Re-examination of the major personality-type factors in the Cattell, Comrey and
Eysenck scales: Were the factor solutions by Noller et al. optimal?
By Gregory J. Boyle
Institute of Education, University of Melbourne, Parkville, Victoria 3052, Australia
Abstract:
A recent higher-order factor analysis of the Cattell, Comrey and Eysenck personality
scales by Noller, Law and Comrey (1987) in the J. Person. Soc. Psychol. 53, 775–
782 provided a useful account of the number and nature of normal personality-type
dimensions measured within the questionnaire, self-report domain. The analyses
reported were based on an exemplary sample of Australian adults, matched carefully
across sex, age, and social class, thereby providing a sound basis for investigating
personality structure. Noller et al. extracted and rotated seven separate factors using
procedures suggested by Comrey (A First Course in Factor Analysis. Academic
Press, New York, 1973), thereby attaining moderate approximation of the final
rotated solution to maximum simple structure. In an attempt to improve on the
approximation to simple structure criteria, the present study reanalysed the Noller et
al. data set and presents the results of a 6-factor oblique solution derived from the
intercorrelations of all 33 variables included in the Noller et al. study, as well as a 5factor solution based on the intercorrelations of the 25 personality scales alone. The
present findings agree strongly with the conclusions of Noller et al. that there are
indeed five major personality-type dimensions within the normal trait sphere, but
suggest a slightly different interpretation of these structural dimensions.
Many studies have investigated the higher-order factors discernible in self-report
personality instruments (cf. Cattell and Johnson, 1986; Kline, 1979; Zuckerman,
Kuhlman and Camac, 1988). Most of these studies have been based on analyses of
three personality instruments which have focused on different levels of the
hierarchical structural model of normal personality traits. The Eysenck Personality
Inventory (EPI-Eysenck and Eysenck, 1963) has proven itself over the years as a
reliable indicator of the two most important higher-order personality-type dimensions:
Extraversion and Neuroticism. These structural dimensions have consistently emerged
from higher-order analyses in virtually all psychometric studies of the normal
personality sphere (cf. Boyle, 1986, 1988b). Indeed, so robust have these two
dimensions proven, that the Eysenckian school has labelled these as ‘superfactors’
(e.g. Eysenck, 1981, 1983; Eysenck and Eysenck, 1985). The Cattellian school by
contrast has focused its measurement more at the primary source trait level of the
normal personality structural model although provision for calculation of higher-order
factors at the Eysenckian level of analysis has also been incorporated into Cattell’s
Sixteen Personality Factor Questionnaire (16PF-Cattell, Eber and Tatsuoka, 1970).
Clearly, as Boyle (1986) demonstrated, both Eysenck and Cattell are measuring
common factor variance within the normal personality trait domain, although at
different levels of the hierarchical structure. As Boyle (p. 583) stated, “The
behavioural scientists Eysenck and Cattell have much in common, having investigated
intrapersonal psychological structure, albeit at different levels in its hierarchical
structure. Eysenck has studied the personality domain in terms of a typological model
involving a small number of important super factors. In contrast, Cattell has typically
resorted to analyses of a considerably larger number of primary factors. . . . It is
moreover, readily apparent that Eysenck’s super factors are represented among the
second-order 16PF and SSQ factors. . .“. Eysenck (1984, pp. 335-336) has himself
conceded that, “it is unusual to discover such close correspondence between authors
so distinct in their methods, procedures, evaluations and premises . . . . The Cattell
and Eysenck constructs and theories should be seen, not as mutually contradictory,
but as complementary and mutually supportive”. The 16PF, like the EPI has stood the
test of scrutiny over time (see Mitchell, 1985) and has survived critical analysis
remarkably well. (Some currently popular instruments which are being promulgated
and aggressively marketed in the commercial world as ‘superior’; to the 16PF have
not withstood such critical scrutiny over lengthy periods of time).
At an intermediate measurement level, Comrey’s Personality Scales (CPS-Comrey,
1970, 1980) have provided an alternative structure for describing personality traits.
Noller et al. (1987) have provided a comprehensive description of the CPS scales,
along with a detailed review of the studies into personality structure based on analyses
of data from the EPI, 16PF and CPS respectively. The CPS take as their starting point,
Factor Homogeneous Item Dimensions (FHID’s), each of which is comprised of four
separate items, balanced for direction of scoring, and designed specifically to measure
important factor dimensions. This intermediate approach (in terms of the Cattellian
and Eysenckian structural models) has some merit. Citing Comrey and Schiebel
(1985), Noller et al. (p. 776) indicated that use of FHID’s enables “the more
comprehensive, less numerous, and potentially more theoretically meaningful traits . .
