The Structure
of
Phenotypic Personality Traits
Lewis
R. Goldberg
This personal historical article traces
the
development
of
the Big-Five factor structure, whose growing acceptance
by personality researchers has profoundly influenced
the
scientific study of individual differences.
The
roots of this
taxonomy
lie in the
lexical hypothesis
and
the insights
of
Sir Francis Gallon, the prescience of
L.
L.
Thurstone,
the
legacy of Raymond B. Cattell,
and the
seminal analyses
of Tupes
and
Christal. Paradoxically,
the
present popu-
larity of this model owes much
to its
many critics, each
of whom tried
to
replace
it,
but
failed.
In
reaction, there
have been
a
number of attempts to assimilate other models
into
the
five-factor structure. Lately, some practical
im-
plications of the emerging consensus can
be
seen
in
such
contexts
as
personnel selection
and
classification.
O
nce upon
a
time,
we had no
personalities
(Mi-
schel, 1968). Fortunately times change,
and the
past decade
has
witnessed
an
electrifying burst
of interest
in
the most fundamental problem of the field—
the search
for a
scientifically compelling taxonomy
of
personality traits. More importantly,
the
beginning
of a
consensus
is
emerging about
the
general framework
of
such
a
taxonomic representation.
As a
consequence,
the
scientific study of personality dispositions, which had been
cast into
the
doldrums
in the
1970s,
is
again
an
intellec-
tually vigorous enterprise poised on the brink of a solution
to
a
scientific problem whose roots extend back
at
least
to Aristotle.
The Lexical Hypothesis
Sir Francis Galton
may
have been among
the
first
sci-
entists
to
recognize explicitly
the
fundamental "lexical
hypothesis"—namely, that the most important individual
differences in human transactions will come to be encoded
as single terms
in
some
or all of
the world's languages.
Moreover, Galton (1884)
was
certainly
one of the
first
scientists
to
consult
a
dictionary as
a
means of estimating
the number of personality-descriptive terms
in
the lexicon
and
to
appreciate
the
extent
to
which trait terms share
aspects of their
meanings.
Galton's estimate of the number
of personality-related terms in English was later sharpened
empirically, first by Allport and Odbert (1936), who culled
such terms from
the
second edition
of
Webster's
Un-
abridged Dictionary,
and
later
by
Norman (1967),
who
supplemented
the
earlier list with terms from
the
third
edition. Galton's insight concerning the relations among
personality terms
has
been mirrored
in
efforts
by
later
investigators
to
discover the nature
of
those relations,
so
as
to
construct
a
structural representation
of
personality
descriptors.
One of the first of these investigators was L. L. Thur-
stone,
a
pioneer
in the
development
of
factor analysis;
the report
of
his initial findings reads today with almost
haunting clairvoyance:
Sixty adjectives that
are in
common
use for
describing people
.
.
. were given
to
each
of
1300 raters. Each rater
was
asked
to
think
of
a person whom
he
knew well
and to
underline every
adjective that
he
might
use in a
conversational description
of
that person.
... the ...
correlation
. . .
coefficients
for the
sixty personality traits
were
then analyzed by means of multiple
factor methods and
we
found that
five
[italics
added]
factors are
sufficient
to
account for the coefficients..
. .
It is of considerable psychological interest to know that the
whole list of sixty adjectives can
be
accounted for
by
postulating
on\y
five
independent common factors.
...
we did not foresee
that the list could be accounted for by as few
.
. .
factors. This
fact leads
us to
surmise that
the
scientific description of person-
ality
may
not
be
quite
so
hopelessly complex
as it is
sometimes
thought
to be
[italics added]. (Thurstone, 1934,
pp.
12-14)
The Big Five
Curiously, Thurstone never followed up his early anal-
ysis of these 60 adjectives and instead elected to rean-
alyze the questionnaire scales developed by Guilford.
Oblique rotations of 13 Guilford scales (Thurstone,
1951) led to the development of the seven factors in the
Thurstone Temperament Schedule (Thurstone, 1953),
two of whose scales intercorrelated over .70. Thur-
stone's devotion to oblique rotations in factor analysis
was mirrored by Raymond B. Cattell, who began his
personality explorations with a perusal of the approx-
imately 4,500 trait-descriptive terms included in the
Allport and Odbert (1936) compendium. Cattell (1943)
used this trait list as a starting point, eventually devel-
oping a set of
35
highly complex bipolar variables, each
pole of which included a composite set of adjectives
and phrases. These variables were then used in various
studies, in each of which the correlations among the
Lewis
R.
