Download as pdf or txt
Download as pdf or txt
You are on page 1of 14

Personality and Individual Di€erences 29 (2000) 907±920

www.elsevier.com/locate/paid

The NEO-FFI: emerging British norms and an item-level


analysis suggest N, A and C are more reliable
than O and E
Vincent Egan a,*, Ian Deary b, Elizabeth Austin c
a
East Midlands Centre for Forensic Mental Health, Arnold Lodge, Cordelia Close, Leicester LE5 0LE, UK
b
Department of Psychology, University of Edinburgh, Edinburgh, Scotland, UK
c
Biomathematics and Statistics Scotland, Edinburgh, Scotland, UK
Received 2 June 1999; received in revised form 26 October 1999; accepted 12 November 1999

Abstract

The NEO Five Factor Inventory (NEO-FFI) was given to 1025 British subjects as part of three
independent research studies. Data from these studies were pooled and subjected to item-level analyses.
Using standard scoring criteria from the measure provisional British norms were produced which were
broadly equivalent to those obtained in the USA. The individual subscales showed good internal
consistency. However, the item-level principal components analysis using varimax and oblique rotation
and con®rmatory factor analysis revealed that only the Neuroticism, Agreeableness and
Conscientiousness traits were coherently represented in the main factors derived by the analysis.
Openness and Extraversion factors did not show such stability or consistency. It is argued that as a
result of these diculties, thoughtlessly embracing the NEO-FFI as a quick and ecient instrument for
measuring the `Big Five' personality traits is perhaps premature, as the instrument requires modi®cation
and improvement before it can truly be regarded as measuring ®ve independent personality traits. 7 2000
Elsevier Science Ltd. All rights reserved.

Keywords: `Big Five'; Personality; Psychometrics; Item-analysis; NEO-FFI; Norms

* Corresponding author. Fax: +44-116-246-0379.


E-mail address: vince.egan@arnoldl.cnhc-tr.trent.nhs.uk (V. Egan).

0191-8869/00/$ - see front matter 7 2000 Elsevier Science Ltd. All rights reserved.
PII: S 0 1 9 1 - 8 8 6 9 ( 9 9 ) 0 0 2 4 2 - 1
908 V. Egan et al. / Personality and Individual Di€erences 29 (2000) 907±920

1. Introduction

The recognition of the `Big Five' personality traitsÐNeuroticism (N), Extraversion (E),
Openness (O), Agreeableness (A), and Conscientiousness (C)Ðallegedly re¯ects a converging
general consensus in di€erential psychology (Brand & Egan, 1989). The `Big Five' model can
incorporate a number of other personality theories, suggesting that it provides a useful general
framework for viewing human behaviour (Costa & McCrae, 1995). Evidence for similar `Big
Five' components has been found in other excursions into questionnaire and lexical research
(Goldberg, 1993).
Measurement of the `Big Five' is possible using a number of instruments, of which the most
standard is the revised NEO Personality Inventory (NEO PI-R, Costa & McCrae, 1992). The
NEO PI-R comprises 240 statements, to which the individual responds by stating whether they
`strongly disagree', `disagree', are `neutral', `agree', or `strongly agree' with a given proposition
about themselves. Items are summed to provide an overall measure of the ®ve broad traits.
Each broad trait has six facet scores (i.e., elements of the trait which converge to give an
overall trait description), for example `Anxiety', `Angry Hostility', `Depression', Self-
Consciousness', `Impulsiveness' and `Vulnerability' all contribute to overall N.
In many clinical and research settings subjects are unable or unwilling to complete a lengthy
questionnaire, and general information about personality is regarded as sucient. For this
reason, a short-form of the NEO PI-R was developed; the NEO Five Factor Inventory (NEO-
FFI, Costa & McCrae, 1992). The NEO-FFI comprises 60 items derived from a factor analysis
of the 1986 administration of the NEO-PI. For each domain, the 12 items with the highest
positive loading on the corresponding trait were taken. The resulting short scales correlated
upwards of 0.68 (and mostly substantially higher) with the full NEO-PI trait scales, and
demonstrated good internal reliabilities (McCrae & Costa, 1989). This brief personality
instrument accounts for about 85% of the variance in convergent validity criteria, as derived
from ratings of similar traits using adjective endorsement, and spouse and peer ratings (Costa
& McCrae, 1992, p. 54). Clearly, for a quick and e€ective general measure of personality the
NEO-FFI appears to be more than adequate.
While the `Big Five' model is useful conceptually, there continue to be rumblings of
discontent and uncertainty about NEO-PI's empirical and theoretical underpinnings (Eysenck,
1991). For example, NEO-PI Openness is not orthogonal to intelligence, as Openness appears a
predictor of IQ as measured by the revised Wechsler Adult Intelligence Scale (McDonald,
1995), and Conscientiousness may be little more than a facet of Eysenck's `Psychoticism' factor
(Draycott & Kline, 1995). The psychometry behind the items and factor structure of the NEO-
FFI also appear more ambiguous than one would perhaps desire; for example, despite the
wealth of empirical information in the NEO PI-R Manual (Costa & McCrae, 1992), no
original item-level analyses of their scales were provided. Subsequent studies have attempted to
redress this shortcoming. For example, a study carried out using Canadian female student
volunteers examined the NEO-FFI at an item level found that some items in the O and A
scales did not load highly on their corresponding component, and that their observed scores
deviated signi®cantly from the norms for women in the manual (Holden & Fekken, 1994). In
another study, con®rmatory factor analysis seeking to con®rm the purported structure of the
NEO-FFI found that whilst 35% of the observed variance could be explained by ®ve factors,
V. Egan et al. / Personality and Individual Di€erences 29 (2000) 907±920 909

