Vous êtes sur la page 1sur 16

Int J Educ Vocat Guidance (2008) 8:111125

DOI 10.1007/s10775-008-9141-0

Hollands hexagonal personality model for a sample

of Greek university students

Despina Sidiropoulou-Dimakakou Kostas Mylonas

Katerina Argyropoulou

Received: 22 November 2006 / Accepted: 2 February 2008 / Published online: 10 June 2008
! Springer Science+Business Media B.V. 2008

Abstract The aim of this study was to describe the hexagonal person-environment
fit for the Holland personality types for a Greek sample of 156 university students.
The statistical analysis followed both exploratorysuch as multidimensional
scalingand confirmatory methodssuch as covariance structure models. These
methods were employed in an exploratory sense, in a descriptive attempt to portray
Hollands hexagon structure for this Greek sample. The findings of this study are
comparable to the overall theory and other international samples, and also serve as a
first psychometric testing of the Self-Directed Search in a Greek population.

Resume. Le mode`le hexagonal de la personnalite de Holland applique a` un

echantillon detudiants universitaires grecs. Le but de cette etude etait de decrire le
degre dajustement personne-environnement selon le mode`le des types de personnalite
de Holland dans un echantillon grec de 156 etudiants universitaires. Lanalyse statis-
tique a combine des methodes exploratoirescomme le multidimensional scalinget
confirmatoirescomme des mode`les de structure de covariances. Ces methodes ont ete
utilisees dans une perspective exploratoire, en vue de depeindre descriptivement la
structure de lhexagone de Holland de cet echantillon grec. Les resultats de cette etude
sont comparables a` la theorie generale et a` ce que lon trouve sur dautres echantillons
internationaux; ils constituent egalement le premier testing psychometrique du
Self-Directed Search sur une population grecque.

Zusammenfassung. Hollands hexagonales Personlichkeitsmodell bei einer

Stichprobe griechischer Studenten. Das Ziel dieser Untersuchung war die
Bestatigung der Anwendbarkeit des hexagonalen Person-Umwelt-Konzepts der
Personlichkeitstypen nach Holland fur eine Stichprobe von 156 griechischen

D. Sidiropoulou-Dimakakou (&) ! K. Mylonas ! K. Argyropoulou

Department of Psychology, Faculty of Philosophy, Pedagogy and Psychology,
University of Athens, Panepistimiopolis, Ilissia, 157 84 Athens, Greece
e-mail: dsidirop@psych.uoa.gr

112 Int J Educ Vocat Guidance (2008) 8:111125

Universitatsstudenten. Die statistische Analyse verwendete sowohl befragendewie

z.B. multidimensionale Einschatzungenals auch bestatigende Methodenwie
z.B. Kovarianz-Strukturmodelle. Diese Methoden wurden in einem erforschenden
Sinn angewendet, in einem beschreibenden Versuch zur Darstellung von Hollands
Hexagonal-Struktur fur diese griechische Stichprobe. Die Ergebnisse dieser
Untersuchung sind vergleichbar mit der zu Grunde liegenden Theorie und anderen
internationalen Ergebnissen, und sie dienen als erste psychometrische Testung der
SDS (Self-Directed Search) bei einer griechischen Gruppe.

Resumen. El Modelo Hexagonal de Holland aplicado a una Muestra de

Estudiantes Universitarios Griegos. La finalidad de este estudio fue describir el
ajuste entre la persona y el ambiente de los tipos de personalidad de Holland
(modelo hexagonal) en una muetra de 156 estudiantes universitarios. Paea el
analisis estadstico se aplicaron metodos tanto exploratorioscomo el escalado
multidimensionalmultidimensional scalingcomo confirmatorioscomo los
modelos estructurales de co-varianza. Estos metodos se usaron en un sentido
exploratorio, con la intencion de describir la estructura hexagonal en esta muestra
griega. Los resultados de este estudio son comparables con la teora general y con su
aplicacion a otras muestras internacionales, y sirven tambien como una primera
evaluacion psicometrica de la Busqueda Auto-Dirigida (Self-Directed Search) en
una poblacion griega.

