The Big Five as States: How Useful Is the Five-Factor Model to Describe Intraindividual Variations over Time?
←
→
Page content transcription
If your browser does not render page correctly, please read the page content below
JOURNAL OF RESEARCH IN PERSONALITY 32, 202–221 (1998)
ARTICLE NO. RP972206
The Big Five as States: How Useful Is the Five-Factor Model
to Describe Intraindividual Variations over Time?
Peter Borkenau
Martin-Luther-Universität, Halle, Germany
and
Fritz Ostendorf
Universität Bielefeld, Bielefeld, Germany
This study investigated the similarity between the factor structure of longitudinal
variations in states and the factor structure of individual differences in traits. On
90 consecutive days, 22 students self-administered 30 self-report items that were
markers of the Big Five. Most participants showed good discrimination among the
90 measurement occasions. Correlations were computed between items across mea-
surement occasions. These P-correlations were factored, and the factor matrices
were target rotated toward a reference factor structure of individual differences,
using orthogonal Procrustes rotations. A substantial match was obtained between
the factor structure of longitudinal correlations that had been averaged across partici-
pants and the factor structure of individual differences. For individual participants,
this factor match was worse. It is concluded that the Big Five are useful to describe
longitudinal variations in states. A sharp distinction between the five-factor model
and the trait approach is recommended. 1998 Academic Press
The five-factor model of personality has received much attention during
the last decade, and it has contributed substantially to a revitalization of trait
theory in the eighties and nineties. According to Goldberg, the five-factor
model reflects an emerging consensus about the general framework of a taxo-
nomic representation of personality traits. ‘‘As a consequence, the scientific
study of personality dispositions, which has been cast into doldrums in the
We are indebted to the research participants for their patient cooperation, to Nadine Mauer
for collecting the data and for her help in the data analysis, and to Lewis R. Goldberg, Susanne
Hempel, Willem K. B. Hofstee, Robert R. McCrae, and Uwe Wolfradt for helpful comments
on earlier drafts of this article.
Address correspondence and reprint requests to Peter Borkenau, Institut fuer Psychologie,
der Martin-Luther-Universitaet, Postfach 1108, D-06099 Halle, Germany. E-mail: p.borkenau
@psych.uni-halle.de.
202
0092-6566/98 $25.00
Copyright 1998 by Academic Press
All rights of reproduction in any form reserved.INTRAINDIVIDUAL STRUCTURES 203
1970s, is again an intellectually vigorous enterprise’’ (Goldberg, 1993,
p. 26). And McCrae and Costa grant the five factors the status of explanatory
constructs that account for individual differences in personality. They ‘‘pro-
pose that the traits of the five-factor model are best viewed as explanations
for an intermediate category of characteristic adaptations, which in turn pro-
vide explanations for behavior’’ (McCrae & Costa, 1995, p. 247). Thus
distinguished proponents of the five-factor model1 suggest that support for
this model implies support for a trait approach to personality as well.
THE FIVE-FACTOR MODEL AND THE TRAIT APPROACH
This view probably reflects the historical fact that the five-factor model
was established in studies on individual differences and has been used almost
exclusively to account for individual differences in traits. But this may reflect
a historical accident more than a phenomenon about personality. It might
well be that the five factors are not only useful to describe enduring individ-
ual differences between persons, but also to describe the variability within
persons over time. This would imply that the five-factor model is not ‘‘wed-
ded’’ to a trait approach to personality, but rather is a structural model that
is useful for higher-level descriptions of enduring differences between per-
sons as well as of longitudinal variations within persons.
There is evidence from various sources that suggests this hypothesis: First,
Cattell (1946) proposed various factor-analytic techniques, among them the
common cross-sectional R-technique (i.e., factoring of correlations between
variables across persons) and the longitudinal P-technique (i.e., factoring of
intraindividual correlations between variables across occasions). In several
factor-analytic studies of individual cases, Cattell and his associates used P-
technique to uncover the factor structure of intraindividual fluctuations in
temperament (Cattell, Cattell, & Rhymer, 1947) and in motivation (Cattell,
1951; Cattell & Cross, 1952). Cattell (1955) summarized the results of these
studies to the effect that: (a) the factors identified by P-technique are ‘‘rea-
sonably invariant from study to study’’ and (b) that these factors ‘‘are for
the most part matchable with major factors found in R-technique studies’’
(Cattell, 1955, p. 340). As an explanation, Cattell (1955) suggested that func-
tional unities that vary between persons may also fluctuate within persons,
resulting in similar (but not identical) patterns of R-correlations and P-corre-
lations among the surface variables.
Second, similar results of R-factor analyses and of P-factor analyses have
been found in studies on mood (Russell, 1980; Watson & Tellegen, 1985;
Zevon & Tellegen, 1982). Zevon and Tellegen (1982) administered a 60-
1
Some authors (e.g., Goldberg, 1993) distinguish between the lexically derived Big Five
and Costa and McCrae’s five-factor model. We do not make this distinction here but use both
terms interchangeably.204 BORKENAU AND OSTENDORF
item mood-adjective checklist to 23 undergraduates for 90 consecutive days
and, using P-factor analysis, identified the factors Positive Affect and Nega-
tive Affect in the self-reports of 21 of these participants. Moreover, when
they averaged the factor loadings from their P-factor analyses across their
23 participants, the average loadings correlated .97 (Positive Affect) and
.94 (Negative Affect) with the factor loadings from an R-factor analysis.
Reanalyses of other studies yielded similar results (Watson & Tellegen,
1985). Feldman (1995), however, found systematic individual differences
in the P-correlations between her participants’ mood ratings. Thus it might
be that high congruences between R-factors and P-factors is obtained only
if the P-factor loadings (or the P-correlations) are averaged across par-
ticipants.
Finally, the structure of intraindividual variations in states should resemble
the structure of individual differences in traits as far as factors of personality
reflect higher-level dimensions that underlie the categorization of behavior
(Borkenau, 1990, 1992; Hogan, 1983; Norman & Goldberg, 1966; Romer &
Revelle, 1984). The Big Five are found not only in self-reports and ratings
by knowledgeable informants, but also in ratings by strangers (Passini &
Norman, 1966), in semantic similarity ratings for trait pairs (D’Andrade,
1974), in co-occurrence likelihood ratings for trait pairs (Hakel, 1969), and
in correlations among prototypicality ratings of acts for traits (Borkenau,
1988). This suggests that the Big Five may reflect conceptual relations
among personality-descriptive terms, and it is reasonable that these concep-
tual relations do not depend very much on whether the terms are used to
describe long-lasting traits or temporary states.