the comparative ®t index (an indication of how closely the theoretical model ®ts the observed
data) was just 0.66, suggesting a weak factor structure (Mooradian & Nezlak, 1996). A further,
and perhaps fundamental problem with the NEO-FFI is that the scales are not independent;
for example Deary et al. (1996) found that the N scale of the NEO-FFI correlated with E at
ÿ0.42 (P < 0.001), and with C at ÿ0.39 (also P < 0.001). As the dimensional traits are
correlated, a 5-factor solution may not be optimal, and analyses of the NEO-FFI at a trait
dimension level have suggested that 5, 4, 3 and even 2-factor solutions are possible (Ackerman
& Heggestad, 1997; Ferguson & Patterson, 1998). To the degree that the NEO-FFI derives
from the most discriminating items of the NEO PI-R, these various problems, uncertainties
and practical problems may be perpetuated amongst those researchers who continue to use
either of these scales.
The current study sought to generate British norms for the trait scores which would enable
researchers and clinicians working in the UK to consider individual records, and no longer
have to rely on possibly invalid American norms. It also sought to examine the psychometric
properties of the NEO-FFI within a British cohort in order to ascertain whether similar item
ambiguity could be observed in a British sample.

2. Subjects and method

2.1. Subjects

The study cohort comprised 1025 subjects derived from the pooled information from studies
conducted by Willock et al. (1999), Deary et al. (1996), and Egan et al. (1999). The Willock et
al. study (1999) comprised 252 individuals being examined as part of a study of individual
di€erences in¯uencing decision-making in Scottish farmers. The farmers in the study had an
income greater than £16,000 per annum. The Deary et al. study (1996) involved 454 consultant
doctors in Scotland across a range of major specialities being studied as part of an
investigation into work-related stress. The Egan et al. study (1999) comprised 301 individuals
(of whom 112 were clinical referrals to the Regional Forensic Psychology Service for
assessment and treatment) recruited as subjects in a study of sensational interests and
personality traits. The control group in this latter study comprised cleaners, security men,
®shermen as well as more educated professionals. Pooling the three studies provided a cohort
with a good range of skills, mental ability and putative psychopathology, making it a more
representative cross-section of British society than cohorts comprising one occupation, or
student samples. The full sample comprised 803 males (78.1%), and 221 females (21.5%); four
individuals did not provide information on their sex. Nine hundred and sixty-two individuals
provided information about their age, their mean age being 44.9 years (SD=13.2).

