See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/284610489

Personality types based on the Big Five model: A cluster

analysis over the Romanian population

Article · January 2011

CITATIONS READS

52 4,810

2 authors, including:

Florin Alin Sava

West University of Timisoara
66 PUBLICATIONS 1,815 CITATIONS

SEE PROFILE

All content following this page was uploaded by Florin Alin Sava on 11 January 2016.

The user has requested enhancement of the downloaded file.

Cognition, Brain, Behavior. An Interdisciplinary Journal
Copyright © 2011 ASCR Publishing House. All rights reserved.
ISSN: 1224-8398
Volume XV, No. 3 (September), 359-384

PERSONALITY TYPES BASED ON THE BIG FIVE

MODEL. A CLUSTER ANALYSIS OVER THE
ROMANIAN POPULATION
Florin A. SAVA*, Radu I. POPA
West University of Timişoara, Timişoara, Romania

ABSTRACT

This paper reports two independent studies investigating personality types using
cluster analysis based on the five-factor personality model. In the first study, we
focused on identifying the most appropriate cluster number by checking the
replicability of two to eight clusters across two random halves of a nationally
representative Romanian sample. The second study established the external
validity of the more replicable three-cluster and five-cluster solution by studying
their relations to specific self-reported behaviors. The results indicate good
validity for a five-cluster solution describing the following personality types:
resilient, undercontrolled, strain, overcontrolled, and passive. The findings are
discussed in relation to a suggested classification based on degrees of self-control
and of adaptation / flexibility.

KEYWORDS: Five-factor personality model, Big Five, cluster analysis,

personality type, resilient type, DECAS Personality Inventory

TRAITS VS. PERSONALITY TYPES

Personality psychology is currently dominated by a dimensional approach to

personality description in which the scores for each personality trait are usually
considered in isolation. This type of analysis reflects initial intuition because
individuals have a tendency to evaluate others by focusing on only one dimension
(e.g., George is friendly). Although there is often a correspondence between the
score obtained for a given dimension and a particular behavioral descriptor (e.g.,
individuals with high scores on Agreeableness Scale are often perceived as being
friendly), this is not always the case because other relevant factors are ignored. For

*
Corresponding author:
E-mail: [email protected]
360 F. A. Sava, R. I. Popa

example there is a much higher probability that an individual will be described by

others as being friendly if he or she obtains high scores on both the Agreeableness
and Extraversion scales. By contrast an individual with high agreeableness but low
extraversion will tend to be perceived as docile or conformist.
Combinations of factors can therefore provide us with more information
than we can obtain by interpreting a given scale in isolation. A personality type,
from this perspective, is described in terms of unique combinations of traits. In
practice, the best known are probably the 16 personality types described by the
Myers-Briggs Type Indicator (MBTI) (Briggs Myers, McCaulley, Quenk, &
Hammer, 1998), but this typology has been criticized in the academic world due to
a series of shortcomings concerning its scientific validity (Hunsley, Lee, & Wood,
2004). Other familiar personality type descriptions are those proposed in the
California Personality Inventory (CPI) - alpha, beta, gamma, and delta (Gough &
Bradley, 1996, as cited in Pitariu, Iliescu, Tureanu, & Peleasa, 2006), those
proposed by Eysenck (1967), based on an ancient temperament typology, and those
proposed by Block and Block (1980) – brittle undercontrollers, brittle
overcontrollers, resilient undercontrollers, and resilient overcontrollers.
Research on personality types belongs to the person-centered tradition in
personality research and focuses on those types that are frequently observed in a
population. Previous studies comparing the predictive power of trait vs. type
approach reported mixed results (Huey & Weisz, 1997) or suggested similar
predictive power both in terms of the multivariate percentage of explained variance
and the number of significant longitudinal correlates (Asendorpf & Denissen,
2006), while a study conducted by Costa, Herbst, McCrae, Samuels, and Ozer
(2002) favored the dimensional approach. Asendorpf and Denissen (2006)
considered that the type approach did better when predicting longitudinal compared
to concurrent outcomes due to the fact that a type is dependent on pattern of traits
and not on the mean-level tendencies that might fluctuate across time as it is the
case for traits.
In our opinion, the fact that the type approach was relatively on par with
the variable approach is remarkable. After all, the dimensional approach allows for
more predictors (usually those included in the five-factor model), which are
assessed in a fine-grained manner (continuous variables), while the most common
type approach (Block’s typology) usually involves only three types measured as
categorical variables.

PERSONALITY TYPES AND THE FIVE-FACTOR PERSONALITY MODEL

Although the Big-Five model is an atheoretical dimensional model, various

attempts have been made to establish a personality typology based on particular
combinations of the five personality traits (Asendorpf, Borkenau, Ostendorf, & van

Cognition, Brain, Behavior. An Interdisciplinary Journal

15 (2011) 359-384
 F. A. Sava, R. I. Popa 361

Aken, 2001; Asendorpf & Van Aken, 1999; Avdeyeva & Church, 2005;
Barbaranelli, 2002; Block, 1995; de Fruyt, Mervielde, & van Leeuwen, 2002;
Herzberg & Roth, 2006, Pulkkinen, 1996; Robins, John, Caspi, Moffitt, &
Stouthamer-Loeber, 1996). Almost all these attempts refer back to Block and
Block’s (1980) personality typology.
The best-known classification of personality types, based on the Big-Five
model, identifies only three gender-indifferent personality types (Asendorpf & Van
Aken, 1999; de Fruyt, Mervielde, & van Leeuwen, 2002, Robins et al., 1996). This
typology differentiates between personality types described as resilient,
overcontrolled, and undercontrolled. This classification is based on Block and
Block’s proposal (1980), which focused, in one dimension, on Ego resiliency (the
degree of an individual’s flexibility and adaptation towards the demands of the
environment) and in another on Ego control (the intensity with which impulses and
wishes are expressed – measured along an axis that has self-discipline at one end
and impulsivity / explosive behavior at the other).
The specific typology for the resilient personality is indicated by scores
above the average for all the Big-Five dimensions, except for neuroticism - a
dimension in which resilient people obtain low scores. The typology for the
overcontrolled personality typically involves below average scores for extraversion
combined with above average ones for the traits of neuroticism and
conscientiousness, the other two trait dimensions being less relevant for describing
this type. Finally, the profile of the undercontrolled personality type is usually
characterized by low agreeableness and conscientiousness in combination with high
levels of neuroticism and extraversion.
However, this typology suffers from shortcomings, concerning
(a) differences between studies in the number of identified clusters,
(b) inappropriate labeling of clusters, (c) differences in cluster-analytic methods,
and (d) disregard for possible cultural differences. We will address these four issues
in turn.
Although the most commonly proffered solutions suggest a three-cluster
typology, Pulkkinen (1996) describes seven personality types (with twelve
personality subtypes) while a recent study using a nationally representative German
sample suggests a five-cluster solution (Herzberg & Roth, 2006).
Even those studies that provide a three-cluster solution seem to suffer from
a labeling problem. One conclusion of Herzberg and Roth (2006), in their review of
the literature, is that in many instances prototypes with differently detailed profiles
have been assigned the same names as used in previously described work,
apparently in order to align the new findings to a three-cluster solution. For
instance, the undercontrolled type as described by Asendorpf et al. (2001) had a
high level of neuroticism, a very high level of extraversion and an average level of
agreeableness, whilst the same type label was used by de Fruyt, Mervielde, and van
Leeuwen (2002) for a type having an average level of neuroticism and lower levels

