Personality and Individual Di€erences 28 (2000) 879±886

www.elsevier.com/locate/paid

The Big Three or the Big Five? A replication study

Aristide Saggino*
Seconda Universita' di Napoli, Napoli, Italy
Received 2 July 1998; received in revised form 7 April 1999; accepted 12 June 1999

Abstract

This study addresses the question of the dimensionality of personality, in particular comparing the
three- and ®ve-factor models and trying to replicate the ®ndings of Draycott and Kline (Draycott, S. G.,
& Kline, P. (1995). The Big Three or the Big Five Ð the EPQ-R vs the NEO-PI: a research note,
replication and elaboration. Personality and Individual Di€erences, 18, 801±804). The Italian edition of
the Eysenck Personality Questionnaire-Revised (EPQ-R) and the Big Five Questionnaire (BFQ) were
compared by means of principal component factor analysis and canonical correlation analysis. Results
con®rm that the EPQ-R and BFQ share much variance, even if only four factors account for the
correlations among the eight scales. Eysenck's hypothesis that Agreeableness and Conscientiousness are
facets of the Psychoticism factor was partially supported. Only by using external criteria (e.g. the
correlation with biological variables) can the question of dimensionality be resolved. # 2000 Elsevier
Science Ltd. All rights reserved.

Keywords: Personality; Five-factor theory; Three-factor theory; Factor analysis; Canonical correlation analysis;
Eysenck Personality Questionnaire-Revised; Big Five Questionnaire

1. Introduction

Eysenck's three-factor theory (e.g. Eysenck & Eysenck, 1985) and the Big Five theory
(Digman, 1990) have emerged as the two most important psychometric theories in the ®eld of
personality. According to the ®rst, there are three main factors: Extraversion, Neuroticism and
Psychoticism, whereas the Big Five theory claims that ®ve factors are needed to account for

* Tel.: +39-81-758-0461; fax: +39-81-758-0461.

E-mail address: asaggino@iol.it (A. Saggino).

0191-8869/00/$ - see front matter # 2000 Elsevier Science Ltd. All rights reserved.
PII: S 0 1 9 1 - 8 8 6 9 ( 9 9 ) 0 0 1 4 6 - 4
880 A. Saggino / Personality and Individual Di€erences 28 (2000) 879±886

most of the variance in the ®eld of personality: Extraversion, Neuroticism, Agreeableness,