Keywords Self-Directed Search ! Hexagon structure ! Circumplex model

It has been over 40 years since Holland first proposed his theory of vocational
choice, which continues to be widely accepted by the career guidance counselors
(Herr, Cramer, & Niles, 2004). Hollands (1997) theory of vocational interests
postulated the existence of six broad personality types and environmental models
that describe peoples personalities and the environments in which they are
exhibited. In Hollands theory the six vocational personality typesRealistic (R),
Investigative (I), Artistic (A), Social (S), Enterprising (E), and Conventional (C)
(RIASEC)compose the points of an hexagon. Along the hexagon continuum, the
types that are closer to each other are more alike than those that are more distant or
A number of researchers have recently extended the investigation of Hollands
theorised structure of interests to different populations (Einarsdottir, Rounds,
Egisdottir, & Gerstein, 2002; Fouad & Dancer, 1992; Haverkamp, Collins, &
Hansen, 1994). In this assessment researchers have attempted to determine whether
Hollands six types are identifiable and whether the observed ordering and shape of
the RIASEC configuration is comparable. According to Fouad and Dancer (1992)
interest structure is not universal and may be influenced by culture or ethnicity.
Rounds and Tracey (1996) also suggested that culture system can influence the
strength and interrelations of preferences.
Even though Hollands model has intrigued the interest of Greek scientists who
are involved in vocational assessment and guidance (Panagiotou, 2003;

Int J Educ Vocat Guidance (2008) 8:111125 113

Sidiropoulou-Dimakakou, Touloumakou, & Papadakou, 2004; Tetradakou, 1997),

there is lack of relevant research in Greece. The study of vocational interest
structure in a social and cultural setting that is different from the United States has
practical and theoretical implications for vocational guidance in Greece. Testing
Hollands theory in Greece may shed light on some culture-bound assumptions and
can also be informative about the possible influence of social, cultural, and
contextual factors in career choice and development (Hesketh & Rounds, 1995).
Hollands model would be an excellent springboard (Furnham, 2001) in the
framework of developing and refining their own ideas, tackling the difficult
problems of measuring person-environment structure and understanding its
consequences in another culture outside the United States.
Structural studies of vocational interests in the United States have supported the
invariance of Hollands model across gender (Anderson, Tracey, & Rounds, 1997;
Tracey & Rounds, 1993). No consistent gender differences were detected in the
cross-cultural meta-analysis of Hollands theory (Tracey & Rounds, 1996). Gender
differences in the structure of vocational interests have yet to be evaluated in
Greece, since no such attempt has been made, at least in terms of structure.
The aim of this study was to closely picture the hexagonal person-environment fit
for the personality types that Holland has operationalised (Holland, 1985) for a
Greek sample of undergraduate and graduate university students. This approach
relied on both exploratory methods for the hexagon circumplex but also on
statistical fit testing of the theory through structural equation modelling addressing
the questions of which is the circumplex structure of the personality types for this
Greek sample. The findings of the present studyeven the confirmatory models
should be regarded as first stage exploratory attempts, and should be comparable to
the overall theory and other international samples in order to serve in describing
psychometric characteristics of the Self-Directed Search (SDS) in a Greek



The sample consisted of 156 university students, 84 at the undergraduate level

(53.8%) and 72 at the postgraduate level of their studies (46.2%). The undergrad-
uate students median age was 21 years; for the postgraduate students, though,
median age, as expected, was higher (26 years). Approximately 60% of the sample
were females (94 females, 62 males), which is close to the current gender
representation in Greek Universities.
Undergraduate students were mainly (78.6%) studying Pedagogy and Philosophy
their future employment being high-school teachersand 21.4% were studying
Psychology. Postgraduate students were attending either the School Psychology M.Sc.
course (47.2%) or the Counselling and Career Guidance M.Sc. course (52.8%). A cross-
tabulation of the level of studies (undergraduate or postgraduate) and students gender,
resulted in 52 female students (33.3%) at the undergraduate level, and 42 (26.9%)

114 Int J Educ Vocat Guidance (2008) 8:111125

females at the postgraduate level of their studies; respectively, 32 students (20.5%)

were males at the undergraduate level, and 30 (19.2%) were males at the postgraduate
level of studies.