Cognition, Brain, Behavior. An Interdisciplinary Journal

15 (2011) 359-384
362 F. A. Sava, R. I. Popa

of extraversion and agreeableness. The significance of these disparities is all the

greater given that the trait measures were obtained in both studies using the same
instrument, the NEO PI-R, (the most widely-used inventory for assessing the five-
factor model of personality).
Another interesting finding is reported by Rammstedt, Riemann,
Angleitner, and Borkenau (2004), who found that a three-cluster solution with
clearly identified profiles in terms of resilient, undercontrolled and overcontrolled
types was found only when self-report measures were used, and that personality
types which emerge from the analysis of data depend strongly on personality
measures and informants.
Some authors, such as Costa et al. (2002), argue that the cluster-analytic
methods affect the results. For instance, cluster analysis is sensitive to sample size
and composition. Most of the findings that support a three-cluster solution have
been based on samples which were rather small (< 500) and of relatively
homogeneous composition (e.g., college students, rather than a representative
population sample). In fact the only published study based on a national
representative sample was conducted in Germany (Herzberg & Roth, 2006) and its
results suggest that a five-cluster solution is a better fit than a three-cluster solution.
Another technical issue was raised by Breckenridge (2000), who demonstrated that
selecting the number of clusters on the basis of Cohen’s kappa replication index, as
suggested by Asendorpf et al. (2001), introduces a strong bias towards choosing
fewer clusters. A solution to these limits would be the use of a more liberal kappa
value. Although Asendorpf et al. (2001) recommended .60 as a cut off value to
justify a cluster solution, such a value seems to be rather technically conservative,
and therefore we a priori chose a cut off value of .50 for mainly two reasons:
(a) kappa coefficients are highly sensitive to sample composition and tend to get
lower values for more heterogeneous samples (as in the case of our national
representative sample) (Boehm, Asendorpf, & Avia, 2002); (b) kappa indices are
biased towards choosing fewer clusters, as (Breckenridge, 2000) demonstrated, thus
leading to an underestimate of the correct number of clusters if conservative cut off
values are used.
Last, but not the least, the cultural argument discussed by Avdeyeva and
Church (2005) addresses the cross-cultural generalizability of such findings. The
authors note that none of the previously published studies had been conducted in a
non-western culture, and more seriously that the instruments used in previous
studies had been developed originally in American English which may bias the
results. Moreover, the same authors have conducted the only published study on a
non-western culture and found only partial support for the usual three-cluster
solution, in that the profiles for the undercontrolled and overcontrolled types
differed from the previously found profiles of these types.

Cognition, Brain, Behavior. An Interdisciplinary Journal

15 (2011) 359-384
 F. A. Sava, R. I. Popa 363

OBJECTIVES AND RESEARCH DESIGN

In our first study, we focused on identifying the most appropriate cluster number
starting from Block’s (1980) typology, as well as establishing a proper label for
each cluster in a Romanian sample.
As Asendorpf (2001) affirmed, the consistency of personality types across
different studies is far from being perfect, mainly because different studies may
differ in (i) language, (ii) culture, (iii) selectivity and size of the sample,
(iv) instrument of trait assessment, (v) judge (self or others), and (vi) method of
deriving types.
In order to look for the impact of five out of six above-mentioned aspects
(we based our study entirely on self-report measures) and to overcome the previosly
mentioned limitations: (a) we conducted the analysis on a nationally representative
population sample in order to be in a position to extract the best cluster number,
especially from the point of view of replicability; (b) we used a more liberal cut off
value for Cohen's kappa for identifying the most appropriate number of clusters as
more appropriate when studying a large, heteregogenous sample; (c) we conducted
our study in a non-English-speaking country, using an indigenous instrument to
measure the Five-factor model of personality, and (d) we addressed the issue of
labeling the resulting clusters based on a more objective basis, by computing their
degree of similarity with a typical personality type (profile) as resulted from
literature review related to Block and Block’s (1980) model.
Based on these characteristics we will provide further evidences related to
the opportunity to continue considering the Block and Block’s (1980) typology a
reference framework for interpreting the personality types derived from a five-
factor model of personality.
Finally, considering arguments by Herzberg and Roth (2006) and by Hair,
Anderson, Tatham, and Black (1998), we did not only study the replicability of
cluster solutions, but also checked in a second study how well cluster membership
of participants predicted specific behaviors. This would provide evidence that the
different uncovered types are predictive of real life differences between types.

STUDY 1

METHOD

Participants
The randomly selected sample of 1,073 participants was representative of the
Romanian population in terms of distribution of gender, age, ethnographically
significant regions, and milieu (urban vs. rural). However the questionnaire was
administered only to a population of participants between 16 and 60 years of age

Cognition, Brain, Behavior. An Interdisciplinary Journal

15 (2011) 359-384
364 F. A. Sava, R. I. Popa

and who had at least ten years of formal education. The percentages obtained for
each sample category were: 49.3% (urban) and 50.7% (rural); 26.5% (16-25 years),
34.2% (26-40 years), and 39.3% (41-60 years). The gender distribution was 50.1%
males and 49.9% females. The wider context of the research had as its main
objective the establishing of national representative norms for the DECAS
Personality Inventory through the use of a stratified multi-stage probability design
as a sampling procedure. Thus, in the first stage explicit stratification was employed
by using several strata variables that are usually taken into account when running
national surveys (e.g., geographical region, the milieu, the population size for the
cities etc.). This measure allows a random selection of independent samples for
each stratum. During the second stage, a simple random sample of localities was
drawn in order to cover all combination of strata (e.g., cities with less than 50,000
inhabitants from Transylvania region). The next stage involved a systematic
probabilistic sample procedure to select households from the selected locations
(e.g., every fifth household), by using multiple randomly chosen starting points to
cover both center and the suburbs of the selected locations. Finally, the last step
used gender and age as strata, in order to reproduce the proportions of these
demographics in accordance with data provided by the last national census. In this
last step, one person per selected household was asked to respond, based on a
simple probabilistic procedure (e.g., interviewing the person who has celebrated his
or her birthday more recently). The design of the sampling procedure and the
implementation of the field survey (collection of data) were subcontracted by two
specialized survey companies.

Instruments
All participants in the study were tested using the DECAS Personality Inventory
and a questionnaire addressing demographic characteristics. DECAS (Sava, 2008)
is a 95 item test which solicits dichotomic (“true” / “false”) answers to items
equally distributed among five content scales: openness; extraversion,
conscientiousness, agreeableness and emotional stability (as opposed to
neuroticism) that have been developed based on the factor analysis framework . The
internal consistency of the scales within this sample ranges from .70 for
conscientiousness to .75 for emotional stability, and the six week test-retest stability
coefficients range from .79 to .91, the later being obtained on a smaller convenience
sample. The questionnaire also comprises three validity scales measuring social
desirability, random responding, and acquiescence. The validity of the DECAS
scales was confirmed in a study that included, among others, the DECAS, the BFQ
(Caprara, Barbaranelli, & Borgogni, 2008), Goldberg's IPIP items (Goldberg,
1999), and the NEO-FFI (Costa & McCrae, 1992). Table 1 reports the mean and the
range of the correlations between corresponding scales in the different instruments
to measure the Big Five, the left column referring to correlations of the five DECAS
scales with the corresponding scales in the other instruments, and the right column

Cognition, Brain, Behavior. An Interdisciplinary Journal

15 (2011) 359-384
 F. A. Sava, R. I. Popa 365

referring to the correlations among the corresponding scales in the other

instruments. None of the differences between the averaged correlations in the same
row was significant. Likewise, a confirmatory factor analysis including scale scores
from three instruments which were developed based on the big five model: DECAS
(Sava, 2008), Goldberg’s IPIP (Goldberg, 1999), and BFQ-2 (Caprara, Barbaranelli,
& Borgogni, 2008) revealed that the five correlated factors solution provides a
rather good model fit, with all specific scales (e.g., extraversions scales) loading
more than .40 on their expected corresponding dimension (e.g., extraversion latent
variable). Supplementary data describing the psychometric features of the test can
be found in Sava (2008).