Conscientiousness and Openness to Experience. The Big Five model has derived mainly from
the lexical approach to the study of personality (for example, Digman, 1990; John, 1990;
McCrae & John, 1992). Subsequently, this approach has been applied to the study of
personality by means of self-report inventories.
According to McCrae and Costa (1985), the two models have di€erent aims. The three-factor
model has the primary aim of identifying those dimensions of individual di€erences which are well
grounded in biological processes, while the aim of the ®ve-factor model is ``the classi®cation of all
major sources of individual di€erences in personality'' (McCrae & Costa, 1985, p. 588).
Di€erent authors have suggested the presence of ®ve large, stable second-order factors in
factor analyses of various questionnaires. For example, Krug and Johns (1986) factor analyzed
Cattell's Sixteen Personality Factors (16 PF; Cattell, Eber & Tatsuoka, 1970) ®nding ®ve
second-order factors; McCrae and Costa (1989) found four out of the ®ve big personality
factors in the Myers±Briggs Type Indicator (MBTI; Myers & McCaulley, 1985). Costa and
McCrae (1995) found ®ve factors in a validation study of the Eysenck Personality Pro®ler
(EPP; Eysenck & Wilson, 1991). Others have found more support for Eysenck's `giant three'.
For example, Saggino and Kline (1996) found the Myers-Briggs Type Indicator scales resemble
Eysenck's three P, E and N dimensions. Indeed, Draycott and Kline (1995) found only three,
corresponding to Eysenck's three factors, in a joint factor analysis of the NEO Personality
Inventory (NEO-PI; Costa & McCrae, 1985) and the Eysenck Personality Questionnaire-
Revised (EPQ-R; Eysenck & Eysenck, 1991). They concluded that `` . . .the NEO-PI does
indeed account for variance over and above that accounted by the EPQ-R but that this
residual variance fails to form appropriately sized factors representative of the Big Five. Of the
variance shared between the two instruments a high proportion of it is attributable to the
robust dimensions of extraversion and neuroticism'' (Draycott & Kline, 1995, p. 803).
McCrae and Costa (1985) showed that the E and N factors of the NEO-PI were similar to
Eysenck's E and N factors. Eysenck's P was essentially covered by their Conscientiousness and
Agreeableness scales. Eysenck (1991, 1992a) interpreted A and C as facets of the low pole of P,
whereas Costa and McCrae considered P as ``a relatively arbitrary con¯ation of two
independent dimensions'' (Costa & McCrae, 1995, p. 310). Costa and McCrae (1995)
considered the pattern of correlations they found between Agreeableness, Conscientiousness,
their sum and the facets of the EPP incompatible with Eysenck's (1991, 1992a) hypothesis that
A and C are facets of the Psychoticism dimension.
While many researchers support the ®ve-factor position, Eysenck (e.g. 1991, 1992a) suggests
that only three factors are basic and are the best account for personality variance.
Furthermore, they are underpinned by biological functioning. According to Eysenck, the
remaining factors can be only ®rst-order factors.
Di€erent methods are possible to compare the three-factor and the ®ve-factor systems. It is
possible, for example, to conduct joint factor analyses of homogeneous item parcels or joint
factor analyses at item level. In the present study we are concerned with the two systems as sets of
scales, not as item pools. Therefore, the present paper addresses the question of the
dimensionality confronting the three- and the ®ve-factor models at the scale level, comparing the
Italian edition of the EPQ-R with the Big Five Personality Questionnaire (BFQ; Caprara,
Barbaranelli, Borgogni & Perugini, 1993). A second aim of this paper is to verify how the EPQ-R
 A. Saggino / Personality and Individual Di€erences 28 (2000) 879±886 881

relates to the BFQ, whether the BFQ data account for variance over and above that accounted for
by the EPQ-R. In other words, we are interested in the nature of the shared variance between the
BFQ and the EPQ-R, trying to replicate the ®ndings of Draycott and Kline (1995).

2. Method

2.1. Tests administered

Two tests were administered in this study:

1. The Eysenck Personality Questionnaire-Revised is an improvement of the Eysenck
Personality Questionnaire (Eysenck & Eysenck, 1975). The main improvement over the EPQ
is the further development of the P scale. The Italian edition of the EPQ-R is a 100 forced-
choice items self-report questionnaire which measures Eysenck's three second-order factors:
Extraversion (E), Neuroticism (N) and Psychoticism (P). It also has a lie scale (L). The six
items appended at the end of the original edition of the EPQ-R, scored only for the
Addiction and Criminality scales, were excluded from the Italian edition. All the items use a
dichotomous format (yes/no). The reliabilities and the factor structure of the Italian edition
are reported in San Martini, Mazzotti and Setaro (1996).
2. The Big Five Questionnaire is a 132 items self-report inventory which measures ®ve factors
corresponding to the Big Five: Energy (E), Friendliness (F), Conscientiousness (C),
Emotional Stability (ES) and Openness (O). Unlike Eysenck's Neuroticism factor, a high
score on the BFQ Emotional Stability scale has to be interpreted as low neuroticism. The
BFQ, like the EPQ-R, also has a Lie scale (L). Every dimension is organized in two facets:
Dynamism and Dominance for Energy; Cooperativeness and Politeness for Friendliness;
Scrupulousness and Perseverance for Conscientiousness; Emotion Control and Impulse
Control for Emotional Stability; Openness to Experiences and Openness to Culture for
Openness. Every facet scale contains 12 items. Half are positively worded and half are
negatively worded. So each of the ®ve scales contains 24 items. The Lie scale, designed to
assess social desirability, contains 12 items. The subjects have a 5-choice answer scale to
reply to the items, ranging from complete disagreement (1=`very false for me') to complete
agreement (5=`very true for me'). If for any reason a subject does not answer a question,
his/her answer is coded 3 (intermediate), according to the instructions given in the BFQ
manual (Caprara, Barbaranelli & Borgogni, 1993). According to the authors, the BFQ has
good validity and it seems to measure the same ®ve constructs measured by the NEO
Personality Inventory (McCrae & Costa, 1985). All the necessary information on the
reliability and validity of the BFQ are reported in Caprara et al. (1993a, b).