To summarize, it has been extensively investigated how well the Big Five
account for individual differences in traits. In contrast, the reasonable hy-
pothesis that the Big Five also account for intraindividual variations in states
has not been tested. This is the purpose of the present study.
OVERVIEW
Over a period of 90 consecutive days, 22 participants filled out a rating
sheet each evening. The rating sheet comprised 30 adjective scales that had
been established as marker items of Neuroticism, Extraversion, Agreeable-
ness, Conscientiousness, and Intellect in a large-scale German-language
study on individual differences (Ostendorf, 1990). We instructed the partici-
pants of the present study to indicate how well the 30 adjectives described
their behavior on that particular day. We then correlated the 30 daily self-
ratings separately for each participant across the 90 measurement occasions
and factor analyzed these P-correlations. We then compared the resulting P-
factors to the R-factor structure of corresponding trait ratings. Moreover,
we averaged the P-correlations for the 22 participants, factor analyzed theseINTRAINDIVIDUAL STRUCTURES 205
averaged P-correlations, and compared the resulting P-factors to the R-
factors.
Method
Participants
The participants were 22 first-year psychology students (19 women and 3 men) at the Uni-
versity of Halle, Germany. They received course credit for their participation. Most of them
were 19 years old or in their early twenties, except three females who were in their thirties
or forties.
Measures and Instruction
Self-report measures. A diary method was used. The participants were instructed to describe
their behavior on 90 consecutive days by filling out the same 30 adjective scales in the evening
hours of each day. The instruction read:
This rating sheet comprises 30 scales that range from not at all (0) to extremely (6).
. . . Please indicate on these scales how you retrospectively appraise your behavior
this day. . . . It is important that you respond to each of the 30 scales daily, even
if some of the questions may be difficult for you to answer. Moreover, please indicate
the date and the hour that you fill out the rating sheet. It follows from the nature
of the rating task that you should fill out the rating sheets in the evening hours.
The 30 adjectives were marker variables of either Neuroticism, Extraversion, Agreeableness,
Conscientiousness, or Intellect. More precisely, we used the factor matrices published by Os-
tendorf (1990) to identify six German marker items, three with positive and three with negative
loadings, for each of the five factors. Ostendorf (1990) administered 830 unipolar adjective
scales to a sample of 408 participants, and 394 of these participants were also described on
the same scales by at least one peer. A second criterion for the choice of the 30 items was
that they could be used for trait as well as state descriptions. The distinction between states
and traits is fuzzy rather than definite, and there are terms that can be used to describe both
states and traits (Chaplin, John, & Goldberg, 1988). However, as this is not true of all marker
variables of the Big Five, adjectives like gifted, a marker of the Intellect factor in Ostendorf’s
(1990) study, had to be discarded as the participants could not sensibly be asked 90 times
whether they had acted in a gifted way on that particular day.
To prevent boredom, to discourage stereotypic responding, and to control for order effects,
ten different forms of the rating sheet were produced in which the same 30 items were arranged
in ten different orders. The same number of copies of each of the ten forms was printed. It
was intended (and approximately achieved) that each participant responded nine times to each
of the ten versions of the rating sheet.
Semantic similarity ratings. To check the influence of the individual participants’ meaning
systems on the pattern of their individual P-correlations, the participants also judged the seman-
tic similarity of the 435 pairings of the 30 items. This was done on seven-point rating scales
with the endpoints opposite in meaning (⫺3) and highly similar in meaning (⫹3).
Procedure
The data were collected from April 1996 to July 1996. After having signed an informed
consent form, the participants received a booklet that comprised the scales for the semantic
similarity ratings. Moreover, they received an envelope that contained one copy of each of206 BORKENAU AND OSTENDORF
the ten forms of the self-rating sheet. They were instructed to reappear in the office the next
week, to return the envelope with the rating sheets they had filled out during the last week,
to fetch another envelope with seven new exemplars of the rating sheet for the next week,
and so forth, until they had received and filled out all 90 rating sheets.
The individual participants marked each copy of their 90 rating sheets with an unique self-
generated code. This made it feasible to identify which rating sheets stemmed from the same
person, without uncovering that person’s identity.
Results
Reference Factor Structure of Individual Differences
As we intended to compare the P-factor structure of variations in states
to the R-factor structure of individual differences in traits, we first established
the R-factor structure of the 30 adjectives as trait-descriptive terms. This was
accomplished by a reanalysis of Ostendorf’s (1990) data in which only the
30 items used in the present study were included. The self-ratings and the
peer ratings on the 30 adjectives were submitted to separate principal compo-
nents analyses and varimax rotations. Both analyses suggested the extraction
of five factors that accounted for 52.8% (self-ratings) or 57.1% (peer ratings)
of the variance. Note that we factor analyzed one-item scales that have only
moderate reliabilities. Thus the explained proportion of variance was highly
satisfactory. Moreover, the varimax-rotated factors could clearly be identi-
fied as Neuroticism, Extraversion, Agreeableness, Conscientiousness, and
Intellect. The varimax-rotated factor pattern of the self-ratings is reported in
Table 1.
The factor pattern of the peer ratings was highly similar. To quantify this
similarity, congruence coefficients, as suggested by Wrigley and Neuhaus
(1955), were calculated subsequent to an orthogonal Procrustes rotation of
the factor pattern of the peer ratings toward the factor pattern of the self-
ratings. Calculating congruence coefficients after a Procrustes rotation is a
test of how well an observed loading pattern can be recovered in a new set
of data. It differs from confirmatory factor analyses mainly in that an ob-
served factor pattern is compared to another observed pattern instead of to
a hypothetical factor pattern.
The SPSS version of the computer program by McCrae, Zonderman,
Costa, Bond, and Paunonen (1996) was used. This program reports congru-
ence coefficients for each factor, for each item, and for the entire factor
matrix. The lowest factor congruence was .97, the lowest item congruence
was .96, and the matrix congruence was .98. Thus in the factor analyses of
individual differences, the factor patterns of the self-ratings and of the peer
ratings were almost identical.