The SDS consists of four scales named Activities, Competencies, Occupations and
Self-estimates of Traits, all designed to evaluate each of six personality types. The
first scale consists of 11 items with Activities for each of the six subclasses based on
the six personality types for which respondents indicate whether they like or dislike
them. The second scale refers to 11 Competencies with the respondents indicating
the activities they can perform well or at a competent level. The third scale refers to
14 for which the respondents identify those occupations that appeal to or interest
them. Finally, based on their general experience, the respondents evaluate
themselves for 12 traits (two per personality type) on a 7-point scale. All four
scales provide the necessary information to calculate the total score for each one of
the six personality types and evaluate the magnitude of the three highest total scores
resulting into the three-letter Holland code.
Estimates of internal consistency indicated (Anastasi & Urbina, 1997) that the
range of summary-scale Cronbach a coefficients varies by age and gender, but in
general these estimates range from .81 to .92 for females and from .81 to .93 for
males, regardless of age. Testretest reliabilities (three-to four-week intervals) for
high school students have yielded a median reliability coefficient of .81 for boys and
a median coefficient of .83 for girls.
For use in this project the authors translated the SDS into Greek and a back
translation was carried out by a qualified native English speaker translator.

Procedure and statistical analysis

The Greek version of the SDS was administered during regular class hours to all
participants in this study. After the administration the participants were given a
short debriefing about the purpose of the study and thanked for their participation.
The statistical analysis followed two main axes: multidimensional scaling and
structural equation modeling through covariance structure testing. These were
preceded by data transformations to achieve normality and by internal consistency
assessment to verify the suitability of the final six SDS scores for analysis. The
multidimensional scaling techniques were used in order to arrive at an interpretable
and comparable system of personality types for the overall sample, and for females
and males separately. This was supported by trigonometric transformations of the
coordinates reached by the solution in each model (Veligekas, Mylonas, & Zervas,
2007). Covariance structure analysis was then employed to test for four different
structures of the six SDS personality types, comparing them to the null model and to
the data driven structure. Circulant models (Tracey, 2000) and Predigers hypothesis
(Prediger, 1982) were tested through this structural equation modelling attempt in
an exploratory sense, since formation of hypotheses within the present study should
obviously be avoided.

Int J Educ Vocat Guidance (2008) 8:111125 115


Internal consistency and correlational analyses

Several analyses were conducted to compute the internal consistency indices for the
current data, first in respect to each individual personality type across the three out of
four subscales and then within each of these subscales (Table 1). With analyses
performed on each of the three subscales separately for each of the six types, 18 (six
types by three subscales) internal consistency coefficients were computed with an
average coefficient of .71 (ranging from .61 to .85). When each of the three SDS
subscales was examined separately for each personality type, average internal
consistency for all 36 items of each personality type reached .85 but also some lower
reliability coefficients were observed. The drop for the separate subscales might
be an indication of the necessity of aggregating the information from all subscales in
order to arrive at stable measures. Even more, this aggregate of four subscales is not
unique for the Greek version of the SDS but it is considered, internationally, a
standard procedure when the SDS is employed (Einarsdottir et al., 2002).

Table 1 Internal consistency indices for subscale scores and sets of summated scores
Realistic (11 itemsSubscale 1) .74 All 36 (R) itemsa .88
Realistic (11 itemsSubscale 2) .75
Realistic (14 itemsSubscale 3) .77

Investigative (11 itemsSubscale 1) .68 All 36 (I) itemsa .86

Investigative (11 itemsSubscale 2) .66
Investigative (14 itemsSubscale 3) .78

Artistic (11 itemsSubscale 1) .76 All 36 (A) itemsa .89

Artistic (11 itemsSubscale 2) .63
Artistic (14 itemsSubscale 3) .85

Social (11 itemsSubscale 1) .71 All 36 (S) itemsa .80

Social (11 itemsSubscale 2) .64
Social (14 itemsSubscale 3) .71

Enterprising (11 itemsSubscale 1) .74 All 36 (E) itemsa .87

Enterprising (11 itemsSubscale 2) .63
Enterprising (14 itemsSubscale 3) .83

Conventional (11 itemsSubscale 1) .67 All 36 (C) itemsa .79

Conventional (11 itemsSubscale 2) .70
Conventional (14 itemsSubscale 3) .61

All six types (12 itemsSubscale 4) .64

z-score transformations applied to the raw data
Note: Subscale 1 = Activities, Subscale 2 = Competencies, Subscale 3 = Occupations, Subscale
4 = Self estimates of traits