2.2. Subjects

A total of 217 volunteer Ss took part in the study. Seven Ss were not considered because the
test data were not complete, so only 210 Ss (32 males and 174 females) are utilized in the
882 A. Saggino / Personality and Individual Di€erences 28 (2000) 879±886

present study. Four subjects did not indicate their sex. Ss were all third and ®fth year
psychology undergraduate students: 155 were students in clinical psychology and 49 in
developmental psychology (six students did not indicate their study subject). The present
sample is representative of the Italian population of psychology students in that females are in
the majority. The mean age of the 210 Ss who completed the questionnaires was 23.63 yr
(S.D.=3.12). It was 23.61 yr for females (S.D.=3.26) and 23.78 yr for males (S.D.=2.50). All
the subjects had the senior high school diploma. Six already had a graduation diploma in a
di€erent subject.

2.3. Procedure

Subjects were asked to complete both the EPQ-R and the BFQ in one group session as part
of their course module. They were told that all data would be used for research purposes only
and that the two questionnaires had to be ®lled in as quickly as they could. Both
questionnaires were completed anonymously.

2.4. Analyses

Two combined analyses with males and females were carried out, because we are interested
not in sex di€erences but in comparing the two di€erent personality models. In the ®rst, we
factor analyzed the three EPQ-R scales together with the ®ve BFQ scales using a principal
component analysis. In the second, the scale scores of the two tests were subjected to a
canonical correlation analysis. In the ®rst analysis factors were rotated to simple structure both
by the orthogonal Varimax procedure and by the oblique Direct Oblimin procedure. Because
the two procedures yielded the same results, only the Direct Oblimin procedure is presented.
Furthermore, the Direct Oblimin rotation was shown to be highly ecient by Kline and
Barrett (1983).

3. Results

According to the scree test, four factors were rotated to simple structure by the oblique
Direct Oblimin procedure. The four factors were rotated to simple structure also by the
Varimax rotation but similar results were obtained.
Table 1 reports the structure matrix of a four factor Direct Oblimin solution. This solution
explains 78.49% of the variance. Absolute loadings equal or greater than 0.30 were regarded as
salient. Factor 1 is clearly an Extraversion factor, loading both on the EPQ Extraversion and
the BFQ Energy scales. BFQ Openness is the other loading of note. There is also a secondary
loading on the BFQ Friendliness scale. Factor 2 is the anxiety factor, loading both on the EPQ
Neuroticism scale and the BFQ Emotional Stability scale. Factor 3 is the tough-mindedness
(Psychoticism) factor. The Psychoticism and Friendliness scales have high loadings on this
factor. Factor 4 seems to be the conscientiousness factor, because it loads on the BFQ
Conscientiousness scale. EPQ Extraversion and BFQ Openness scales have secondary loadings
on this factor. Therefore, our results appear to be only in part similar to those obtained by
 A. Saggino / Personality and Individual Di€erences 28 (2000) 879±886 883

Table 1
Structure matrix of a principal components solution for the EPQ-R and the BFQ scales

Componentsa

Factor 1 Factor 2 Factor 3 Factor 4 h2

BFQ-E 0.844 ÿ0.171 0.251 0.749

EPQ-E 0.796 0.248 ÿ0.307 0.819
BFQ-O 0.616 ÿ0.254 ÿ0.209 0.470 0.618
BFQ-ES ÿ0.945 0.104 0.136 0.900
EPQ-N ÿ0.197 0.943 ÿ0.148 0.902
EPQ-P 0.150 0.130 ÿ0.869 0.812
BFQ-F 0.440 ÿ0.190 0.663 0.612
BFQ-C 0.153 0.912 0.867
Eigenvalue 2.377 1.540 1.225 1.137
% of variance 29.71 19.25 15.31 14.22
Cumulative % of variance 29.71 48.96 64.27 78.49
a
The absolute values less than 0.10 have been omitted.

