 56, 453– 484. Löckenhoff, C. E., Terracciano, A., Bienvenu, O. J., Patriciu, N. S.,
Chapman, B. P., Duberstein, P. R., Sörensen, S., & Lyness, J. M. (2007). Nestadt, G., McCrae, R. R., et al. (in press). Ethnicity, education, and the
Gender differences in Five Factor Model personality traits in an elderly temporal stability of personality traits in the East Baltimore Epidemio-
cohort. Personality and Individual Differences, 43, 1594 –1603. logic Catchment Area study. Journal of Research in Personality.
Costa, P. T., Jr., & McCrae, R. R. (1982). An approach to the attribution of McCartney, K., & Rosenthal, R. (2000). Effect size, practical importance,
aging, period, and cohort effects. Psychological Bulletin, 92, 238 –250. and social policy for children. Child Development, 71, 173–180.
Costa, P. T., Jr., & McCrae, R. R. (2006). Age changes in personality and McCrae, R. R., & Costa, P. T., Jr. (2003). Personality in adulthood: A
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

five-factor theory perspective (2nd ed.). New York: Guilford Press.

This document is copyrighted by the American Psychological Association or one of its allied publishers.

their origins: Comment on Roberts, Walton, and Viechtbauer (2006).

Psychological Bulletin, 132, 26 –28. McCrae, R. R., Costa, P. T., Jr., Pedroso de Lima, M., Simões, A.,
Costa, P. T., Jr., McCrae, R. R., Zonderman, A. B., Barbano, H. E., Ostendorf, F., Angleitner, A., et al. (1999). Age differences in person-
Lebowitz, B., & Larson, D. M. (1986). Cross-sectional studies of per- ality across the adult life span: Parallels in five cultures. Developmental
Psychology, 35, 466 – 477.
sonality in a national sample: 2. Stability in neuroticism, extraversion,
McCrae, R. R., Terracciano, A., & the Personality Profiles of Cultures
and openness. Psychology and Aging, 1, 144 –149.
Project. (2005). Universal features of personality traits from the observ-
Costa, P. T., Jr., Terracciano, A., & McCrae, R. R. (2001). Gender
er’s perspective: Data from 50 cultures. Journal of Personality and
differences in personality traits across cultures: Robust and surprising
Social Psychology, 88, 547–561.
findings. Journal of Personality and Social Psychology, 81, 322–331.
Mroczek, D. K., Spiro, A., III, & Griffin, P. W. (2006). Personality and
Digman, J. M. (1997). Higher-order factors of the Big Five. Journal of
aging. In J. E. Birren and K. W. Schaie (Eds.), Handbook of the
Personality and Social Psychology, 73, 1246 –1256.
psychology of aging (6th ed., pp. 363–377). New York: Academic Press.
Donnellan, M. B., Conger, R. D., & Burzette, R. G. (2007). Personality
Rammstedt, B., & John, O. P. (2007). Measuring personality in one minute
development from late adolescence to young adulthood: Differential
or less: A 10-item short version of the Big Five Inventory in English and
stability, normative maturity, and evidence for the maturity–stability
German. Journal of Research in Personality, 41, 203–212.
hypothesis. Journal of Personality, 75, 453– 472.
Roberts, B. W., Walton, K. E., & Viechtbauer, W. (2006a). Patterns of
Donnellan, M. B., Oswald, F. L., Baird, B. M., & Lucas, R. E. (2006). The
mean-level change in personality traits across the life course: A meta-
Mini-IPIP scales: Tiny-yet-effective measures of the Big Five factors of analysis of longitudinal studies. Psychological Bulletin, 132, 1–25.
personality. Psychological Assessment, 18, 192–203. Roberts, B. W., Walton, K. E., & Viechtbauer, W. (2006b). Personality
Donnellan, M. B., & Trzesniewski, K. H. (2008). How should we study traits change in adulthood: Reply to Costa and McCrae (2006). Psycho-
generational change— or should we? A critical examination of the logical Bulletin, 132, 29 –32.
evidence for “Generation Me.” Unpublished manuscript. Roberts, B. W., Wood, D., & Smith, J. L. (2005). Evaluating the five factor
Feingold, A. (1994). Gender differences in personality: A meta-analysis. theory and social investment perspective on personality trait develop-
Psychological Bulletin, 116, 429 – 456. ment. Journal of Research in Personality, 39, 166 –184.
Goldberg, L. R., Sweeney, D., Merenda, P. F., & Hughes, J. E., Jr. (1998). Srivastava, S., John, O. P., Gosling, S. D., & Potter, J. (2003). Develop-
Demographic variables and personality: The effects of gender, age, ment of personality in early and middle adulthood: Set like plaster or
education, and ethnic/racial status on self-descriptions of personality persistent change? Journal of Personality and Social Psychology, 84,
attributes. Personality and Individual Differences, 24, 393– 403. 1041–1053.
Haisken-DeNew, J. P., & Frick, R. (2005). Desktop companion to the Steunenberg, B., Twisk, J. W. R., Beekman, A. T. F., Deeg, D. J. H., &
German Socio-Economic Panel Study (GSOEP). Berlin, Germany: Ger- Kerkhof, A. J. F. M. (2005). Stability and change of neuroticism in
man Institute for Economic Research (DIW). aging. Journal of Gerontology: Psychological Sciences, 60B, P27–P33.
Helson, R., & Kwan, V. S. Y. (2000). Personality development in adulthood: Terracciano, A., McCrae, R. R., Brant, L. J., & Costa, P. T., Jr. (2005).
The broad picture and processes in one longitudinal sample. In S. Hampson Hierarchical linear modeling analyses of the NEO-PI-R scales in the
(Ed.), Advances in personality psychology (Vol. 1, pp. 77–106). London: Baltimore Longitudinal Study of Aging. Psychology and Aging, 20,
Routledge. 493–506.
Helson, R., Kwan, V. S. Y., John, O. P., & Jones, C. (2002). The growing Twenge, J. M. (2000). The age of anxiety? Birth cohort change in anxiety
evidence for personality change in adulthood: Findings from research and neuroticism, 1952–1993. Journal of Personality and Social Psychol-
with personality inventories. Journal of Research in Personality, 36, ogy, 79, 1007–1021.
287–306. Weiss, A., Costa, P. T., Jr., Karuza, J., Duberstein, P. R., Friedman, B., &
Hofer, S. M., & Sliwinski, M. J. (2006). Design and analysis of longitu- McCrae, R. R. (2005). Cross-sectional age differences in personality
dinal studies on aging. In J. E. Birren and K. W. Schaie (Eds.), Hand- among Medicare patients aged 65 to 100. Psychology and Aging, 20,
book of the psychology of aging (6th ed., pp. 15–37). New York: 182–185.
Academic Press.
Hu, Y., Stewart-Brown, S., Twigg, L., & Weich, S. (2007). Can the Received June 17, 2007
12-item General Health Questionnaire be used to measure positive Revision received May 13, 2008
mental health? Psychological Medicine, 37, 1005–1013. Accepted May 23, 2008 䡲