JOURNAL OF RESEARCH IN PERSONALITY

ARTICLE NO.

32, 431453 (1998)

RP982225

Heritabilities of Common and Measure-Specific Components

of the Big Five Personality Factors
John C. Loehlin
University of Texas at Austin

Robert R. McCrae and Paul T. Costa, Jr.

Gerontology Research Center, National Institute on Aging, National Institutes of Health

and
Oliver P. John
University of California, Berkeley
Three different measures of the Big Five personality dimensions were developed
from the battery of questionnaires used in the National Merit Twin Study: one from
trait self-rating scales, one from personality inventory items, and one from an adjective check list. Behavior-genetic models were fit to what the three measures had in
common, and to the variance distinctive to each. The results of the model fitting
agreed with other recent studies in showing the Big Five dimensions to be substantially and about equally heritable, with little or no contribution of shared family
environment. Heritabilities for males and females did not differ significantly. For
Agreeableness and Conscientiousness, some effect of shared environment was found
for measure-specific variance on the personality inventory, and for Extraversion and
Neuroticism, models involving nonadditive genetic variance or twin contrast effects
provided slightly better fits. 1998 Academic Press
Key Words: Big Five; heredity; environment; twins; model fitting; sex differences.

In recent years, there has been considerable interest in the so-called Big
Five dimensions of personality, which were originally derived from the
vast array of trait terms used in natural languages, and which appear to have
We thank Lewis R. Goldberg and Robert C. Nichols for their helpful comments on an earlier
version of the manuscript.
Correspondence and reprint requests should be addressed to J. C. Loehlin, Department of
Psychology, University of Texas, Austin, TX 78712. E-mail: loehlin@psy.utexas.edu.
431
0092-6566/98 $25.00

Copyright 1998 by Academic Press

All rights of reproduction in any form reserved.

432

LOEHLIN ET AL.

considerable generality across measuring instruments, languages, and methods of analysis (for reviews, see Digman, 1990; John, 1990; and McCrae,
1989). The names given to these dimensions sometimes differ: we will refer
to them as Extraversion (E), Agreeableness (A), Conscientiousness (C), Neuroticism (N), and Openness (O).
Although not all psychologists are enamored of the Big Five (e.g., Block,
1995), and many others prefer to work at different levels of detail or with
other types of constructs than traits, there has been sufficient interest in the
Big Five to justify looking at genetic and environmental contributions to
their variation. There have been several recent assessments of the extent to
which genes or family environments account for individual differences on
the Big Five dimensions. Most of these come to the conclusion that all five
are moderately and about equally heritable, and that shared family environmental factors play a minor role, if any, in accounting for individual differences on them.
Loehlin (1992) summarized the then-existing evidence from twin, adoption, and family studies for a variety of questionnaire scales that he classified
according to the Big Five. He fit behavior-genetic models to the available
correlations from the literature, and concluded that under various assumptions the genes contributed from 28% to 59% of the variance of the Big Five
traits, and the shared environment of family members 0% to 11%. Among
the studies included in the model fitting, the only one that included explicit
measures of A,C, and O, a Swedish study of elderly twins, produced typical
results for C and O, but a different result for A. The genetic contribution
to measured Agreeableness in this study, estimated at 12%, did not differ
significantly from zero, and the family environment contribution, estimated
at 21%, was statistically significant (Bergeman et al., 1993).
A recent heritability analysis of the Big Five dimensions as measured by
the Revised NEO Personality Inventory of Costa and McCrae (1992) used
a total of 660 monozygotic (MZ) pairs and 380 dizygotic (DZ) pairs from
pooled Canadian and German twin samples (Jang, McCrae, Angleitner, Riemann, & Livesley, 1998). For all five traits a simple model involving only
additive genes and nonshared environment fit the data. Estimates for the
heritabilities of factor scales for E, A, C, N, and O were .50, .48, .49, .49,
and .48, respectively. These estimates are obviously very similarin particular, the heritability of Agreeableness is quite in line with the rest.
Another recent twin study, this time in California, measured 313 MZ pairs
and 91 DZ pairs from a volunteer community sample (Waller, in press).
Waller used seven factor scales, five of them corresponding fairly closely
to the Big Five. The heritabilities were .49, .33, .46, .42, and .58, for scales
corresponding to E, A, C, N, and O. Again, Agreeableness shows clear evidence of heritability, although it is a little lower than the others. Evidence

HERITABILITIES OF THE BIG FIVE

433

for shared family environmental effects was not found for Agreeableness,
being present only for Neuroticism, and there not large (12%).
In the German study, ratings were obtained from two peers of the twins
as well as from the twins themselves. The authors carried out a joint model
fitting to the self- and peer-report data (Riemann, Angleitner, & Strelau,
1997). From this, the estimates of the genetic contribution to the Big Five
dimensions were .60, .57, .71, .61, and .81, respectively, with the effects of
shared environment again being negligible. The estimates of genetic effects
are substantial for each of the Big Five dimensions. Furthermore, the heritabilities were higher than those typical for twin self-report data alone. Apparently, the common element distilled from various perspectives on a given
individual was more heritable than that unique to a particular viewpoint.
Global and Specific Levels of Analysis
Most models of personality are hierarchical, with broad factors (such as
the Big Five) defined by more narrow and specific traits. Heritability can
be studied at either level. Loehlin (1992) fit data for two components of
ExtraversionDominance and Sociabilityfrom several twin, family, and
adoption studies. Similar estimates were obtained for the additive effects of
genes (36% in each case) but different estimates for nonadditive genes (22%
for Dominance, 8% for Sociability).
In the combined German-Canadian study, the authors were interested in
examining what was distinctive to various facets of each Big Five dimension,
in addition to what the facets had in common (Jang, et al., 1998). To this
end the authors fit models to the residual facet scores in each domain, after
regressing out the common factors. For the most part, these residual scores
still fit a genes-and-unshared-environment model, but with lower heritabilitiestypically in the .20s and .30s. One of the reasons for the lower heritability estimates is no doubt the relatively large proportion of variance in these
residual scores that represents measurement error. However, the residuals
from a couple of facets in the Agreeableness domain, Altruism and Modesty,
appeared to fit a shared environment model, as did Achievement Striving
and Deliberation from the Conscientiousness domain.
The present study looks at archival data from a large U.S. twin sample
from the perspective of the Big Five. These were data that had been gathered
prior to the widespread use of this approach to describing personality. The
sample consisted of the 839 twin pairs of the National Merit twin study, as
described in Loehlin and Nichols (1976). Thus this study allows us to extend
heritability estimates for the Big Five to a somewhat different population
and age group: college-bound U.S. high school seniors in the early 1960s.
Moreover, it proved possible to estimate the Big Five dimensions separately
from three different sources within this data set: (1) from a series of 47