Draycott and Kline (1995). The present factor solution seems to account very neatly for the
EPQ Psychoticism and Neuroticism variance and for the BFQ Energy, Emotional Stability and
Conscientiousness variance. In fact, all these scales load on the designated factors and their
communalities are high. The same could be said for the EPQ Extraversion scale, even if it has
a secondary loading on Factor 4. The remaining scales appear to be more problematic. In fact,
they have secondary loadings and only about two thirds of their variance is accounted for by
the four-factor solution.
Table 2 sets out the canonical correlation analysis of the EPQ-R and the BFQ scales. It
reports the percentage of variance accounted for by the canonical variates, the redundancy, the
canonical factor loadings and the canonical weights. The ®rst canonical correlation is 0.811.
The second, after removing the ®rst root, is 0.660; the third, after removing the ®rst two
canonical roots, is 0.438. Therefore, there is an overlapping variance between the canonical
variate pairs of 66, 44 and 19% respectively. All the three canonical correlations are high ( p <
0.001), indicating that all the three canonical roots are interpretable and that there is a
signi®cant proportion of overlapping variance. If a correlation of 0.30 is taken as cut-o€, it is
clear that BFQ-E, BFQ-ES and EPQ-N are implicated in the interpretation of the ®rst
canonical variate. In fact, those with low scores on BFQ-E and BFQ-ES tend to have high
scores on EPQ-N, as would be expected. Similarly, BFQ-E and EPQ-E are implicated in the
second canonical variate, because students with low scores on the BFQ-E scale tend to have
low scores on the EPQ-E scale. As regards the third canonical variate, scores on BFQ-
Friendliness, BFQ-Conscientiousness, BFQ-Openness and EPQ-Psychoticism scales tend to
correlate with the third canonical variate indicating that those with high scores on BFQ-O and
low scores on BFQ-F and BFQ-C tend to have high scores on EPQ-P. Therefore, it appears
that variates 1 and 2 are interpretable as the two dimensions of Neuroticism and Extraversion.
Variate 3 has high correlations with EPQ-P, BFQ-F, BFQ-C and BFQ-O scales. Variate 3 is
more dicult to interpret, though it appears that EPQ-P is a combination of BFQ-F, BFQ-C
884 A. Saggino / Personality and Individual Di€erences 28 (2000) 879±886

Table 2
Canonical correlation analysis for the BFQ and the EPQ-R scales

First canonical variate Second canonical variate Third canonical variate

canonical factor canonical canonical factor canonical canonical factor canonical

loadings weights loadings weights loadings weights

BFQ-E ÿ0.311 ÿ0.215 ÿ0.804 ÿ0.918 ÿ0.018 ÿ0.022

BFQ-F ÿ0.222 ÿ0.031 ÿ0.272 ÿ0.207 ÿ0.767 ÿ0.879
BFQ-C 0.009 0.194 0.254 0.452 ÿ0.341 ÿ0.378
BFQ-ES ÿ0.960 ÿ0.943 0.264 0.349 ÿ0.064 0.031
BFQ-O ÿ0.274 ÿ0.067 ÿ0.276 0.004 0.322 0.615
Percentage of variance 22.883 18.620 16.264 Total=57.766
Redundancy 15.052 8.117 3.120 Total=26.290
EPQ-P 0.058 ÿ0.100 ÿ0.244 ÿ0.246 0.968 0.977
EPQ-E ÿ0.266 ÿ0.101 ÿ0.928 ÿ0.981 ÿ0.260 ÿ0.241
EPQ-N 0.990 0.989 ÿ0.119 ÿ0.249 0.071 ÿ0.122
Percentage of variance 35.152 31.179 33.668 Total=100.00
Redundancy 23.122 13.593 6.460 Total=43.175