Note that these R-factor analyses were run to establish a reference factor
structure for the P-factor analyses. Thus a decision had to be made in favor
of either the self-rating structure or the peer rating structure. We decided inINTRAINDIVIDUAL STRUCTURES 207
TABLE 1
Reference Factor Structure of Individual Differences in Traits
Trait adjective N E A C I
Irritable .77 ⫺.07 .01 ⫺.06 .04
Bad-tempered .53 .01 ⫺.27 ⫺.25 ⫺.15
Vulnerable .80 ⫺.04 .12 .00 ⫺.04
Emotionally stable ⫺.63 .11 .16 .25 .15
Calm ⫺.61 ⫺.02 .19 .21 .18
Resistant ⫺.74 .17 .05 ⫺.05 ⫺.06
Dynamic ⫺.16 .58 .10 .18 .29
Sociable ⫺.13 .72 .20 .13 ⫺.06
Lively .21 .71 .04 .03 .09
Shy .31 ⫺.73 .06 ⫺.09 ⫺.07
Silent .01 ⫺.68 .24 .09 ⫺.08
Reserved .12 ⫺.73 .28 .11 .04
Good-natured ⫺.12 ⫺.07 .67 .08 .01
Helpful .10 .22 .66 .20 ⫺.08
Considerate .02 ⫺.02 .74 .13 ⫺.01
Selfish .05 .04 ⫺.60 ⫺.12 ⫺.01
Domineering .06 .07 ⫺.72 .14 ⫺.05
Obstinate .15 .08 ⫺.62 ⫺.06 ⫺.01
Industrious ⫺.10 .11 .12 .71 .05
Persistent ⫺.20 .06 .03 .59 .18
Responsible .01 .08 .18 .69 ⫺.01
Lazy .06 ⫺.10 ⫺.13 ⫺.69 ⫺.01
Reckless ⫺.06 .34 .11 ⫺.60 .02
Changeable .28 .04 ⫺.06 ⫺.65 ⫺.18
Witty ⫺.13 .06 .04 .02 .74
Knowledgeable ⫺.24 ⫺.03 ⫺.03 .16 .72
Prudent ⫺.09 ⫺.07 ⫺.06 .02 .72
Unresourceful ⫺.20 ⫺.26 ⫺.14 .02 ⫺.56
Uninformed .07 .04 .07 ⫺.20 ⫺.67
Unimaginative ⫺.11 ⫺.30 ⫺.03 .03 ⫺.63
Note. N, Neuroticism, E, Extraversion, A, Agreeableness, C, Conscientiousness, I, Intellect.
Markers are set in italics.
favor of the self-rating structure to hold the source of the ratings (self- versus
peer) constant across the two types of analyses (individual differences versus
longitudinal variations). Because the factor structures of the self-ratings and
of the peer ratings were highly similar, however, the effects of this decision
on the results of the subsequent analyses are likely to be negligible.
Intraindividual Variation across Time
Concerning the 90-day diary study, we checked first how strongly the 22
participants distinguished among the 90 measurement occasions. This is not
a trivial issue in this kind of study because participants might well perceive208 BORKENAU AND OSTENDORF the 30 adjectives as descriptors of enduring traits that hardly vary across time. To check this, the means and standard deviations across the 90 mea- surement occasions were separately calculated for each participant and each item. A participant’s range of means (per item across measurement occa- sions) indicated the extent to which his or her responses to the 30 items differed systematically, and a participant’s standard deviation indicated the extent to which he or she discriminated among the 90 measurement occa- sions. Theoretically, the item means could vary between 0.00 and 6.00, and the participants differed remarkably in how far they exhausted this possible range. For the participant with the largest range, the item means varied be- tween 0.08 and 5.82. Thus this participant responded to some items in an extreme and uniform way. For example, for her 90 responses to the adjective unresourceful, this participant used 84 times the rating category 0, 5 times the rating category 1, and once the rating category 2, resulting in an item mean of 0.08. In contrast, for the participant with the smallest range, the item means varied between 2.48 and 3.38, indicating that this participant responded to no item in an extreme and uniform way. On average, the differ- ence between an individual participant’s highest and lowest item mean was 3.42. Moderate means were a necessary but not sufficient prerequisite for sub- stantial within-item standard deviations that indicated the extent of a partici- pant’s discrimination among occasions. The 660 (30 items ⫻ 22 participants) intraindividual standard deviations varied between 0.11 and 2.48 and their median was 0.99. Note that the intraindividual standard deviations varied systematically between participants. The participant who distinguished least among the 90 occasions had a median standard deviation of 0.63, and the participant who distinguished most had a median standard deviation of 2.02; the latter participant used all seven rating categories in her 90 responses to each of the 30 items, resulting in small differences between item means and large standard deviations across occasions. Factor Analyses of Longitudinal Correlations Eigenvalues. Separately for each participant, we calculated the P-correla- tions between his or her responses to the 30 items across the 90 measurement occasions and submitted them to principal components analyses. The eigen- value plots varied considerably between the 22 participants. The proportion of variance accounted for by the first unrotated principal component ranged from 8.6 to 56.8%, and the proportion of variance accounted for by the first five principal components ranged from 35.3 to 81.7%. These differences in the eigenvalues reflected large differences in the size of the individual partici- pants’ P-correlations, that in turn reflected large differences in the size of their intraindividual standard deviations. However, the two most extreme cases mentioned above were outliers in
INTRAINDIVIDUAL STRUCTURES 209 terms of the variance accounted for. Among the 20 remaining participants, the proportion of variance accounted for by the first five principal compo- nents ranged from 46.3 to 66.3%, thus resembling the proportion of variance accounted for by the first five principal components in the R-factor analyses. Evaluation of fit. Could the same factors be identified in the P-correlations and the R-correlations? To check this, we rotated the principal components that had been obtained for each participant in the P-factor analyses toward the reference factor structure reported in Table 1, using orthogonal Procrustes rotation. We then evaluated the match between the P-factors and the R- factors via congruence coefficients. This method capitalizes on chance to some extent, and we therefore calculated the probability that any observed congruences might reflect random fits. McCrae et al. (1996) reported that, in their analyses of random data, the means of the distributions