The Role of Factor Analysis in The Devel
The Role of Factor Analysis in The Devel
The Role of Factor Analysis in The Devel
Abstract
ITie purpose of this paper is to examine the usefulness of factor analysis in
developing and evaluating personality scales that measure hmited domain con-
structs The approach advocated follows from several assumptions that a smgle
scale ought to measure a smgle construct, that &ctor analysis ought to be applied
routinely to new personality scales, and that the factors of a scale are important
if it can be demonstrated that they are diiFerenbally related to other measures
A detailed study of the Self-Monitonng Scale illustrates how &ctor analysis can
help us to understand what a scale measures A second example uses the self-
esteem hterature to illustrate how factor analysis can clarify the proliferation of
scales within a single content domain Both examples show how fJEictor analysis
can he used to identify important conceptual distinctions Confirmatory tech-
niques are also introduced as a means for testing specific hypotheses It is con-
cluded that &ctor analysis can make an important contnbuhon to programmatic
research m personality psychology
cedures have proven to be useful and important tools in this regard be-
cause they allow an investigator to augment, refine, and test (in some
cases statistically) his or her intuitive grasp of an area and because they
provide a means by which to deal vwth vanables that are not only unob-
served but usually unobservable (latent vanables)
Factor analysis was first introduced by Spearman (1904) in his studies
of intelligence, and much of the early work with factor analysis dealt with
the structure of mtelhgence and the development of performance meas-
ures Personality measurement has always been somewhat of a younger
sibling to thefieldof ability testing, and thus it was not long before fector
analytic procedures were also being used to develop and refine global
personality measures and to define the structure of the trait universe
Although personality theonsts still disagree as to the central dimensions
of personality, there now seems to be at least some agreement as to the
nature of the disagreement Studies using adjective lists to describe
peers generally suggest that there arefiveor six dimensions that are cen-
tral to the trait universe (e g , see Digman & Inouye, m press, Digman
& Takemoto-Chock, 1981, Goldberg, 1981, Hogan, 1983, McCrae &
Costa, m press, and Wiggins & Broughton, 1985), whereas studies that
employ self-report lnventones tend to yield a larger number of pnmary
dimensions (e g , Cattell's 16 dimensions or Guilford's 13) which can in
tum be summanzed in terms of 2 or 3 higher-order dimensions or su-
perfectors (Eysenck & Eysenck, 1969, Royce & Powell, 1983)
In recent years, personality researchers have seemed less interested
m the development of global measures of personality and have instead
focused their attention more intently on single psychological constructs
of hmited domain (Maddi, 1984) Popular t(q)ics have included locus of
control, masculinity and femininity, need fw achievement, machiavelli-
anism, authoritananism, sensation-seeking, loneliness, self-esteem,
self-<xMisciousness, and self-monitonng to name but a few The purpose
of this paper ts to exarmne the usefulness of factor analysis as a meth-
odohgtcal toolfor advancing research on these constructs ofltnuted do-
main
A Bnef Overview erf Factor AralytK Methods
Mach erf" what we have to say m this pjq>er rests cm an lmpcHtant as-
sumptHm that a mtgile scale ought to measure a single construct TTus
IS Kit a novel CH- idH}syncratic point, nor CHie that we can attempt to take
110 Bnggs & Cheek
credit for Guilford made the point 30 years ago "any test that measures
more than one common factor to a substantial degree yields scores that
are psychologically ambiguous and very difficult to interpret" (1954, p
356) McNemar (1946) also speaks clearly on this issue "Measurement
implies that one charactenstic at a time is being quantified The scores
on an attitude scale are most meaningful when it is known that only one
continuum is involved Only then can it be claimed that two individuals
with the same score or rank can be quantitatively and, withm limits,
qualitatively similar in their attitude towards a given issue" (p 268)
Although we believe this premise to be quite reasonable, and one with
which most scale developers (as well as most readers) will be able to
agree, it raises two sorts of problems First, there is the problem of how
to specify the notion of a single construct both conceptually and opera-
tionally Several concepts are typically used when evaluating the ade-
quacy of a scale unidimensionahty, internal consistency, and homogene-
ity These concepts are all related to the notion that a scale ought to be
focused on a single construct, and they are often used interchangeably
when in fact they are not synonymous (McDonald, 1981) For mstance, a
scale can achieve a generally acceptable level of internal ccMisistency (as
measured by coeflScient alpha) and yet be relatively heterogeneous (as
measured by the mean correlation across pairs of items) Similarly, a
scale can reach generally acceptable levels of internal consistency and
homogeneity and still yield multiple factors We will explore this issue
more fuUy m a later example, but for now the point is that there is some
disagreement as to what specific indices and decision cntena are appro-
pnate when evaluating the adequacy of a scale (Hattie, 1984, McDcaudd,
1981)