Table 1.
Correlations of the DECAS with other instruments (BFQ-2, IPIP, NEO FFI) and
correlations of the other instruments among each other (mean N – 163)

DECAS Scale r mean and range for correlating r mean, correlations of z **

DECAS with other scales the other instruments
among each other *
DECAS N .65 (.54 - .72) (BFQ S – NEO FFI S) .73 1.37 ns
DECAS E .67 (.59 - .81) (BFQ E – NEO FFI E) .68 0.16 ns
DECAS O .64 (.54 - .74) (IPIP O – NEO FFI O) .68 0.63 ns
DECAS A .50 (.45 - .57) (BFQ A – NEO FFI A) .56 0.75 ns
DECAS C .65 (.62 - .66) (IPIP C – NEO FFI C) .66 0.15 ns

* Correlations involving scales to measure emotional stability were inverted to prevent

negative correlations (N - neuroticism, E - extraversion, O - openness, A - agreeableness, C
- conscientiousness)
** z value testing for differences between two average correlation coefficients

Procedure
The first step consisted in eliminating invalid cases from the sample. The two
criteria for considering a case as invalid were: (1) cases having a social desirability
score over two standard deviations above the average; (2) unusual cases
(“outliers”), at a p < .00 level based on Mahalanobis distance statistics. The second
step consisted of creating standardized z scores following a procedure described in
Urbana (2004), specific to situations in which the shape of data distribution is not
normal. However, such a procedure goes against the suggestion made by Asendorpf
et al. (2001) who explicitly recommend analyzing the raw scores in order to
increase the chance of replicating a three-cluster solution. This recommendation
was challenged by Costa et al. (2002), who consider it a potential source of
methodological artifact; they also state that a cluster analysis solution based on z-
standardized data is more appropriate when calculating kappa indices.

Cognition, Brain, Behavior. An Interdisciplinary Journal

15 (2011) 359-384
366 F. A. Sava, R. I. Popa

The remaining cases (1039 participants), were randomly allocated to two

sub-samples: sample A (N = 519), and sample B (N = 520). The working algorithm
required double cross-validation (Barbaranelli, 2002; Hair, et al., 1998). In this
procedure one sub-group is used to generate the results and the other to confirm
them. The procedure is then reversed so that each sub-group is considered in turn as
the “primary” sample and the other as the “validation” sample. K-means analyses
were performed on both sub-samples as a follow-up to the application of Ward’s
algorithm. This was based on Asendorpf et al. (2001) recommendations which were
applied in most subsequent similar studies (Barbaranelli, 2002; Rammstedt et al.,
2004; Schnabel, Asendorpf, & Ostendorf, 2002), and became conventional for such
purpose (Roth & Herzberg, 2007). More detailed information about the problems
that must be considered when choosing various clustering methods is presented in
Mandara (2003).

Results
For processing the data SPSS (SPSS 11.0, 2001) was used for selecting the
optimum number of clusters.

Table 2.
Descriptive indicators for different solutions, ranging from 2 to 8 clusters

Task involved Indices 2 3 4 5 6 7 8

Cross-validation samples
Primary A sub-sample (N = 519) kappa .63 .63 .47 .50 .34 .35 .29
Primary B sub-sample (N = 520) kappa .53 .43 .47 .58 .46 .51 .43
Average kappa values (N = 1039) kappa .58 .53 .47 .54 .40 .44 .31

Table 2 presents the main findings from analyzing three sets of data: (a) the
cluster solution based on replicating the findings from sub-sample A on sub-sample
B (cross-validation); (b) the cluster solution based on replicating the findings from
sub-sample B on sub-sample A (double cross-validation); (c) the cluster solution
resulted from a k-means analysis of the entire sample (N = 1019) taking as starting
points the means of the cluster seeds for the two sub-samples.
We can notice from table 2 that only three of the solutions (two, three and
five clusters) have kappa values for the entire sample solution greater than the pre-
established level of .50, although the three-cluster solution has a wider range for
kappa values if sub-samples are taken into account (from .43 to .63).

Cognition, Brain, Behavior. An Interdisciplinary Journal

15 (2011) 359-384
 F. A. Sava, R. I. Popa 367

Figure 1.
Prototype expressed in z scores for the five identified personality types
(N - Neuroticism; E - Extraversion; O - Openness; A - Agreeableness;
C - Conscientiousness)

Therefore, when we consider the degree of stability of solutions (kappa

values) for the entire sample only three of the cluster solutions: with two, three, and
with five clusters show good internal validity. It seems, however, that less
confidence can be placed in the two-cluster than in the other two remaining
solutions, given that in a two-cluster solution the trait profile of the one type would
then be the mirror image of the trait profile of the other type, therefore an artefact.
The resulting modal values of the personality types are presented in figures 1 and 2.
As a means of further checking the appropriate labeling for the clusters in
our solution we computed two additional parameters: the Euclidean distance and
linear correlations between the z values that characterize each resulted cluster and
those of the five-cluster solution provided by Herzberg and Roth (2006), as well as
the median z values for the three-cluster solution resulting from eleven different
profiles reviewed by the same authors (e.g. for resilient people we estimated a z
score of .37 for extraversion based on the fact that this value represents the median
E score of the resilients across all 11 studies reviewed by Herzberg and Roth
(2006). Pearson correlations (r) between two profiles beyond .70 and Euclidian
distances (d) less than 1.50 (corresponding approximately to the 25th percentile)
were considered a good match.

Cognition, Brain, Behavior. An Interdisciplinary Journal

15 (2011) 359-384
368 F. A. Sava, R. I. Popa

Five-cluster solution

Undercontrolled type. The percentage of individuals included in the first cluster

was 16.7% of the participants, representing 174 cases. According to the two criteria
we used to confirm the label for this scale only one profile fitted the required
standard, namely the undercontrolled type from a three-cluster solution (r = .83,
d =.89). However, the resulting profile can also be seen to be a mixture of high
undercontrolled and low reserved - following Herzberg and Roth’s (2006) five-
cluster solution. From the Five Factor profile perspective it can be observed that the
distinguishing elements for this cluster are E+ and C-.

Strain type. The second cluster includes 17.2% of the participants, a total of 179
persons. None of the existing literature-derived profiles meet the criteria to justify a
claim of sufficiently close similarity with this cluster. The specific traits, on a Five-
Factor personality model, indicate a prototype with disharmonic features: scores
above the average for O, E, N and C, and scores below average (e.g.: z < -0.50) on
the A scale.

Resilient type. Cluster three includes 21.4% of the participants representing 222
cases. The distinctive traits of these people are found in high scores on all five
scales of the Big-Five model except for neuroticism. The label used to name this
typology is consistent with previously described profiles (both with the resilient
type in a three-cluster(r = .84, d = .86) and (r = .81, d = 1.22) with the resilient in a
five-cluster typology). However this cluster is also rather similar to the confident
(well-adjusted) type as proposed by Herzberg and Roth (2006) (r = .72, d = .89).

Overcontrolled type. Cluster four comprehends 22.3% cases, representing 232

participants. The only profile that fitted this type was the one for the overcontrolled
type rendered in a three-cluster solution (r = .74, d =.79).The overcontrolled type
obtained in a five-cluster solution (Herzberg & Roth, 2006) had the profile closest
to this cluster, although at a lower degree of similarity than the required standards
(r = .52, d = 1.77).

Passive type. The fifth cluster also accounts for 232 cases, representing 22.3%
observations. No previously reported profile seems to fit this type well. However
the closest profiles in terms of the two criteria mentioned above are the reserved
type in Herzberg and Roth’s (2006) five-cluster solution (r = .44, d = 1.10) and the
resilient type in our literature-based three-cluster solution (r = .45, d = 1.51). On the
other hand, this cluster could also be seen as an opposite type for the strain type
(r = - .97, d = 3.27 these being the highest absolute values obtained from all
possible paired comparisons). On a Five-Factor personality model, the prototype
profile indicates high scores for A and low scores for dimensions N, E, and C.