2.2. Method

The 60 items of the NEO-FFI were recorded for all participants. Dimensional scores derived
from the standard NEO-FFI scoring system were calculated and used to produce mean,
standard deviation, and alpha reliability values for a large and relatively representative British
910 V. Egan et al. / Personality and Individual Di€erences 29 (2000) 907±920

sample. T-scores based on British ranges for the individual scales broken down by sex were
calculated. The NEO-FFI items were then subjected to principal components analysis with
varimax and oblique rotation. A scree test was used to identify components to be retained. To
vigorously test the structure of the NEO-FFI, we examined the factors obtained against those
for other samples, and conducted a con®rmatory factor analysis (CFA) using structural
equation (SEQ) modelling of the items to see how well they ®tted together in relation to the
model proposed to underlie the measure.

3. Results

3.1. British norms

For the purposes of generating provisional British NEO-FFI norms, the items were summed
according to standard scoring procedures in the instrument manual (Costa & McCrae, 1992).
Table 1 presents a summary of trait means, standard deviations and alpha reliabilities for the
full sample. All scales were highly reliable, with N being particularly so. Table 1 also presents a
comparison of male and female scores on the ®ve NEO-FFI scales; all but C show signi®cant
sex di€erences, with males being lower on average in N, E, O and A than females; there was
no di€erence for C. The large number of subjects in the sample allowed us to provide T-score
norms for men and women in the UK. These are presented in a format similar to the pro®le
printed on the inside of Form S of the NEO-FFI simplifying interpretation of the raw scores
by converting them to T-values: i.e., they have a mean of 50 and an SD of 10. These tables are
presented in Appendices A and B.

3.2. Intercorrelations between the trait scores

Table 2 presents the intercorrelations between the ®ve trait scores derived from the NEO-
FFI. These indicate that the traits are indeed correlated, with N being substantially associated
with lower E, lower A, and lower C, and E being associated with higher levels of C.
Exploratory factor-analysis of these ®ve `orthogonal' dimensions revealed, after varimax

Table 1
Means, standard deviations, and alpha reliabilities for a large British sample tested using the NEO-FFI (n = 1025),
and broken down by sex (raw scores)

All subjects (n = 1025) Men (n = 802) Women (n = 221) t P<

Mean SD Alpha reliability Mean SD Mean SD

N 19.5 8.6 0.87 19.04 8.3 21.36 9.3 ÿ3.36 0.001


E 27.1 5.9 0.74 26.83 5.9 28.3 6.0 ÿ3.21 0.001
O 26.5 6.5 0.72 26.02 6.4 28.6 6.3 ÿ5.28 0.001
A 29.7 5.9 0.74 29.25 5.8 31.5 5.6 ÿ5.10 0.001
C 32.1 6.6 0.84 32.09 6.4 31.9 7.5 0.36 n.s.
V. Egan et al. / Personality and Individual Di€erences 29 (2000) 907±920 911

rotation, a two factor solution which converged in three iterations and explained 59.9% of the
variance, the eigenvalues being 1.84 and 1.15. Factor 1 had high positive loadings for E (0.74),
C (0.67), and A (0.50), and a high negative loading for N (ÿ0.77). This dimension thus
encompasses outgoing, orderly, good-natured and emotionally-stable features into a single
continuum representing optimally non-psychopathological features. The second factor was
primarily de®ned by O, which loaded at 0.90 with the underlying dimensional construct. This
factor had lower loadings for C (ÿ0.38) and A (0.32), and could be interpreted as representing
something other than general adjustment of personality.

3.3. Item-level-analysis of the NEO-FFI

To examine from where the sources of variance in the scale the observed factor solution
derived, principal components analysis of the 60 test items was conducted. This extracted 14
factors with eigenvalues over 1, which explained a total of 55.8% of the observed variance in
the NEO-FFI. These factors underwent varimax rotation and converged in 35 iterations. A
scree test of these factors suggested that the ®rst ®ve factors extracted (explaining a total of
36.9% of the variance) represented the main sources of variance in the NEO-FFI data matrix
(Fig. 1).