of factor congruences ranged from .32 to .34 and that the 95th percentiles ranged from .52 to .55. Paunonen, Jackson, Trzebin- ski, and Foersterling (1992), however, reported somewhat higher random fits; their distribution of random fits had a mean of .47 and a standard devia- tion of .15, implying a 95th percentile of .76. The most straightforward expla- nation of this discrepancy is that distributions of random fits depend upon the number of factors and the number of variables (Paunonen, 1997). We therefore did an independent Monte-Carlo analysis to find out the distribution of random fits for 30 variables, 5 factors, and 90 observations. First, 1000 datasets were generated by a computer; each of these datasets consisted of random values of 90 cases on 30 normally distributed variables. We factor analyzed these random data, retained five factors, and submitted the 1000 five-factor patterns to orthogonal Procrustes rotations towards the target five-factor structure reported in Table 1. We obtained factor congru- ences with a mean of .34 and a standard deviation of .12, implying a 95th percentile of .54 and a 99th percentile of .61. For the item congruences, the mean of the distribution of random fits was .33 and the standard deviation was .40, implying a 95th percentile of .87 and a 99th percentile of .95. Fi- nally, the mean of the distribution of random fits for the entire factor matrices was .34 and the standard deviation was .05, implying a 95th percentile of .42 and a 99th percentile of .46. These figures indicate the critical values that congruence coefficients must pass to reject the null hypothesis of zero congruence. Note, however, that just as the statistical significance of a correlation does not imply that the correlation is high, the rejection of the null hypothesis of zero congruence does not imply a factor match. Rather, according to a rule of thumb, congruence coefficients should exceed .90 to define a matching factor (Barrett, 1986). This was the criterion that we actually applied. P-factors obtained for individual participants. The factor and matrix con- gruences for the individual participants (with the R-factor pattern reported in Table 1) are reported in the first six data columns of Table 2. Most factor
210 BORKENAU AND OSTENDORF
TABLE 2
Congruence of Individual P-Factor Patterns with the R-Factor Pattern
Factor congruence coefficients for
Participant N E A C I Matrix Reliability
A .43 .74 .31 .46 .58 .51 .41
B .41 .56 .82 .38 .46 .53 .58
C .78 .62 .61 .77 .82 .72 .77
D .87 .78 .62 .75 .59 .72 .59
E .87 .80 .63 .66 .67 .73 .75
F .83 .76 .72 .63 .71 .73 .74
G .77 .82 .68 .72 .76 .75 .82
H .73 .79 .87 .83 .59 .75 .93
I .75 .92 .62 .65 .84 .76 .76
J .89 .87 .75 .77 .52 .76 .82
K .84 .85 .83 .59 .74 .77 .74
L .71 .90 .90 .67 .70 .77 .84
M .90 .85 .67 .75 .71 .78 .81
N .86 .91 .72 .75 .62 .78 .84
O .89 .80 .76 .73 .77 .79 .82
P .84 .93 .81 .76 .71 .81 .80
Q .78 .88 .82 .72 .83 .81 .81
R .87 .90 .82 .72 .78 .82 .86
S .90 .85 .81 .77 .80 .83 .74
T .81 .86 .81 .83 .84 .83 .85
U .83 .86 .78 .88 .89 .85 .75
V .85 .94 .87 .90 .89 .89 .81
Median .84 .85 .77 .74 .73 .77 .80
Note. N, Neuroticism, E, Extraversion, A, Agreeableness, C, Conscientiousness, I, Intellect.
Participants are ordered according to their matrix congruence. The last column reports the
factor match for ratings on odd versus even days.
congruences and all matrix congruences exceeded chance levels, but only
10 out of 110 passed the .90-criterion of matching factors. Altogether, the
factor congruences for the individual participants were moderate: The me-
dian overall fit was .77, as indicated in the bottom row of Table 2. To illus-
trate the P-factor pattern of an individual participant, Table 3 reports the
Procrustes-rotated P-factor pattern for Participant K, a case with a median
overall fit of .77.
Although the P-factors for this participant were for the most part match-
able with major factors found in R-technique studies, this match was far
from perfect. Factor 1 (Neuroticism) showed unexpectedly high loadings of
the adjectives changeable and witty (reversed) and an unexpectedly low load-
ing of the adjective calm. Factor 2 (Extraversion) showed unexpectedly high
loadings of the adjectives witty, industrious, responsible, and considerate
(reversed), whereas Factor 3 (Agreeableness) showed an unexpectedly lowINTRAINDIVIDUAL STRUCTURES 211
TABLE 3
Procrustes-Rotated P-Factor Pattern of Participant K
State adjective N E A C I
Irritable .73 ⫺.30 .05 ⫺.22 ⫺.20
Bad-tempered .64 .13 ⫺.37 ⫺.19 .16
Vulnerable .79 .06 ⫺.03 ⫺.32 ⫺.06
Emotionally stable ⫺.69 .04 .17 .36 .11
Calm ⫺.39 ⫺.26 .16 .15 .42
Resistant ⫺.59 .29 ⫺.13 .33 .40
Dynamic .12 .50 ⫺.12 .07 ⫺.56
Sociable ⫺.03 .71 ⫺.03 .23 .22
Lively ⫺.19 .79 ⫺.06 .19 .01
Shy .12 ⫺.63 .05 ⫺.15 ⫺.30
Silent .09 ⫺.75 .12 ⫺.23 ⫺.34
Reserved .38 ⫺.53 .13 ⫺.29 ⫺.34
Good-natured ⫺.01 .05 .41 ⫺.10 ⫺.08
Helpful .18 .09 .66 .23 ⫺.11
Considerate .07 ⫺.42 .11 .37 ⫺.02
Selfish ⫺.10 ⫺.20 ⫺.60 ⫺.30 .01
Domineering .16 .20 ⫺.57 .47 ⫺.10
Obstinate .36 .26 ⫺.49 .21 ⫺.03
Industrious .01 .46 .13 .24 .18
Persistent ⫺.41 .27 ⫺.30 ⫺.09 .04
Responsible ⫺.16 .45 ⫺.02 .30 .04
Lazy ⫺.14 ⫺.35 .05 ⫺.55 ⫺.06
Reckless .14 .25 .13 ⫺.55 ⫺.24
Changeable .72 ⫺.14 ⫺.15 ⫺.36 ⫺.10
Witty ⫺.58 .54 ⫺.21 .11 .28
Knowledgeable ⫺.06 ⫺.01 ⫺.09 .26 .83
Prudent .04 ⫺.11 ⫺.01 .11 .87
Unresourceful ⫺.05 ⫺.09 ⫺.09 ⫺.06 ⫺.87
Uninformed .27 ⫺.09 ⫺.05 .25 ⫺.75
Unimaginative ⫺.05 ⫺.22 .10 ⫺.41 ⫺.50
Note. N, Neuroticism, E, Extraversion, A, Agreeableness, C, Conscientiousness, I, Intellect.
loading of the adjective considerate. The worst match, however, was ob-
tained for Factor 4 (Conscientiousness) which had unexpectedly low load-
ings of the adjectives industrious, persistent, and responsible but an unex-
pectedly high loading of the adjective domineering. Finally, Factor 5
(Intellect) had unexpectedly high loadings of the adjectives calm, resistant,
and dynamic (reversed) and an unexpectedly low loading of the adjective
witty. Thus the P-factors for Participant K in Table 3 somewhat resembled
the R-factors in Table 1, but the match was far from perfect.