434

LOEHLIN ET AL.

bipolar trait self-ratings, (2) from the items of a standard personality questionnaire, the California Psychological Inventory (CPI, Gough, 1957), and
(3) from a version of Goughs Adjective Check List (ACL, Gough & Heilbrun, 1965). Thus we are in position to examine for each trait both the heritability of the factor common to these different response modes, and the heritabilities of the distinctive contributions of each mode. Although all three
response modes (trait ratings, CPI, and ACL) were designed to measure the
five factors at a global level, they differ somewhat in the specific component
traits they emphasize, so the analysis of specific variance in each mode is
in some respects like Jang et al.s analysis of specific facet variance. The
three modes also differ in their approach to measurement, and in that respect
our design resembles the analysis of variance common and distinctive to
self- and peer-ratings. Most basically, these analyses address the question
of whether different indicators of a global trait are interchangeable from a
behavior genetics viewpoint, or whether results are likely to vary depending
on the indicator used.
METHOD
Participants
The sample was drawn from among the roughly 600,000 high school juniors who took the
National Merit Scholarship Qualifying Test in 1962. An item on the face sheet asked the testtaker if he or she was a twin. Pairs of individuals answering yes to this question who were
from the same high school, of the same sex, and had the same surname and home address
were matched by computer. They were then sent a brief questionnaire confirming that they
were indeed twins and inviting them to participate further in the study, as well as asking
questions about physical resemblance, mistaken identity, and the like, which could be used
to diagnose the pair as MZ or DZ. Those agreeing to participate were individually sent a
battery of questionnaires assessing their personalities, activities, interests, and values, which
they returned by mail. A parent of the twins responded to a questionnaire about the twins
early behavior and experiences. Further details and copies of the questionnaires may be found
in Loehlin and Nichols (1976). Most results to be reported are from 807 pairs of twins, 490
diagnosed as MZ and 317 as DZ.

Measures
Trait ratings. The questionnaire filled out by the twins contained a set of 47 self-rating
scales. Each was a 7-point scale identified by a word or phrase at each end, such as Messy
Neat, PopularUnpopular, ReligiousNonreligious. The testee was instructed to circle
one of the seven numbers along the scale, which were successively labeled Very, Fairly,
and Slightly at each end, with Neither or Both in the middle.
Four expert judges 1 experienced with the Big Five classified the 47 scales into up to two
categories each (including a residual category Beyond the Big Five). Scales were developed
based on these classifications. In most cases, there was unanimous agreement among the four

The judges, in addition to the fourth author, were Lewis R. Goldberg, John M. Digman,
and Maureen Barckley. We are grateful for their expert assistance.

HERITABILITIES OF THE BIG FIVE

435

TABLE 1
Big Five Trait-Ratings Scales, with Sample Items
Extraversion (5 items; .75)
ShyOutgoing a
TimidBold a
Agreeableness (6 items; .63)
Easily angeredGood-natured a
Kind a Cruel
Conscientiousness (9 items; .80)
Responsible a Irresponsible
Dependable a Undependable
Neuroticism (7 items; .68)
HappyUnhappy a
Well-adjustedMaladjusted a
Openness (3 items; .20)
Original a Unoriginal
Sophisticated a Unsophisticated
a

Direction of response scored for the trait.

judges on the classification of the ratings used for each scale. Ratings on the subscales were
individually standardized and summed to yield Big Five scale scores.
The scales are described in Table 1. (The sample items given in this and the following
tables are the two having the highest item-total correlations.) Considering their relative brevity,
four of the five scales have reasonable alpha reliabilities. The fifth, Openness, had very few
rating items judged to represent it, and its low of .20 in part reflects the shortness of this
three-item scale.
CPI scales. Included with questionnaire battery sent to the National Merit twins was a
standard personality inventory, the California Psychological Inventory (CPI). Big Five scales
were developed for the CPI based on a sample of 348 adults from the Baltimore Longitudinal
Study of Aging (BLSA; Shock et al. 1984) who had been given both the NEO-PI and the
CPI. The sample included 153 men aged 27 to 92, and 195 women aged 19 to 89 (McCrae,
Costa, & Piedmont, 1993).
For each of the five factors, each CPI item was first correlated with the NEO-PI criterion
factor within gender. Only items that were significantly correlated ( p .05) with the factor
in both groups were retained. In the next step, these items were correlated with all five factors
in the combined sample, and items that correlated more highly with one of the other factors
were eliminated, to yield nonoverlapping CPI scales. These final scales contained between 16
and 75 items, and had reliabilities between .72 and .93, as shown in Table 2. As might be
expected, the reliabilities tend to vary inversely with scale length, with the lowest occurring
for the 16-item scale for Agreeableness.
Some evidence on the reasonableness of these Big Five CPI scales is given by correlations
with other personality measures collected in the BLSA, such as the MyersBriggs Type Indicator (MBTI, Myers & McCaulley, 1985) and the Jackson (1984) Personality Research Form
(PRF). Correlations with the MBTI corresponded very closely both in pattern and size of
correlations to those obtained in an earlier study using the NEO-PI (McCrae & Costa, 1989).
Correlations with the PRF scales were also generally similar to those obtained earlier with
the NEO-PI (Costa & McCrae, 1988), although those involving the A and C scales tended to
be somewhat smaller than the correlations obtained in that study. Further details on the sam-

436

LOEHLIN ET AL.

TABLE 2
Big Five CPI Scales, with Sample Items
Extraversion (51 items; .92)
I am a good mixer (T)
It is hard for me to find anything to talk about when I meet a new person (F)
Agreeableness (16 items; .72)
I have one or more bad habits that are so strong that it is no use fighting against
them (F)
I am often said to be hotheaded (F)
Conscientiousness (29 items; .82)
I set a high standard for myself and I feel others should do the same (T)
I always try to do at least a little better than what is expected of me (T)
Neuroticism (75 items; .93)
I certainly feel useless at times (T)
I often get disgusted with myself (T)
Openness (42 items; .83)
Disobedience to any government is never justified (F)
I read at least ten books a year (T)
Note. Keyed response in parentheses after each item. CPI California Psychological
Inventory.