Goldberg
is at the
Department
of
Psychology, University
of
Oregon
and at the
Oregon Research Institute.
This article was
a
keynote address to the Sixth European Conference
on Personality, sponsored
by the
European Association
of
Personality
Psychology on June 16-19, 1992,
in
Groningen, The Netherlands. Work
on this address was supported
by
Grant MH-39077 from
the
National
Institute
of
Mental Health.
I
am
much indebted
to
Paul T. Costa, Jr., John
M.
Digman, Oliver
P.
John, Robert R. McCrae, Warren T. Norman, Dean Peabody, Gerard
Saucier, Auke Tellegen,
and
Jerry
S.
Wiggins
for
their many thoughtful
suggestions.
Correspondence concerning this article should be addressed to Lewis
R. Goldberg, Oregon Research Institute, 1899 Willamette Street, Eugene,
OR
97401.
Electronic mail may
be
sent
to
lewg@ori.org.
26
January
1993 •
American Psychologist
Copyright 1993
by the
American Psychological Association, Inc. 00O3-O66X/93/S2.0O
Vol. 48,
No. I,
26-34
variables were factored using oblique rotational
pro-
cedures (e.g., Cattell, 1947).
Cattell
has
repeatedly claimed
to
have identified
at
least
a
dozen oblique factors. However, when Cattell's
variables were later analyzed
by
others, only five factors
have proven
to be
replicable (e.g., Digman
&
Takemoto-
Chock, 1981;Fiske, 1949; Norman, 1963; Smith,
1967;
Tupes
&
Christal, 1961). Similar five-factor structures
based
on
other sets
of
variables have been reported
by a
number of other investigators (e.g., Borgatta, 1964a; Dig-
man
&
Inouye, 1986; Goldberg,
1990,
1992; McCrae
&
Costa, 1985a, 1987),
and
these studies have
now
been
reviewed extensively (e.g., Digman,
1990;
John,
1990;
McCrae & John, 1992; Wiggins &Pincus, 1992; Wiggins
& Trapnell,
in
press).
These Big-Five factors have traditionally been num-
bered
and
labeled, Factor
I,
Surgency
(or
Extraversion);
Factor
II,
Agreeableness; Factor
III,
Conscientiousness;
Factor
IV,
Emotional Stability
(vs.
Neuroticism);
and
Factor
V,
Culture.
1
More recently, Factor
V has
been
reinterpreted
as
Intellect (e.g., Digman
&
Takemoto-
Chock, 1981; Peabody
&
Goldberg, 1989)
and as
Open-
ness
to
Experience (e.g., McCrae
&
Costa, 1987). Given
the strong consensus that
has
been emerging about
the
general nature of the Big-Five domains, the disagreement
about
the
specific nature
of
Factor
V is
somewhat
of a
scientific embarrassment.
Although there
is
some disagreement about the pre-
cise nature
of
these five domains, there
is
widespread
agreement that some aspects of the language of personality
description
can be
organized hierarchically (e.g., Cantor
& Mischel, 1979; Hampson, John,
&
Goldberg, 1986).
In such
a
representation, the Big-Five domains are located
at
the
highest level that
is
still descriptive
of
behavior,
with only general evaluation located
at a
higher and more
abstract level (John, Hampson, & Goldberg, 1991). When
thus viewed hierarchically,
it
should
be
clear that
pro-
ponents
of
the five-factor model have never intended
to
reduce the rich tapestry of personality
to a
mere five traits
(e.g., Shweder
&
Sullivan, 1990). Rather, they seek
to
provide
a
scientifically compelling framework
in
which
to organize
the
myriad individual differences that char-
acterize humankind.