The moderate congruence coefficients at the level of individual partici-
pants may either reflect true differences between the structure of individual
fluctuations in states and the structure of individual differences in traits, or212 BORKENAU AND OSTENDORF
they may reflect sampling error as the number of observations in each longi-
tudinal analysis was only 90. If sampling error were small, the factor patterns
of individual participants should be reliable. To check their reliability, we
subdivided the 90 measurement occasions into two observation periods, the
odd days (Day 1, Day 3, . . . , Day 89) and the even days (Day 2, Day 4,
. . . , Day 90). We then computed the P-correlations separately for the two
45-day observation periods and each individual participant, factored these
P-correlations, and compared the factors (using again Procrustes rotations)
obtained for the same participant between the two observation periods.
The congruence coefficients that indicate the reliability of individual P-
factor matrices are reported in the last column of Table 2. The median reli-
ability was .80, that is, not much higher than the average match between the
individual P-factor structures and the R-factor structure.2 Furthermore, to
check whether individual differences in the reliabilities of the P-factor struc-
tures predicted their match with the R-factor structure, we calculated the
correlation, across participants, between the reliability of the 22 P-factor
structures (reported in the last column of Table 2) and their match with the
R-factor structure (reported in the second-last column of Table 2). This corre-
lation was .76. Thus lack of match between the individual P-factor structures
and R-factor structures mostly reflected the unreliability of the individual P-
factor structures.
Averaged Intraindividual Correlations
Thus, to obtain more stable estimates of the structure of longitudinal varia-
tions in states, we averaged the corresponding P-correlations across the 22
participants, using Fisher’s Z-transformation for correlations. This procedure
is highly similar to the chain-P technique suggested by Cattell (1973).3
Factor analyses and congruence coefficients. We then submitted these
averaged P-correlations to a principal components analysis and found six
eigenvalues larger than 1.0. To check which number of P-factors yielded the
best match with the R-factors, we extracted 2, 3, 4, 5, 6, 7, and 8 factors
from the averaged P-correlations. Moreover, we extracted and varimax ro-
tated the same number of R-factors from the individual differences data and
target rotated the P-factors towards these R-factors, using again orthogonal
Procrustes rotations. The resulting congruence coefficients are reported in
Table 4.
2
It is likely that the sampling error of P-correlations for a 90-day observation period is
smaller than the sampling error of P-correlations for a 45-day observation period, and thus
the median congruence of .80 underestimates the reliability of individual participants’ P-factor
structures. But we are not aware of any statistical technique to adjust the congruence coeffi-
cients for this reduction in the number of observations.
3
We are indebted to an anonymous reviewer for bringing Cattell’s chain-P technique to
our attention.INTRAINDIVIDUAL STRUCTURES 213
TABLE 4
Congruence of the Factors in the Averaged P-Correlations with Varimax-Rotated R-factors
for 2 to 8 Factor Solutions
Number of factors extracted
2 3 4 5 6 7 8
Factor 1 .97 .80 .89 .94 .95 .93 .95
Factor 2 .86 .65 .93 .96 .95 .92 .97
Factor 3 .87 .96 .88 .87 .92 .95
Factor 4 .85 .93 .89 .92 .92
Factor 5 .88 .93 .91 .71
Factor 6 .57 .80 .73
Factor 7 .50 .81
Factor 8 .86
Matrix .92 .78 .90 .92 .88 .87 .88
The match was best for the two-factor solution and the five-factor solution,
whereas it was worst for the three-factor solution. When the two-factor solu-
tion of the averaged P-correlations was Procrustes rotated, the first factor
combined high Extraversion and high Intellect, and the second factor com-
bined high Neuroticism and low Agreeableness. The loading pattern was less
clear for Conscientiousness.4
The Procrustes-rotated five-factor loading pattern of the averaged P-corre-
lations is reported in Table 5. It resembled the R-factor pattern (see Table
1) in that 27 of the 30 adjectives had their highest loading on the expected
factor. Moreover, 26 of the 30 item congruences were higher than .90, and
28 of the 30 item congruences were higher than .88. Finally, the matrix con-
gruence of .92 indicated that the Big Five were well-suited to account for
the individual differences in trait-descriptions as well as for the longitudinal
variations in state descriptions.
But there were differences in detail, and most of them affected the Consci-
entiousness factor. The adjectives reckless and changeable did not load sub-
stantially on Conscientiousness and obtained very low item congruences of
.52 (reckless) and .73 (changeable). In contrast, the adjectives dynamic and
helpful had unexpectedly high loadings of .38 on this factor. Thus, the Con-
scientiousness factor had a stronger activity component in the P-analyses
than in the R-analyses. Other differences affected the Intellect factor in that
the adjectives unresourceful and unimaginative had substantial negative
4
One anonymous reviewer suggested that the P-factors in the two-factor solution might
reflect Positive Affect and Negative Affect. This suggestion is supported by our findings as
far as the strongest marker items for the first factor referred to Extraversion and the strongest
marker items for the second factor referred to Neuroticism. Note, however, that the two-factor
solution accounted for not more than 34% of the variance.214 BORKENAU AND OSTENDORF
TABLE 5
Procrustes-Rotated Factor Pattern of the Averaged P-Correlations
State adjective N E A C I
Irritable .79 ⫺.18 ⫺.13 ⫺.09 ⫺.09
Bad-tempered .47 ⫺.07 ⫺.48 ⫺.11 ⫺.15
Vulnerable .78 ⫺.18 ⫺.08 ⫺.12 ⫺.09
Emotionally stable ⫺.72 .12 .11 .23 .15
Calm ⫺.63 ⫺.02 .20 .05 .11
Resistant ⫺.74 .10 .02 .07 .05
Dynamic ⫺.08 .62 ⫺.01 .38 .14
Sociable ⫺.18 .73 .14 .10 .16
Lively ⫺.04 .74 ⫺.12 .03 .18
Shy .24 ⫺.63 ⫺.02 .03 ⫺.25
Silent .15 ⫺.71 ⫺.04 .04 ⫺.19
Reserved .13 ⫺.68 .07 ⫺.01 ⫺.24
Good-natured ⫺.18 .15 .48 .20 ⫺.03
Helpful .02 .25 .42 .38 .02
Considerate ⫺.14 .06 .57 .23 .00
Selfish .17 .07 ⫺.60 ⫺.13 ⫺.06
Domineering .15 .27 ⫺.64 .08 ⫺.06
Obstinate .14 .26 ⫺.63 .07 ⫺.02
Industrious .07 .15 .06 .77 .19
Persistent ⫺.16 .22 ⫺.13 .54 .14
Responsible ⫺.01 .11 .22 .65 .10
Lazy ⫺.13 ⫺.15 ⫺.13 ⫺.72 ⫺.18
Reckless .04 .39 ⫺.33 ⫺.20 ⫺.18
Changeable .59 ⫺.07 ⫺.21 ⫺.27 ⫺.14
Witty ⫺.17 .25 ⫺.02 .17 .57
Knowledgeable ⫺.10 .04 ⫺.16 .33 .68
Prudent ⫺.10 .06 ⫺.01 .31 .66
Unresourceful ⫺.02 ⫺.42 ⫺.20 .15 ⫺.52
Uninformed .17 .01 .05 ⫺.23 ⫺.62
Unimaginative .00 ⫺.40 ⫺.13 ⫺.04 ⫺.57
Note. N, Neuroticism, E, Extraversion, A, Agreeableness, C, Conscientiousness, I, Intellect.