Tlie second problem is that a number of well known and widely used
sc^es seem to run contrary to our assumption that single scales ought to
measure single constructs For example, popular scales such as the
Eysenck Personality Questionnaire, the Marlowe-Crowne Social Desir-
abihty Scale, and the Ho^n Empathy Scale all have mean mtentem cor-
relaticns of less than 10 In feet, Fiske (1971) suggests that the mean cxir-
relation between pairs of items for the typical test is less than 10
Furt}Krmc»-e, many pc^pular scales (including the three hsted above) are
known to be multifectonal (e g , see Howarth, 1976, Jc^nscHi, Cheek, &
Smthea-, 1983, I^ulhus, 1984)
In part, this traicfencyfiMrsc^es to be heterogei^ous and multifecto-
nal has to do wath the breadth w level of the construct being measured
Tlus lssiK can be well illustrated using the dimensicm of extraversion-
mtroversicm Eysenck (l!%3) pomts out that "One ofthe central problems
in perscmahty research has been the question of wtethra" higher-CM-der
as extraversicm can be regarded m any meaningfol sense as
Factor analysis 1-11
with Nunnally's (1978) assertion that "a measure should spring from a
hypothesis regarding the existence and nature of an attnbute" (p 277)
The Scale
Intentem Relatumshtps
Snyder constructed his scale from a pool of 41 self-descnptive items
which he culled to 25 using an index that discnminates individuals with
high total scores from those with low total scores This procedure pro-
vides a type of item-total correlation, and it is designed to maxtmiTX in-
ternal conststency (Nunnally, 1978, Snyder, 1972) Unfortunately, how-
ever, Nunnally also points out that this type of procedure does not
necessanly result m a homogeneous set of items
Indeed, the major cnticism that can be made of selecting items
in terms of correlations with total scores is that the method appar-
ently works as well when several groups of items relate strongly to
different factors as when all items relate only moderately to the
same fector Thus if there are several prominent factors m the
items, the problem is not so much that the item analysis will fail,
but rather that it will work deceivingly well (pp 284-285)
The Self-Monitonng Scale seems to be a case in point Although the
scale satisfies the usually accepted standards for internal consistency
with Kuder-Richardson Formula 20 coefficients of around 70 (e g , Sny-
der, 1974), even a simple reading ofthe scale suggests a fundamental dis-
similanty m the content ofthe items For instance, cxMnpare item 18 (I
have considered being an entertainer) with item 19 (In order to get along
and be liked, I tend to be what pec^le expect me to be rather than any-
thing else) Tme answers to both are scored in the directiCHi of high self-
monitonng, even though the first dearly invdves scK;iaI self-confidence
whereas the second suggests a dependency on others or a lack of inter-
personal assurance Moreover, this item heterogeneity can be docu-
mented statistically The mean ofthe 300 possible pairwise correlations
among the 25 items is only about 08 (Bnggs, Cheek, & Buss, 1980) In
part, this average coeflBcient is near zero because roughly one-quarter of
the intentem conrelations are negative The correlations vary from
armind — 30 to about + 50
Tlius, we are ccMifrcmted with a paradox The Sdf-Monitonng Scale
meets the normal recjuirements for internal rehabihty, yet many of its
items are unrelated This apparent cxmtradiction (xx;urs be<^use stan-
dard reliabihty estimates of internal ccmsistency are not pure measures
(tfitem h(Hnoga»eity Ku^-Richardscm Pmnula 20 is a ccxivafitumal m-
dex. of lnt^tial ccmsisteiK^, the vaost genersd tcxcta ofv ^ c h is Crcmbach's
Factor analysis 115
Acting tactor
I would probably make a good actor
I have considered being an entertainer
I have never been good at games like charwles or improvisational acting (R)
"Extraversion" fcK^tw
I feel a bit awkward in company and do not ^xm up quite as well as I should (R)
At a party I let others keep the ]okes and stones gmng (R)
In a group of peo(^ I am rarely the center of attention (R)
C^ther-directedness feKstor
In different situations arxt with different people, I often act like very different persons
In order to get along and be liked, I tend to be what people expect me to be rather than
anything else
I guess I put on a show to impress or entertain people
Notes —(R) denotes items to t>e recoded for scoring in the direction of high Self-Monitoring Items
based on Bnggs, Cheek, and Buss (1980)—s«e also Leary etal (1982), Sparacino etal (1983), and
Tobey&Tunnell(1981)
Factor Structure
Given that the items ofthe Self-Momtonng Scale are only minimally
related on average and given the heterogeneity ofthe item intercorrela-
tions, we can use fector analysis to detect pattems m the correlations
among items Analyses ofthe Self-Momtonng Scale have demonstrated
that items can indeed be grouped together into several distinct fectors
In our own research, we identified three replicable fectors and labelled
them acttng, "extraversion," and other-directedness (Bnggs et al ,
1980) Representative items from the^e fectors are presented in Ikble 1
Scalesformedfromthe highest loading items on each ofthe factors were
relatively independent (for extraversion and other-directedness, r =
— 11, ior extraversion and acting, r = 31, and for other-directedness
and acting, r = 13) TTie items cm these subscales also proved to be
more iMmogeneous than the items on the full Self-Mcmitonng Scale The
mean intentem correlatimis for the tcAal samjde m tl^ Bnggs et al study