Indeed, these broad domains incorporate hundreds,
if
not
thousands,
of
traits: Factor
I
(Surgency
or
Extra-
version) contrasts such traits
as
talkativeness, assertive-
ness,
and
activity level with traits such as silence, passivity,
and reserve; Factor
II
(Agreeableness
or
Pleasantness)
contrasts traits such
as
kindness, trust,
and
warmth with
such traits as hostility, selfishness,
and
distrust; Factor
III
(Conscientiousness
or
Dependability) contrasts such traits
as organization, thoroughness,
and
reliability with traits
such as carelessness, negligence,
and
unreliability; Factor
IV (Emotional Stability
vs.
Neuroticism) includes such
traits
as
nervousness, moodiness, and temperamentality;
and Factor
V
(whether labeled
as
Intellect
or
Openness
to Experience) contrasts such traits
as
imagination,
cu-
riosity,
and
creativity with traits such
as
shallowness
and
imperceptiveness.
From Critic to Proponent
It might
be
argued that
the
hallmark
of a
compelling
structural model
is
that
it is
initially disliked, thereby
stimulating numerous attempts
to
replace
it
with some-
thing more attractive—all
of
which fail.
In any
case,
so
it has been with
the
Big-Five model
of
perceived person-
ality trait descriptors. Most of the present proponents
of
the model were once
its
critics,
and
some
of
its present
critics contributed
to its
success. Indeed,
as
sympatheti-
cally articulated
by
Wiggins
and
Trapnell
(in
press),
the
intellectual "father"
of the
Big-Five factors, Raymond
Cattell,
has
consistently denied
his
paternity
and has yet
to embrace
the
model.
The Accidental Discoverer (Fiske)
Whereas Thurstone (1934) found
the
correct number
of
broad personality factors,
his
collection
of
60
trait adjec-
tives
was too
idiosyncratically assembled
to
have
pro-
duced today's Big-Five structure. Instead,
the
honor
of
first discovery must
be
accorded
to
Donald Fiske (1949),
who analyzed
a set of
22 variables developed
by
Cattell
and found five factors that replicated across samples
of
self-ratings, observer ratings,
and
peer ratings. Fiske's
la-
bels
for his
factors, like those proposed
by
subsequent
investigators, were never perfectly successful attempts
to
capture the prototypical content of these broad domains:
Confident Self-Expression
(I),
Social Adaptability
(II),
Conformity (III), Emotional Control (IV),
and
Inquiring
Intellect (V). Like Thurstone before him, however, Fiske
did
not
follow
up his
initial findings. Indeed, these early
histories read like that
of
Leif Erikson,
who
made
one
voyage
of
discovery, found
a
continent,
but
never
re-
turned.
The True Fathers (Tupes
and
Christal)
To the extent that anyone other than Cattell deserves
the
real credit
for
initiating this complex saga, that honor
belongs
to
Tupes and Christal (1958, 1961), who analyzed
the findings from
a
number
of
studies that used sets
of
variables developed
by
Cattell (including Fiske, 1949)
and
found five replicable factors. Their classic 1961
Air
Force
technical report has now been reprinted (Tupes & Chris-
tal,
1992), along with
an
appreciatory editorial by McCrae
(1992)
and a
historical introduction
by
Christal (1992).
In
the
latter, Christal described
the
remarkable series
of
U.S.
Air Force studies carried out between 1954 and 1961,
including investigations of the long-term predictive valid-
1
As was
pointed
out by
Peabody
and
Goldberg (1989),
the
inter-
pretation
of
Factor
V as
Culture arose from
a
historical accident:
Al-
though Cattell
had
initially constructed
a
subset
of
variables relating
to
Intellect,
in the
seminal studies of Cattell (1947)
he
omitted
all of
those
variables
in
favor
of an
intelligence test.
In
turn, this test
was
omitted
from his later studies, leaving
no
direct representation
of
most Intellect
variables.
In
their absence, Factor
V was
called Culture
by
Tupes
and
Christal (1961)
and
some later investigators. However, when variables
related
to
Intellect were reintroduced (e.g., Goldberg, 1990),
it
became
clear that Intellect
was the
more appropriate label
for the
fifth broad
factor.
January
1993 •
American Psychologist
27
ity of peer ratings, a study of the effect of length of ac-
quaintance on the accuracy of such ratings, a comparison
of the factor structures derived from variables presented
in three different response formats, and—of course—the
seminal comparisons of factor structures across diverse
samples.