. ” to be more readily discerned. As they correctly pointed out, the 16PF provides
measures of more primary trait dimensions than is typically useful for most
practitioners (with necessarily lowered factor loadings on items as compared with
those for FHID’s). Noller et al. were correct to suggest that the CPS provide a more
parsimonious account of structural dimensionality for practical utility. While Cattell
attempted to be more comprehensive by locating primary trait factors at the item
level, in essence, the decision as to the starting point for primary factors is somewhat
arbitrary. Nevertheless, Mershon and Gorsuch (1988) have shown that the primary
16PF factors do account for approximately twice the variance of the eight personality
scales measured in the CPS instrument.
In their recent article in the J. Person. Sot. Psychol. Noller et al. (1987) demonstrated
support for at least four of the ‘Big Five’ higher-order personality dimensions
reported by Digman and Inouye (1986) for data derived from personality ratings, and
by McCrae and Costa (1987) for self-report data (cf. Zuckerman et al., 1988).
However, as Noller et al. pointed out, McCrae and Costa used a self-report instrument
designed to measure these ‘Big Five’ dimensions, in the first place. Noller et al. (p.
776) also hypothesised that “Data from self-report inventories are likely to be much
more complex, because one is answering questions about oneself and not merely
evaluating others, or even oneself, in a simplistic way… Our study was designed to
test the relation empirically”. Noller et al. further stated in regard to the Cattell,
Comrey, and Eysenck personality scales that “these three inventories should provide a
good test of the j-factor hypothesis because one inventory involves many factors (16),
one involves a few (2), and the third involves an intermediate number (8). Numerous
studies… have suggested that the 16PF overestimates the number of primary factor
constructs needed to assess personality, whereas the EPI underestimates the number
needed (Digman, 1979; Digman and Inouye, 1986)“. However, the EPI does not
actually index primary personality traits, but instead measures broader typological
dimensions.
The primary source trait dimensions measured in the 16PF are not directly
comparable with the personality-type dimensions measured in the EPI. The more
appropriate comparison is between the 16PF second-stratum typological factors and
those quantified in the EPI (Krug and Johns, 1986, have actually demonstrated the
factorial validity of five 16PF secondaries using a large sample of 17,381 Ss). Much
research has shown that the two Eysenckian superfactors correspond closely with the
first two higher-order 16PF factors. Even though it is quite legitimate to include the
16PF primaries and the EPI personality-type variables in the same factor analysis, as
Noller et al. did (and as the present reanalysis of their data does), the possibility for
confounding the primary and secondary factors nevertheless remains potentially
problematic. Despite this difficulty, Noller et al. were able to demonstrate a
remarkable similarity between the ‘Big Five’ constructs across both the rating and
self-report domains. Their first 5 factors were similar to Karson and O’Dell’s (1976)
16PF secondaries. Their rationale was sound-“to test for the presence of these factors
by using self-report questionnaires designed without a conceptual framework
involving their existence” (Noller et al., p. 776). In addition, their review of the
psychometric literature pertaining to the delineation of personality structure was
informative, and provided a sound basis for their study.
Noller et al. also cited a number of studies which had failed to adequately replicate the
primary factor structure of the 16PF, thereby suggesting that the 16PF measures too
many primary dimensions. Certainly, on the basis of the common factor model alone,
the first 8-10 primaries must surely account for by far the greater proportion of the
variance (cf. Karson and C)‘Dell, 1976). By definition, the later factors must account
for significantly reduced and perhaps even trivial proportions of the common factor
variance measured in the 16PF. Even so, the various studies referred to above have
been subjected to critical scrutiny by Kline (1979) who has concluded, in the main,
that the reported factor analyses of the 16PF were deficient on various methodological
grounds (cf. Cattell and Krug, 1986, as well as Mershon and Gorsuch, 1988, for
detailed reviews of the evidence supportive of the 16PF primary factor structure).
Likewise, Gillis (1988, p. 157) has questioned the factor analytic methodology of
investigators failing to replicate Cattellian psychometric structure. This is not,
however, a criticism of the Noller et al. study, as their work was based on a clearcut
rationale and sound methodological procedures, even though they employed factor
analytic techniques differing from those advocated by Cattell (1973, pp. 282-287;
1978).
In reviewing the relationships of the 16PF primary factors with those of Guilford,
Comrey, Eysenck and Grygier, Kline (1979, Chap. 5) indicated the general
inadequacy of Guilford’s personality factors, which he argued were invalid (pp. 140141). He also reported that the Comrey factors were essentially the Cattellian 16PF
second-order factors (p. 143). On a different note, and in accord with the findings of
Mershon and Gorsuch (1988), Krug (1978) reported that the 2-factor EPI accounts for
approximately one-third of the common factor variance measured in the 16PF, as
would be expected. Therefore, while the EPI is a solid measure of Extraversion and
Neuroticism, the 16PF provides a wider sampling of the normal personality trait
domain, measuring in addition, factors labelled Tough Poise, Independence, and
Control at the Eysenckian level of analysis (cf. Krug and Johns, 1986).
Nolier et al. referred to a study by Kline and Barrett (1983) on the 16PF which
purportedly verified only four of the primaries. Perusal of the Kline and Barrett
(1983) article, as well as the Barrett and Kline (1982) study indicates that the four
factors were obtained from a higher-order factoring of the scale intercorrelations (cf.