ples, scale construction, and correlations with other measures are available from the second
author.
Adjective scales. A 159-item abbreviated version of Gough and Heilbruns (1965) Adjective
Check List (ACL) was part of the battery of questionnaires filled out by the twins. In a study
by John and Roberts (1993), adjectives from the full ACL were classified according to the
Big Five dimensions. From the ACL adjectives identified by John and Roberts as belonging
to one of the Big Five factors, those included in the version filled out by the twins were taken
to define the Big Five adjective scales. Further procedural details may be obtained from the
fourth author.
The tendency of an individual to mark many or few adjectives is a pervasive artifact in
adjective check list studies. Therefore, total endorsement frequency, the number of items out
of the 159 that each individual endorsed as self-descriptive, was removed from the raw scores
by regression.
The properties of the Big Five adjective checklist scales are shown in Table 3. Except for
A, the scale lengths lie between those for the CPI scales and the trait self-rating scales, and
for the most part the alpha reliabilities are intermediate as well. Note that there is some direct
content overlap between the trait-rating and the adjective scales. Several of the end-point labels
of the trait scales, for example, Outgoing, Responsible, and Original, also appear as adjectives
scored on the checklist scales.
Intelligence and Openness. There has sometimes been discussion of the fifth of the Big
Five as a possible manifestation of intelligence in the personality domain. In our sample
there was a substantial correlation between the National Merit Scholarship Qualifying Test
(NMSQT) score and Openness as measured by the CPI items (r .50). Correlations of the
NMSQT with the other two Openness scales were also positive, but lower (.22 and .08, for
adjectives and self-ratings, respectively). Because we wished to emphasize the status of Openness as a personality dimension, overall intellectual ability, as measured by NMSQT total
score, was removed from the three Openness scores by regression.

437

HERITABILITIES OF THE BIG FIVE

TABLE 3
Big Five Adjective Checklist Scales,
with Sample Items
Extraversion (21 items; .72)
Outgoing (T)
Talkative (T)
Agreeableness (16 items; .75)
Helpful (T)
Pleasant (T)
Conscientiousness (23 items; .81)
Responsible (T)
Efficient (T)
Neuroticism (20 items; .81)
Worrying (T)
Tense (T)
Openness (12 items; .65)
Original (T)
Imaginative (T)
Note. Keyed response in parentheses
after each item.

It is recognized that something may be lost by doing thisthe observed personality trait
of Openness has legitimate intellectual aspects, and in some sense one might be removing
too much by regressing intelligence out. Nevertheless, for our purposes the gain in clarity
justifies this step. Much is known concerning the heritability of intelligence, but less concerning the personality side of Openness.

RESULTS
Simple Correlations
Table 4 gives MZ and DZ twin correlations for the Big Five measures
from the three data sources. It will be noted that the MZ correlations are
uniformly higher than the DZ correlations, suggesting a genetic contribution
TABLE 4
MZ and DZ Twin Correlations for the Three Sets of Big Five Scale Scores
Ratings

CPI

ACL

Factor

MZ

DZ

MZ

DZ

MZ

DZ

Extraversion
Agreeableness
Conscientiousness
Neuroticism
Openness
Mean

.47
.32
.42
.43
.39
.41

.01
.06
.21
.17
.19
.13

.60
.46
.53
.53
.49
.52

.30
.34
.34
.25
.27
.30

.39
.29
.37
.44
.36
.37

.06
.18
.14
.06
.08
.08

438

LOEHLIN ET AL.

to these measures. For the CPI scales, the DZ correlations are typically about
half as large as the MZ correlations, consistent with a model of additive
genes and little effect of shared family environmentthe common finding
in earlier studies. The DZ correlations for the trait self-rating scales and the
adjective check list are often considerably less than half the MZ correlations,
suggesting the possible presence of some additional factor such as nonadditive genetic variance or contrast effects. However, the measures in Table 4
contain both common and unique factors, and the model-fitting to be undertaken will show more clearly whether the data are consistent with simple
models or require more complex ones.
Factor Analysis of Measures
To verify that the 15 measures did indeed define five factors corresponding
to the Big Five, a preliminary factor analysis was carried out, based on the
1614 individuals in the sample. Five principal factors were extracted from
a correlation matrix, using squared multiple correlations as communality estimates, and rotated by Varimax. The resulting factor loadings are shown in
Table 5.
On the whole, the factor loadings correspond to those expected for Big
Five factors (boldface type). There are a few fairly weak defining loadings,
but all are over .30; for every factor, at least two of the three exceed .50.
TABLE 5
Factor Analysis of the 15 Measures
Factor
Measure

1(E)

2(A)

3(C)

4(N)

5(O)

Rating-E
Rating-A
Rating-C
Rating-N
Rating-O
CPI-E
CPI-A
CPI-C
CPI-N
CPI-O
ACL-E
ACL-A
ACL-C
ACL-N
ACL-O

.90
.00
.23
.29
.15
.73
.20
.04
.12
.08
.75
.05
.01
.20
.01

.02
.85
.26
.47
.04
.00
.32
.06
.21
.05
.04
.60
.11
.45
.00

.06
.16
.81
.22
.11
.11
.13
.55
.24
.11
.02
.07
.65
.17
.02

.02
.06
.10
.35
.03
.08
.34
.21
.71
.06
.04
.21
.27
.53
.05

.14
.01
.09
.04
.62
.21
.08
.06
.16
.41
.03
.16
.01
.10
.59

Note. Five principal factors extracted from correlation matrix with squared multiple correlations as communalities, and rotated via Varimax. Identifying loadings in boldface.

HERITABILITIES OF THE BIG FIVE

439

There are three off-pattern loadings of .30 or more, all involving a negative
relationship between A and N. This probably reflects the presence of an
evaluative component, as A tends to be the most positively evaluated of the
Big Five, and N the most negative.
Model Fitting
It is usual in behavior-genetic analyses of complex multivariate data to
fit explicit models to covariance matrices among the variables involved. This
permits one to ask whether the data are consistent with simple models, or
whether more complex models are required. If models are found that do fit
the data, estimates of their parameters are available, from which estimates
of the relative contributions of genes and environments to the variation of
the trait may be obtained (Neale & Cardon, 1992).
The model fitting was based on three measures for each of the Big Five
dimensions. One of these measures was from the trait ratings, one from the
CPI, and one from the ACL, obtained as described in the preceding section.
All three measures were standardized over the total set of individuals in the
sample, and covariance matrices calculated separately for MZ and DZ pairs,
using these standardized scores. (Had the scores been separately standardized
for the MZs and DZs and for twins 1 and 2 these would be correlation matrices, but we did not do this in order to retain the statistical advantages of
covariance matrices for model fitting.)
The model fitting was done separately for each of the Big Five dimensions
because they are in theory independent. The data to be fit in each case consist
of two six-variable covariance matrices, one for MZ twins and one for DZ
twins, containing the three measures for the first twin of each pair followed
by the same three measures for the second twin. (The covariance matrices
are given in the Appendix.) As a first step, a traditional twin model involving
additive genes (h), shared or common environment (c), and unshared environment (e), was fit to what the three scales measuring the factor had in
common, and to the residual variance distinctive to each type of scale.
A path diagram of the model is given in Fig. 1. One twin, indicated by
subscript 1, is represented in the lefthand part of the figure, the other, by
subscript 2, on the right. The top part of the figure represents the genetic
and environmental causation of the latent Big Five factor (labeled F). The
factor is a latent trait indexed by the three measures R (for ratings), Q (for
questionnaire items), and A (for adjectives). At the top of the figure are the
genotypes, G, connected to F via a path h, and correlated across twins 1.0
for MZs and .5 for DZs. (The value of .5 for DZs assumes that the genetic
variance in G is purely additive and the twins parents mated randomly with
respect to this trait. Assortative mating tends to be fairly slight for personality
traits [e.g., Price & Vandenberg, 1980]; possible nonadditive genetic effects
will be examined later.) Also at the top of the figure are shown the twins