Variables that were markers in the R-factor analyses are set in italics.
loadings not only on this factor but also on Extraversion. This may indicate
that for this sample of predominantly young women, interpersonal situations
were perceived as important opportunities to be witty and imaginative. Fi-
nally, the adjectives bad-tempered and reckless had substantial negative
loadings on Agreeableness. This may indicate that the participants of the
longitudinal study perceived themselves as agreeable only when they con-
trolled their behavior.
These similarities and dissimilarities between the R-factor pattern and the
averaged P-factor pattern were obtained quite independently of the methods
of factor extraction and factor rotation. If the averaged P-correlations wereINTRAINDIVIDUAL STRUCTURES 215
TABLE 6
Varimax-Rotated Factor Pattern of the Averaged P-Correlations
State adjective N E A C I
Irritable .79 ⫺.20 ⫺.14 ⫺.02 ⫺.07
Bad-tempered .46 ⫺.12 ⫺.50 ⫺.06 ⫺.10
Vulnerable .79 ⫺.19 ⫺.10 ⫺.05 ⫺.08
Emotionally stable ⫺.74 .12 .13 .15 .16
Calm ⫺.62 .01 .23 ⫺.04 .10
Resistant ⫺.74 .10 .04 ⫺.01 .04
Dynamic ⫺.13 .56 ⫺.07 .46 .10
Sociable ⫺.19 .73 .08 .19 .05
Lively ⫺.06 .73 .18 .14 .08
Shy .23 ⫺.66 .01 ⫺.03 ⫺.16
Silent .15 ⫺.73 .01 ⫺.05 ⫺.08
Reserved .14 ⫺.70 .11 ⫺.09 ⫺.15
Good-natured ⫺.19 .15 .46 .22 ⫺.07
Helpful ⫺.02 .23 .39 .42 ⫺.02
Considerate ⫺.15 .07 .56 .23 ⫺.03
Selfish .16 .03 ⫺.61 ⫺.11 ⫺.03
Domineering .11 .18 ⫺.67 .12 ⫺.03
Obstinate .11 .19 ⫺.66 .11 .01
Industrious ⫺.01 .06 .05 .77 .24
Persistent ⫺.23 .14 ⫺.13 .53 .18
Responsible ⫺.08 .04 .21 .65 .14
Lazy ⫺.05 ⫺.07 ⫺.12 ⫺.73 ⫺.22
Reckless .04 .36 ⫺.38 ⫺.12 ⫺.22
Changeable .61 ⫺.07 ⫺.24 ⫺.20 ⫺.14
Witty ⫺.18 .31 .01 .14 .54
Knowledgeable ⫺.13 .09 ⫺.10 .25 .71
Prudent ⫺.12 .12 ⫺.04 .24 .67
Unresourceful ⫺.05 ⫺.54 ⫺.20 .10 ⫺.41
Uninformed .18 ⫺.06 ⫺.01 ⫺.15 ⫺.64
Unimaginative ⫺.01 ⫺.49 ⫺.14 ⫺.06 ⫺.50
Note. N, Neuroticism, E, Extraversion, A, Agreeableness, C, Conscientiousness, I, Intellect.
Variables that were markers in the R-factor analyses are set in italics.
submitted to a principal axes analysis (with communality estimates) instead
of a principal components analysis, the matrix congruence dropped negligi-
bly from .92 to .91. Moreover, the factor matrices that resulted from varimax
rotations resembled those that resulted from orthogonal Procrustes rotations.
The varimax-rotated factor pattern of the averaged P-correlations is reported
in Table 6. It shows that the factor pattern of the averaged P-correlations
resembled the factor pattern of the R-correlations even without a target rota-
tion.