3.4. Varimax rotation of the NEO-FFI items

Table 3 presents the factor loadings between the NEO-FFI items (reordered and labelled for
ease of reading) and the ®rst ®ve varimax factors (also re-ordered for ease of reading). Factor
1 is clearly and unequivocally N, with all items loading highly and positively on the factor; it
also contains two items from the E dimension (E9 and E12), and individual items from the A
and C traits (A6, and C11, respectively). Factor 2 contains seven of the E items, but also
contains the A7 item `Most people I know like me'. Factor 3 contains nine of the 12 O items;
the item O8 (`I believe we should look to our religious authorities for decisions on moral
issues') does not load on any dimension within the current ®ve-factor solution. Factor 4
contains all 12 A items, but also has a signi®cant negative loading for item N8 (`I often get
angry at the way people treat me'). Factor 5 comprises all C items, but also positive loadings
for items E11 (`I am a very active person') and A10 (`I generally try to be thoughtful and

Table 2
Intercorrelations (Pearson's r ) between NEO-FFI trait scores (n = 1025)a

N E O A C

N ± ÿ40 07 ÿ22 ÿ36


E ± 16 22 30
O ± 08 ÿ15
A ± 13
C ±

a 
Decimal point dropped; two-tailed test; =P < 0.001; =P < 0.02.
912 V. Egan et al. / Personality and Individual Di€erences 29 (2000) 907±920

considerate'); two O items correlate negatively with this C factor, O1 (`I don't like to waste my
time daydreaming'), and O2 (`Once I ®nd the right way to do something, I stick to it').
Overall, item-level analysis suggested that the N, A, and C dimensions represent coherent
separate traits, whilst E and O were more problematic, with three items within these scales not
relating to the ®rst ®ve factors whatsoever. In total, only 46 of the 60 items in the NEO-FFI
represented relatively unique items relating to particular traits. This may account for the
apparent non-orthogonality of the dimensions.

3.5. Oblique rotation of the NEO-FFI items

As a result of this non-orthogonality of factors and the correlated scale scores, an oblique
rotation was applied to the NEO-FFI items. Loadings were accepted as meaningful if they
were equal to or greater than 0.30. This generated a very similar solution in which the ®rst
factor was again unambiguously N (albeit with a minor loading for E12 (`I would rather go
my own way than be the leader of others')). Factor 2 contained nine of the 12 O items. Factor
3 contained all 12 C items, but also had signi®cant loadings for O1 and O2, E11, and A10.
Factor 4 held 10 of the A scale items. Factor 5 had seven of the 12 E items, along with item 7
of the A scale (`Most people I know like me'). Nevertheless, ®ve items of the NEO-FFI did not
load signi®cantly on the ®ve major factors extracted (E6, E7, E10, E12, and O8). Nine of the
items did not load on their ostensibly originating factors. The oblique rotation was able to
allocate four of the wrongly-factored items to their theoretical source, but left six items (E11,
E12, O1, O2, A7 and A10) misplaced.

3.6. Coecients of congruence

The coecient of congruence is used as a measure of the similarity between factors

Fig. 1. NEO-FFI item-analysis. Factor scree plot.


V. Egan et al. / Personality and Individual Di€erences 29 (2000) 907±920 913

Table 3
Factor-analysis of the ®rst ®ve varimax factors extracted from analysis of the NEO-FFI in a sample of 1025 British
subjectsa

Factor 1 Factor 2 Factor 3 Factor 4 Factor 5

N1 44 ÿ14 10 03 05
N2 52 ÿ16 ÿ02 06 ÿ06
N3 70 00 05 ÿ04 ÿ15
N4 57 ÿ16 09 ÿ08 ÿ06
N5 71 ÿ11 07 ÿ08 ÿ10
N6 73 ÿ07 03 ÿ03 ÿ14
N7 60 16 08 07 ÿ06
N8 58 03 ÿ09 ÿ36 03
N9 63 ÿ01 ÿ08 ÿ04 ÿ25
N10 62 ÿ23 04 ÿ01 ÿ06
N11 70 ÿ01 ÿ13 ÿ03 ÿ22
N12 60 05 05 ÿ13 ÿ17

E1 ÿ04 61 13 01 03
E2 ÿ13 64 08 13 02
E3 ÿ13 57 ÿ04 03 ÿ16
E4 ÿ06 63 14 13 16
E5 ÿ17 36 10 ÿ25 24
E6 ÿ22 28 18 26 ÿ02
E7 ÿ14 05 01 ÿ07 14
E8 ÿ16 76 00 05 15
E9 ÿ33 54 04 14 07
E10 09 26 30 ÿ22 27
E11 ÿ21 26 10 ÿ13 41
E12 ÿ31 08 21 ÿ04 15