440

LOEHLIN ET AL.

FIG. 1. Path model of common and measure-distinctive genetic and environmental causation of a Big Five factor and scales measuring it. (Circles latent variables, squares manifest variables; straight arrows causes, curved arrows correlations; 1,2 the two twins
of a pair; F Big Five factor; G, C, E additive genes, common environment, unshared
environment; R, Q, A trait ratings, questionnaire items, adjective scales; lower case letters corresponding causal paths. Only a few paths are labeled in the bottom part of the
figure; analogous labels apply for the rest.)

shared environments C, connected to the latent trait by a path c and correlated

(by definition) 1.0 between the twins. A third path to F, e, accounts for the
residual variation, E, reflecting environmental events, pre- or postnatal, that
are unique to the particular twin, plus genotype-environment interaction
but not measurement error, because F is a latent trait. Effects general to the
testing situation but specific to the individual, such as current mood, could
also be part of E.
The observed measures are shown in the squares, labeled R for the trait
ratings scale, Q for the CPI items scale, and A for the ACL scale. They
are connected to the factor via the paths r, q, and a, which represent factor
loadingsthe extent to which the common factor F contributes to the particular observed scores.
The bottom part of the figure represents the causation of the specific parts
of each of the observed measures of the trait, i.e., those aspects independent
of the common Big Five factor. Below the squares are Gs, representing the
genetic contribution to the specific part; Cs, representing the shared environ-

441

HERITABILITIES OF THE BIG FIVE

TABLE 6
Genetic and Environmental Components of the Variance of the Latent Big Five
Dimensions, from Fitting the Simple h, c, e Model
Dimension

h2

c2

e2

Extraversion
Agreeableness
Conscientiousness
Neuroticism
Openness
Mean

.57
.51
.52
.58
.56
.55

.00
.00
.00
.00
.00
.00

.44
.49
.48
.42
.44
.45

95% CI for h2
.50
.42
.43
.51
.45

to
to
to
to
to

.64
.61
.61
.66
.68

Note. h2 genetic variance; c2 shared environmental variance; e2 nonshared environmental variance. h, c, e corresponding paths. CI confidence interval. 95% CI is approximate, obtained as (h 1.965SEh)2, from model with c fixed to zero.

mental contribution; and Es, representing unique environmental contributions, genotype-environment interactions, situational factors, and, this time,
measurement error (because R, Q, and A are observed scores). Corresponding Gs are again assumed correlated 1.0 or .5 across twins, depending on
zygosity, and the Cs correlated 1.0 by definition. The specific factors for the
three measures are represented as mutually independent. This is so by definition within twins. There are logically possible scenarios whereby specifics
for one measure for one twin might become correlated with specifics for a
different measure for the other twin, but our modeling assumes that if any
such effects exist, they are trivial enough to be ignored without seriously
distorting the analysis.
The model was fit to the five pairs of MZ and DZ covariance matrices
using the computer program LISREL (version 8Joreskog and Sorbom,
1993). The common factor variance was constrained to equal 1.0, and the
variances of G, C, and so on were fixed to 1.0. The variances of the observed
variables are close to 1.0 because of the preliminary standardization over the
combined groups, so the various estimates may be interpreted as standardized
values, to a reasonable approximation.
The common factors. The variance contributions for the five latent factors
are shown in Table 6. For genes and shared environment they are the squares
of the h and c path values; the residual variances e 2 were obtained as such
in the solution.
As is evident from the table, these values proved to be quite similar across
the five factors: They range from .51 to .58 for h 2, from .42 to .49 for e 2,
and are uniformly estimated as zero for c 2 the minimum value the model
permits. This model, then, suggests that, on average, about 55% of individual
personality trait variation in this population is associated with genetic differences, 45% with environmental events unique to each twin, situational fac-

442

LOEHLIN ET AL.

tors, or genotype-environment interaction, and none to shared environment.

Moreover, these contributions are quite similar across the Big Five factors,
including an Openness factor that is independent of measured intelligence.
(Out of curiosity, we also fit the model to the unpartialed data for Openness.
The result was a somewhat poorer overall fit, a higher h 2, 69%, a lower e 2,
32%, and c 2 still zero.) Because the analysis is being made for latent variables, the unique environmental component will not include random errors
of measurement, but, as noted, it could include transient factors such as mood
that might influence a given individual differently from his or her twin during
the period of filling out the questionnaire battery.
Even with these fairly large twin samples the parameter estimates are not
extremely precise. Approximate 95% confidence intervals for h 2 are given
in the table, obtained via the standard errors estimated by the fitting program
(see table footnote for details). Obviously, these confidence intervals would
be consistent with modest differences in heritability among the Big Five
factors, although (particularly in conjunction with similar results in other
twin studies) they suggest that any very large differences of this kind are
unlikely.
Table 7 provides a breakdown for each of the three measures into the
variance attributable to the common factor and that to the specifics of the
measuring instrument, with the specific part divided into the contributions
of the genes, the shared environment, and a residual including unique environment, situational factors, genotype-environment interaction, and error.
The first column of Table 7 gives the squares of the paths from the latent
variable to each of the three measures. These indicate the extent to which
the variance of each measure reflects the common factor, as opposed to specifics and error. Across the Big Five, the trait rating and adjective scales
tend to share more in common and the CPI scales less. This is perhaps not
surprising, because of the overlap between the trait ratings and the adjective
check listrecall that many of the trait rating scales are anchored by descriptive adjectives, some of which also occur in the ACL. The fact that the loadings of the CPI scales are lower presumably reflects distinctive variance in
these scales, not error, because the scales are fairly reliable. The three openness scales tend to have lower path values than the others. In this case, this
may well represent a somewhat greater proportion of measurement error,
given the rather short scales. (It was not a result of the partialing out of
ability variancethe unpartialed version did not differ greatly in this respect,
with values of .30, .17, and .53.)
The measure-specific factors. The righthand portion of Table 7 contains
the heredity-environment analyses for the specifics. More often than not,
there are nonnegligible genetic components to the specific variancesfor
Conscientiousness, Neuroticism, and Openness for the trait ratings; for all
but Agreeableness for the CPI; for Extraversion, Conscientiousness, and