was 29 tor acting, 27 iar extraversion, and 18 for other-directedness
compared to 08ft»"the full scale Gabrenya & Arkin (1980) identified a
similarfec^tca-structure altlwv^ they settled on a four-fiK:tor soluticm
Tl^se basicfectc»^have now been re^dicated m a number of {Kibhshed
studies (Cegala, &ivage, &unner, & Ccmrad, 1982, Ecklmann, 1985,
Fumham & Capcm, 1983, Leary, Silver, Darby, & Schlenker, 1982, Sen-
ior & Wymer, ]J983, Riggio & Fhedman, 1982, Sparacino, Rcmdu, l&g-
ley, Flesch, & Kuhn, 1983. and Tdbey & Tunnell, 1981)
Hiese lnvestigatitms reach surpisingly sunilar ccmdusicms given the
number of investigators involved and the variety of techmques em-
Factor analysis 117
ployed Although the studies differ with respect to at least four proce-
dural issues—type of item format, type of factor solution, type of rota-
tion, and the number of factors retained m thefinalsolution—as shown
in Table 2 mterpretations as to the nature ofthe denved factors show a
decided resemblance
The first difference concems what type of response format to use
when administenng the Self-Momtonng Scale—dichotomous or multi-
point Snyder introduced the scale with a true-false format and most of
the subsequent empincal research has retained that format In our
study, however, we employed a 5-point format in order to avoid limiting
the size of the correlations among items due to skewed frequencies of
endorsements (see Nunnally, 1978, pp 141-146) The correlation be-
tween total scores on the tme-false form and total scores on the 5-point
form (administered 45 days apart) is roughly equivalent to the reliability
of the item sample as measured by the alpha coefficient (both hover
around 70)
The second source of vanation m Table 2 involves the method of ex-
traction Although the vanous methods employed (e g , principal com-
ponent analysis, principal axis factonng, and maximum likelihood) differ
markedly in terms of their fundamental assumptions and computational
algonthms, apparently the factor structure ofthe Self-Momtonng Scale
IS sufficiently robust to emerge regardless of the particular procedures
emjdoyed This finding is not unusual, the different procedures repre-
sented often seem to yield similar solutions (e g , Velicer, Peacock, &
Jackson, 1982)
The third type of difference involves rotation ofthe extracted factors
Recall from our earlier discussion that there is no unique solution to the
problem of rotation in factor analysis Various statistical cntena have
been introduced to assist in the rotation to a "correct" solution, but no
one method is appropnate for all situations The vanous methods can be
divided into those which require that the factors be orthogonal or inde-
pendent (e g , Vanmax) and those which allcnvfectorsto correlate some-
what (e g , Direct Oblimm and Promax) Because the vanables in this
case all belong to the same scale (and thus should be related despite the
feet that the mean intentem correlation is low), an obbque form of rota-
tion would seem more appropnate smc^ any fectors to emerge would
presumably be some>^t correlated The similanty m findings across
tbe studies m Table 2 despite vanations m the method of rotation sug-
gests tiiat thefectCR-sare not highly correlated since forcing thefectorsto
be indepeiKlent did noi alter the results m any meaningful way
ThefinaldtSererkce among these vanous studies concems the metlxx}
for extracting the "proper" number of factors frora a correlaticm matnx
A number of objective imx^dures have been suggested fw use with ex-
Bnggs & Cheek
118
o o
c c c c
o o o o
0 o
o 01 u> *^ O)
CO ^ tf) fc_ W
T5
it £
A 5 "
® c£ c.
'
O 2 oo "
go. © ra c 2
ffTfffT
1
•s il"
I
I if if I I Ii
5 I 11 If I 5 «i
1 c
I I
•5
1 a.
o
o EI I I I
eg
S-
o
a)
fi-
p
a)
S-
o £
in t l t i l t •?
I a>
iiv C\l
iin
•a
I!
Factor analysis 119
set of items, even though the scale was developed so as to maximize in-
ternal consistency, and (2) part of the item heterogeneity can be ex-
plained m terms of several fectors that are present m the items In this
context, it IS interesting to note the origin of some ofthe items At least
one third of the items on the Self-Monitonng Scale can be traced back
to other published mventones Three of the items (6, 16, and 19) were
acquired from items 14 and 26 of Berger's (1952) Expressed Acceptance
of Self Scale Two others (3 and 17) resemble items 4 and 8 from a self-
esteem scale by Phillips (1951) and four more (4, 5, 8, and 21) denve from
measures of lnternality and externality reported by Collins, Martm,
Ashmore, and Ross (1973) In fact, all nme of the borrowed items were
included m the study by Collins et al which identified dimensions of
internal and extemal onentation For our purposes, the important find-
ingfromtheir study is that the nine items loaded on two distinct factors
One factor included items 19, 6, 16, 17, and 3 from the Self-Monitonng
Scale All of these items load on our other-directedness factor (Collins
et al also labelled their factor Other-directedness, a remarkable coin-
cidence ) The other factor included items 5, 8, 21, and 4 from the Self-
Mcmitonng Scale Two of these items load on our Acting factor, one on
the Extraversion fector, and one not at all Colhns et al labelled their
factor lack qf constraints on behavior and showed that it correlated 37
with Eysenck's Extraversion Scale As m our research, these two factors
were relatively unrelated It would seem, therefore, that the multidi-
mensionality ofthe Self-Monitonng Scale can be traced directly back to
the pool of items from which it was constmcted