Barrett and Kline, p. 259). As the derived factors were therefore higher-order
dimensions, the finding of only four factors could not cast doubt on the primary 16PF
structure, at least. Indeed, as Noller et al. acknowledged, some investigators (e.g.
Adcock and Adcock, 1977; Burdsal and Bolten, 1979) have essentially replicated the
primary factor structure of the 16PF (cf. Boyle, 1988a, b; Cattell and Krug, 1986;
Mershon and Gorsuch, 1988). Noller et al. factor analysed the scale intercorrelations
for the EPI, 16PF and CPS instruments using a large sample of 669 Australian adults
as Ss. This sample was carefully matched across sex, socioeconomic status, and
across no fewer than five different age levels ranging from 16 through to 65 yr. Their
sample therefore, was structly controlled in regard to composition, unlike many of the
published studies involving college or university students. Since their sample was so
carefully matched across several important independent variables, a particularly good
test of the relationships between the Cattell, Comrey, and Eysenck psychometric
instruments was possible. However, in basing their factor analysis on scale
intercorrelations, Noller et al. automatically precluded any possibility of deriving
primary factors as measured in the 16PF. Only higher-order personality-type factors
which loaded on various combinations of primaries could therefore be obtained.
Hence, their conclusion (p. 777) that “substantial overlap exists in the Cattell
scales…” must be viewed in the light that no attempt was made to derive primary
factors in the first place.
The particular factor analytic methodology employed by Noller et al. involved the
minimum residual method, followed by principal factoring and subsequent orthogonal
rotation by Tandem Criterion procedures. According to Comrey (1973), the minimum
residual method implicitly extracts an upper-bound number of factors, which is
narrowed down in the Criterion I method, thereby providing an approximate
indication of the appropriate number of factors. Generally, orthogonal rotation
provides a special rotational solution, in contrast to the wide array of possible
solutions obtainable through use of oblique rotational strategies (Boyle, 1985b; Loo,
1979). Examination of the ±0.10 hyperplane count for the Criterion II solution
indicates that 36.8% of the scale variables were in the hyperplane, and that the
solution may have provided a less than optimal approximation to simple structure
criteria (cf. Cattell, 1978; Child, 1970; Kline, 1987). It should however be noted that
with higher-order factoring, the obtainable hyperplane counts are typically
considerably lower than are those for first-order factor pattern solutions. One
possibility is that Noller et al. extracted too many factors (their seventh factor
accounted for a very small percentage of the common factor variance-cf. Walkey,
1983), which combined with the use of an orthogonal rotational strategy, may have
reduced the hyperplane count (usually though, the hyperplane count increases with
extraction of a greater number of factors). In order to clarify these issues, the present
study reanalysed the Noller et al. (1987) data using a different, but generally accepted
factor analytic methodology (see below), in the hope of providing an even clearer
picture of the major personality-type dimensions within the normal personality
sphere.
Method
Factor analytic methodology
In the first approach, the 33 x 33 matrix of scale intercorrelations taken as the starting
point by Noller et al., also served as the point of departure for the present factor
analysis. The 33 variables included the 26 personality scales for the EPI, 16PF and
CPS instruments together, as well as the validity and response distortion scales (EPI:
L-Lie Scale; 16PF: FG-Faking Good; FB-Faking Bad; CPS: V-Random Responding;
R-Social Desirability), and also scores for sex and age. In the second approach, the 25
x 25 matrix of scale intercorrelations, excluding the response distortion scales, as well
as those pertaining to sex, age and 16PF Factor B (Intelligence), served as the starting
point for a separate factor analysis of the personality dimension scales alone. By
excluding the non-personality scales from this second factor analytic approach, it was
hoped that a somewhat clearer picture of the major higher-order typological factors
might emerge. In both instances, an iterative principal factoring procedure was
employed along the lines advocated both by Cattell (1973, 1978) and by Lee and
Comrey (1979). While it was recognised that the determination of highly accurate
communalities is not as important as it is with very small factor matrices (Nunnally,
1978), for the sake of maximum accuracy though, it was decided to utilise an iterative
procedure, together with estimation of the pertinent number of factors in each instance
by the psychometric Scree test (Cattell and Vogelmann, 1977; Hakstian, Rogers and
Cattell, 1982) rather than by the eigen values greater than unity rule (Kaiser-Guttman
method--(cf. Yeomans and Golder, 1981). The Scree test has been shown in studies of
plasmodes to be appreciably more accurate than the K-G criterion when the number
of variables is either very low (< ~ 20), or very high (> ~ 50) (cf. Child, 1970, pp. 4344). Finally, the extracted factors in each approach were rotated to oblique (direct
Oblimin) simple structure via the procedures provided in the standard Statistical
Package for the Social Sciences (SPSS)—(Nie, Hull, Jenkins, Steinbrenner and Bent,
1975). The principal factoring procedure has been recommended by Cureton and
D’Agostino (1983), among others. Approximation to simple structure criteria of the
resultant factor pattern solutions in each instance was assessed in terms of the
corresponding ±0.01 hyperplane counts, as recommended by Cattell (1978) and
Gorsuch (1983).