O1 19 03 14 ÿ04 ÿ39
O2 ÿ09 ÿ20 23 00 ÿ45
O3 08 03 57 05 07
O4 ÿ19 ÿ05 48 09 ÿ06
O5 04 01 62 11 ÿ07
O6 03 25 33 ÿ04 00
O7 10 08 47 10 ÿ09
O8 ÿ04 ÿ02 13 ÿ20 ÿ13
O9 12 07 69 09 00
O10 03 00 67 11 ÿ03
O11 ÿ10 15 60 ÿ13 23
O12 03 04 61 ÿ19 ÿ04

A1 07 27 08 41 25
A2 ÿ26 05 03 51 09
A3 ÿ09 20 ÿ15 54 12
A4 16 06 07 43 00
A5 ÿ26 10 15 50 06
A6 ÿ30 ÿ05 28 46 ÿ08
(continued on next page)
914 V. Egan et al. / Personality and Individual Di€erences 29 (2000) 907±920

Table 3 (continued )

Factor 1 Factor 2 Factor 3 Factor 4 Factor 5

A7 ÿ14 ÿ37 01 33 17
A8 ÿ03 25 ÿ03 58 03
A9 08 ÿ07 11 62 ÿ23
A10 09 21 18 31 34
A11 ÿ12 ÿ11 15 39 ÿ10
A12 06 ÿ08 ÿ21 59 02

C1 09 05 ÿ11 12 49
C2 ÿ16 ÿ03 ÿ10 ÿ01 58
C3 ÿ22 ÿ10 ÿ01 05 48
C4 06 08 07 24 59
C5 ÿ17 05 ÿ08 ÿ11 67
C6 ÿ28 ÿ08 ÿ22 10 50
C7 ÿ11 08 20 00 65
C8 01 11 09 02 57
C9 ÿ28 ÿ06 ÿ17 15 49
C10 ÿ14 18 08 ÿ04 69
C11 ÿ34 ÿ10 ÿ04 ÿ02 56
C12 ÿ02 04 24 ÿ10 62
a
Loadingsr0.30 in bold and underlined, decimal point dropped.

obtained from di€erent samples (Harman, 1967). Values of this coecient range from +1
to ÿ1. A value of +1 indicates perfect agreement between factors (all loadings the same
or proportional) whilst a value of ÿ1 indicates perfect agreement but with a relative
negative sign. Values near zero indicate that there is no agreement. In practice
congruences are found to be large when the factors show a clear visual match and the
number of variables involved is small (Harman, 1967). In the present study, congruences
greater than 0.9 can be taken to indicate a good match of the observed and theoretical
factors. Since the NEO manual does not provide a table of factor loadings for a
standardisation example, our NEO-FFI congruences were calculated with reference to the
NEO-FFI scoring key (e.g., the comparison factor for N had loadings of +1 for all N
items and zero for all other items (Holden & Fekken, 1994)).

Table 4
Coecients of congruence for the NEO-FFI in a large British sample

Orthogonal solution CFA solution

N 0.99 0.99
E 0.88 0.92
O 0.92 0.88
A 0.98 0.98
C 0.99 0.99
V. Egan et al. / Personality and Individual Di€erences 29 (2000) 907±920 915

Table 4 presents the coecients of congruence arising from orthogonal and SEQ solutions. As
is apparent by inspection of the table, the N, A and C scales are highly coherent. The results from
the orthogonal rotation suggest there is a problem with E and hint that O is not quite as good a
scale as N, A or C. In the coecients of congruence generated by the CFA, O is identi®ed as
problematic, and E less reliable than N, A, or C. (The results of the two analyses are not
equivalent, since one model is orthogonal and the other oblique.) Our values for the orthogonal
analysis are not the same as those suggested by one of our referees. This may be because they used
a di€erent reference matrix or they used a di€erent congruence formula. Our analysis was based
upon and checked against the formula in Harman (1967) and is correct!