As we pointed out earher, however, identifying a replicable factor
struciture m no way ensures the importance or usefulness of those fectors
(Lanyon & Goodstein, 1982) The vabdity of the factors must be dem-
onstrated in the same way as any other new scale Evidence supporting
the validity ofthe factors ofthe Self-Mcmitonng Scale is presented in the
next three sections, each summanzing a different level of validation
Conceptual Validation
When examining a factor structure we want to know whether the
items that load on separatefectorsformmeaningful and distinct groups
Our confidence m a sdutum is bolstered by coherent and exjdic»ble fec-
tors For instance, early analyses ofthe Self-Consciousness Scale re-
vealed a meaningful distinction in item content—pnvate vs public self-
ccmsciousness, ar^ social anxiety—which resulted m three separate
subscales (Feningstem, Scheier, & Buss, 1975) Similarly, Collins (1974)
was able to show that Rotter's Locus of Control Scale ccmsisted of four
mter^etablefectcH^cwfourtypes of extemahty belief that the wcMrld is
difficult, unjust, govCTned by "luck," aiKl pohtrcally unrespcmsive
Factor analysis 121
Validation by Self-Report
A number of studies have reported such correlations for the factors of
the Self-Momtonng Scale A sample of these correlations is presented m
Table 3 This sample is biased m two ways First, it involves only those
studies which have used the three factors identified by Bnggs et
al (1980) (This hmitation greatiy simplifies the organization ofthe table
and IS not tembly self-serving smce that solution is currently the one
most commonly used ) Second, other self-report dimensions are re-
ported only if there is a significant correlation vnth the total score on the
Self-Monitonng Scale or vwth at least one ofthe fectors Many self-report
vanables do not ccarelate with the Self-Momtonng Scale or its factors,
iKM- IS there any reason to expect that they should
TTie table is divided mto several parts The top section of the table
contams measures of poor adjustment (e g , anxiety and neuroticism)
TTie next section (fovra involves measures of extraversion and scxaal sur-
gency, whereas the third section consists of self-esteem and self-cx)nfi-
dence measures TTie last part contains measures that are ccmceptually
related to the construct of self-mcmitonng such as machiavellianism and
empathy Also lnducfed at the bottom ofthe table is the Crowne-Mar-
lowe measure of social desirability
TTu-ee things should be apparentfromthe table First, consistent with
Sn)^r's earher assoticms (1974,1979), total scores on tte Self-M<mitor-
mg ScaJe are relatively uncorrelated with other personality measures
122 Bnggs & Cheek
Self-Monitoring Scaies
Sample Total Extra- Other-
Self-report measures size score Acting version directed
Adjustment measures
Shyness 1020 - 10 - 23 - 56 37
Social anxiety 221 ns - 27 - 46 20
Manifest anxiety 62 30 06 - 11 49
Trait anxiety 70 ns ns - 26 26
PRF social recognition 67 16 - 05 - 25 41
Neuroticism 67 12 - 01 - 36 45
[2nd sample] 55 30 ns ns 46
Soaal surgency measures
Extraversion 55 41 25 51 ns
[2nd sample] 67 19 20 43 -19
Sociability 1020 20 12 36 05
PRF affiliation 67 12 03 37 - 14
PRF dominance 67 29 43 42 - 16
PRFexhiisttion 67 34 40 54 - 17
Persuasive ability 72 26 36 46 00
Communication effectiveness 72 36 46 43 14
Self-esteem measures
Self-esteem 1020 - 17 ns 36 - 49
Texas Social Behavior 776 27 38 65 - 32
Inventory
Coopersmfth SEI 62 - 24 - 05 21 - 47
Jams-Field self-esteem 136 - 03 24 35 - 34
Rosenberg self-esteem 136 - 11 11 23 - 31
Measures (x»K»ptually related to self-monrtonng
Machiavellianism 77 24 23 02 26
Affective Communication Test 67 19 31 56 - 29
Empathy 61 20 32 36 - 10
[2nd sample] 221 25 31 31 ns
Ring performance scales
People 100 - 46 - 39 - 61 - 09
Roles 100 35 45 57 - 09
Chameleon 100 03 - 15 02 09
Sodal desirability 62 - 24 - 13 06 - 22
(and sample] 100 - 26 - 01 - 06 - 36
F£K:tor analysis 123
and acting factors, whereas the relationship with empathy involves the
extraversion and acting factors
Finally, total scores on the Self-Monitonng Scale are substantially re-
lated to two ofthe three Ring and Wailston Performance Scales Snyder
(1974) reports a nonsignificant correlation with the chameleon (c) sc^e (r
= — 25), but does not mention the role (r) or person (p) scales A study
by Dabbs, Evans, Hopper, & Purvis (1980) found a substantial correla-
tion between self-monitonng and both the role scale (r = 52, p < 01)
and the person scale (r = — 51, p < 01) Similarfindingsare presented
in Table 1, but here it is clear that the extraversion and acting factors are
responsible for the substantial overlap in these two measures, the other-
directedness factor IS unrelated to all three performance measures
One other set of findings, not reported in Table 1, deserves attention
In recent years, Jerry W i ^ n s (1979, 1980) has advocated a circumplex
model for the representation of interpersonal behavior He has devel-
qped a set of interpersonal adjective scales which form a 2-dimensional
circle that can be partitioned into eight octants One ofthe pnnciple ad-
vantages of this model is that it provides a conceptual framework for de-
scnbing the universe of interpersonal constmcts, although this approach
IS not without its cntics (e g , Jackson & Helmes, 1979) Wiggins &
Broughton (1985) have attempted to locate the fectors of the Self-Mom-
tonng Scale withm the Interpersonal Circle The acting subscale fialls
squarely m the ambitious-dominant octant, and the extraversion sub-
scale lands close by between the ambitious-dominant octont and the gre-