Cognition, Brain, Behavior. An Interdisciplinary Journal

15 (2011) 359-384
 F. A. Sava, R. I. Popa 369

From analysis of demographic data, we obtained significant statistical

associations between our five prototypes and education level - χ² (16) = 77.84,
p < .001, urban vs. rural milieu - χ² (4) = 35.08, p < .001, and the exact location of
urban residence - χ² (4) = 19.52, p < .001.
Analysis of standardized residuals to study the effect of education level
shows that there is a higher presence of the resilient typology among university
graduates and postgraduates, and a low presence of this type in the group comprised
of those with only 10 years of education and those who have no education beyond
the completion of vocational training school. The reverse of this situation can be
found for the overcontrolled type; this is more frequent among those who have less
than 12 years of education, by contrast with university graduates. Another
interesting aspect is that among people who have a postgraduate degree (the highest
form of education classified in our dataset) there is a relatively high frequency of
the passive type and a relatively low frequency of the undercontrolled type.
Analysis of standardized residuals shows relatively higher percentages for
the resilient and undercontrolled types in the urban milieu, as well as higher
percentages for overcontrolled and strain type in the rural milieu. A supplementary
analysis, focusing on the precise location of residence within the urban area,
indicates a higher presence of the passive type in the centre of cities and a
significant presence of the strain type in the suburbs of cities.

Figure 2.
Prototype expressed in z scores for the three identified personality types
(N - Neuroticism; E - Extraversion; O - Openness; A - Agreeableness;
C - Conscientiousness)

Cognition, Brain, Behavior. An Interdisciplinary Journal

15 (2011) 359-384
370 F. A. Sava, R. I. Popa

Three-cluster solution
Resilient type. From the profile analysis of the three-cluster solution, in the first
group we may find the prototype corresponding to the resilient in the five-cluster
classification. This cluster comprehends 29.6% of the participants (308 cases). The
specific traits of these people involve high scores in E, A, O, average scores for C,
and a low N score. This profile fits well with the literature-based profiles for a five-
cluster solution; thus for resilient and confident types r = .77, d = 1.37, and r = .78,
d = .77, respectively. This profile is also similar to the one described as the resilient
type in the three-cluster literature-based solution (r = .78, d = .81).
The psychological profile for this three-cluster type is similar to the resilient type
presented in the five-cluster solution. Seen from a five-cluster solution viewpoint
96% of five-cluster resilients, 36% of five-cluster undercontrolleds and 14% of
five-cluster passives are individuals who, on a three-cluster solution, would be
categorized as three-cluster resilients.
Strain type. The second cluster invites direct comparison with the strain type in the
five-cluster classification, which included 332 participants (32.0% of those
questioned). In this segment of the population we can observe that the prototype
shows scores above the average for E, C and N, and below average for A. The
psychological description is similar to that presented in the five-cluster solution for
the strain type. None of the previous literature-based profiles satisfied the statistical
criteria to justify a potential overlap with this cluster. From a structural perspective
this cluster contains over 99% of the strain individuals, 41% of the undercontrolled
and 32% of the overcontrolled type as identified in the five-cluster solution.
Passive-overcontrolled type. This cluster contains those individuals identified as of
passive type falling within the overcontrolled category. This type comprehends
39.4% of the participants, totaling 399 cases. From the Five-Factor personality
model perspective, this prototype shows scores below average for E and N, and
slightly below average for O and C. The significance of this cluster can be better
deduced from the composition of the group tested. On a five-cluster solution
analysis, as described above, this heterogeneous type includes approximately 86%
of the five-cluster passive types, and almost 67% of the members of the
overcontrolled type, along with 23% of the undercontrolled type. None of the
previous literature-based profiles fitted the standards to justify a potential overlap
with this cluster.
An analysis of demographic data allows us to reach similar conclusions
concerning the association of the three personality types with: (1) education level -
χ² (8) = 45.17, p < .00; (2) rural vs. urban milieu - χ² (2) = 22.77, p < .00; and
(3) exact location of residence - χ² (2) = 15.89, p < .00.
We can observe a higher incidence of the resilient type among university
graduates and a lower frequency of this type in the section of the sample with less

Cognition, Brain, Behavior. An Interdisciplinary Journal

15 (2011) 359-384
 F. A. Sava, R. I. Popa 371

than a high school level of education. A higher percentage of the resilient type is to
be found in the urban milieu and a higher percentage of the passive-overcontrolled
and strain types in the rural milieu. As for the more precise location of residence in
the urban area, the resilient type can be found more often in the central areas of
cities, while the strain type appears more often in the suburbs.
Finally, in order to obtain a better description of the specific characteristics
of each type discussed in this study and to see how well does the Block and Block’s
(1980) model can provide a meaningful interpretation for the resulted clusters, we
mapped various personality types using a multidimensional scaling approach. More
specifically, we have plotted on the same chart types derived from four different
sources: (1) our five-cluster and three-cluster solutions (Ro5 and Ro3); (2) Herzberg
and Roth’s five-cluster solution obtained using a German, nationally representative
sample (Ger5); (3) Avdeyeva and Church’s first study (2005) that used NEO PI-R
and provided a three-cluster solution to data from a non-western, Philippine sample
(Phil3); (4) a literature based, three-cluster solution (Std3). This last was obtained
by aggregating the findings from seven studies, including 11 different analyses
involving a total of more than 5000 people, conducted in a variety of western
countries by various researchers. The aggregated results were derived by averaging
data presented in Herzberg and Roth’s (2006) Table 1.

Figure 3.
A multidimensional scaling map containing several proposed prototypes

Cognition, Brain, Behavior. An Interdisciplinary Journal

15 (2011) 359-384
372 F. A. Sava, R. I. Popa

A visual inspection of how the various personality types plot on the map
suggests two different possible classification systems: (1) one (inside arrows)
reliant on a classical perspective that takes a U-shape perspective on personality
types (Asendorpf et al., 2001), considering the levels of resilience (flexibility)
(resilient type at the top) and of control (undrecontrolled at the bottom left and
overcontrolled at the bottom right) as the main axes along which to categorize
various personality types, in accordance with Block and Block’s (1980) theory;
(2) one that adopts a complementary perspective which takes into account the
degree of behavioral internalization (externalization) and the degree to which
healthy psychological resourcefulness is available to allow for adaptive coping (for
a discussion that links mental disorders to internalization, externalization, and
personality, see Krueger 1999 or Krueger, McGue, & Iacono, 2001).