The effects of sampling error. We conducted additional analyses to esti-
mate the extent that these congruence coefficients were attenuated by sam-
pling error: We split the sample of 22 participants into two subsamples of216 BORKENAU AND OSTENDORF 11 participants and averaged the P-correlations of the 11 participants in the same subsample, using again Fisher’s Z-transformation for correlations. Thus we obtained two sets of averaged P-correlations that were based on the responses of two mutually exclusive samples of 11 participants. We then extracted the first five principal components from the averaged P-correlations within each subsample. They accounted for 51.9% of the vari- ance in both subsamples. Then we submitted the factor pattern from one subsample to a varimax rotation and that from the other subsample to an orthogonal Procrustes rotation toward the varimax-rotated loading pattern. We obtained a matrix congruence of .97 with factor congruences of .98 (Neu- roticism), .99 (Extraversion), .95 (Agreeableness), .96 (Conscientiousness), and .98 (Intellect). Note that these congruences were about .05 higher than the congruences that were found in comparisons between the factor structure of averaged P-correlations and the R-factor structure (see Table 4). Thus the discrepancies between the P-factor structure and the R-factor structure reflected sampling error to some extent but not entirely. Finally, to check the reliability of the finding that the factor structure of averaged P-correlations resembled the factor structure of R-correlations, we rotated the P-factor pattern obtained in the two subsamples toward the R- factor pattern reported in Table 1. For one subsample, we obtained a matrix congruence of .91 and factor congruences of .93 (Neuroticism), .94 (Extra- version), .87 (Agreeableness), .89 (Conscientiousness), and .92 (Intellect). For the other subsample, we obtained a matrix congruence of .92 and factor congruences of .95 (Neuroticism), .96 (Extraversion), .90 (Agreeableness), .86 (Conscientiousness), and .93 (Intellect). Thus the factor congruences for the combined sample and for the two subsamples were highly similar. Semantic Similarities and Longitudinal Correlations We had collected semantic similarity ratings to check whether differences between the meaning systems of the participants might explain individual differences in their P-correlations. According to various studies (e.g., Bor- kenau, 1986; Gara & Rosenberg, 1981; Norman & Goldberg, 1966; Shweder, 1975), semantic similarity ratings for pairs of personality-descriptive terms predict the R-correlations among corresponding ratings quite well. In this study, we checked how well a participant’s longitudinal correlations could be predicted from (a) this participant’s own semantic similarity estimates, (b) the other participants’ semantic similarity estimates, and (c) the means of the semantic similarity estimates by all participants. The relevant correlations were calculated across the 435 item pairings. First, when each participant’s P-correlations were correlated with his or her own semantic similarity estimates, the 22 correlations ranged from .23 to .65, and their average was .53. Second, when each participant’s P-correla- tions were correlated with the semantic similarity estimates by the other 21
INTRAINDIVIDUAL STRUCTURES 217
individual participants, the 462 correlations ranged from .14 to .70, and their
average was .48. Thus the participants’ own similarity estimates predicted
their P-correlations somewhat better than the other participants’ similarity
estimates. This rather moderate difference probably reflected the impact of
unique meaning systems. Third, when each participant’s P-correlations were
correlated with the averaged semantic similarity estimates by all participants,
the 22 correlations ranged from .28 to .76 and their mean was .63. Finally,
when the averaged P-correlations were correlated with the averaged semantic
similarity estimates, the correlation was .81. Thus to increase the reliability
of the similarity ratings and of the P-correlations had much stronger effects
on the prediction of the P-correlations than to consider individual differences
between the participants’ meaning systems.
DISCUSSION
In numerous previous studies, the Big Five have been identified as recur-
rent factors in self- and peer reports on individual differences in traits
(McCrae & John, 1992). In the present study, we recovered the Big Five
also in longitudinal variations in states if averaged P-correlations, but not if
individual participants’ P-correlations, were factor analyzed. The latter find-
ing probably reflects that factor structures get more robust if they are based
on more data points, as correlations from small samples have large confi-
dence intervals. This phenomenon is well known from studies on the factor
structure of individual differences (Hofstee, DeRaad, & Goldberg, 1992).
But whereas it is possible in R-factor analyses to extend the number of items
to several hundred and to extend the number of participants to several thou-
sand, it is hardly possible in P-factor analyses to extend the number of re-
peated measurements far beyond 90. Thus data by several participants have
to be combined, as was done in the present study.
This, however, creates an ambiguity because the factor structure of aver-
aged P-correlations may reflect a phenomenon that exists only at the group
level and has no bearing whatsoever on the individual case. Indeed, the mod-
erate congruences at the level of individual participants might reflect reliable
differences between individuals, and this explanation is supported by the
finding that a participant’s semantic similarity estimates predicted his or her
own P-correlations a little bit better than the other participants’ P-correla-
tions. But the main source of mismatches was lack of reliability of the indi-
vidual P-factor patterns. Three findings support this view: (a) an odd-even
split of the 90 measurement occasions showed that the P-factor patterns of
individual participants were only moderately reliable, (b) the more reliable
an individual participant’s P-factor pattern was, the better was its match with
the R-factor reference structure, and (c) the semantic similarity estimates
predicted the averaged P-correlations more precisely than the individual P-
correlations.218 BORKENAU AND OSTENDORF Because we started with marker variables of the Big Five, it might be argued that our study is biased in favor of the five-factor model. Indeed, we did not conduct a state-taxonomic study that would encompass (a) identi- fying thousands of state-descriptive terms in a lexicon, (b) reducing their number by excluding synonyms and near-synonyms, (c) collecting self- descriptions by administering the reduced set of terms to the same partici- pants several hundred times, and (d) P-factoring these self-descriptions at the level of individual participants. Rather, we imported marker variables that had been established in a German large-scale trait-taxonomic study (Angleitner, Ostendorf, & John, 1990; Ostendorf, 1990) and checked whether these variables turned out as marker variables in P-factor analyses too. They do so pretty well, although better at the aggregate level than at the level of the individual participants. But even at the level of the averaged P-correlations, some discrepancies from the R-factor structure were found. There are at least two explanations for the similarities and discrepancies between the R-factor structure and the P-factor structure. These two explana- tions reflect a latent variable view and a conceptual similarity view. The latent-variable view suggests that latent tendencies give rise to correlations between observable variables and that the latent variables may differ between individuals in a similar way as they fluctuate within individuals over time. For example, enduring individual differences in Conscientiousness may pro- duce an R-correlation between persistence and responsibility, and fluctua- tions in Conscientiousness within individuals may produce a similar P-corre- lation. That the match between the Conscientiousness R-factor and the Conscientiousness P-factor is substantial but not perfect may then reflect that individual differences and longitudinal variations in Conscientiousness have somewhat different effects on observable behavior. For example, individual differences in trait conscientiousness may affect individual differences in recklessness and changeability, whereas longitudinal variations in state con- scientiousness may not affect fluctuations in these two observable character- istics. According to the conceptual similarity view, ratings on adjective scales correlate because they refer to overlapping sets of observations. For example, ratings of industrious and persistent would be correlated because instances of diligence tend to be instances of persistence too (Borkenau, 1986, 1988; Romer & Revelle, 1984). Moreover, these conceptual relations should be quite independent of whether the adjectives are used to describe enduring individual differences or longitudinal variations within persons. Thus per- sons who are more diligent than others tend to be also more persistent than others and, for the same reason, a person tends to be more persistent today than yesterday if that person is more diligent today than yesterday. But to some extent the meaning of personality-descriptive terms also de- pends on context. For example, if the adjective changeable is used to de-
INTRAINDIVIDUAL STRUCTURES 219
scribe individual differences, its meaning may differ from a context in which
this term is used to describe variations over time. According to the present
findings, this adjective is likely to refer to an aspect of Conscientiousness
in trait descriptions, but is likely to refer to an aspect of Emotional Stability
in state descriptions.