ganous-extraverted octant In contrast, other-directedness is lcxated cm
the other side ofthe circle and off of the circumference It falls closest to
the aloof-introverted octant, directiy opposite from the greganous-ex-
traverted octant (as well as the other twofecrtors)Wiggins and Brough-
ton conclude that "Although it is not clear what is measured by the items
ofthe Other-directedness Scale, it is dear that such items should not be
combined with those from the other two subscales to form a sii^e scale"
(P 31)
It IS apparent, therefcH-e, both in this case and m those reported m
Table 3, that an analysis by subscale reveals considerably more than an
analysis using the total score alone Of course, there are also instances
m which the subsc^es have all correlated similarly with another vana-
ble R)r example, in cme sample, Lennox and WcJfe's (1984) Ability to
Modify Self-Presentation Scale correlated 45 with total sccwes cm the
Seif-Mcmitonng Sc^e, 39 with acting ability, 32 with extraversion, and
28 with other-directedness (all p's < 01, Johnson. Jewell, & Tirrell,
1984)
Factor analysis 125
Behavtoral Validation
Although personally we are persuaded by the evidence reported m
the last section, we realize that for many readers behavioral evidence is
the sine qua non If the factor structure has no behavioral implications,
so what'' The third and most compelling type of validation, therefore,
involves charting the relationships between factors and behavior Several
studies have provided such evidence Because the self-monitonng con-
struct has yielded a vanety of testable hypotheses, these studies can be
grouped into four sections accordmg to the kind of hypothesis being
tested communication ability, self-attention, situational variability, and
moderator variables
Sendtng messages and reading cues Four published articles have
looked specifically at how communication skills are related to the factors
of the Self-Monitonng Scale Two of these reports (Riggio & Fnedman,
1982, 1983) provide complementary analyses of a single data set in which
three types of performance measures were collected ability to send
emotional messages, ability to deceive, and abihty to detect nonverbal
cues Total scores on the Self-Monitonng Scale did not relate to any of
the behavioral or ratmg measures reported in these two studies Results
with the factors also were modest, but they outperformed the lull scale
Ia particular, an individual's ability to convey emotional messages was
related to scores on the acting factor (r = 24, p < 10), whereas scores
on a measure of facial animation (speech rate, head movements, and
smihng) correlated with both the extraversion factor (r = 39, p < 01)
and the acting factor (r = 28, p < 05), although tte facial animation
measure itself did not prove to be a reliable indicator of deception In
additum, the extraversion and acting fectors were related to a discrep-
ancy in the amount of nervous behavior displayed when telling a he vs
telhng the truth High compared to low scorers on both the extraversion
fector (r = 30, p < 05) and the acting fector (r = 25, p < 05) were
more likely to inhibit the leakage of nervous mannensms (e g , leg
movements, posture shifts, and hand-to-head contacts) vAvsa febncabng
a story than when speakmg truthfully
A third article in this area contains two separate studies examining
the relationship between self-monitonng and deception abilities (Sieg-
man & Reynolds, 1983) In the first investigation, participants re-
sponded to a senes of personal or impersonal questions and were ln-
stnKrted to answer truthfully to some and felsely to others Tl» majcH-
ifependent vanaUes were measures of verbal fluency Total scores cm the
Self-Monitonng Scale were unrelated to the fluency measures, but
126 Bnggs & Cheek
sion they are making on others To test this notion, Tobey and Tunnell
(1981) videotaped women talbng about their views on a topic and then
asked them to predict the impression they would make on others The
dependent measure was the extent to which an individual's predictions
agreed with the judgments of raters Total scores on the Self-Momtonng
Scale were unrelated to the index of agreement (r = 08, ns), but scores
on the acting factor were somewhat related (r = 25, p < 05) Subse-
quent analyses suggested that "high actors made accurate predictions
because they knew they had made good impressions" (p 667, italics ong-
inal)
In a recent study, Miell and LeVoi (in press) asked participants to m-
ter-act with either a fhend or a stranger Participants talked together for
about 10 minutes and then filled out a set of ratmgs, the partners antici-
pated a future interaction Other-directedness was positively related to
participants' rejHjrts that they used their partner's behavior as a guide to
what to do dunng the interaction, scores on the other two factors and on
the total scale were not related to these reports High scores on the act-
ing and extraversion factors, however, were negatively related to how
self-conscious an individual felt
Sttuattonal vartabdtty According to the self-monitonng construct,
individuals who score high on the scale should exhibit greater cross-sit-
uational variability than those who score low They should modify their
behavior to fit the demands of a particular situation Evidence for this
claim was reported by Snyder and Monson (1975), but several other stud-
ies have failed to corroborate thisfinding(Greaner & Peaner, 1982, San-
tee & Maslach, 1982, Schneiderman, 1980, Tunnell, 1980, Zanna, Olson,
& Fazio, 1980)
A study by Brookmgs, Flood, Hessmge, Kuhls, Miller, & Wright