Other Early Explorers (Borgatta and Smith)
Borgatta (1964a, 1964b) and Smith (1967, 1969) both
reacted to the work of Tupes and Christal (1958, 1961)
with their own independent studies and found much the
same five-factor structure. Borgatta (1964a) compared the
structures derived from self-ratings with those from peer
rankings and peer ratings in two samples and found five
robust factors, which he labeled Assertiveness (I), Lik-
ability (II), Responsibility (HI), Emotionality (IV), and
Intelligence (V). Smith (1967) compared the structures
derived from three large samples
(TV
= 583,
521,
and 324)
and found five robust factors, which he labeled Extraver-
sion (I), Agreeableness (II), Strength of Character (HI),
Emotionality (IV), and Refinement
(V);
moreover, Smith
found that scores derived from the Strength of Character
factor correlated .43 with college grades. In a later peer-
nomination study, Smith (1969) recovered Factors I-IV
in male and female samples at both the junior high school
and high school levels; moreover, he found high corre-
lations with smoking status, smoking being negatively re-
lated to Factors II (Agreeableness) and III (Strength of
Character), and positively related to Factor I (Extraver-
sion).
Neither investigator, however, seems to have carried
out any systematic follow-up research, and their role in
the history of the Big-Five factors is now typically rele-
gated to a footnote.
The First Serious Critic (Norman)
Warren Norman, often erroneously labeled the father of
the Big-Five structure, spent much of his early research
career as a skeptic. Paradoxically, although his importance
in the history of the five-factor model is universally ac-
knowledged, his refusal to become a true believer has
typically been overlooked. Yet, after his seminal studies
confirming the five-factor model with a selected set of
Cattell variables (Norman, 1963), he instituted an exten-
sive research program aimed at replacing that model with
a more comprehensive one. He began by expanding the
corpus of English personality terms assembled by Allport
and Odbert (1936), then classifying the terms in the ex-
panded pool into such categories as states, traits, and roles,
and finally collecting normative information about some
2,800 trait-descriptive terms (Norman, 1967). Norman
was convinced that because of the inevitable computa-
tional and other technical limitations of research in the
1930s and 1940s, Cattell's variables left much to be de-
sired, and therefore that studies using a representative
subset of the total English personality-trait lexicon would
uncover dimensions beyond the Big Five. Although Nor-
man himself never tested this appealing conjecture, others
did (e.g., Goldberg, 1990), and it has proven to be wrong.
The Second Serious Critic (Digman)
The first computer was coming to the University of Ha-
waii, and John Digman (an experimental psychologist
with no interest in either personality or developmental
psychology) set out to learn to program it. As an initial
data set to be used to test his ability to program the new
machine, Digman decided to try to replicate the child
personality factors obtained by Cattell and Coan (1957),
inasmuch as he had easy access to teacher ratings at Ha-
waii's University Laboratory School. Using the Cattell
variables, Digman believed he had found eight oblique
factors, some of which differed from those found by Cat-
tell and Coan.
2
Intrigued by the discrepant findings, Dig-
man (1965) conducted a second teacher-rating study,
adding new variables to the original set. Some of Digman's
10 new oblique factors matched neither those from Cattell
and Coan nor those from his own first study (Digman,
1963).
In an attempt to make sense of these dissonant
findings, Digman reanalyzed the data from a number of
previous studies of teachers' ratings of children and con-
cluded that 7 oblique factors were robust across the sam-
ples (Digman, 1972). In further studies, using different
sets of variables, Digman became convinced that there
were at least 10 oblique factors of child personality (and
perhaps more at the adult level), a view he espoused as
recently as 1977.
In the spring of
1978,
however, Digman was to teach
a course in factor analysis, for which he obtained a num-
ber of correlation matrices from classic studies of abilities
and personality traits, including those previously analyzed
by Tupes and Christal (1961) and by Norman (1963).
Before providing them to his students for reanalysis,
however, he checked them carefully and found clerical
errors in the matrices of two of Cattell's studies. More
importantly, he discovered that when six or more factors
were rotated from the various matrices, the factors did
not correspond, whereas when five factors were rotated,
there was striking interstudy correspondence (Digman &
Takemoto-Chock, 1981). Having set out to prove the cor-
rectness of
a
10-factor model, Digman regretfully became
convinced of the robustness of the Big Five. Later teacher-
rating research by Digman and Inouye (1986) again ob-
tained the Big-Five factors, now with a revised set of child
personality variables.