Results
(I) Factor analysis on all 33 variables
Application of the Scree test suggested that only 6 factors should be extracted,
whereas Noller et al. took out 7 factors. Convergence of communality estimates
(starting from the squared multiple correlations which are lower-bound estimates of
communality) required 15 iterations. Examination of the hyperplane counts associated
with both the 6- and 7-factor oblique solutions indicated that 48.0 and 49.8%
(compared with 36.8% in the Noller et al. study for their 7-factor Tandem Criterion
solution) were in the ±0.10 hyperplane band. While hyperplane counts usually
increase with extraction of additional factors (Boyle, 1985a), the 1.8% increase in
going from 6 to 7 factors did not represent a significant improvement in the
approximation of the factor pattern solutions to simple structure criteria, however.
There was a difference of 0.08 between eigen values for the seventh and eighth
factors, but a difference of 0.27 for the sixth and seventh factors, thereby suggesting
that the seventh factor extracted by Noller et al. was a trivial, non-significant one.
Moreover, the appropriateness of their orthogonal rotation methodology was unclear
from comparison of the hyperplane counts, which in the present instance was 13%
higher for the comparable 7-factor solution. Even the 6-factor solution (presented in
Table 1) exhibited a hyperplane count some 11%higher than their ‘I-factor solution,
indicating a closer approximation to simple structure criteria in the present instance.
As is evident, Factor 1 (which accounted for 17.1% of the variance associated with
the unrotated principal components – 32.9% of the common factor variance after
rotation) represents the Extraversion dimension with significant loadings on
Eysenck’s Extraversion scale, the several scales comprising the second-order 16PF
Exvia dimension (the equation given by Krug and Johns, 1986, for calculating Exvia
involves A + , F + , H + and Q2 - ), and the Activity and Extraversion scales of the
CPS. The high loading on 16PF Factor E of 0.59 suggests an association of
Dominance with Extraversion in the present sample, which may be characteristic of
Australians more than of Americans, in general. It is possible that Americans have a
more extraverted style of social interaction, than do Australians, who may be more
restrained in interpersonal situations, so that those who are markedly extraverted may
tend to be simultaneously dominant also.
Factor 2 (accounting for 29.0% of the rotated common factor variance) represents the
Neuroticism dimension with significant loadings on Eysenck’s Neuroticism scale, the
16PF primaries which contribute to the Cattellian second-order Anxiety factor (Krug
and Johns, 1986, indicated that this dimension involves C -, H - , L+,O+,Q3andQ4+),as well as on the Trust and Stability scales on the CPS. This factor clearly
contrasts neuroticism with stability. While the signs of the factor loadings are
reversed as compared with those given in the prediction equation by Krug and Johns
(the reversal of the signs is an artifact of the factor analysis itself), it is evident that
16PF Factor H (Boldness vs Timidity) did not exhibit a significant loading, as was
also the case for Factor Q3 (Self-Sentiment). Hence, the pattern of 16PF primary
factors contributing to this higher-order dimension differs somewhat from that given
both in the 16PF Handbook (Cattell et al., 1970), and that indicated by Krug and
Johns, which may be largely a function of the particular sample used. Nevertheless,
interpretation of this factor as Neuroticism is quite clear.
The third factor (involving 14.4% of the rotated common factor variance) represents
the Cattellian second-order dimension labelled Tough Poise, which contrasts ToughMindedness with Tender-Mindedness and emotional sensitivity. In the prediction
equation given by Krug and Johns (1986), this dimension exhibits some different
16PF primary factor loadings according to sex (for males: A-, F+, I -, M- and Ql-; for
females: A-, E+, F+, I-, L+ and M-), which given their large sample size of 17,381 Ss
(9,222 males; 8,159 females), must be regarded as robust differences in the expression
of Tough Poise among males and females respectively. Indeed, this higher-order
factor exhibited a highly significant loading on the variable Sex (-0.69) thereby
indicating clearcut sex differences in the sample of Australian adults who served as Ss
in the study by Noller et al. (1987). Moreover, loadings for the CPS scales labelled
Trust and Empathy, along with the 16PF Factor A (Warmth), and Factor I (TenderMindedness) contrasted with the CPS scales of Acitivity and Masculinity. Although
the specific factor loadings on the 16PF primaries differed somewhat from those
provided by Krug and Johns, there is little doubt that this third higher-order factor
represents Tough Poise.
The fourth factor (contributing 11.7% of the rotated common factor variance)
represents the Cattellian second-order dimension labelled Control by Krug and Johns
(1986), which Cattell et al. (1970) referred to as Superego Strength. This higher-order
factor exhibited significant loadings on the CPS scales of Orderliness, Conformity and
Activity, as well as on the 16PF scales of Superego/Conscientiousness, and SelfSentiment (defined as ‘controlled, exacting will power, socially precise’), and also on
the EPI Lie scale. There were also significant loadings on Age and Social Desirability
which suggests that self-control and the ability to respond in socially accepted ways
may improve with increasing age.