443

HERITABILITIES OF THE BIG FIVE

TABLE 7
Genetic and Environmental Components of the Variance Specific to Particular Measures of
the Latent Big Five Dimensions, from Fitting the Simple h, c, e, Model
Specific part
Dimension & Measure
Extraversion
Ratings
Questionnaire
Adjectives
Agreeableness
Ratings
Questionnaire
Adjectives
Conscientiousness
Ratings
Questionnaire
Adjectives
Neuroticism
Ratings
Questionnaire
Adjectives
Openness
Ratings
Questionnaire
Adjectives

General part

h2

c2

e2

.87
.53
.57

.00
.15
.08

.00
.04
.03

.15
.24
.36

.52
.23
.53

.04
.01
.00

.00
.30
.03

.42
.46
.44

.50
.29
.59

.17
.15
.09

.00
.18
.00

.30
.36
.33

.52
.41
.70

.15
.24
.02

.00
.01
.00

.34
.33
.30

.37
.14
.42

.22
.36
.10

.00
.04
.00

.42
.45
.48

Note. h2 genetic variance; c2 shared environmental variance; e2 nonshared environmental variance.

Openness for the ACL. (We are considering as nonnegligible those components contributing 5% or more to the total variance of a measure; typically,
these are also significant at the .05 level by a differential 2 test.) By contrast
to the common factors, there are some nontrivial shared environmental factors among the specificsnone for the trait ratings or the ACL, but for
Agreeableness and Conscientiousness among the CPI scales. There tend to
be substantial e 2s throughout for the specifics; it will be recalled that these
now include errors of measurement.
We can test whether the specifics truly differ in composition across the
three measures by fitting models constraining the hs and cs to be equal across
measures, and seeing if the goodness-of-fit chi square rises significantly. Indeed it does, in all five cases: the hs and cs the of the specifics differ across
measures.
Sex differences. Could better fits be obtained by fitting the model separately for males and females? Covariance matrices were calculated for the
two sexes separately, and the basic h, c, e model was fit twice, once with

444

LOEHLIN ET AL.

the 14 free parameters of the model allowed to differ between the sexes, and
once constrained to be the same. Chi square tests of the difference in fit were
not statistically significant for A, C, or O, suggesting that sex differences
were not important for these three Big Five dimensions.
The overall chi square tests were significant for E and N, indicating the
presence of sex differences. However, in both cases, the parameters of greatest interesth, c, and e for the latent traitcould be equated across the two
sexes without a significant rise in 2. This also proved to be the case for the
hs, cs, and es for the specifics. Thus the significant sex differences appeared
to lie in the factor loadings. In the case of E, for r, q, and a, respectively,
these were .98, .72, and .79 for males and .85, .76, and .70 for females (with
the rest of the parameters equated across the sexes). For N the corresponding
values were .67, .67, and .72 for males and .76, .60, and .88 for females.
Obviously, in all cases the common factor is expressed substantially in all
three indicators, so no gross differences in interpretation are called for. For
males, the ratings seem to carry more weight for E, and for females, the
ACL for N. Any more elaborate interpretation of these sex differences should
probably await their replication in other samples.
Alternative models. Even though the simple model of Fig. 1 fits fairly well
by tests to be discussed in the next section, the presence of lower-bound
estimates (zeroes) for the shared environment may be resulting from DZ twin
correlations less than one-half MZ twin correlations, a situation in which
models that include nonadditive genetic variance or contrasts between twins
may provide better fits. Two such alternative models are considered. One
introduces a nonadditive genetic parameter (i) in lieu of shared environment.
This was modeled as epistasis (interactions involving several gene loci) by
replacing the Cs at the top of Fig. 1 by Is, and setting the correlation between
the epistatic genes (I) of the two twins to 1.0 for MZs and zero for DZs.
This represents the hypothesis that these traits may be emergenic, in the
sense of Lykken (1982). Models that represented the nonadditive genetic
variation as due to dominance rather than epistasis were also fitted, but are
not separately reported. The dominance and epistasis models were pretty
much equivalent in terms of overall fits and estimates of broad heritability,
although the dominance models tended to place more of the genetic variance
in the nonadditive category. For a discussion of some other possible genetic
models, see Eaves (1988).
The contrast effects model retained a shared environment C, but added a
contrast parameter b between the two twins, representing the possibility that
each twin might be affected in his judgment of himself (or in his actual
development) by the other twins status on the trait. If this parameter is positive, the effect is one of assimilation, making the twins more alike; if negative, the effect is one of contrast, making the twins less alike. This model
isnt quite as straightforward to report as the others, because it implies vari-

HERITABILITIES OF THE BIG FIVE

445

ances that differ somewhat between MZs and DZs because of covariances
involving the additional path. We present results standardized on the basis
of the DZs, who are expected to be more like the general population.
The various alternative models were applied only at the level of the general
factor, although in principle one could fit such models at the level of the
specifics as well. (Readers wishing to explore such alternatives can do so
using the covariance matrices given in the Appendix.)
Tests of the model fitting. The results from fitting the three models to the
data from each of the five factors are shown in Table 8. Information about
overall model fit is given in the three rightmost columns of the table. In the
first of these columns is the goodness-of-fit chi square, which allows a test
of the null hypothesis that the model fits exactly in the population. Next is
the root mean square error of approximation (RMSEA), an index that estimates the goodness of approximation to a true model in the population. One
rule of thumb for interpreting the RMSEA is that values below .10 represent
a good fit, values below .05 a very good fit (Steiger, 1989, p. 81). By this
criterion, all the model fits are good, and all but those for E are very good.
Because the sample sizes are large, we can reject the null hypothesis of an
exact fit for E or A for any of the models (by a 2 test with 27 or 28 df,
p .05). However, exact fits for C, N, or O cannot be ruled out. The final
column of the table gives the upper limit of the 90% confidence interval for
RMSEA, which allows us to test the null hypothesis that the model provides
a poor approximation in the population (i.e., RMSEA .10). Mostly, we
can reject this null hypothesis of poor fit. Only in the case of E do values
of RMSEA above .10 fall within the 90% confidence interval; for the other
four factors we can be 90% confident that the fit in the population is a good
one for any of the three models.
Can we distinguish among the fits of the three alternative models? In the
case of the h,c,e,b model, the difference in 2s between it and the h,c,e model
can be tested directly as a chi square with 1 df. In the case of the h,i,e model,
the difference from an h,e model, which will have the same chi square as
the h,c,e model because of the zero c parameter, can be similarly tested.
Such tests indicate that for Extraversion and Neuroticism either of the two
alternatives fits significantly better than the original model; this is not true
for A, C, or O.
In short, one could justify interpreting any of the three models for any of
the five factors, although one might prefer one of the two alternative models
for E and N. How do these differ in their results from the simple h,c,e model?
In general, the h,i,e model allows some of the genetic variance to shift from
additive (h 2 ) to nonadditive (i 2 )the bulk of it in the case of E, about half
of it for N, and lesser amounts for A, C, and O. The total amount of genetic
variance, in the column h 2b (the so-called broad heritability) stays about
the same.