More Recent Critics (Peabody and Goldberg)
Like Cattell, both Norman and Digman had initially as-
sumed that the dimensionality of phenotypic personality
traits was quite large—certainly larger than five. The social
psychologist Dean Peabody, on the other hand, had used
trait-descriptive adjectives in a series of trait-inference
2
An interesting historical footnote: Digman presented these findings
at a meeting attended by Cattell, who requested a copy of Digman's
correlations. Without notifying Digman, Cattell then re-factored these
correlations, rotated 12 oblique factors (which he claimed replicated his
own),
and published his solution before Digman had submitted his own
report for publication (Digman, 1963). Indeed, Digman only discovered
Cattell's (1963) article when he was sent a reprint.
28
January 1993 • American Psychologist
studies, in each of which he uncovered only three broad
dimensions. Peabody's ideas and procedures were ingen-
ious:
Using a dictionary as a starting place, Peabody
(1967) concocted various sets of four terms, so that within
each set the descriptive and evaluative aspects of trait
meanings were systematically unconfounded. For ex-
ample, one set of four included the traits generous, thrifty,
extravagant, and stingy—the first two traits being desir-
able and the last two undesirable, the first and third re-
ferring to loose features and the second and fourth to
tight features descriptively. Subjects made trait inferences
within each set of four traits (e.g., If one is generous, how
likely is one to be thrifty [same evaluative valence but
opposite as descriptively] as compared with extravagant
[opposite valence but similar descriptively]?).
In a series of factor analyses of subjects' trait infer-
ences across various sets of
traits,
Peabody (1967, 1970,
1978,
1984) uncovered a dimension of general evaluation,
plus two descriptive dimensions, which he labeled asser-
tiveness and tight versus loose (impulse control versus
expressiveness). In other studies, Peabody (1968, 1985)
applied this model to national characteristics, finding
widespread agreement on the descriptive features asso-
ciated with a particular nationality, but disagreement on
the evaluations (e.g., We are thrifty [generous], whereas
they are stingy [extravagant]).
Although Peabody's three-factor structure had little
impact on the general field of personality structure, it had
a profound impact on my own thinking. Both Peabody
and Norman were among my closest professional col-
leagues, and I followed their divergent paths with growing
consternation. Peabody appeared to become increasingly
wedded to his three-factor model, whereas Norman (and
later Digman) were convincing many of us that there must
be at least five orthogonal trait dimensions. How was one
to reconcile these two structural representations? On the
one hand, one could simply disregard Peabody's three-
factor model in favor of the more inclusive five-factor
structure. On the other hand, to my scientific tastes, the
Peabody model was elegant and beautiful, whereas the
five-factor structure was an aesthetic nightmare: All of
the Big-Five factors but the first (Surgency) were highly
related to evaluation, and the dimensions themselves had
little appeal to me. In contrast, the Peabody model iso-
lated general evaluation as an important dimension of
personality perception (which seemed logically compel-
ling),
thereby assuring that the remaining dimensions
would be evaluation free (which seemed desirable).
Moreover, although Peabody has forcefully and consis-
tently denied any similarities between his three factors
and the three E-P-A (evaluation, potency, and activity)
dimensions of affective meaning discovered in the se-
mantic-differential studies of Osgood, Suci, and Tannen-
baum (1957), I was far from convinced. General evalu-
ation is identical in both representations, whereas potency
and activity in the Osgood model share important features
with Peabody's dimensions of assertiveness and impulse
expression. I found these theoretical links intriguing.
In my early work (e.g., Goldberg, 1982), I cham-
pioned the Peabody model over the Big Five. I expanded
on the general idea behind Peabody's sets of four traits
to develop clusters of quasi-synonyms and quasi-anto-
nyms grouped into tables we called Peabody plots, in
which descriptively similar trait adjectives were listed on
the right-hand side of a page, all their antonyms were
listed in the left-hand side, and all terms of both types
were ordered vertically by their evaluations (indexed by
their mean social desirability values from Norman, 1967).