The fifth factor to emerge (accounting for 7.3% of the rotated common factor
variance) represents the Cattellian second-order dimension labelled Independence.
The factor loadings for this higher-order personality-type dimension contrasted the
CPS Conformity scale with the 16PF scales labelled Dominance, Imagination,
Radicalism (Experimenting), and Self-Sufficiency, respectively. Once again, Krug
and Johns (1986) provided separate prediction equations for males and females (the
16PF factors for males being: E+, G-, H+, L+, N-, O-, Ql+ and Q2+; for females: E+,
G- , H+, M + , Q1 + and Q2 +), although there was no evidence of comparable sex
differences in the present instance. In view of the failure of significant loadings to
occur for the 16PF primaries G and H (for both males and females), as well as L and 0
(for males) it would appear that this fifth higher-order factor was the least welldefined of the personality-type dimensions, consistent with the fact that it was the last
to emerge from the analysis, accounting for the least amount of the common factor
variance. It is possible though, that the lack of significant loadings on Factors L and 0
may have been partly due to the use of the combined-sex sample in the present
instance. On this evidence, it would have been quite interesting had Noller et al.
(1987) conducted separate higher-order factor analyses on the data obtained from the
326 males, and that for the 343 females in their sample of 669 adults from the general
population.
The sixth factor (which contributed only 4.7% of the common factor variance) clearly
represents a response distortion dimension, having significant loadings on the
Random Responding scale of the CPS, the 16PF Faking Bad scale, as well as the EPI
Lie scale. These loadings contrasted with the 16PF Factor B (intelligence), suggesting
an inverse relationship between response distortion on self-report personality
inventories, and intelligence level. Thus, less intelligent individuals might exhibit
higher mean scores on the various response distortion scales of the three instruments,
and vice versa. In any event, this sixth higher-order factor was a fairly trivial one, in
so far as it accounted for such a small proportion of the common factor variance.
Nevertheless, while not representing a personality-type dimension as such, it does
display a certain functional unity enabling clear interpretation.
Interestingly, the intercorrelations for the six higher-order factors were all fairly low
(see Table 2), thereby providing the necessary a posteriori justification for the
application of orthogonal rotation strategies in the Noller et al. study. It is important to
recognise though, that to assume orthogonality a priori, without first checking on the
degree of obliquity of the factor correlations using an oblique rotational strategy, and
varying the degree of obliquity systematically from very low to very high using say
the delta shift parameter in the SPSS package, is potentially problematic. However, in
working with higher-order factor solutions, the production of an essentially
orthogonal solution is largely to be expected as an artifact of the methodology alone,
because of the artificial compression of variance into higher-order factors. This
statistical effect must be recognised as a determinant of the apparent independence of
personality-type dimensions. It has long been accepted for example, that Eysenck’s
Extraversion and Neuroticism factors are essentially unrelated dimensions. However,
this apparent finding may have emerged largely as a statistical artifact of the higherorder factoring itself. Nevertheless, it is evident that the present factor solutions
provided a somewhat closer approximation to simple structure criteria (there was an
increase of 13.0% for the present 7-factor solution, and for the &factor solution
shown in Table 1, the increase in hyperplane count was still 11.2%, despite the
extraction of one less factor).
(2) Factor analysis on the 25 personality variables alone
In the second approach, application of the Cattellian Scree test suggested that 5
factors should be extracted and rotated. Communality estimates (again starting from
the squared multiple correlations-SMC’s) required 20 iterations to reach convergence
at the third decimal place. Given that the hyperplane count decreases significantly
with smaller numbers of extracted factors, and with smaller factor matrices (in the
present instance, a reduced number of variables was included in the analysis), it is
interesting to note that the +_±0.01 hyperplane count associated with the 5-factor
solution was still slightly higher (40.0%) than that obtained for the Noller et al. 7factor solution (36.8%), which had been derived from the intercorrelations of all 33
variables.
Interpretation of the five obtained higher-order factors is essentially identical to that
provided above for the first 5 factors included in Table 1. However, in this second
approach-see the factor pattern solution in Table 3-the order of extraction of these
factors has altered appreciably, with the Control/Superego personality dimension now
emerging in the second position, after Extraversion, but before Neuroticism. In terms
of the amount of rotated common factor variance, Control accounted for 28.3%, while
Extraversion accounted for 37.0% of the variance. In this instance, Neuroticism was
associated with only 15.0% of the variance, while Tough Poise accounted for 11.9%,
and Independence accounted for 7.8% respectively. Accordingly, the major finding
from this reanalysis of the 25 personality variables alone was that the
Control/Superego dimension emerged as the second most sizeable of the personalitytype factors. In effect, the factors representing Control and Neuroticism switched
positions, with the Neuroticism factor appearing to be of considerably less importance
in the overall personality than previously suggested. This finding would imply that the
prevalence of a high Neuroticism level may have somewhat different implications,
depending on the level of the individual’s Control factor. As for the degree of
intercorrelation among the five higher-order personality-type dimensions, the
obtained correlations were essentially trivial, thereby providing further support for the
use of orthogonal rotational strategies with higher-order factoring of personality traits
(see Table 4), although the statistical effects due to compression of variance into
higher-order factors mentioned above, should be recognised.