and BFQ-O scales. These results are substantially similar to those obtained by Draycott and
Kline (1995). It is important to note that the number of three roots is just an artefact of the
analysis, because the maximum number of roots that can be extracted is equal to the minimum
number of variables in either set.
Particularly interesting is the percentage of variance each variate pair extracts from the
variables of the other set, that is `redundancy'. As can be seen from Table 2, the EPQ-R
variance accounts for 43.175% of BFQ variance and the ®ve factors of the BFQ account for
26.290% of EPQ-R variance. Therefore, 57% of BFQ variance is not accounted for by the
EPQ-R, even though these two questionnaires have variance in common. Our unaccounted
percentage of variance is higher than that of Draycott and Kline (1995). Therefore, we can
conclude that the variance of the ®ve BFQ dimensions is only partially subsumed by the three
Eysenck scales.

4. Discussion and conclusion

From the results of the present study it appears that the EPQ Extraversion and BFQ Energy
scales measure the same construct. The same is true of the BFQ Emotional Stability and EPQ
Neuroticism scales. This is a signi®cant ®nding if we consider the di€erent origins of the EPQ-
R and the BFQ. Therefore, these constructs appear to represent stable, reliable traits of
personality.
As demonstrated by the factoring of all the eight scales, a four factor solution best accounts
for the correlations among them. The variance which remains unaccounted for does not form
signi®cant, independent factors. Our results are generally similar to those of Draycott and
Kline (1995), although they found only three factors using the NEO-PI and EPQ-R. In their
 A. Saggino / Personality and Individual Di€erences 28 (2000) 879±886 885

factor solution the BFQ Conscientiousness scale loaded on the tough-mindedness factor.
Otherwise, our results are very similar.
A point of relevance is the high loading of the BFQ-F scale on the Psychoticism factor. This
result is con®rmed by the canonical correlation analysis (Table 2), according to which the EPQ
Psychoticism scale is essentially a combination of BFQ-C, BFQ-F and BFQ-O scales. These
combined results seem to con®rm in part Eysenck's (1991, 1992a) hypothesis that
Conscientiousness and Agreeableness are facets of the Psychoticism factor. Similar results were
obtained by Draycott and Kline (1995) and also by Costa and McCrae (1995), though in their
case only C had salient correlations with some facets of the EPP, mainly with facets of the
Psychoticism factor. Costa and McCrae (1995) consider that their results demonstrate the
inconsistency of Eysenck's (1991, 1992a) position. We disagree, since their main correlations of
Sum A+C are with the Psychoticism factor. According to Costa and McCrae (1995, p. 310),
``if A and C are best regarded as facets of P, they should show a similar pattern of correlations
with external criteria and summing A and C should yield a variable with even stronger
correlations than either component show''. Furthermore, Eysenck (1992b) arms A and C are
only two of the Psychoticism facets. Therefore, A and C would cover only a small area in the
rich ®eld of personality primary traits. If that is true, it could help to explain the apparent
contradiction.
If we look at the canonical correlation analysis of all eight BFQ and EPQ-R scales, it is
clear that the BFQ accounts for variance over and above that accounted for by the EPQ-R.
But in the present study this residual variance fails to form further representative factors, as is
the case in Draycott and Kline's (1995) study. Anyway, a high proportion of variance seems to
be attributable to the two Extraversion and Neuroticism factors.
In our opinion, it is only by using external validation criteria that the question of
dimensionality (Big Three or Big Five) can be ®nally sorted out. The recovery of the ®ve-factor
model from an instrument designed to measure it is hardly convincing. The same is true if we
obtain three factors using an instrument designed to measure the three-factor theory. In fact,
factor analysis can be useful but cannot be used alone to decide among di€erent theoretical
orientations (Eysenck, 1992b). Therefore, only if we use external criteria, such as biological
measures, can this question be resolved. In this respect the three-factor model has been better
researched (e.g. Eaves, Martin & Eysenck, 1989).

Acknowledgements

This work was supported by a grant from the Italian National Research Council (CNR). I
gratefully acknowledge the comments of Dr. Sybil B.G. Eysenck and Dr. Glenn Wilson on
earlier drafts of this article.