DISCUSSION

The objective of the first study was to identify the number of clusters and assign to
them the most appropriate labels by which the Romanian adult population might be
segmented, based on a typology derived in relation to the five-factor personality
model that take into account the Block and Block’s (1980) typology model. The
results of the cluster analysis indicate two possible solutions: one with three and the
other with five clusters. Both solutions have an acceptable level of stability,
showing acceptable replicability under a double cross-validation procedure.
The five-cluster solution based on the five-factor personality replicates a
series of personality types already well known in the research literature. This
solution combines the five-factor model of personality with Block’s typology: the
undercontrolled type (with insufficient control of the impulses), the overcontrolled
type (with excessive control of impulses) and the resilient type. However two
additional types can be characterized, inviting further refinement in their
definitions. For the present we have labeled them the strain type [comparable to
Friedman’s (1996) A type and to Block and Block’s (1980) brittle undercontrolled
type] and the passive type [somewhat similar to Friedman’s (1996) B type and to
Block and Block’s (1980) resilient overcontrolled type]. The strain type was also
found as more prone to possess irrational beliefs and maladaptive schemas that
might trigger psychopathological problems (Sava, 2009). Another interesting
finding is that these two newly distinguished clusters seem to be antagonistic. In
addition there is no correspondence between our strain type and the remaining two
types (confident and reserved) found in the German representative sample, but there
is some correspondence between the passive types described in our study and
Herzberg and Roth’s reserved type (compared with their reserved type our passive
type people are more reserved, less active, and relatively confident in their actions).
When comparing the two possible solutions provided in this study, it may
be observed that our three-cluster solution reproduces two of our five clusters,

Cognition, Brain, Behavior. An Interdisciplinary Journal

15 (2011) 359-384
 F. A. Sava, R. I. Popa 373

specifically the resilient and the strain types. The third cluster, which has been
labeled the passive-overcontrolled type, seems in fact to be formed mainly as a
combination of two clusters (the two independent types being the passive and the
overcontrolled). Finally we should note that the undercontrolled type in a five-
cluster solution cannot be simply mapped onto a three-cluster solution; membership
of this type becomes partitioned in an almost equal way between the three types of
the three-cluster solution when a reassignment is made. Even though the three-
cluster solution seems to be the most favored solution for investigators using the
Five-Factor model (Herzberg & Roth, 2006), we can observe a partial
correspondence between our three-cluster types and the classical typology of Block
(1995). Based on the results from this study it would be a fallacy of nomenclature to
label, for instance, the passive-overcontrolled type as overcontrolled on the basis of
its specific profile characteristics. So, while the resilient type is practically identical
as a psychological profile with that seen in other studies and the strain type seems to
be a particular subtype of the theoretical undercontrolled type, the passive-
overcontrolled type is not similar to clusters described in other research, even
though certain similarities and correspondences do exist.
From such a graphical representation as the one provided in Figure 3 we
can appreciate the richness and the diversity of types assigned to the same class. For
instance, we are able to notice at least three different undercontrolled subtypes: the
strain type (N+, E+, A-), closer to the choleric temperament or to borderline
personality disorder (self-regulating failure due to negative affect), the prototypical
undercontrolled type (E+, C-), closer to the demonstrative, impulsive-hedonistic
type and to histrionic personality disorder (self-regulating failure due to excessive
reward seeking, immediate positive stimuli), and the undercontrolled type obtained
in the German sample (N+, A-, C-), more similar to the antisocial personality (self-
regulating failure due to inconsistent aims and lack of perseverance). It is worth
noting that all three of these personality disorders belong to the B cluster according
to the DSM IV classification (Widiger & Frances, 2002). Additionally, it can be
seen that the three undercontrolled subtypes are plotted rather closed to each other,
which implies they have a more similar profile than other types / subtypes of
personality. This chart could also provide a clue that differences in the profile of
various prototypes which were given an identical name actually represent various
subtypes of a theoretical type, which have been obtained solely on statistical basis
due to the characteristics of the samples which were analyzed. But most important,
the chart suggest that we are dealing with fuzzy types and not with discrete types,
due to several possible subtypes that might result due to variance in sample
characteristics of various studies.
Unfortunately, a limitation of this analysis is that no independent measures
of resiliency and control were included in the study; further research would
therefore be needed to validate such a taxonomy that suggests that personality types
based on the five-factor model of personality could be classified on their self-

Cognition, Brain, Behavior. An Interdisciplinary Journal

15 (2011) 359-384
374 F. A. Sava, R. I. Popa

regulation capabilities (degree of resilience and degree of control), as Gramzow et

al. (2004) have suggested.

STUDY 2

The high level of heterogeneity found in the third cluster, the passive-
overcontrolled type, might lead us to expect that to discard the five-cluster solution
in favour of the three-cluster solution would result in a loss of valuable information
about the behavior of the adult members of a specific type.
In order to verify in what manner a five-cluster solution may furnish
additional information of practical relevance we carried out a second study, a non-
experimental one, to test whether specific associations exist between the various
identified types and certain self-reported behavioral preferences. This procedure
was aimed at external validation of the cluster solutions in the first study. For
instance, Caspi et al. (1997) found that undercontrolled people tend to engage more
frequently in health-risky behaviors. Similarly, Atkins and Hart (2008) looking at
data from a longitudinal project found that those who were identified as
undercontrolled at 5 or 6 years of age were more likely than those resilient or over-
controlled to have sexual intercourse before the age of 16 and they were more likely
to report they had hurt someone seriously at least once in the past year at the age of
12 (Atkins, 2007). Moreover, a significant inverse association was found between
likelihood of childhood unintentional injury and the overcontrolled prototype score
(Atkins and Matsuba, 2008), while children identified as resilients at 5 or 6 years of
age were more likely than children characterized by the overcontrolled and
undercontrolled personality types to volunteer 8 and 10 years later in adolescence
(Atkins, Hart, & Donnelly, 2005). Finally, Vollrath, and Torgersen (2008) showed
that types combining high extraversion and low constraint (a personality profile that
is similar to the one for the undercontrolled type in our five cluster solution),
engage in more risky health behaviors such as tobacco cunsumption or a higher
number of new sexual parteners. So, unless a shift of the analysis from the three-
cluster solution to the five-cluster solution can be found to offer additional
information related to the self-reported behaviors of the people surveyed we should
consider the more parsimonious solution, the three-cluster one, to be better. On the
other hand, if segmentation of the population into five psychological types could be
shown to have a superior predictive value then that would provide support for the
view that the five-cluster solution is to be preferred.

Cognition, Brain, Behavior. An Interdisciplinary Journal

15 (2011) 359-384
 F. A. Sava, R. I. Popa 375

METHOD

Participants
Data used in the second study was collected from 150 participants in the course of a
routine annual psychological examination that was obligatory for compliance with
work safety legislation. After application of the screening procedures described in
study one, 133 valid observations remained for analysis. The resulting sample
comprised 94 males and 39 females, with ages between 18 and 59 (M = 41,
SD = 11.4), classified according to three levels of education: those who had not
completed high school (46%), those who had completed high school (45%), and
those with university degrees (9%).

Instruments
Two sets of collected data are relevant: the DECAS Personality Inventory and the
Bio-data. This latter refers to responses to seven questions examining self-reported
behavioral aspects: (1) general state of health (alcohol and tobacco consumption,
perception of state of health) and (2) childhood upbringing and general lifestyle (the
strictness with which they were raised in childhood, the regularity of their practice
of sport; exercise of voting rights; frequency of change of employment).
The bio-data was measured either nominally or ordinally (with only a few
categories). For instance, in relation to smoking habits, the alternative answers to
the question: “How many cigarettes do you usually smoke per day?” were a) none, I
do not smoke b) up to 10 cigarettes per day; c) more than 10 cigarettes per day.

Procedure
The instruments were applied in groups of 15 to 25 participants. The remaining 133
valid cases were cluster analyzed directly by a k-means nonhierarchical method,
using as cluster seeds the values found in our first study that had been conducted on
a national representative sample.

Results
First, we checked whether the resulting clusters in the second study were similar to
those obtained in the first study. We found support for the existence of a similarity
between the profiles obtained in the first study and those observed in the second
(e.g. the similarity between the strain type in the five-cluster solution obtained in
the first study and the strain type for the five-cluster solution in the second study
was .90 in terms of Euclidian distance dissimilarity and .83 in terms of profile
similarity correlation). For all cases, the similarity between a specific type obtained
in the first study and the corresponding type obtained in the second study met the
acceptance criteria [Pearson correlations (r) between two profiles beyond .70 and
Euclidian distances (d) less than 1.50)], a result showing a good replicability of the
types.