116 Int J Educ Vocat Guidance (2008) 8:111125

The final six personality type aggregates, as described previously, also included
information from each respondents self-evaluations; thus, two more scores were
included in the final sum per personality type. For this procedure there is a metric
problem. The first three subscales of the SDS are scored, item by item, by the
respondent on a yesno basis. This binomial level of measurement is different
from the ordinal level used for the self-evaluation fourth subscale (scores ranging
from one to seven for each self-evaluation). Although arithmetically the maximum
values seem compatible, the respondents may rate themselves only twice for each
type in the self-evaluation subscale. This may introduce some form of construct bias
(van de Vijver & Leung, 1997), or some alternative treatments confounding
variable (Christensen, 1988). However, these self-evaluations are an important part
of the original assessment procedure and cannot be neglected altogether. For each
personality type separately all correlation coefficients were computed between each
self-evaluation and the first three subscale scores for the respective personality
types. The average correlation was .40 (ranging from .34 to .47). For this reason,
and for reasons of comparability, the self-evaluation information was included into
the final personality type scores and the analysis.
Descriptive statistics were computed for each personality type to test for deviation
from normality for the six distributions. With two exceptions (the Realistic type score
and the Social type score) the distributions followed the normal distribution (the
Kolmogorov-Smirnov z criteria did not reach statistical significance). For the Social
personality type, the distribution was slightly negatively skewed, deviating slightly,
but at a significant level from the normal distribution. For the Realistic type the
students mean was rather low in comparison to the other five personality type means.
This was verified through an analysis of variance with repeated measures design, by
contrasting a-priori the Realistic type to the other five types. All comparisons (for 1 and
155 degrees of freedom) were statistically significant at the .001 level and the g2 values
ranged from .33 to .88, showing systematic differences in the students preferences for
the Realistic type in comparison to the other five types. These differences can be
statistically attributed to the positively skewed distribution and can also be interpreted
in respect to the specific sample preferences. Additionally, skewness itself could be
attributed to the characteristics of the sample used in this study.
To remedy for the skewness problems in the data the summated scores were
transformed for the Realistic and the Social personality types. For the Realistic type, a
square root transformation was used. For the Social type scores the cubic scores for
152 casesfour cases were excluded as outlierswere computed. Reducing
accordingly the number of cases for further analysis to 152. The transformed scores
for the Social personality type followed the normal distribution. Final of all six scores
were transformed into z-scores and then again to T-scores, for interpretability reasons.
Before proceeding to the second stage of the analysis, the intercorrelations for the
six personality types were computed (Table 2) on the transformed data (T-scores).
As expected, the correlations for adjacent type in Hollands hexagonal model (e.g.
Artistic and Social types) were reasonably high. However, some more rather high
correlations, such as the Investigative-Conventional one were present and this could
be an indication of differential group functioning producing these correlations and
required further investigation in the analysis.

Int J Educ Vocat Guidance (2008) 8:111125 117

Table 2 Intercorrelations among all subscale aggregate T-scores, and initial mean scores (raw, non-
transformed) for the six personality types
(N = 152) R I A S E C

R 11.62 .46 .16 .21 .30 .36

I .54 21.67 .34 .30 .16 .16
A .15 .30 23.90 .42 .35 .11
S .05 .24 .33 37.99 .54 .38
E .21 .33 .21 .45 22.32 .68
C .30 .39 -.11 .14 .39 17.53

Notes: (1) Above the diagonal, Hollands correlations (computed via metric adjustments by the authors,
assuming N = 156)
(2) Following a Fisher-z transformation of both observed (current) and Hollands correlation indices (and
assuming N = 156, with a lower N being a difference suppressor), a direct comparison of each pair or
correlations resulted into z criteria. Bold numbers in the table denote differences between the observed
correlations in this sample and correlations as reported by Holland. Overall, the Conventional type
correlations with the I, S and E types differ significantly for the two sets of correlations

Multidimensional scaling analysis

In order to explore the underlying structure, at least two general methods of analysis
can be followed: exploratory factor analysis and/or multidimensional scaling. Direct
application of exploratory or confirmatory factor analysis models would test the
latent traits characterising the six personality types, but this was not the main aim of
the analysis in this study. The main interest lies in exploring the relative positions of
the types in the structure. Additionally, factor analysis would be inappropriate for
such a limited (less than 10) number of measures (Kline, 1993). To circumvent these
limitations, multidimensional scaling was employed both under an exploratory and
also a confirmatory fashion to first describe the relative position of each of the six
personality types in respect to the other five and then test for the expected
dimensions of the circumplex, as depicted by expected multidimensional scaling
solutions and related proposed models (Prediger, 1982; Tracey, 2000). This method
of analysis was expected to yield a solution which would be comparable to the
theoretical and empirical description given by Hollands theory and also would be in
accordance with the methods used by Holland himself during his theory formation.