(1982) compared scores on the total Self-Monitonng Scale and its fectors
with ratings of vanabihty by self and by others Self-related variability
was significantly related to scores on the other-directedaess fector (r =
36, p < 001) but not to scores on the total scale (r = 16, p > 10)
Other-rated vanabihty was not related to the total scores or the fectors
A more recent study also su^ests a relationship between situational
vanabihty and the otlier-directedness factor johnsoa, Jewell, & Turell
(1984) assessed ihe degree to which individuals acquiesed m an attitude
change paradigm 'E>tal scraps oa the Self-Mraiitonng Scale cwrelated
positively with acquiescence (r = 30, p < 01) as did scores on the
otl^r-directedness fector ( r = 31, p < 01) ScoresOTtte other two fec-
tors were not significantly related to acquiescence (r = 06 fra- extra-
version, and r = 18 for actmg)
A moderator variable Accordmg to Snyder's conceptualization of
setf-mcmitormg, low sdf-mrautOTS regard themselves as "ratlrer
128 Bnggs & Cheek
pled beings who value congruence between their actions m social situa-
tions and relevant underlying attitudes, feelings, and dispositions " In
contrast, high self-monitors see themselves as "ratherflexibleand adap-
tive creatures w^io shrewdly and pragmatically tailor their social behav-
ior to fit situational and interpersonal specifications of appropnateness"
(Snyder & Campbell, 1982, pp 186-187) Thus, self-monitonng should
moderate the relationship between internal dispositions and social acts
(e g , Ajzen, Timko, & White, 1982, Snyder & Tanke, 1976)
Two recent studies of moderator variables have examined the Self-
Monitonng Scale as well as its fectors Cheek (1982) compared self-rat-
ings with peer-ratings on four personahty dimensions The Self-Moni-
tonng Scale felled to show meanmgful moderating efifects on the rela-
tionship between self-ratmgs and peer-ratings The acting factor,
however, did moderate this relationship, but in the direction opposite
from that suggested by Self-Monitonng theory High scorers showed
stronger agreements between self-ratings and peer-ratings than did low
scorers Scores on other-directedness followed a similar pattern al-
though somewhat less consistently, whereas scores on extraversion
tended to work m the opposite direction (l e , m the direction predicted
by the construct of self-monitonng)
Wymer and Penner (m press) not only replicated this pattem of find-
mgs, they further clarified the issue by distinguishing between (and em-
pincally assessing) two types of predictability the congruence between
self-ratings and peer-rahngs and the congruence between mtemal dis-
positicms (attitudes and traits) and actual behavior In their study, the
two types of predictability were unrelated to each other Individuals
sconng high cm the acting fector displayed more self-peer congruence
than did those sconng low on acting, a replication of the Cheek (1982)
study FOT attitude-behavior congruence, however, h i ^ sccM-ers on the
extraversiraifectoras well as the other-directedness fector displa>«d less
attitude-behaviOT ccmgruence than did those w^w scored low Thus, two
fectors were related to attitude-behavior consistency in the direction
jM-edicted by ihe craistmct of self-mtMutcHing, and the third fector was
related to agreement betwerai self aiKl rther ratmgs (albeit m the direc-
tKm c^posite to that predicted by Snydter) Interestingly, scores on the
full Self-Mcmitcmng Scale failed to moderate either kind of {nredictabil-
lty Once again, then, we have seen how thefectorsof the Self-Monitw-
lng Scale clarify the results of a study and prove more useful than the
^ sccnres alcme
Summing Up
The pmnt (^this extended review has been to demcmstrate the utihty
of AtefectcH^retracted bam the Self-Mcmibmng Scale Simjdy {wt, ^
Factor analysis 129
fectors work They resolve the ambiguity of the item pool and its low
lntentem correlations, they are mterpretable, and they are related to
other measures of vanous sorts in sensible ways Not only do factors clar-
ify how the scale works psychometncally, they reveal something crucial
about the construct itself Individuals who attend to social cues and reg-
ulate their behavior accordingly may do so for one of two fundamentally
diflFerent reasons They may do so because they are comfortable m social
situations and can work to maximize social interactions (perhaps m a way
that IS pragmatic, shrewd, or even manipulative), or they may attempt
to minimize the chances of error, a self-defensive move to cover-up and
fit in In his essay "On Facework," Coflniaa (1967) distinguishes the use
offeceworkas an avoidance process (consisting of defensive and protec-
tive maneuvers)fromthe aggressive use of facework for "making points "
More recently, Arkin (1981) has distinguished between self-presenta-
tional strategies that he labels protective and acquisitive
In factor analyses of the Self-Monitoring Scale, this distinction
emerges as the other-directedness factor vs the extraversion and acting
fectors (for example, recall the pattems of correlations m Table 3) We do
not want to suggest, however, that these factors adequately measure this
two-component model of self-presentation Factor analysis provided a
basic insight, but the factors themselves are no elixir Sometimes they
work, sometimes they do not (e g , Caldwell & O'Reilly, 1982, Snyder &
Cangestad, 1982), which is not surpnsing since they aever went through
a process of careful scale development and refinement The proper next