Cognition, Brain, Behavior. An Interdisciplinary Journal

15 (2011) 359-384
376 F. A. Sava, R. I. Popa

In order to see which of the two competing solutions is better in terms of

external validity, in predicting various behaviors and attitudes, we used a non-
parametric statistical method (χ² test of homogeneity), with a statistical power to
detect small to average effect sizes for 133 participants (e.g., w = .20), ranging from
31 to 53, in most cases taking values between .40 and .50. In order to increase the
statistical power of the analysis, we also took into consideration marginal results,
having a p < .10.

Table 3.
Comparison of health, psychological and social status for various prototypes (values refer to
percentages)

5 clusters Resilient Under- Over- Passive Strain χ² df V Residual

controlled controlled analysis
Perceived health problems 9.1 15.0 36.0 8.6 25.7 8.75† 4 .25 O > P, R
Tobacco consumption 54.5 70.0 38.5 33.3 51.4 8.24† 4 .24 U > O,P
Alcohol consumption 18.2 20.0 23.1 16.7 27.8 1.46 4 .10 --
Regular (weekly) sport 41.7 23.8 22.2 24.3 33.3 2.50 4 .13 --
Strict upbringing 0.0 0.0 18.5 8.1 30.6 14.57** 4 .33 S > R, U
Exercise of voting rights 100.0 75.0 44.4 83.8 63.9 17.91** 4 .37 R,P > O
Number of past jobs (> 3) 16.7 42.1 11.1 8.3 11.1 15.77* 8 .24 U > R,P,O,S
Higher education degree 25.0 4.8 0.0 13.5 5.6 13.95† 8 .23 R > U,O,S
3 clusters Resilient Passive- Strain χ² df V Residual
overcontrolled analysis
Perceived health problems 12.5 22.5 22.2 1.45 2 .10 --
Tobacco consumption 45.5 34.1 57.4 5.10† 2 .19 S > PO
Alcohol consumption 15.2 19.5 27.3 1.95 2 .12 --
Regular (weekly) sport 33.3 19.5 30.4 2.13 2 .12 --
Strict upbringing 0.0 14.6 23.2 9.42** 2 .27 S>R
Exercise of voting rights 77.1 75.6 62.5 2.98 2 .15 --
Number of past jobs (> 3) 27.3 9.8 12.5 6.37 4 .15 --
Higher education degree 13.9 9.8 3.6 3.57 4 .11 --

** for ps < .01, * for ps < .05, and † for ps < .10
R – Resilient; S – Strained; PO – Passive-Overcontrolled; U – Undercontrolled;
O – Overcontrolled; P – Passive
For illustrative purposes, the Cramer’s V values indicating various intensities for effect sizes
are: V = .10 (small effect); V = .30 (medium effect); and V = .50 (strong effect) (Ferguson,
2009; Sava and Maricutoiu, 2007).

Cognition, Brain, Behavior. An Interdisciplinary Journal

15 (2011) 359-384
 F. A. Sava, R. I. Popa 377

DISCUSSION

As shown in Table 3, the five-cluster prototype is a solution that allows more

discrimination between self-reported behaviors than does the three-cluster solution.
For example there is a higher frequency of smokers among the undercontrollers
(45% smoke at least 10 cigarettes a day) than among other categories, particularly
the passive type. The undercontrollers also tend to change employment more often
than any other personality type, while the overcontrollers are more likely to
perceive themselves as subject to frequent health problems than are the resilient and
the passive types.
The strain category includes more people who agreed they were strictly
raised as children, particularly when compared to the resilient and the
undercontrolled types. Likewise, the resilient cluster contains a higher proportion
who exercised their vote during the most recent elections as well as a higher
percentage of people educated to university degree level.
The objective of the second study was to achieve external validation of the
personality types identified in the first study. The data we obtained support the five-
cluster, over against the three-cluster, solution. The five-cluster solution can be seen
to provide better differentiation between the participants’ behavior styles. For
example, considering employment stability, only the five-cluster solution
successfully highlights the fact that it is the undercontrolled type which manifests
the highest degree of job instability.
It remains an admitted limitation of the results obtained in the second study
that both the classification of participants into personality type and the criteria for
classifying behavioral differences were based on self-report data.

GENERAL DISCUSSION

The purpose of the research reported in this paper is to validate, internally and
externally, a typology of personality based on the dimensions of the five-factor
model, measured in a large sample representative of the Romanian population. The
results show that five types can be identified and this solution has been shown to
have both external and internal validity.
The first three discussed types (resilient, overcontrolled and
undercontrolled types) are consistent findings of research studies of personality
typology based on the Big-Five model (Asendorpf & Van Aken, 1999; Barbaranelli,
2002; Block, 1995; de Fruyt, Mervielde, & van Leeuwen, 2002; Herzberg & Roth,
2006). However the only two studies conducted using nationally representative
samples (the present study and Herzberg & Roth’s 2006 study) converged on the
view that a five-cluster solution may possibly furnish a better typology based on the
five-factor personality model. One possible explanation for this is that when using

Cognition, Brain, Behavior. An Interdisciplinary Journal

15 (2011) 359-384
378 F. A. Sava, R. I. Popa

national representative samples which are more heterogeneous more clusters can
emerge, reflecting the higher degree of type heterogeneity.
Fruyt, Mervielde, and van Leeuwen (2002) have argued, research in this
area should aim to go beyond simple replication of the three major personality types
described by Block (1995) and should try to identify new personality types, beyond
those already described. Hertzberg and Roth (2006) were among the first authors to
propose an extension of personality prototypes beyond the big-three types to
include confident and reserved types also. In the present study we obtained two
additional antagonistic types: the strain type and the passive type, where the later
has some correspondence with the reserved type described by Herzberg and Roth
(2006) and to a resilient subtype which was entitled - well adapted - found by
Boehm, Asendorpf, and Avia (2002).
Based on psychological similarities between various profiles, as presented
in figure 3, we may tentatively reply to a question raised by Avdeyeva and Church
(2005) that is, whether resiliency and control are actually orthogonal dimensions as
suggested by Block and Block (1980) or whether the relationship between resilience
and control is in fact quadratic (e.g., optimal resiliency is associated with an optimal
level of control) as has been argued by Asendorpf et al. (2001). It would appear that
all types based on a five-factor personality model can be delineated by combining
various degrees of control and resilience, treated as orthogonal dimensions. For
instance, the reserved, the passive and the passive-overcontrolled types can be seen
as exhibiting at different degrees of intensity the type described by Block and Block
(1980) as resilient overcontrolled.
However we should also point out the real difficulties that exist in
replicating any standard typology of personality that is based solely on cluster
analysis. As Costa et al. (2002) have noted, the same familiar cluster names have
been used in presentation of the results from various studies so as to conform to the
nomenclature in reports of previous findings even though the Five-Factor
personality profile pattern was not similar for the specific types being so labeled. A
possible solution to this limitation is proposed by Herzberg and Roth (2006). They
suggested that the cluster analysis should only be applied to representative
population samples, as the initial step to extract the appropriate number of clusters
in a given country. Then the next step would be to assign the participants to a
specific type on the sole basis of population parameters (rather than on the basis of
any new sample), using discriminant functions as a statistical method rather than
cluster analysis. This approach is intended enhance cluster replicability and reduce
the possibility of labeling error. However, regardless the cluster solution found in a
study, it seems that all resulted personality types can be interpreted as a specific
combination of the two self-regulation processes: the degree of resilience and the
degree of control. In our case, the strain type shares some characteristics of the
undercontrolled type, but the former has even a lower degree of resilience.