3.7. Con®rmatory factor analysis

A ®nal test of the NEO-FFI factor structure was done by conducting a CFA with the EQS
program (Bentler, 1995). An initial model was chosen in which the ®ve factors were orthogonal
and each item loaded only on its `own' factor. EQS provides statistical tests to indicate which
paths should be added to or deleted from the model. The Wald test indicated that the item O8 did
not load signi®cantly on its own or any other factor, so it was omitted from further analyses. The
standard procedure in CFA is to use the output of the Lagrange Multiplier test to suggest paths to
be added to the model (item cross-loadings and correlations between factors) until the goodness-
of-®t indices reach acceptable values. This method is not so useful if the sample size and/or the
degrees of freedom of the model are large (Raykov, 1998). For the current data both the sample
size and degrees of freedom are large, so an alternative approach was devised to provide a
`stopping criterion' which produced a parsimonious model. This was done by plotting the chi-
squared increments obtained from the multivariate Lagrange Multiplier test and applying a scree
test criterion to di€erentiate between `large' and `small' increments. Using this method the
following paths were added to the model: (negative) E/N and C/N correlations and cross-loadings
of items E11, O1 and O2 onto the C factor. Full results for the model are shown in Table 5. It can
be seen that the CFA model has a very similar structure to the ®ndings reported in Table 3. The
main di€erence between the two models is that the CFA results strongly suggest that N fails to be
orthogonal to E and C in the NEO-FFI, a ®nding that has been reported previously, see for
example Deary et al. (1996), and is suggested by the sample correlation matrix (Table 4). The
correlations obtained from the CFA model suggest correlations between N and E of ÿ0.32 (P <
0.0001) and, between N and C, ÿ0.34 (P < 0.0001).

4. Discussion

While the NEO-PI is a more complex instrument and contains many subdimensions within
the `Big Five' traits, the brevity of the shorter NEO-FFI makes this instrument practical for
settings where rapid screening is necessary, or where subjects are unable or unwilling to
engage in lengthy assessment procedures, such as for clinical research. As such it is
important that the NEO-FFI meets good standards of reliability and validity, internal
consistency and factorial stability. Provisional British norms for the NEO-FFI suggest that
the trait scores within the instrument have high internal consistencies. Known sex di€erences
916 V. Egan et al. / Personality and Individual Di€erences 29 (2000) 907±920

Table 5
Con®rmatory factor analysis of the NEO-FFI items in a British sample

N E O A C

N1 0.38
N2 0.49
N3 0.69
N4 0.55
N5 0.72
N6 0.73
N7 0.57
N8 0.52
N9 0.62
N10 0.61
N11 0.69
N12 0.59

E1 0.52
E2 0.62
E3 0.49
E4 0.59
E5 0.36
E6 0.34
E7 0.11
E8 0.75
E9 0.63
E10 0.29
E11 0.26 0.38
E12 0.23

O1 0.12 ÿ0.37
O2 0.14 ÿ0.36
O3 0.61
O4 0.31
O5 0.65
O6 0.26
O7 0.39
O8 0.00
O9 0.76
O10 0.61
O11 0.50
O12 0.53

A1 0.41
A2 0.49
A3 0.56
A4 0.33
A5 0.52
A6 0.43
A7 0.41
V. Egan et al. / Personality and Individual Di€erences 29 (2000) 907±920 917

Table 5 (continued )

N E O A C

A8 0.58
A9 0.47
A10 0.35
A11 0.33
A12 0.45

C1 0.42
C2 0.59
C3 0.50
C4 0.51
C5 0.67
C6 0.52
C7 0.64
C8 0.52
C9 0.53
C10 0.69
C11 0.62
C12 0.54

between men and women were recovered, although these were perhaps more marked than
Costa and McCrae (1992) might have anticipated. The T-score distributions calculated for
men and women were similar to those observed in American samples when converted back
into raw scores (Table 6). The similarity between American and British norms (even in what
one anonymous referee called an unrepresentative but ``advantageous conglomeration of
individuals'') ostensibly suggests that information arising from studies using this scale is
readily applicable to both nations.
A more ®ne-grained analysis revealed greater diculties. On the one hand, the N, A, and C
scales seem to measure the latent traits they are supposed to. On the other hand, O and, in
particular, E traits had their variance split amongst many of the trivial factors lying amongst
the scree factors of the analysis. That this was demonstrated using two di€erent methods of
rotation (one arguably more appropriate for the correlated subscales of the NEO-FFI) and

Table 6
Comparison of British and American raw scores for the NEO-FFI when T-score=50