Using a variety of different sets of
terms,
I produced nine
rounds of taxonomies of trait-descriptive adjectives
(Goldberg, 1982), as well as a taxonomy of trait-descrip-
tive nouns (Goldberg, 1980a).
However, at the same time as I was developing these
essentially armchair taxonomies, I was administering
large sets of trait-descriptive adjectives to samples of
sub-
jects for self-ratings, peer ratings, or both ("external" data)
and administering smaller subsets of trait adjectives to
other samples for ratings of semantic similarity
("inter-
nal"
data). During the decade roughly from 1975 to 1985,
I was continuously carrying out analyses of these various
data sets in an effort to discover a scientifically compelling
taxonomic structure. It was as if I were looking through
a glass darkly: In each analysis, I would discover some
variant of the Big-Five factors, no two analyses exactly
the same, no analysis so different from the rest that I
couldn't recognize the hazy outlines of the five domains.
For nearly a decade I wandered as if in a fog, never certain
how to reconcile the differences obtained from analysis
to analysis.
During all of that period I kept searching for a short-
hand notational system for labeling variants of the Big
Five—something akin to the names of classic chess po-
sitions—so that for each analysis I would have a way to
refer to its particular factor locations. Ultimately, I wanted
to link each of the particular Big-Five variants I had been
finding with characteristics of the data—such as the type
of item pool, the nature of the subject sample, the use of
unipolar as compared with bipolar variables, and the kind
of data-analytic procedure. In the absence of such a no-
tational system, I published none of the findings from
my structural analyses, but continued to collect additional
data and reflect on the findings from each new analysis.
3
These analyses eventually led me away from my in-
fatuation with the three Peabody factors; I couldn't shake
the fact that analyses of any reasonably representative
pool of common trait adjectives always provided evidence
for five broad factors, rather than for three. Peabody, on
the other hand, was still not convinced. To resolve our
disagreement, Peabody proposed that we mutually agree
on a representative set of bipolar trait scales and then that
3
Some of these analyses of my inventories of
566
and 587 adjectives
were presented at a Western Psychological Association symposium or-
ganized by John Digman (Goldberg, 1980b) and then eventually pub-
lished in Goldberg (1990), whereas some of the findings from smaller
item pools, including both bipolar and unipolar Big-Five factor markers,
were published in Goldberg (1992).
January 1993 • American Psychologist
29
we apply that representative set in a variety of samples
of external and internal data. The representative set of
scales was developed by Peabody (1987), and the findings
from our analyses of those scales in seven data sets were
reported in Peabody and Goldberg
(1989).
In that article,
we showed that the factor structures derived from external
and internal data were quite similar, but not identical,
and that they included five (external) or six (internal) or-
thogonal dimensions. Moreover, we were able to incor-
porate the three Peabody factors into the Big-Five struc-
ture.
What was still unclear to me, however, was how to
secure agreement on the exact positioning of the factor
axes in this five-dimensional space. In analyses of external
data, the simple-structure position as obtained by varimax
rotation will inevitably vary somewhat from sample to
sample, even when the same variables are analyzed; one
probably needs samples of around 1,000 or more to at-
tenuate such intersample perturbations. Moreover, the
simple-structure position is affected enormously by the
selection of variables, and no increase in sample size can
counteract the effect of this important determinant of
factor location. In addition, as Peabody and Goldberg
(1989) have demonstrated, the simple-structure location
will change as a function of the evaluative homogeneity
versus heterogeneity of the targets being described (e.g.,
self-ratings or ratings of friends vs. ratings of both liked
and disliked targets). And, as the analyses reported in
Goldberg (1992) suggested, factor locations may depend
on whether the ratings are obtained using unipolar or
bipolar scales.
In a way, this situation is similar to that faced by
early cartographers as they struggled to provide maps of
the emerging world. Because the earth is a sphere, any
set of orthogonal three-dimensional coordinates could be
used to map that world with equal precision. And, just
as cartographers eventually settled on a standard system
with north-south and east-west axes, so personality re-
searchers must settle on a standard set of locations for
the Big-Five dimensions. My efforts to develop factor
markers (Goldberg, 1992) are meant to be a step in that
direction. Ultimately, the field will form a consensus about
the "best" factor locations, a consensus that will be in-
fluenced by both aesthetic and practical considerations
(Briggs, 1992).