As far as we recovered the Big Five in our P-factor analyses, our findings
are at variance with one argument by critics of the five-factor model, namely,
that this model ignores the malleability of individual behavior (McAdams,
1992; Pervin, 1994). Because trait approaches emphasize the enduring as-
pects of individual differences, this would be a strong argument if a five-
factor approach were necessarily a trait approach. But our findings show
that the five-factor model is not ‘‘wedded’’ to the trait approach. A more
appropriate analogy is that the five-factor model is a child of the trait ap-
proach: It would probably not exist without the trait approach; it keeps more
or less intimate bonds to the trait approach, but it is capable of becoming
fully independent. And as far as the five-factor model is useful to describe
variations in psychological states over time, it has to be sharply distinguished
from the trait approach.
REFERENCES
Angleitner, A., Ostendorf, F., & John, O. P. (1990). Towards a taxonomy of personality de-
scriptors in German: A psycho-lexical study. European Journal of Personality, 4, 89–
118.
Barrett, P. (1986). Factor comparison: An examination of three methods. Personality and
Individual Differences, 7, 327–340.
Borkenau, P. (1986). Toward an understanding of trait interrelations: Acts as instances for
several traits. Journal of Personality and Social Psychology, 51, 371–381.
Borkenau, P. (1988). The multiple classification of acts and the big five factors of personality.
Journal of Research in Personality, 22, 337–352.
Borkenau, P. (1990). Traits as ideal-based and goal-derived social categories. Journal of Per-
sonality and Social Psychology, 58, 381–396.
Borkenau, P. (1992). Implicit personality theory and the five-factor model. Journal of Person-
ality, 60, 295–329.
Cattell, R. B. (1946). The description and measurement of personality. New York: World
Book.
Cattell, R. B. (1951). P-technique, a new method for analyzing the structure of personal moti-
vation. New York Academy of Sciences Transactions, 2(14), 29–34.
Cattell, R. B. (1955). The chief invariant psychological and psycho-physical functional units
found by P-technique. Journal of Clinical Psychology, 11, 319–343.
Cattell, R. B. (1957). Personality and mood by questionnaire. San Francisco: Jossey-Bass.
Cattell, R. B., Cattell, A. K. S., & Rhymer, R. M. (1947). P-technique demonstrated in de-
termining psychophysiological source traits in a normal individual. Psychometrika, 12,
267–288.
Cattell, R. B., & Cross, K. P. (1952). Comparison of the ergic and self-sentiment structures
found in dynamic traits by R- and P-techniques. Journal of Personality, 21, 250–271.220 BORKENAU AND OSTENDORF
Chaplin, W. F., John, O. P., & Goldberg, L. R. (1988). Conceptions of states and traits: Dimen-
sional attributes with ideals as prototypes. Journal of Personality and Social Psychology,
54, 541–557.
D’Andrade, R. G. (1974). Memory and the assessment of behavior. In H. M. Balock (Ed.),
Measurement in the social science (p. 159–186). Chicago: Aldine–Atherton.
Feldman, L. A. (1995). Valence focus and arousal focus: Individual differences in the structure
of affective experience. Journal of Personality and Social Psychology, 69, 153–166.
Gara, M. A., & Rosenberg, S. (1981). Linguistic factors in implicit personality theory. Journal
of Personality and Social Psychology, 41, 450–457.
Goldberg, L. R. (1993). The structure of phenotypic personality traits. American Psychologist,
48, 26–34.
Hakel, M. D. (1969). Significance of implicit personality theories for personality research and
theory. Proceedings of the 77th Convention of the American Psychological Association,
4, 403–404.
Hofstee, W. K. B., De Raad, B., & Goldberg, L. R. (1992). Integration of the Big Five and
circumplex approaches to trait structure. Journal of Personality and Social Psychology,
63, 146–163.
Hogan, R. (1983). A socioanalytic theory of personality. In M. Page (Ed.), Nebraska Sympo-
sium of Motivation (pp. 55–89). Lincoln, NE: University of Nebraska Press.
McAdams, D. (1992). The five-factor model in personality: A critical appraisal. Journal of
Personality, 60, 329–361.
McCrae, R. R., & Costa, P. T. (1995). Trait explanations in personality psychology. European
Journal of Personality, 9, 231–252.
McCrae, R. R., & John, O. P. (1992). An introduction to the five-factor model and its applica-
tions. Journal of Personality, 60, 175–215.
McCrae, R. R., Zonderman, A. B., Costa, P. T., Bond, M. H., & Paunonen, S. V. (1996).
Evaluating replicability of factors in the revised NEO Personality Inventory: Confirma-
tory factor analysis versus Procrustes rotation. Journal of Personality and Social Psychol-
ogy, 70, 552–566.
Norman, W. T., & Goldberg, L. R. (1966). Raters, ratees, and randomness in personality
structure. Journal of Personality and Social Psychology, 4, 681–691.
Ostendorf, F. (1990). Sprache und Persönlichkeitsstruktur: Zur Validitaet des Fünf-Faktoren-
Modells der Persoenlichkeit [Language and personality structure: On the validity of the
five-factor model of personality]. Regensburg, Germany: Roderer.
Passini, F. T., & Norman, W. T. (1966). A universal conception of personality structure?
Journal of Personality and Social Psychology, 4, 44–49.
Paunonen, S. V. (1997). On chance and factor congruence following orthogonal Procrustes
rotation. Educational and Psychological Measurement, 57, 33–59.
Paunonen, S. V., Jackson, D. N., Trzebinski, J., & Foersterling, F. (1992). Personality structure
across cultures: A multimethod evaluation. Journal of Personality and Social Psychology,
62, 447–456.
Pervin, L. A. (1994). A critical analysis of current trait theory. Psychological Inquiry, 5, 103–
113.
Romer, D., & Revelle, W. (1984). Personality traits: Fact or fiction? A critique of the Shweder
and D’Andrade systematic distortion hypothesis. Journal of Personality and Social Psy-
chology, 47, 1028–1042.INTRAINDIVIDUAL STRUCTURES 221
Russell, J. A. (1980). A circumplex model of affect. Journal of Personality and Social Psychol-
ogy, 39, 1161–1178.
Shweder, R. A. (1975). How relevant is an individual difference theory of personality? Journal
of Personality, 43, 455–484.
Watson, D., & Tellegen, A. (1985). Toward a consensual structure of mood. Psychological
Bulletin, 98, 219–235.
Wrigley, C. S., & Neuhaus, J. O. (1955). The matching of two sets of factors. American
Psychologist, 10, 418–419.
Zevon, M. A., & Tellegen, A. (1982). The structure of mood change: An idiographic /nomo-
thetic analysis. Journal of Personality and Social Psychology, 43, 111–122.You can also read