Exploratory stage

A two-dimensional scaling analysis was employed, based on Euclidean distance

transformations of the six summated personality scores (T-scores). The distance
matrix was analysed for the total sample and also for males and females separately.
These results, in respect to their stimuli relative coordinates, are plotted in Fig. 1.
For the overall multidimensional solution (total sample), Youngs Stress = .003 and
R2 = .99. These statistics were hardly surprising given the strength of the theoretical
model and also the small number of measures in respect to the number of
dimensions requested. For the female sample only (N = 93), Youngs Stress = .003

118 Int J Educ Vocat Guidance (2008) 8:111125


1 A
Artistic i R
Investigative I

S c
s E e
-1 Social C

-2 -1 0 1 2
Upper case typeface: Females Lower case typeface: Males

Fig. 1 Two-dimensional scaling solutions for the six personality types (all subscale aggregate T-scores),
overall and by gender

and R2 = .99. For the male sample only (N = 59), Youngs Stress = .0003 and
R2 = .99. Both male and female personality structures seem to generally follow the
overall pattern; nevertheless, there are discrepancies between the two genders as
depicted for their trigonometric transformations on the circle periphery.
The main reason for trigonometric transformations in this study was that
exploratory multidimensional scaling solutions often suffer from interpretation
problems, specifically when the dimensions do not immediately portray constructs
or latent traits. For example, one cannot easily suggest that the first dimension
represents a contrast between a group of three personality types (namely Realistic,
Investigative, and Artistic) and another group (namely Conventional, Enterprising,
and Social), since such a contrast does not seem to be easily readable in the initial
two-dimensional plot (Fig. 1); in fact, under this eye-balling procedure (Tracey,
2000), any contrast between two groups of personality types does not readily reflect
a specific latent factor, since the theoretical presupposition of placement of all
personality types on a hexagon continuum, is not clearly met. A possible way to
circumvent this problem is to constrain the model by estimating the measures
relative position regardless of their distances from the central (0, 0) point on the
axes. Thus, we would use only the information given by their orientation in regard
to this central point. The angles formed by this orientation (the coordinates on the
two axes), were computed for each personality type through an arctangent
transformation and were then transformed to degrees on the circumference, a
method of analysis also successfully employed by Mylonas (in Veligekas et al.,

Int J Educ Vocat Guidance (2008) 8:111125 119


c .30

-.11 R
.39 r R .54
.39 i I
.21 I

e .05

.15 .24
E .30
.21 A

S .33

Fig. 2 Two-dimensional scaling solutions (all subscale aggregate T-scores, trigonometric

transformation) and Pearson correlations for the six personality types, overall and by gender. Key:
Outside the circumference: total sample (n = 152), Inside the circumference: upper case typeface:
females (n = 93); lower case typeface: males (n = 59)

2007). These degrees, for the total sample (and also for males and females
separately), were plotted in Fig. 2. This figure partially enhances interpretation of
the relative position of the six personality types on the circumference in respect to
constructs involved. Also, by rotating these relative positions in respect to an
arbitrary point (for Fig. 2, the Realistic position) the problem of reversed scaling on
the axes could be circumvented, and all groups under consideration (overall,
females, males) become comparable in respect to their two-dimensional scaling
solutions. The correlation coefficients between all pairs of personality types are also
reported in Fig. 2, for the total sample only.
The overall solution seems to generally follow Hollands model, but there is
some difference; the Realistic and the Investigative types are very closely linked,
whereas one would expect some distance between them, as is the case for the
remaining four personality types. Also, for both genders the pattern remains
approximately the same with small differences in the relative distance of some
personality types (e.g., the Conventional-Enterprising-Social subsystem is rather
loose for the female sample). The main finding at this stage is obviously the
Realistic-Investigative subsystem of homogeneous measures, with the two
personality types closely linked also when each gender is examined separately.