Summary And Conclusions
The present reanalyses of the scale intercorrelations between the Cattell, Comrey, and
Eysenck personality instruments for all 33 variables included in the Noller et al.
study, as well as for the 25 personality scales alone (i.e. with validity and response
distortion scales for the 16PF, CPS, and EPI not included), suggest that there are
indeed five robust higher-order dimensions in the normal personality sphere, as Noller
et al. reported. Only the specific interpretation of some of these personality-type
factors differs across studies, to some extent. It is interesting to note that the ‘Big
Five’ factors still held up, even when the intercorrelations of only the 25 actual
personality scales served as the basis for the factor analysis.
As is evident, the five secondaries (personality-type dimensions labelled Extraversion,
Neuroticism, Tough Poise, Independence and Control), which were reported by Krug
and Johns (1986) for the 16PF using a very large sample of well over 17,000 Ss, were
replicated to some extent in the work of Noller e al. and more particularly from the
present reanalyses of their study. Given that these factors are at the Eysenckian
typological level of measurement, it would seem that these five higher-order
dimensions demonstrate the essential compatibility of the Eysenck, Cattell and
Comrey psychometric models. Arguments against the importance of hierarchical
structural models of personality, and against the use of factor analysis in discovering
and confirming personality structure, cannot be justified on the superficial assertion
that Eysenck, Comrey and Cattell have proposed different numbers of trait
dimensions. This spurious argument fails to acknowledge that each investigator has
focused his attention on different levels within the hierarchical structural model of
personality traits. While Cattell focused more on the underlying, source trait primaries
derived from item intercorrelations, Eysenck concentrated predominantly on the
broader personality type dimensions which loaded on the Cattellian primaries.
Comrey on the other hand, chose to locate his primary factors at an intermediate level
between those of Cattell and Eysenck, by taking FHID’s (item parcels in Cattellian
terminology) as his starting point. Clearly, criticism of these structural models on the
grounds of apparent discrepancies between the above investigators, must be regarded
as at best, a rationalisation, or even worse, as a lack of sophistication on the part of
such critics. Even considering the two extreme positions, both the Cattellian and
Eysenckian schools have much more in common than is sometimes evident at first
glance. As for the Comrey scales, some seem to relate closely to the higher-order
dimensions above. For example, the CPS factor labelled Social ConformityRebelliousness (C) would seem to relate in the present instance to the higher order
dimensions labelled Control/Superego, Independence, and Tough Poise. More closely
related would seem to be CPS factors labelled Emotional-Stability (S) and
Extraversion-Introversion (E) with the higher-order dimensions of Neuroticism and
Introversion respectively. This suggests that the CPS scales comprise a combination
of Cattellian primary and second-order factors.
As the factor intercorrelations between the ‘Big Five’ were generally quite low in the
present reanalyses, irrespective of whether or not this was largely a statistical artifact
of the higher-order factoring procedure itself, it is apparent that the use of othogonal
rotational strategies by Noller et al. was appropriate. In the present reanalyses, only
when the SPSS delta shift parameter was increased to 0.5 or higher was the obliquity
of the resulting factors increased significantly. However, the concomitant factor
pattern solution (in the present study) exhibited a slightly less adequate approximation
to simple structure criteria, as evidenced by a reduced ±0.10 hyperplane count
(44.95% as compared with 47.98% respectively), for the 6-factor solution derived
from the intercorrelation of all 33 variables. Accordingly, the above factor-pattern
solutions presented in Tables 1 and 3 were derived in each instance with the delta
shift parameter set at zero, thereby allowing for a moderate degree of obliquity to
emerge, although in fact, the obtained factors were only trivially correlated.
Overall, the present findings provide further support for the five personality-type
dimensions demonstrated by Krug and Johns (1986), which were labelled
Extraversion, Neuroticism, Tough Poise, Independence, and Control (this last factor
may be more important than the other ones, with the exception of Extraversion, for
any given individual). Moreover, perusal of the higher-order factors obtained in the
Noller et al. study, indicates that four of the Krug and Johns factors were also clearly
found in their study. In the Noller er al. study, Factor 1 represented Control, Factor 2
represented Neuroticism, Factor 3 represented Tough Poise, Factor 4 represented
Extraversion, while Factor 5, which was partially characterised by the CPS Social
Conformity-Rebelliousness (C) scale, which Digman and Inouye (1986) suggested
might be labelled Friendly Compliance- Hostile Noncompliance, seemed to at least
partially represent Independence (cf. Krug and Johns).