h,c,e
h,i,e
h,c,e,b
h,c,e
h,i,e
h,c,e,b
h,c,e
h,i,e
h,c,e,b
h,c,e
h,i,e
h,c,e,b
h,c,e
h,i,e
h,c,e,b

.57
.04
.82
.51
.36
.58
.52
.40
.51
.58
.31
.74
.56
.36
.53

h
.00
.00
.00
.00
.00
.13
.00
.00
.22
.00
.00
.00
.00
.00
.38

.00
.57
.00
.00
.17
.00
.00
.13
.00
.00
.30
.00
.00
.22
.00

b
.00
.00
.18
.00
.00
.12
.00
.00
.15
.00
.00
.10
.00
.00
.22

Estimates

.43
.39
.23
.49
.47
.35
.48
.47
.32
.42
.39
.31
.44
.41
.22

e2
.57
.61
.82
.51
.53
.58
.52
.53
.51
.58
.61
.74
.56
.58
.53

h b2
125.38
101.62
99.28
48.88
47.68
47.21
38.21
37.44
36.82
39.39
34.59
34.72
32.00
30.48
29.29

2
.09
.08
.08
.04
.04
.04
.03
.03
.03
.03
.02
.03
.02
.02
.01

RMSEA
.11
.10
.10
.06
.06
.06
.05
.05
.05
.05
.05
.05
.04
.04
.04

UL

Note. E, A, C, N, O are the Big Five factors Extraversion, Agreeableness, Conscientiousness, Neuroticism, and Openness. The three models
are, respectively, a simple model with genes (h), common environment (c), and unshared environment (e); a model with nonadditive genetic effects
(i), and a model with contrast effects (b). 2 goodness-of-fit chi square for the model in question. RMSEA root mean square error of approximation. UL upper limit of 90% confidence interval for RMSEA.

Model

Factor

TABLE 8
Results from Fitting Three Alternative Models to the Data

446
LOEHLIN ET AL.

HERITABILITIES OF THE BIG FIVE

447

The h,c,e,b model estimates a twin interaction parameter b. It proves to

be small and negative in the case of all five factors, consistent with a modest
effect of contrast between the twins. The admission of this parameter still
leaves the c 2 estimate at zero for Extraversion and Neuroticism, but allows
for a positive estimate of the effect of shared family environment (13% to
38%) for Agreeableness, Conscientiousness, and Openness. The estimate of
the broad heritability drops a bit for O, rises for E, A, and N, and stays about
the same for C. Thus there is more difference in heritability across the five
factors with the third model than with the first two: highest for Extraversion
and Neuroticism (82% and 74%), intermediate for Agreeableness (58%), and
lowest, though still substantial, for Conscientiousness and Openness (51%
and 53%). Note that for this model the totals of genetic and environmental
variances exceed 1.0. This is because negative covariance terms involving
the path b combine with the simple totals in a complete variance accounting.
But it also means that some caution should be exercised in interpreting heritabilities from this modelthey are one step more hypothetical than in the
other two.
DISCUSSION
There is a simple model of genetic and environmental influences that often
fits behavior-genetic data on personality: genes plus individual environment,
with little or no effect of shared environment. In the large National Merit
twin sample, this model provides an acceptable fit to the common factor
underlying the three measures of each of the Big Five traits. The estimated
path values under this model imply that 51% to 58% of individual difference
variation along the Big Five dimensions is genetic in origin, 42% to 49% is
due to experience unique to the individual, to temporary situational factors,
and to gene-environment interaction, and none is due to effects of environment shared by the twins.
We can complicate this view a little. When we look at the variance distinctive to each of the measures, as opposed to what they have in common, there
seem to be shared effects of the twins environments for the CPI measures
of Agreeableness and Conscientiousness. The latter is consistent with results
from a different approach using these same data, which found a significant
shared environmental component for a norm-favoring scale of the CPI
(Loehlin & Gough, 1990). These results are also in line with the finding of
shared environment effects in four residual facet scales of the Revised NEO
(Jang et al., in press). Two such facets lay in the Agreeableness domain
(Altruism and Modesty), two in the Conscientiousness domain (Deliberation
and Achievement Striving). Note that the CPI Big Five measures of the present study were derived via the NEO. Also consistent with the results of the
Jang et al. study are significant genetic components to the measure-specific
variance in several of the domains. In the present study, this might in part

448

LOEHLIN ET AL.

reflect differential involvement of the different facets in the different measures.