Cognition, Brain, Behavior. An Interdisciplinary Journal

15 (2011) 359-384
 F. A. Sava, R. I. Popa 379

Similarly, the passive type shares many characteristics of the resilient prototype, but
is less assertive and active.
A further point needs to be made concerning these two new clusters,
namely the strain type and the passive type. These types seem to be novel, with no
previous findings fitting the personality profiles associated with them. We are
therefore left with a need to consider whether our finding of two novel types might
be accounted for due to features peculiar to this study. The two features concerned
are: the indigenous (DECAS) personality measure used and the specific Romanian
cultural context. Since the DECAS personality inventory has been previously
validated as an adequate measure for use in investigating the standard five-factor
personality model, it is more likely to be the cultural context that is the source of the
two previously unremarked clusters. In this regard, Adamesteanu (2008) affirmed
that there are two types of Romania (the urban – forward-looking and culturally
orientated to the West, which is modernized, and the rural – which is less
developed, and more traditional and archaic). Post hoc analysis has revealed that a
significantly higher percentage of people categorized as strain type live in a rural
milieu (61%) and a similar tendency is seen in regard to the passive type. A
contrasting situation has been found for the more commonly recognized clusters, in
particular for the resilient type and for the undercontrolled type, of whom
respectively 59% and 60% live in an urban milieu. Based on the above figures, it
may be the case that these two novel types, which go beyond the Big-Three
typology, manifest as a consequence of sensitivity of the prototype patterns to the
cultural context.
Another limitation is related to our decision to conduct our analysis without
taking into account the potential implications of gender differences that some
previous findings (e.g., Pulkkinen, 1996) suggest may be significant. We based our
decision on technical grounds, being primarily interested in ensuring an appropriate
sample size to overcome most of the technical limitations that would have been
involved had we run a cluster analysis on a smaller sample.
A possible additional limitation of the study is the decision to use an
alternative instrument for the Big Five, one that may underestimate the type
equivalence across cultures. However, when we compare our results with those
obtained in the German population (Herzberg & Roth, 2006), we can see that we
have replicated a five-cluster solution rather well. This resulted even though we
employed a different means of operationalizing the five factor personality model as
a strategy to increase the construct validity of these findings, in particular the
finding that a five-cluster solution might be a better solution than a three-cluster
solution when using nationally representative samples.
Another limitation is the decision to extract the clusters solely on the five
global scores which affects the reliability of the results (Mandara, 2003). The use of
another instrument such as the NEO PI-R (Costa & McCrae, 1992), which provides
scores for facets in addition to the five global dimensions, might have offered a

Cognition, Brain, Behavior. An Interdisciplinary Journal

15 (2011) 359-384
380 F. A. Sava, R. I. Popa

better way to develop reliable clusters. However, in this research we found a very
good replication of the five clusters obtained on a Romanian representative sample
in the second study that was done on an independent sample consisting of 150
adults. This finding supports the suggestion made by Herzberg and Roth (2006) that
in order to get replicable clusters in subsequent studies based on convenient
samples, it is recommended to use as cluster seeds those values that represent
population parameters resulted from a probabilistic representative sample.
In a similar vein, this paper provided limited opportunities for discussing
the personality types derived from the Big Five model through the lenses of other
theoretical approaches [see for instance Cloninger (1994); Thomas, Chess, & Birch
(1970)]. For instance, the concept of novelty seeking proposed by Cloninger (1994)
can be theoretically linked to the undercontrolled type, while harm avoidance seems
to be a less developed trait among people perceived as resilient. Future studies
should empirically explore the link between the five-factor model derived typology
and other relevant theories regarding personality and temperament.
Despite these limitations, using personality types does have its utility in
that the convenient labels used to summarize combinations of traits make sense to
many practitioners as well as to naïve specialists working in human resources and in
clinical contexts, even if such labels might have less predictive power than an
approach using traits (Costa et al., 2002; Roth & von Collani, 2007). In an
analogous way the four personality types proposed by Gough (1987, as cited in
Pitariu et al., 2006) do not have as much predictive power as the California
Personality Inventory’s twenty folk scales on which those types are based.
However, due to their practical utility, they have been widely accepted within the
community of practitioners and are now routinely included in electronically
generated reports. In short, despite the existence of some technical limitations in
their use, personality types have proved user-friendly for the average consumer of
psychological testing without surrendering too much of the predictive power which
is to be found in a dimensional approach. Based on these practical considerations,
we found it useful to look at different personality types on the basis of a five-factor
personality model. The explanatory complexity and ecological validation of such an
approach brings additional benefits, allowing us to move beyond a classical
perspective that remains focused on the one-dimensional treatment of personality.
To summarize our main findings, we can conclude that a choice of five,
rather than three, personality prototypes is more appropriate when seeking a
personality typology based on the five-factor personality model. This finding is
particularly relevant for studies where representative samples are used rather than
for studies based on convenient medium- or small-sized samples. It is also the case
that the five-cluster solution provides a better fit when care is taken to avoid the
bias towards a smaller cluster number solution that some technical decisions can
otherwise generate.

Cognition, Brain, Behavior. An Interdisciplinary Journal

15 (2011) 359-384
 F. A. Sava, R. I. Popa 381

Beyond the question of how many prototypes it is better to extract, there is

an additional issue related to labeling such types. It is not easy to make direct
comparison of the findings of reported studies due to the variety of their cultural
contexts, sampling procedures and sizes, instruments and types of research (e.g.,
self-reports). To this picture we must add concerns that have been voiced about the
consistency of labeling of types. We may therefore feel that it is time to challenge
the idea that there are just three universal types with well-established five factor
personality profiles. Yet, as Block and Block’s (1980) suggested, all prototypes
seem to combine, with various degrees of intensity, two important axes: the degree
of resilience / flexibility and the degree of control.

ACKNOWLEDGMENTS

This research was supported by CNCSIS – UEFISCSU, project number

PNII - IDEI 1076/2009. This organization had no role in the design and implementation of
the study.

REFERENCES

Asendorpf, J. B. (2001). Editorial: The Puzzle of Personality Types. European Journal of

Personality, 16, S1-S5.
Asendorpf, J. B., Borkenau, P., Ostendorf, F., & van Aken, M. A. G. (2001). Carving
personality description at its joints: Confirmation of three replicable personality
prototypes for both children and adults. European Journal of Personality, 15, 169–
198.
Asendorpf, J. B., & Denissen, J. J. A. (2006). Predictive validity of personality types versus
personality dimensions from early childhood to adulthood: Implications for the
distinction between core and surface traits. Merrill-Palmer Quarterly Journal of
Developmental Psychology, 52, 486–513.
Asendorpf, J. B., & Van Aken, M. A. G. (1999). Resilient, overcontrolled, and
undercontrolled personality types in childhood: Replicability, predictive power and
the trait-type issue. Journal of Personality and Social Psychology, 77, 815–832.
Atkins, R. (2007). The association of personality type in childhood with violence in
adolescence. Research in Nursing and Health, 30(3), 308-319.
Atkins, R., & Hart, D. (2008). The under-controlled do it first: childhood personality and
sexual debut. Research in Nursing and Health, 31(6), 626-639.
Atkins, R., Hart, D., & Donnelly, T.M. (2005). The association of personality type in
childhood with volunteering during the adolescence. Merrill-Palmer Quarterly
Journal of Developmental Psychology, 51(2), 145-162.
Atkins, R., & Matsuba, M.K. (2008). The association of personality and the likelihood of
serious unintentional injury during childhood. Journal of Pedriatic Nursing, 23(6),
451-459.