step would be to take the insights that now are apparent and carefully
construct a new set of items that measure these concepts specifically and
systematically (e g , see Lennox & Wolfe, 1984, Wolfe, Lennox, & Cut-
ler, m press) Simply shortening the scalefrom25 items to 18 or 8 items
(e g , Cangestad & Sayder, 1985) does not provide an adequate solution
because it does not articulate the multiple dimensions inherent m the
scale and m the concept
We also want to state emphatically that we believe it is unacceptable
to contmue using a total score alone when to do so deliberately ignores
distinctions that are conceptually meaningful and empirically useful
(and Sdf-M<Mutoni^ is by no means the only case) The Self-Momtwing
Scale IS a popular measure of personality and has served as the center-
piece for a number of puUisl^ articles The scale has proved successfiil
m predicting a vanety of cntena and has acted to stimulate expenmeatai
social psychologists' interest m the measurement of individual diflfer-
eas^s The jH-oUem IStounderstand tcfe/the scale works As we pointed
out m our 1^0 article, it is diflScult to know how to interpret a high scwe
m Ae Self-MraiitOTU^ Scale "One persoi might score high on actii^
and crilier-directedness, antrtlwr on acting and extraversion, and a third
130 Bnggs & Cheek
tent area Our example for this section of the paper is self-esteem, which
IS unquestionably a fundamental personality construct Coopersmith
(1967) defined self-esteem as "the evaluation which the individual makes
and customanly maintains with regard to himself, it expresses an atti-
tude of approval or disapproval, and indicates the extent to which an in-
dividual believes himself to be capable, significant, successful, and wor-
thy" (pp 4-5) Similar conceptualizations have been presented by other
psychologists (e g , Rosenberg, 1965, 1979), yet research on self-esteem
has been hampered by persistent measurement problems
Construct Explication
fectors in this item pool Fleming and Watts' study provides another ex-
ample of a factor analysis that is conceptually but not statistically confir-
matory
In a replication and extension, Fleming and Courtney (1984) exam-
ined the Shavelson et al (1976) multidimensional model of self-esteem
more fully by adding items assessing self-confidence in physical appear-
ance and physical abilities, and by slightly revising the existing items for
self-regard, social confidence, and self-confidence m school abilities
The 36-item questirainaire was administered to 256 college students,
and these data were factor analyzed with an oblique rotation The scree
test for the number offectorsto be rotated was somewiiat ambiguous, so
the authors examined several possibihfaes before deciding on a five-fec-
tor solution with an oblique rotation The pattem of factor loadings
clearly supported the hypothesized model of five relatively distinct di-
mensions of self-esteem The success of this replication and extension of
Fleming and Watts' (1980) study reinforces their earlier findmgs and
provides further evidence mfevorofthe Shavelson et al (1976) model of
self-esteem A related study by Mamrus, O'Connor, and Cheek (1983)
showed that vocational certainty as assessed by the Vocational Identity
Scale (Holland, Daiger, & Power, 1980) should be added as a sixth di-
mension of self-esteem in college students
We have emphasized that the factors for these dimensions of self-es-
teem are oblique rather than orthogonal In fact, the fector pattem cor-
relations among thefiveobbquefectorsreported by Fleming and Court-
ney (1984) ranged from 15 to 35, with an average of 25, five scales
formed from the highest loadmg items on each ofthe factors had an av-
erage mtereorrelation of 4 Similarly, the six self-esteem dimensions
employed by Mamms et al (1983) yielded an average mterscale correla-
tion of 36 Such results bear on both the convergent and discnmmant
validity of the multidimensional conceptualization of self-esteem The
Shavelson et al (1976) model is explicitly hierarehieal, which means that
measures ofthe self-esteem dimensions sknild have moderate positive
correlations with each oth«- Furthermore, the dimensions should con-
v&rsd craiceptually and statistically into a higher-order amstnict of gen-
eral self-esteem (l e , the sum of scores on all the dimensiwis diould be
mterpretable as an index of overall self-esteem) One way to lode at this
etmvergent vahdity issue is to ctmdiKt a second-order factor analysis
Whea some or all of the first ord^ fectors obtained in an mitial fector
aiuilysis with ( ^ q u e rotatiaa are fonad to be COTrelated, the matnx of
<xsrrdatioas axmmg these factors may be subjected to anod^r, hi^ier-
order fector analysis. Tlus procedure will reveal the extent to which the
first-OTdarfectcnsdiare scwne cmnnKm vanance Tl» seccmd-order ap-
proach to self-esteem dimraisions may be analogous to models of mteUi-
136 Briggs& Cheek
gence which specify separable factors of intellect yet also support a sin-
gle higher-order construct of "general intelligence "
For the reasons just discussed, Fleming and Courtney (1984) hypoth-
esized that all five of their specific self-esteem factors would have sub-
stanbal loadings on a single second-order fector representing the general
self-esteem construct Their results, which are presented in the first col-
umn of Table 4, were consistent with this expectation Another way to
examine this issue is to intercorrelate the self-esteem scale scores, rather
than the first order factors, and to factor analyze that correlation matnx
In the second column of Table 4, we present the factor loadings from an