Another way we can complicate the simple view is by looking at models
that incorporate nonadditive genetic effects or interactions between the
twins. For A, C, and O neither of these alternative models produced a significant improvement in fit. For Extraversion, either one produced a significantly better fit, although still not an outstandingly good one. For Neuroticism, either alternative model led to a significant improvement of an already
very good fit. The nonadditive genetic model assumed the presence of epistasis: interactions among multiple gene loci in their effects on a trait. The
modeling suggested that perhaps half of the genetic effect on N and most
of that on E could be of this type. The twin interaction model resulted in a
small intertwin path that was negative for all five factors, in the range .10
to .22. The effect was one of contrast, an exaggeration of the difference
between the twins.
The two alternative models tended to be about equally effective. They
may be considered to represent alternative interpretations of essentially the
same facts: DZ correlations less than half MZ correlations. This ambiguity
could perhaps be most directly addressed by data from DZ twins reared apart.
For them, correlations could be made lower by epistasis but not by the twins
being contrasted with each other. Some data on DZ twins reared apart exist
for the traits of Extraversion and Neuroticism. For Extraversion, the data
tend to support a nonadditive genetic rather than a contrast model. DZ twins
who had been reared apart from an early age still showed correlations considerably less than half those of MZ twins. This was true in three studies: one
in Finland (Langinvainio, Kaprio, Koskenvuo, & Lonnqvist, 1984), one in
Sweden (Pedersen, Plomin, McClearn, & Friberg, 1988), and one in Minnesota (Tellegen, Lykken, Bouchard, Wilcox, Segal, & Rich, 1988).
The situation for Neuroticism is less clear. In the Finnish study, DZs reared
apart were found to be correlated less than half as much as MZs, and to be
correlated much like DZs reared together. Both results are consistent with
a hypothesis of nonadditive genetic effects. However, the other studies did
not find DZ correlations for Neuroticism to be less than half MZ correlations
to begin with, rendering moot an appeal to alternative models to explain such
a difference. In the Swedish study, the correlation for DZ twins reared apart
was slightly higher than that for DZ twins reared together (.28 vs .24), in
line with a contrast effect, but the difference was in the opposite direction
in the Minnesota study (.29 and .41). No general conclusion concerning alternative models seems justified for this trait.
The differences in h 2 and c 2 among the measure-specific components appear to be real, at least between the CPI measure and the others. The checklist
and trait self-ratings scales were derived by workers in one of the two major
Big Five traditions (Saucier & Goldberg, 1996), the so-called lexical tradi-

HERITABILITIES OF THE BIG FIVE

449

tion, which focuses on the Big Five as dimensions of person description in

natural languages. The other tradition emphasizes the Big Five as fundamental dimensions of personality, and is mainly centered on work with the NEOPI of Costa and McCrae, from which the CPI scales of the present study
were derived. Thus we may have evidence of differences between these approaches at the measure-specific level, along with a consistent core.
The analysis by males and females separately suggested that sex differences in either general or specific heritabilities were not of substantial importance, but measurement differences might be for N and E.
In summary, if we take what various measures of the Big Five have in
common, and fit simple behavior-genetic models, we obtain a simple result:
genes accounting for something over half of individual differences along all
five dimensions, with the rest presumably due to the effects of environmental
inputs that are distinctive to each individual, temporary situational effects,
and genotype-environment interactions (the last three in unknown proportions). The effect of environmental variables that act in the same way on
both twins is estimated as zero for all five domains. We must emphasize that
this does not mean that environment is unimportant for the development of
personality, or as a source of individual differences. It merely means that
whatever happens to individuals that makes a lasting difference is mostly
independent of their families, or depends on their genes, or has effects that
are unique to the individual.
For two of the Big Five, Extraversion and Neuroticism, more complex
models involving gene-gene or twin-twin interactions fit somewhat better,
although our data are not sufficient to distinguish between these alternatives.
Moreover, the aspects of the Big Five measures beyond their common core
are to some extent independently influenced by the genes, and in the case of
the CPI scales for Agreeableness and Conscientiousness, perhaps by shared
environments also.
What does this tell us about the optimum level at which to carry out
behavior-genetic analyses of personality? This issue was considered by
Loehlin (1992, chapter 4). He concluded that analyses below the level of
the Big Five contribute significant added information. The finding in the
present study of differences between common and measure-specific variance,
and the finding of Jang et al. (1998) of differences among residualized Big
Five facets, provide further evidence for this. Recent studies, including this
one, make a case for saying that all five broad factors are substantially heritable and largely unaffected by shared environmental influences. However, the
Big Five factors, or at least commonly-used measures of them, appear not
to be monolithic in how they are influenced by the genes and the environment. Further research could profitably turn to a closer examination of specific subtraits, using multiple measures and observers.
Finally, it should not be forgotten that our results derive from a particular

450

LOEHLIN ET AL.

populationtwin pairs from among the U.S. high school juniors who took
the National Merit Scholarship Qualifying Test in the year 1962. If this study
stood alone, the generality of its results might be in question. Since it does
not, this specificity may be considered a virtue rather than a limitation. In
particular, these data were gathered some decades earlier than those of other
Big Five twin studies, and from a population homogeneous in age and education. Insofar as its results agree with the others, the overall generality of
the findings increase. Further confirmation using nontwin behavior genetic
methods, such as adoption or family studies, remains desirable.

APPENDIX A1
Covariance Matrices for Big Five Factors for MZ Twins
(below Diagonal) and DZ Twins (Above Diagonal)

TR1
CPI1
ACL1
TR2
CPI2
ACL2

TR1

CPI1

.998

.6670
1.048

.968
.6545
.6260
.4507
.4452
.3890
.992

TR1
CPI1
ACL1
TR2
CPI2
ACL2

.910
.3761
.5234
.2915
.2361
.2504
1.104

TR1
CPI1
ACL1
TR2
CPI2
ACL2

.874
.3548
.5298
.3860
.2665
.2214
.996

TR1
CPI1
ACL1
TR2
CPI2
ACL2

1.004
.4715
.5694
.4152
.3493
.3613
1.076

TR1
CPI1
ACL1
TR2
CPI2
ACL2

.946
.2364
.3514
.3762
.1091
.1921

.956
.4960
.4503
.5822
.3860
.2376
.742
1.038
.3532
.1673
.4800
.1959
.4403
1.047
.976
.4176
.2189
.5249
.2087
.4151
.950
.992
.5029
.2331
.5280
.3204
.2214
.953
.984
.2526
.0798
.4857
.1267

ACL1

TR2

Extraversion
.7706
.0140
.5614
.1691
1.107
.0773
1.130
.877
.3049
.964
.3274
.6819
.3727
.6379
Agreeableness
.4922
.0607
.2803
.1020
1.025
.1681
1.148
.993
.2482
.930
.2514
.3473
.2897
.4983
Conscientiousness
.5426
.2230
.5505
.1088
1.003
.1566
1.100
1.038
.2423
.963
.2749
.3672
.3821
.5157
Neuroticism
.6523
.1730
.5302
.1602
1.063
.1127
1.068
.992
.2720
.920
.3792
.4655
.4431
.5540
Openness
.4477
.1971
.3157
.0537
1.092
.0803
1.054
1.039
.1897
.972
.2477
.1786
.3586
.3709

CPI2

ACL2

.1048
.3165
.0317
.8250
1.072

.1279
.0206
.0626
.7634
.5529
1.082

.972
.5736
.1035
.3065
.0845
.3907
1.114
1.056
.3857
.1487
.3448
.1580
.4778
1.003
.991
.4021
.0537
.2438
.1098
.4951
1.103
1.018
.5955
.1113
.2772
.1329
.2834
1.068
1.000
.2374

TR1
CPI1
ACL1
TR2
CPI2
ACL2

1.021
.0537
.1013
.1872
.6120
.3683
.999

TR1
CPI1
ACL1
TR2
CPI2
ACL2

.985
.0729
.0557
.1323
.5807
.3988
.911

TR1
CPI1
ACL1
TR2
CPI2
ACL2

1.027
.0974
.1127
.0560
.6402
.5168
.966

TR1
CPI1
ACL1
TR2
CPI2
ACL2

1.034
.0881
.0199
.0788
.4259
.1867
.929

TR1
CPI1
ACL1
TR2
CPI2
ACL2

.944

Note. TR trait ratings, CPI personality inventory items, ACL adjective check list;
1,2 twins.