Cognition, Brain, Behavior. An Interdisciplinary Journal

15 (2011) 359-384
382 F. A. Sava, R. I. Popa

Avdeyeva, T. V., & Church, A. T. (2005). The cross-cultural generalizability of personality

types: a Philippine study. European Journal of Personality, 19, 475-499.
Barbaranelli, C. (2002). Evaluating cluster analysis solutions: An application to the Italian
NEO Personality Inventory. European Journal of Personality, 16, 43-55.
Block, J. (1995). A contrarian view of the five-factor approach to personality description.
Psychological Bulletin, 117, 187-215.
Block, J. H., & Block, J. (1980). The role of ego-control and ego-resiliency in the
organization of behavior. In W. A. Collins (Ed.), Minnesota Symposia on Child
Psychology (Vol. 13, pp. 39-101). Hillsdale, NJ: Erlbaum.
Boehm, B., Asendorpf, J. B., & Avia, M. D. (2002). Replicable types and subtypes of
personality: Spanish NEO-PI samples. European Journal of Personality, 16, 25–
41.
Breckenridge, J. N. (2000). Validating cluster analysis: Consistent replication and symmetry.
Multivariate Behavioral Research, 35, 261–285.
Briggs Myers, I., McCaulley, M. H., Quenk, N. L., & Hammer, A. L. (1998). MBTI Manual.
A Guide to the Development and Use of the Myers-Briggs Type Indicator (3rd.).
Mountain View, CA: Consulting Psychologists Press.
Caprara, G. V., Barbaranelli, C., & Borgogni, L. (2008). Chestionarul Big Five [Big Five
Questionnaire – BFQ 2]. Cluj-Napoca: Sinapsis.
Caspi, A., Harrington, H. L., Moffitt, T. E., Begg, D., Dickson, N., Langley, J., & Silva, P.
A. (1997). Personality differences predict health-risk behaviors in young
adulthood: evidence from a longitudinal study. Journal of Personality and Social
Psychology, 73(5), 1052-1063.
Cloninger, C. R. (1994). The genetic structure of personality and learning: a phylogenetic
perspective. Clinical Genetics, 46, 124-137.
Costa, P. T. Jr., Herbst, J. H., McCrae, R. R., Samuels, J., & Ozer, D .J. (2002) The
replicability and utility of three personality types. European Journal of Personality,
16, 73-87.
Costa, P. T. Jr., & McCrae, R. R. (1992). Revised NEO Personality Inventory (NEO PI-R)
and NEO Five-Factor Inventory (NEO PI-FFI). Odessa, Fl, USA: Psychological
Assessment Resources.
De Fruyt, F., Mervielde, I., & van Leeuwen, K. (2002). The consistency of personality type
classification across samples and the five-factor measures. European Journal of
Personality, 16, 57-72.
Eysenck, H. J. (1967). The biological basis of personality. Brunswick, NJ: Charles C.
Thomas Publisher.
Ferguson, C. J. (2009). An effect size primer: A guide for clinicians and researchers.
Professional Psychology: Research and Practice, 40(5), 532-538.
Friedman, M. (1996). Type A Behavior: Its Diagnosis and Treatment. New York: Plenum
Press (Kluwer Academic Press).
Goldberg, L. R. (1999). A broad-bandwidth, public domain, personality inventory measuring
the lower-level facets of several five-factor models. In I. Mervielde, I. Deary, F. De
Fruyt, & F. Ostendorf (Eds.), Personality Psychology in Europe, Vol. 7 (pp. 7-28).
Tilburg, The Nederlands: Tilburg University Press.

Cognition, Brain, Behavior. An Interdisciplinary Journal

15 (2011) 359-384
 F. A. Sava, R. I. Popa 383

Gramzow, R. H., Sedikides, C., Panter, A. T., Sathy, V., Harris, J., & Insko, C. A. (2004).
Patterns of self-regulation and the big five. European Journal of Personality, 18,
367-385.
Hair, J. F. Jr., Anderson, R. E., Tatham, R. L., & Black, W. C. (1998). Multivariate data
analysis (5th Ed.). Upper Saddle River, NJ: Prentice Hall.
Herzberg, P. Y., & Roth, M. (2006). Beyond resilients, overcontrollers and undercontrollers?
An extension of personality prototype research. European Journal of Personality,
20, 5-28.
Huey, S. J., Jr., & Weisz, J. R. (1997). Ego control, ego resiliency, and the fivefactor model
as predictors of behavioral and emotional problems in clinicreferred children and
adolescents. Journal of Abnormal Psychology, 106, 404-415.
Hunsley, J., Lee, C. M., & Wood, J. M. (2003). Controversial and questionable assessment
techniques. In S. O. Lilienfeld, S. J. Lynn, & J. M. Lohr (Eds.), Science and
pseudoscience in clinical psychology (pp. 39-76). New York: Guilford.
Krueger, R. F. (1999). The structure of common mental disorders. Archives of General
Psychiatry, 56, 921-926.
Krueger, R. F., McGue, M., & Iacono, W. G. (2001). The higher-order structure of common
DSM mental disorders: internalization, externalization, and their connection to
personality. Personality and Individual Differences, 30, 1245-1259.
Mandara, J. (2003). The typological approach in child and family psychology: A review of
theory, methods, and research. Clinical Child and Family Psychology Review, 6(2),
129-146.
Pitariu, H. D., Iliescu, D., Tureanu, & V., Peleaşă, C. (2006). Inventarul Psihologic
California [California Personality Inventory]. Bucharest: Odyseea.
Pulkkinen, L. (1996). Female and male personality styles: A typological and developmental
analysis. Journal of Personality and Social Psychology, 70, 1288-1306.
Rammstedt, B., Riemann, R., Angleitner, A., & Borkenau, P. (2004). Resilients,
overcontrollers, and undercontrollers. The replicability of the three personality
prototypes across informants. European Journal of Personality, 18(1), 1-14.
Robins, R. W., John, O. P., Caspi, A., Moffitt, T. E., Stouthamer-Loeber, M. (1996).
Resilient, overcontrolled, and undercontrolled boys: three replicable personality
types. Journal of Personality and Social Psychology, 70(1), 157-171.
Roth, M., & Herzberg, P. Y. (2007). The resilient type: ‘simply the best’ or merely an
artifact of social desirability? Psychology Science, 49(2), 150-167.
Roth, M., & von Collani, G. (2007). A head to head comparison of big-five types and traits
in the prediction of social attitudes. Journal of Individual Differences, 28(3), 138-
149.
Sava, F. A. (2009). Maladaptive schemas, irrational beliefs, and their relationship with the
Five-Factor personality model. Journal of Cognitive and Behavioral
Psychotherapies, 9(2), 135-147.
Sava, F. A. (Ed.) (2008). Inventarul de personalitate DECAS [DECAS Personality
Inventory]. Timisoara: ArtPress.
Sava, F. A., & Maricutoiu, L. P. (2007). PowerStaTim 1.0. – Manualul utilizatorului
[PowerStaTim 1.0 – The user’s manual]. Timişoara: Editura Universitatii de Vest.

Cognition, Brain, Behavior. An Interdisciplinary Journal

15 (2011) 359-384
 384 F. A. Sava, R. I. Popa

Schnabel, K., Asendorpf, J. B., & Ostendorf, F. (2002). Replicable types and subtypes of
personality: German NEO-PI-R versus NEO-FFI. European Journal of Personality
Special Issue, 16, S7–S24
SPSS (2001). SPSS Base 11.0. User’s guide. Chicago, USA: SPSS Inc.
Thomas, A., Chess, S., & Birch, H. (1970). The origin of personality. Scientific American,
223, 102-109.
Urbana, S. (2004). Essentials of psychological testing. New Jersey: John Wiley & Sons.
Vollrath, M.E., & Torgersen, S. (2008). Personality types and risky health behaviors in
Norwegian students. Scandinavian Journal of Psychology, 49, 287-292.
Widiger, T. A., & Frances, A. J. (2002). Toward a dimensional model for the personality
disorders. In P. T. Costa Jt. and T. A. Widiger (eds.), Personality disorders and the
five-factor model of personality (2nd ed.) (pp. 23-44). Washignton, DC: American
Psychological Association.

Cognition, Brain, Behavior. An Interdisciplinary Journal

15 (2011) 359-384

View publication stats