Men Women

UK US UK US

N 19 18 20 20
E 27 27 27 27
O 26 27 26 27
A 30 32 30 34
C 32 34 32 35
918 V. Egan et al. / Personality and Individual Di€erences 29 (2000) 907±920

CFA suggests that our interpretation is correct, and that the NEO-FFI should be revised such
that it is made more robust. Moreover, our study demonstrates that at the trait level, the
dimensions indexed by an ostensibly de®nitive measure of the `Big Five' quickly reduce to an
`overwhelming two'. These results make it hard to maintain the view that the NEO-FFI is an
e€ective short-form instrument for deriving separate `Big Five' dimensions. Researchers
requiring a quick and e€ective measure of the major personality dimensions might be
encouraged to instead use the EPQR-A, which has the necessary separation of traits,
conceptual coherence, and good internal and test±retest reliability (Egan, Miller & McLellan,
1998; Francis, Brown & Phillipchalk, 1992).
Our sample was opportunistic, and as such, comparisons between the constituent occupations
and ages are likely to be confounded by intellectual ability, social class, and psychopathology. It is
for this reason that we do not report norms broken down by age or profession. What our study
does show is that when one has a good-sized sample with a broad range of responses, the
purported homogeneity of the NEO-FFI's ®ve factors is by no means as clear as one might desire.
For researchers and academics, the re-embracing of personality at a legitimate level of
explanation for behaviour and psychological phenomena has been encouraging, perhaps
leading to a tendency to prematurely emphasise the Costa and McCrae `Big Five' scheme and
measure over other systems. This bullishness has been positive in that the importance of trait
di€erences is starting to be again accepted by what has often been, at least in recent history, an
unsympathetic audience. Debate is the hallmark of a mature science, and notwithstanding the
narcissism of minor di€erences, a literature is emerging which is more critical of Costa and
McCrae's all-embracing model. This is perhaps characterised by Matthews and Deary's view
that ``it would be more appropriate to speak of the big ®ves, as there is no single set of
identical dimensions agreed upon by all researchers'' (Matthews & Deary, 1998, p. 26); Block
(1995) speaking of the ``intelligent arbitrariness'' of the NEO-PI facet scales; and McKenzie
(1998) questioning the very basis of an ``Openness to Experience'' factor, seemingly
fundamental to a NEO model. It would not do to discourage or disparage the remarkable and
impressive work done by Costa and McCrae in broadening the appeal of personality and
providing ``solid ground in the wetlands of psychology'' (Costa & McCrae, 1995). Nevertheless,
evidence from the current study suggests that, based on an analysis at the item level, the O and
E scales of the NEO-FFI, and thus perhaps the NEO PI-R, requires some modi®cation before
these conceptual and practical diculties are perpetuated into the Millennium.

Appendix A

Normogram to convert adult males' raw scores on the NEO-FFI to equivalent T-scores
based on British norms (n = 802).
V. Egan et al. / Personality and Individual Di€erences 29 (2000) 907±920 919

T-score Raw score

N E O A C

85 ± ± ± ± ±
80 45 45 46 ± ±
75 41 42 43 45 48
70 37 39 39 41 45
65 33 36 36 39 42
60 28 33 33 36 38
55 24 30 30 33 36
50 19 27 26 30 32
45 15 24 23 27 29
40 11 21 20 24 26
35 7 18 17 21 22
30 2 15 14 18 19
25 ± 12 10 15 16
20 ± 9 7 12 12
15 ± 6 ± 9 ±
10 ± ± ± ± ±

Appendix B

Normogram to convert adult females' raw scores on the NEO-FFI to equivalent T-scores
based on British norms (n = 221).