452

LOEHLIN ET AL.

REFERENCES
Bergeman, C. S., Chipuer, H. M., Plomin, R., Pedersen, N. L., McClearn, G. E., Nesselroade,
J. R., Costa, P. T., Jr., & McCrae, R. R. (1993). Genetic and environmental effects on
Openness to Experience, Agreeableness, and Conscientiousness: An adoption/twin study.
Journal of Personality, 61, 159179.
Block, J. (1995). A contrarian view of the five-factor approach to personality description.
Psychological Bulletin, 117, 187215.
Costa, P. T., Jr., & McCrae, R. R. (1988). From catalog to classification: Murrays needs and
the five-factor model. Journal of Personality and Social Psychology, 55, 258265.
Costa, P. T., Jr., & McCrea, R. R. (1992). Revised NEO Personality Inventory (NEO-PI-R) and
NEO Five-Factor Inventory (NEO-FFI) professional manual. Odessa, FL: Psychological
Assessment Resources, Inc.
Digman, J. M. (1990). Personality structure: Emergence of the five-factor model. Annual Review of Psychology, 41, 417440.
Eaves, L. J. (1988). Dominance alone is not enough. Behavior Genetics, 18, 2733.
Gough, H. G. (1957). CPI manual. Palo Alto, CA.: Consulting Psychologists Press.
Gough, H. G., & Heilbrun, A. B., Jr. (1965). The Adjective Check List manual. Palo Alto,
CA: Consulting Psychologists Press.
Jackson, D. N. (1984). Personality Research Form manual (3rd. ed.). Port Huron, MI: Research Psychologists Press.
Jang, K. L., McCrae, R. R., Angleitner, A., Riemann, R., & Livesley, W. J. (1998). Heritability
of facet-level traits in a cross-cultural twin sample: Support for a hierarchical model of
personality. Journal of Personality and Social Psychology, 74, 15561565.
John, O. P. (1990). The Big Five factor taxonomy: Dimensions of personality in the natural
languages and in questionnaires. In L. A. Pervin (Ed.), Handbook of personality: Theory
and research (pp. 66100). New York: Guilford Press.
John, O. P., & Roberts, B. W. (1993). Measuring the Five-Factor Model on the Adjective
Check List. Technical Report, Institute of Personality and Social Research, University
of California, Berkeley, CA.
Joreskog, K. G., & Sorbom, D. (1993). LISREL 8: Structural equation modeling with the
SIMPLIS command language. Mahwah, NJ: Erlbaum.
Langinvainio, H., Kaprio, J., Koskenvuo, M., & Lonnqvist, J. (1984). Finnish twins reared
apart. III: Personality factors. Acta Genetica Medicae et Gemellologiae, 33, 259264.
Loehlin, J. C. (1992). Genes and environment in personality development. Newbury Park,
CA: Sage.
Loehlin, J. C., & Gough, H. G. (1990). Genetic and environmental variation on the California
Psychological Inventory vector scales. Journal of Personality Assessment, 54, 463468.
Loehlin, J. C., & Nichols, R. C. (1976). Heredity, environment, and personality. Austin, TX:
University of Texas Press.
Lykken, D. T. (1982). Research with twins: The concept of emergenesis. Psychophysiology,
19, 361373.
McCrae, R. R. (1989). Why I advocate the five-factor model: Joint factor analyses of the
NEO-PI with other instruments. In D. M. Buss & N. Cantor (Eds.), Personality psychology: Recent trends and emerging directions (pp. 237245). New York: Springer-Verlag.
McCrae, R. R., & Costa, P. T., Jr. (1989). Reinterpreting the Myers-Briggs Type Indicator
from the perspective of the five-factor model of personality. Journal of Personality, 57,
1740.

HERITABILITIES OF THE BIG FIVE

453

McCrae, R. R., Costa, P. T., Jr., & Piedmont, R. L. (1993). Folk concepts, natural language,
and psychological constructs: The California Psychological Inventory and the Five-Factor
Model. Journal of Personality, 61, 126.
Myers, I. B., & McCaulley, M. H. (1985). Manual: A guide to the development and use of
the Myers-Briggs Type Indicator. Palo Alto: Consulting Psychologists Press.
Neale, M. C., & Cardon, L. R. (1992). Methodology for genetic studies of twins and families.
Dordrecht, The Netherlands: Kluwer.
Pedersen, N. L., Plomin, R., McClearn, G. E., & Friberg, L. (1988). Neuroticism, Extraversion
and related traits in twins reared apart and reared together. Journal of Personality and
Social Psychology, 55, 950957.
Price, R. A., & Vandenberg, S. G. (1980). Spouse similarity in American and Swedish couples.
Behavior Genetics, 10, 5971.
Riemann, R., Angleitner, A, & Strelau, J. (1997). Genetic and environmental influences on
personality: A study of twins reared together using the self- and peer report NEO-FFI
scales. Journal of Personality, 65, 449475.
Saucier, G., & Goldberg, L. R. (1996). The language of personality: Lexical perspectives on
the Five-Factor model. In Wiggins, J. S. (Ed.), The Five-Factor model of personality:
Theoretical perspectives (pp. 2150). New York: Guilford.
Shock, N. W., Greulich, R. C., Andres, R., Arenberg, D., Costa, P. T., Jr., Lakatta, E. G., &
Tobin, J. D. (1984). Normal human aging: The Baltimore Longitudinal Study of Aging
(NIH Publication No. 842450). Bethesda, MD: National Institutes of Health.
Steiger, J. H. (1989). EZPath: A supplementary module for SYSTAT and SYGRAPH. Evanston,
IL: SYSTAT, Inc.
Tellegen, A., Lykken, D. T., Bouchard, T. J., Jr., Wilcox, K. J., Segal, N. S., & Rich, S.
(1988). Personality similarity in twins reared apart and together. Journal of Personality
and Social Psychology, 54, 10311039.
Waller, N. G. (in press). Evaluating the structure of personality. In C. R. Cloninger (Ed.),
Personality and psychopathology. Washington, D.C.: American Psychiatric Press.