The present findings confirm the personality-type factors delineated by Krug and
Johns, and support strongly the findings of Noller et al., who themselves stated that
“the largest factors in our study substantially replicate the 16PF second-order”.
(Noller et al., 1987, p. 780). It would be expected that greatest precision of prediction
is to be attained by close scrutiny and interpretation of these ‘Big Five’ in terms of the
particular combinations of primaries contributing to them. This ‘depth Psychometry’
approach (cf. Cattell and Johnson, 1986) adds an important qualitative dimension to
an otherwise quantitative outcome of psychometric measurement within the normal
personality domain.
Acknowledgement – The kind co-operation of Drs Noller and Comrey in willingly providing
the correlation matrix for the present reanalysis is gratefully acknowledged.
References:
Adcock and Adcock, 1977. N.V. Adcock and C.J. Adcock , The validity of the 16PF
personality structure: a large New Zealand sample item analysis. J. Behav. Sci. 2
(1977), pp. 227–237.
Barrett and Kline, 1982. P. Barrett and P. Kline , An item and radial parcel factor
analysis of the 16PF Questionnaire. Person. individ. Diff. 3 (1982), pp. 259–270.
Boyle, 1985a. G.J. Boyle , A reanalysis of the higher-order factor structure of the
Motivation Analysis Test and the Eight State Questionnaire. Person. individ. Diff.
6 (1985a), pp. 367–374.
Boyle, 1985b. G.J. Boyle , Self-report measures of depression: some psychometric
considerations. Br. J. clin. Psychol. 24 (1985b), pp. 45–59.
Boyle, 1986. G.J. Boyle , Intermodality superfactors in the Sixteen Personality Factor
Questionnaire, Eight State Battery and Objective Motivation Analysis Test.
Person. individ. Diff. 7 (1986), pp. 583–586.
Boyle, 1988a. G.J. Boyle , A Guide to the Military use of the 16PF in Personnel
Selection. , Australian Army Psychology Research Unit, Canberra (1988a).
Boyle, 1988b. G.J. Boyle , Elucidation of motivation structure by dynamic calculus.
In: J.R. Nesselroade and R.B. Cattell, Editors, Handbook of Multivariate
Experimental Psychology (2nd edn. ed.),, Plenum Press, New York (1988b), pp.
737–787 Revised .
Burdsal and Bolton, 1979. C.A. Burdsal and B. Bolton , An item factoring of the
16PF-E: further evidence concerning Cattell's normal personality sphere. J. gen.
Psychol. 100 (1979), pp. 103–109.
Cattell, 1973. R.B. Cattell , Personality and Mood by Questionnaire. , Jossey-Bass,
San Francisco (1973).
Cattell, 1978. R.B. Cattell , The Scientific Use of Factor Analysis in Behavioral and
Life Sciences. , Plenum Press, New York (1978).
Cattell and Johnson, 1986. R.B. Cattell and R.C. Johnson , Functional Psychological
Testing: Principles and Instruments. , Brunner/Mazel, New York (1986).
Cattell and Krug, 1986. R.B. Cattell and S.E. Krug , The number of factors in the
16PF: a review of the evidence with special emphasis on methodological
problems. Educ. Psychol. Meas. 46 (1986), pp. 509–522.
Cattell and Vogelmann, 1977. R.B. Cattell and S. Vogelmann , A comprehensive trial
of the scree and KG criteria for determining the number of factors. Multivar.
Behav. Res. 12 (1977), pp. 289–325.
Cattell et al., 1970. R.B. Cattell, H.W. Eber and M.M. Tatsuoka , Handbook for the
Sixteen Personality Factor Questionnaire (16PF). , Institute for Personality and
Ability Testing, Champaign, Ill. (1970).
Child, 1970. D. Child , The Essentials of Factor Analysis. , Holt, London (1970).
Comrey, 1970. A.L. Comrey , Manual for the Comrey Personality Scales (CPS). ,
Educational and Industrial Testing Service, San Diego (1970).
Comrey, 1973. A.L. Comrey , A First Course in Factor Analysis. , Academic Press,
New York (1973).
Comrey, 1980. A.L. Comrey , Handbook of Interpretations for the Comrey
Personality Scales. , Educational and Industrial Testing Service, San Diego
(1980).
Comrey and Schiebel, 1985. A.L. Comrey and D. Schiebel , Personality test correlates
of psychiatric case history data. J. consult. clin. Psychol. 53 (1985), pp. 470–479.
Cureton and D'Agostino, 1983. E.E. Cureton and R.B. D'Agostino , Factor Analysis:
An Applied Approach. , Erlbaum, Hillsdale, N.J. (1983).
Digman, 1979. J.M. Digman The five major domains of personality variables:
Analysis of personality questionnaire data in the light of the five robust factors
emerging from studies of rated characteristics (1979) Paper presented at the
annual meeting of the Society of Multivariate Experimental Psychology, Los
Angeles. .