T-score Raw score

N E O A C

85 48 ± ± ± ±
80 46 ± ± ± ±
75 41 42 43 44 ±
70 37 39 40 42 45
65 33 36 36 39 42
60 28 33 33 36 38
55 24 31 30 33 36
50 20 27 26 30 32
45 15 24 23 27 29
40 11 21 20 24 26
35 7 18 17 21 22
30 3 15 14 18 19
25 ± 13 ± 15 14
20 ± ± ± ± 12
15 ± ± ± ± 9
10 ± ± ± ± 6
920 V. Egan et al. / Personality and Individual Di€erences 29 (2000) 907±920

References

Ackerman, P., & Heggestad, E. (1997). Intelligence, personality and interests: evidence for overlapping traits.
Psychological Bulletin, 121, 219±245.
Bentler, P. M. (1995). EQS structural equations program manual. Encino, CA: Multivariate Software Inc.
Block, J. (1995). A contrarian view of the ®ve-factor approach to personality description. Psychological Bulletin,
117, 187±215.
Brand, C. R., & Egan, V. (1989). The `Big Five' dimensions of personality? Evidence from ipsative, adjectival self-
attributions. Personality and Individual Di€erences, 10, 1071±1165.
Costa, P. T., & McCrae, R. R. (1992). The NEO PI-R professional manual. Odessa, Florida: Psychological
Assessment Resources.
Costa Jr, P. T., & McCrae, R. R. (1995). Solid ground on the wetlands of personalityÐa reply to Block.
Psychological Bulletin, 117, 216±220.
Deary, I. J., Blenkin, H., Agius, R. M., Endler, N. S., Zealley, H., & Wood, R. (1996). Models of job-related stress
and personal achievement among consultant doctors. British Journal of Psychology, 87, 3±29.
Draycott, S. G., & Kline, P. (1995). The Big Three or the Big FiveÐthe EPQ-R vs the NEO-PI: a research note,
replication and elaboration. Personality and Individual Di€erences, 18, 801±804.
Egan, V., Miller, E., & McLellan (1998). Is the personal questionnaire technique more sensitive to cardiac-surgery
related anxiety than a standard pencil-and-paper measure? Personality and Individual Di€erences, 24, 465±473.
Egan, V., Auty, J., Miller, R., Ahmadi, S., Richardson, C., & Gargan, I. (1999). Sensational interests and general
personality traits. Journal of Forensic Psychiatry, 10, 567±582.
Eysenck, H. J. (1991). Dimensions of personality: 16, 5 or 3?Ðcriteria for a taxonomic paradigm. Personality and
Individual Di€erences, 12, 773±790.
Ferguson, E., & Patterson, F. (1998). The ®ve factor model of personality: openness a distinct but related construct.
Personality and Individual Di€erences, 24, 789±796.
Francis, L. J., Brown, L. B., & Phillipchalk, R. (1992). The development of an abbreviated form of the revised
Eysenck Personality Questionnaire (EPQR-A). Personality and Individual Di€erences, 13, 443±449.
Goldberg, L. (1993). The structure of phenotypic personality traits. American Psychologist, 48, 26±34.
Harman, H. H. (1967). Modern factor analysis. Chicago: University of Chicago Press.
Holden, R. R., & Fekken, G. C. (1994). The NEO Five-Factor Inventory in a Canadian context: psychometric
properties for a sample of university women. Personality and Individual Di€erences, 17, 441±444.
Matthews, G., & Deary, I. J. (1998). Personality traits. Cambridge, UK: Cambridge University Press.
McCrae, R. R., & Costa, P. T. (1989). Rotation to maximise the construct validity of factors in the NEO
Personality Inventory. Multivariate Behavioural Research, 24, 107±124.
McDonald, D. A. (1995). The relationship between psychometric intelligence and the ®ve-factor model of
personality in a rehabilitation sample. Journal of Clinical Psychology, 51, 79±88.
McKenzie, J. (1998). Fundamental ¯aws in the ®ve factor model: a re-analysis of the seminal correlation matrix
from which the ``Openness-to-Experience'' factor was extracted. Personality and Individual Di€erences, 24, 475±
480.
Mooradian, T. A., & Nezlak, J. B. (1996). Comparing the NEO-FFI and Saucier's Mini-Markers as measures of the
Big Five. Personality and Individual Di€erences, 21, 213±215.
Raykov, T. (1998). On the use of con®rmatory factor analysis in personality research. Personality and Individual
Di€erences, 24, 291±293.
Willock, J., Deary, I. J., McGregor, M., Sutherland, A., Edwards-Jones, G., Morgan, O., Dent, B., Grieve, R.,
Gibson, G., & Austin, E. (1999). Farmers attitudes, objectives, behaviours, and personality traits: the Edinburgh
study of decision-making on farms. Journal of Vocational Behaviour, 54, 5±36.

You might also like