4
The Assimilators (Costa/McCrae and Wiggins)
At present, one could argue that there are two five-factor
models, one developed by McCrae and Costa (1985a,
1987) and operationalized in the NEO Personality In-
ventory (NEO-PI; Costa & McCrae, 1985) and the other
associated with studies based on the lexical hypothesis
and operationalized in the sets of factor markers provided
by Norman (1963), Peabody and Goldberg (1989), Gold-
berg (1990, 1992), John (1989), Trapnell and Wiggins
(1990),
and Digman and his associates (e.g., Digman,
1989;
Digman & Inouye, 1986). Much is the same in
both models: (a) The number of dimensions is identical,
namely five; (b) the content of Factor IV is essentially the
same, although it is oriented in the opposite direction in
the two models and is thus so labeled (Emotional Stability
versus Neuroticism); and (c) there is considerable simi-
larity, although not identity, in the content of Factor III
(Conscientiousness). On the other hand, at least two of
the differences between the models are quite striking: (a)
The locations of Factors I and II are systematically rotated
so that warmth is a facet of Extraversion in the NEO-PI,
whereas it is a facet of Agreeableness in the lexical model;
and (b) Factor V is conceived as Openness to Experience
in the NEO-PI and as Intellect or Imagination in the
lexical model.
5
These differences stem from the history of the NEO-
PI,
which started out as a questionnaire measure of a
three-factor model, including only Neuroticism, Extra-
version, and Openness to Experience. Whereas other
three-factor theorists, such as Eysenck (1991), have stood
firm as proponents of their original representations, Costa
and McCrae reacted to the events of the early 1980s with
such remarkable openness to experience that by the end
of the decade these investigators had become the world's
most prolific and most influential proponents of the five-
factor model. Their startling transformation was initially
stimulated by two papers from a Digman-organized sym-
posium at the 1980 Western Psychological Association
convention—the first by Digman and Takemoto-Chock
(1981) and the second by me (Goldberg, 1980b)—plus
two chapters I published at that time (Goldberg, 1981,
1982).
As a reaction to those reports, in 1983 Costa and
McCrae invited me to visit them in Baltimore, where I
presented the findings from my first studies of bipolar
factor markers. My efforts to convince them that five or-
thogonal factors were necessary to account for phenotypic
personality differences (Goldberg, 1983) fell on receptive
ears;
indeed, they had already administered 40 of my
factor markers, along with 40 new rating scales of their
own, to their longitudinal sample, and we discussed the
preliminary findings during my visit. The rest, as they
say, is history (e.g., McCrae & Costa, 1985a, 1987). Two
4
One such quasi-aesthetic consideration has guided my own work.
It has long been known that the evaluation, potency, and activity (E-P-
A) dimensions of Osgood, Suci, and Tannenbaum (1957) are only or-
thogonal when a heterogeneous set of concepts are rated on a hetero-
geneous set of bipolar rating scales, and thus the judgments are primarily
metaphorical rather than descriptive. Specifically, when the concepts are
all real persons, the dimensions of potency and activity tend to fuse into
a construct called dynamism by the Osgood team. Surgency (Factor I)
in my preferred rotation of the Big-Five structure represents this fusion
of assertiveness and activity level, which explains its relative independence
of evaluation.
5
Actually, the differences between the two versions of the five-factor
model can be attenuated in the following ways: (a) The trait descriptor
warm has been classified in Facet II+/I+ in some Abridged Big-Five-
dimensional Circumplex (AB5C; Hofstee, de Raad, & Goldberg, 1992)
analyses in the lexical model, whereas warmth is considered a I+/II+
facet in the McCrae and Costa model, suggesting more agreement when
both primary and secondary loadings are considered than when one
considers the primary loadings alone, (b) As suggested by Saucier (in
press),
neither the labels Openness nor Intellect capture well the central
cluster of traits that define Factor V; perhaps a more apt label is Imag-
ination for a factor defined by such traits as creativity and curiosity.
30
January 1993 • American Psychologist