References

Caprara, G. V., Barbaranelli, C., & Borgogni, L. (1993a). BFQ Ð Big Five Questionnaire: manuale. Firenze:
Organizzazioni Speciali.
886 A. Saggino / Personality and Individual Di€erences 28 (2000) 879±886

Caprara, G. V., Barbaranelli, C., Borgogni, L., & Perugini, M. (1993b). The `Big Five Questionnaire': a new
questionnaire to assess the ®ve factor model. Personality and Individual Di€erences, 15, 281±288.
Cattell, R. B., Eber, H. W., & Tatsuoka, M. M. (1970). Handbook for the Sixteen Personality Factor Questionnaire.
Windsor, Berks: National Foundation for Education Research.
Costa, P. T., & McCrae, R. R. (1985). The NEO Personality inventory manual. Odessa, FL: Psychological
Assessment Resources.
Costa, P. T., & McCrae, R. R. (1995). Primary traits of Eysenck's PEN system: three- and ®ve-factor solutions.
Journal of Personality and Social Psychology, 69, 308±317.
Digman, J. M. (1990). Personality structure: emergence of the Five Factor Model. Annual Review of Psychology, 41,
417±440.
Draycott, S. G., & Kline, P. (1995). The Big Three or the Big Five Ð the EPQ-R vs the NEO-PI: a research note,
replication and elaboration. Personality and Individual Di€erences, 18, 801±804.
Eaves, L., Martin, N., & Eysenck, H. J. (1989). Genes, culture and personality: an empirical approach. New York:
Academic Press.
Eysenck, H. J. (1991). Dimensions of personality: 16, 5 or 3 criteria for a taxonomic paradigm. Personality and
Individual Di€erences, 12, 773±790.
Eysenck, H. J. (1992a). Four ways ®ve factors are not basic. Personality and Individual Di€erences, 13, 667±673.
Eysenck, H. J. (1992b). A reply to Costa and McCrae. P or A and C Ð the role of theory. Personality and
Individual Di€erences, 13, 867±868.
Eysenck, H. J., & Eysenck, M. W. (1985). Personality and individual di€erences: a natural science approach. New
York: Plenum Press.
Eysenck, H. J., & Eysenck, S. B. G. (1975). Manual of the Eysenck Personality Questionnaire. London: Hodder &
Stoughton.
Eysenck, H. J., & Eysenck, S. B. G. (1991). Manual of the Eysenck Personality Scales. London: Hodder &
Stoughton.
Eysenck, H. J., & Wilson, G. D. (1991). The Eysenck Personality pro®ler. London: Assessment Network, Ltd.
John, O. P. (1990). The `Big Five' factor taxonomy: Dimensions of personality in natural language and in
questionnaires. In L. A. Pervin, Handbook of personality: theory and research (pp. 66±100). New York: Guilford.
Kline, P., & Barrett, P. (1983). The factors in personality questionnaires among normal subjects. Advances in
Behavioral Research and Therapy, 5, 141±202.
Krug, S. E., & Johns, E. F. (1986). A large sample validation of second-order personality structure de®ned by the
16PF. Psychological Reports, 59, 683±693.
McCrae, R. R., & Costa, P. T. (1985). Comparison of the EPI and psychoticism scales with measures of the ®ve-
factor model of personality. Personality and Individual Di€erences, 6, 587±597.
McCrae, R. R., & Costa, P. T. (1989). Reinterpreting the Myers±Briggs Type Indicator from the perspective of the
®ve-factor model of personality. Journal of Personality, 57, 17±40.
McCrae, R. R., & John, O. P. (1992). An introduction to the Five Factor model and its applications. Journal of
personality, 60, 175±215.
Myers, I. B., & McCaulley, M. H. (1985). Manual: A guide to the development and use of the Myers±Briggs Type
Indicator. Palo Alto, CA: Consulting Psychologist Press.
Saggino, A., & Kline, P. (1996). The location of the Myers±Briggs Type Indicator in personality factor space.
Personality and Individual Di€erences, 21, 591±597.
San Martini, P., Mazzotti, E., & Setaro, S. (1996). Factor structure and psychometric features of the Italian version
for the EPQ-R. Personality and Individual Di€erences, 21, 877±882.