Digman and Inouye, 1986. J.M. Digman and J. Inouye , Further specification of the
five robust factors of personality. J. Person. Soc. Psychol. 50 (1986), pp. 116–123.
Eysenck, 1981. H.J. Eysenck , A Model for Personality. , Springer, New York (1981).
Eysenck, 1983. H.J. Eysenck , Personality as a fundamental concept in scientific
psychology. Aust. J. Psychol. 35 (1983), pp. 289–304.
Eysenck, 1984. H.J. Eysenck , Cattell and the theory of personality. Multivar. Behav.
Res. 19 (1984), pp. 323–336.
Eysenck and Eysenck, 1963. H.J. Eysenck and S.B.G. Eysenck , The Eysenck
Personality Inventory (EPI). , Educational and Industrial Testing Service, San
Diego (1963).
Eysenck and Eysenck, 1985. H.J. Eysenck and M.W. Eysenck , Personality and
Individual Differences: A Natural Science Approach. , Plenum Press, New York
(1985).
Gillis, 1988. J. Gillis , Quixote or Columbus?. In: K.M. Miller, Editor, The Analysis of
Personality in Research and Assessment: In Tribute to Raymond B. Cattell,
Independent Assessment and Research Centre, London (1988).
Gorsuch, 1983. R.L. Gorsuch , Factor Analysis. (2nd edn ed.),, Erlbaum, Hillsdale,
N.J. (1983) Revised .
Hakstian et al., 1982. A.R. Hakstian, W.T. Rogers and R.B. Cattell , The behavior of
number-of-factors rules with simulated data. Multivar. Behav. Res. 17 (1982), pp.
193–219.
Karson and O'Dell, 1976. S. Karson and J.W. O'Dell , A Guide to the Clinical use of
the 16PF. , Institute for Personality and Ability Testing, Champaign, Ill. (1976).
Kline, 1979. P. Kline , Psychometrics and Psychology. , Academic Press, London
(1979).
Kline, 1980. P. Kline , The psychometric model. In: A.J. Chapman and D.M. Jones,
Editors, Models of Man, British Psychological Society, Leicester (1980).
Kline, 1987. P. Kline , Factor analysis and personality theory. Eur. J. Person. 1
(1987), pp. 21–36.
Kline and Barrett, 1983. P. Kline and P. Barrett , The factors in personality
questionnaires among normal subjects. Adv. Behav. Res. Ther. 5 (1983), pp. 141–
202.
Krug, 1978. S.E. Krug , Reliability and scope in personality assessment: a comparison
of the Cattell and Eysenck inventories. Multivar. Exp. clin. Res. 3 (1978), pp.
195–204.
Krug and Johns, 1986. S.E. Krug and E.F. Johns , A large scale cross-validation of
second-order personality structure defined by the 16PF. Psychol. Rep. 59 (1986),
pp. 683–693.
Lee and Comrey, 1979. H.B. Lee and A.L. Comrey , Distortions in a commonly used
factor analytic procedure. Multivar. Behav. Res. 14 (1979), pp. 301–321.
Loo, 1979. R. Loo , The orthogonal rotation of factors in clinical research: a critical
note. J. clin. Psychol. 35 (1979), pp. 762–765.
McCrae and Costa, 1987. R.R. McCrae and P.T. Costa , Validation of the five-factor
model of personality across instruments and observers. J. Person. soc. Psychol. 52
(1987), pp. 81–90.
Mershon and Gorsuch, 1989. B. Mershon and R.L. Gorsuch , Number of factors in the
personality sphere. J. Person. soc. Psychol. (1989) In press. .
Mitchell, 1985. J.V. Mitchell, Editor, The Ninth Mental Measurements Yearbook,
Buros Institute of Mental Measurements, University of Nebraska, Lincoln (1985).
Nie et al., 1975. N.H. Nie, C.H. Hull, J.G. Jenkins, K. Steinbrenner and D.H. Bent ,
Statistical Package for the Social Sciences. , McGraw-Hill, New York (1975).
Noller et al., 1987. P. Noller, H. Law and A.L. Comrey , Cattell, Comrey and Eysenck
personality factors compared: more evidence for the five robust factors?. J.
Person. soc. Psychol. 53 (1987), pp. 775–782.
Nunnally, 1978. J.C. Nunnally , Psychometric Theory. (2nd edn ed.),, McGraw-Hill,
New York (1978).
Walkey, 1983. F.H. Walkey , Simple versus complex factor analyses of responses to
multiple scale questionnaires. Multivar. Behav. Res. 18 (1983), pp. 401–421.
Yeomans and Golder, 1981. K.A. Yeomans and P.A. Golder , The Guttman-Kaiser
criterion as a predictor of the number of common factors. Statistician 31 (1981),
pp. 221–229.
Zuckerman et al., 1988. M. Zuckerman, D.M. Kuhlman and C. Camac , What lies
beyond E and N? Factor analyses of scales believed to measure basic dimensions
of personality. J. Person. soc. Psychol. 54 (1988), pp. 96–107.