analysis ofthe six self-esteem dimensions used by Mamrus et al (1983)
The interpretation of one higher-order factor of general self-esteem is
supported by these results In addition, we included the Personal Iden-
tity Scale (Cheek & Bnggs, 1982) as a means of looking for discnminant
validity evidence That scale was designed to assess self-definition (l e ,
degree of personal identity onentation) rather than self-evaluation, and
therefore should not load on the general self-esteem factor As may be
seen m the table, this test of discnmmant validity was successful
Evaluating Recent Work on the Dimensionality of Self-Esteem
ley, 1977, Gray-Little & Appelbaum, 1979, Stake & Orlofsky, 1981,
Walsh & Taylor, 1982)
The point we are making here about self-esteem research applies to
factor analytic research m general factor analysis is not an end tn itself
but a prelude to programmatic research on a particular psychologtcal
construct As we discuss m the section below on confirmatory factor
analysis, some additional fector analytic work remams to be done m the
self-esteem area The problem that we have encountered in reviewing
the recent literature is that investigators are only too willing to work on
the logical and correlational steps of analysis, yet seem reluctant to be-
come involved m the empincal research necessary for establishing pre-
dictive and discnmmant validity evidence As a result, we are witness-
ing a dramatic proliferation of scales measunng vanous dimensions of
self-esteem (e g , Harter, 1982, Marsh, Rebch, & Smith, 1983, Peterson,
Schulenber, Abramowitz, Offer, & Jarcho, 1984, Phinney & Gough,
1982, Soule, Drummwid, & Mclntire, 1981) There are a few potentially
important differences among these scales, sucb as whether academic
self-confidence should be subdivided mto reading and mathematics
subscales, and some ofthe researchers are seeking extemal validity evi-
dence—for example, the spouse rabngs of typical behaviors m Phinney
(1984) But It may now be time to draw the hue that Grandall prc^)osed
in 1973fiarthe earlier generation of unidimensional self-esteem scales
138 Bnggs & Cheek
Tbe goodness of fit of tbe sample data to eacb model was tested by a
transformation of tbe Cbi Square Significance Test suggested by Muliak
(1975) Tbese tests indicated tbat Model VI provided tbe best fit, and tbe
autbors concluded tbat "significant additional information about tbe un-
derlying muItitrait-muItimetbod matnx was provided by this metbod of
analysis" (Watkins & Hattie, 1981, p 282)
One hmitabon to tbe multimetbod and confirmatory fector analytic
work of Watkins and Hattie (1981) and VanTumen and Ramanaiab (1979)
IS tbat botb studies employed two forms of self-reports as tbeir multtjde
metbods Tbe pnmary metbods for assessmg personality cbaractenstics
not only include self-reports on questionnairesOTmventones, but also
encompass observers' evaluations, data denved from objective measures
m more or less artificial laboratory settings, and "real Me" outcomes
sucb as grades, jwomotions, or beart attacks (Cattell, 1946, Fiske, 1971)
Therefore, we Impe tbat future self-esteem researcb using ccmfirmatory
factor analysis will involve maximally distim^ metlxids m constructii^
multitrait-multimetbod matnces and wiU mvestigate predictive ctmv&r-
gent and discnmmant vabdity tbrougb tbe ap^icatuxi of causal modds
to l<R^tudinal data A substantid portion of tbis demandu^y ideal re-
searcb design bas been approximated by Faulbus (1983) m a study of lo-
cus of c(Hitrd dimensions, and bis paper is wortb readmg as a metbodo-
lopcal example ofthe nAe of amfirmatcn^ analysis as one part of a larger
research eflfort
Factor analysis 141
Conclusion
In this paper we have evaluated the usefulness of factor analysis as a
method for advancing personality research by telling tbe tale of two con-
structs, self-monitonng and self-esteem In tbe first case, we docu-
mented wby Nunnally (1978) was rigbt to argue that factor analysis
should be applied routinely to new personality tests immediately after
tbey are constructed Because tbis was not done, personality psycbolo-
gists are now confronted with a sizeable body of research on self-moni-
tonng that IS empmcally acceptable but tbat is also—to borrow Guil-
ford's expression—"psychologically ambiguous and difficult to
interpret" Tbe process of scale revision has just begun (Gangestad &
Snyder, in press, Lennox & Wolfe, 1984), wbich means that old studies
will have to be reanalyzed and new work planned We are confident that
by 1990 fector analyses ofthe Self-Monitonng Scale will be recognized
as a constructive contnbution to botb tbe tbeory and measurement of
self-presentation
In the second case, we showed how the application offectoranalysis
to multidimensional ccmceptualizations of self-esteem has advanced that
important yet problematic area of personality research More vrork re-
mams to be dme, especially m tbe testmg of bierarcbical facet models
thrwigh confirmatOTy fector analysis and causal modelmg, but the con-
clusion that factor analysis has contnbuted successfully to our under-
standing of self-esteem IS amply supported in the research we reviewed
We beheve that fector analysis is one important hnk m the chain of
bgical, correlatKmal, and expenmental aiuJyses necessary for the full
exphcation of any personahty construct The argument that thorough
142 Briggs& Cheek
R^srances
Ajzen, I ,Tiinlto,C ,&Wlute,J B (1982) Sdf-niomtcRingandtheattitu<k-behavior
—'-^— Journal af Personaltty and Soctal Psi^h(Aogy,*i,"^ ""^
Factor analysts 143