120 Int J Educ Vocat Guidance (2008) 8:111125

Structural equation modeling of Hollands circumplex for the Greek data

To explore the main finding further, confirmatory methods were employed, although
it should be kept in mind that the nature of this study still remained exploratory. The
methods employed have been previously tested in direct relation to Hollands
hexagon (Prediger, 1982; Rounds & Tracey, 1993) and in a more general sense (e.g.
Hofstee, de Raad, & Goldberg, 1992) where factor representations of circumplex
models have been dimensionally tested in a confirmatory sense. Tracey (2000) has
also summarised a large number of confirmatory and other methods when analysing
circumplex structures. Some of the suggested-available options were implemented
in this study as well, namely testing for Predigers models in respect to Hollands
hexagon (Rounds & Tracey, 1993) and also for the circulant and Geometric
Circulant models as described by Tracey (2000). The statistical procedures and their
results are described as follows and in Table 3.
In all, six models were tested. As a point of reference or benchmark model, the
null model with all six personality types considered as independent factors was
tested as the first model. The statistical fit associated was as expected extremely
poor. A data driven model (as suggested by exploratory factor analysis results for
this studys measures) was also tested, although just as another point of reference
because of its ipsative nature. For this data-driven model to be formulated, principal
component analysis and maximum likelihood estimation exploratory factor analyses
were carried out initially, applying orthogonal rotation of the axes to both models.
The maximum likelihood solution was clearer but the principal components solution
seemed closer to the present data and to the original theoretical model with two sets
of adjacent types (C-R-I, S-E-C) loading on factors one and two respectively and
with a third bipolar factor including A and C types in contradistinction. This PCA
three-factor solution was considered useful for comparison with the theoretical
models to be tested and resulted in rather acceptable levels in terms of the Normed-
Fit index and the Tucker-Lewis index, although RMSEA and v2 significance levels
were still far from satisfactory.
The circulant model hypothesis (Tracey, 2000) tests for adjacent types are
related, while the two steps away types are less correlated and the opposite types in
the circumplex are even less correlated. For this model, tested third in order, the
statistical fit was poor and the RMSEA index was very high. The Normed-Fit index

Table 3 Structural equation modeling of Hollands circumplex (summary table)

Model v2 df RMSEA NFI TLI

1. Null (six independent factors) 211.19 9 .39 NA NA

2. Data driven (as computed via PCA, varimax) 51.38 7 .20 .75 .72
3. Circulant 78.17 11 .20 .60 .73
4. Geometric circulant 79.97 10 .22 .56 .69
5. Predigers two-factor (constrained) 44.95 5 .23 .79 .64
6. Predigers two-factor (unconstrained) 44.00 5 .23 .79 .65

Note: All v2 criteria are statistically significant at the .001 level

Int J Educ Vocat Guidance (2008) 8:111125 121

was too low, but the Tucker-Lewis index showed considerable improvement from
the null model. The Geometric Circulant model hypothesis (Tracey, 2000) tests for
the parameters of the circular model was proportional and linearly related to the
distance around the circumference of the circle. The adjacent types are considered
highly correlated, and in the case of the sextant, 3 times more strongly than the
opposite types, where the two-steps away types are twice as strongly correlated than
the opposite types. For this model, the statistical fit was still poor and the RMSEA
index was very high. The Normed-Fit index was too low and the Tucker-Lewis
index showed less improvement from the null model than the previously tested
circulant model.
Predigers two factor model (Rounds & Tracey, 1993) was also tested for the
present data, for which the theoretically assumed coordinates on a two-dimensional
solution are employed to define the two dimensions to be tested. Specifically,
Predigers two-factor model assumes that a first bipolar factor includes I, A, E and C
types, with E and C bearing negative signs, and that a second factor includes R and
S types in contradistinction with I and C loading less strongly on the same factor on
the R side, and with A and E loading less strongly on the same factor on the S
side. Using the coordinates as described by Rounds & Tracey we defined the two
models to be tested under the Prediger hypothesis, one with constrained loading
values and one with unconstrained ones. The constrained model yielded the best
statistical fit in terms of the NFI index, although RMSEA was still very high and the
TLI index did not imply the best levels of departure from the null model. The
unconstrained model proved almost identical in terms of statistical properties.
Overall, Predigers two-factor constrained model seems to present the best fit-in
respect to all models tested- to the data of our study.
To conclude, two models seem to fit our data better than the other models in the
analysis: the data-driven 3-factor model and the constrained two-factor Predigers
model. For the latter it should be noted that no general factor was assumed or tested.
For the data-driven model, it is obvious that it is the closest to the structure in the
present data, but being of ipsative nature it cannot be considered an acceptable
solution but only a point of reference for the next best statistical fit, which seems to
be Predigers two-factor constrained model. For this model, a trade-off in the NFI
and TLI indices seems to exist, but the overall v2 index does not rise extremely with
the loss of the 2 degrees of freedom (the ratios of the v2 over df are 7.34 and 9.00,
respectively), and RMSEA is not very much higher, remaining at approximately the
same levels of around .20, as observed for most models tested.
Finally, two separate models (using Predigers two-factor constrained structure)
were tested separately for males and for females. Unexpectedly enough, the Male
structure was statistically insignificant (v2 reached only 9.36 and for 5 degrees of
freedom was not significant), although RMSEA just approached acceptable levels
(.12). Still, Normed-Fit index was high (.85) and the Tucker-Lewis Index in respect
to improvement (compared to the null model) was very high (.96). Then, the
respective indices for Females only were not so rewarding, since v2 reached 19.11,
and for 5 degrees of freedom it was significant at the .01 level, RMSEA was still far
from acceptable (.17). Still, Normed-Fit index was .74 and the Tucker-Lewis Index
in respect to improvement (compared to the null model) was .87. Such an

122 Int J Educ Vocat Guidance (2008) 8:111125

unexpected discrepancy between genders was not so visible during the

exploratory stage, and then a further multivariate analysis of variance design
between genders showed that a significant multivariate effect was present; for the
Hotellings Trace criterion, F(6, 145) = 3.057, p \ .01, g2 = .11. Univariate differ-
ences were present only for the Realistic personality type with males reporting much
higher scores (a T-score of 52.3, as compared to the female score of 48.5).


The results of this study support the validity of Hollands model in Greece, showing
that the structure of vocational interests in samples of Athens University students is
similar to the structure found in US population. These results also indicate that a
two-dimensional circular order representation of Hollands RIASEC types is
applicable in Greek samples. Interpreting the outcomes, the two-dimensional
circumplex structure as described by Prediger seems to best depict the six types
interrelations in this Greek data set approach. The results support the continuing use
of Hollands theory as a conceptual framework for vocational and educational
counselling in Greece, having important practical implications since they show that
Hollands model is applicable to university student populations in Greece.
The results of this study reveal some inconsistencies in the shape of the
hexagonal model (e.g. the Realistic and the Investigative types are very closely
linked, whereas one would expect some distance between them, as is the case for the
remaining four personality types). The same pattern of relative distance for Realistic
and Investigative types was also indicated in a previous unpublished study
(Touloumakou, 2005) in a sample of Greek undergraduate university students. This
result might possibly be due to Greek cultural parameters or to the demographic
and economic constraints on choice (Furnham, 2001, p. 9). This, in turn, could
affect the occupational interests and perhaps the structure of interests for a group
population in Greece. This means that the demands of the Greek labour market, job
security and developments in technology result in values, interests, skills and
behaviours relevant to those two types (I and R) being perceived as similar and
adjacent in content.
The results are in line with Haverkamp et al. (1994) as well as with the findings
by Ryan, Tracey, and Rounds (1996). As Rounds and Tracey (1996) reported in a
structural meta-analysis of 76 samples from 18 countries, non U.S. samples had a
poorer statistical fit with Hollands model than did U.S. samples. Fifteen of the
eighteen countries failed to follow the model and potential moderators such as
cultural values did not explain model differences between countries. Tracey and
Rounds raised questions about how well the circumplex-hexagonal model describes
interest structure for non-U.S samples. The current research also confirms Hollands
(1992) assumptions that the hexagonal model often resembles a misshapen
polygon more than a regular hexagon in real-world data (Holland, 1992, p. 119)
and also supports Myors (1996) findings, which indicate that if the term misshapen
polygon can be made precise, then new-order relations between the types can be

Int J Educ Vocat Guidance (2008) 8:111125 123

Another issue is the application of the measure to different populations

(Furnham, 2001). As Furnham supported (p. 21), all measures are sensitive to
subtle gender and cultural differences of different populations examined. This
means that the hexagonal structure becomes more sensitive when more populations
are put under study. The present results suggest that culture may influence the
interrelationships between the types and underline the importance of studying not
only the differences that exist between the members of the culturally different
groups but also within the members of the same cultural group (Sidiropoulou-
Dimakakou, 2003).
The differences between men and women could be attributed to the stereotypes
existing in Greek society as well as to the way both girls and boys are raised.
Children according to their gender are raised differently, and therefore go through a
different procedure of career planning (Sidiropoulou-Dimakakou, 2002). In
practice, the results of this study stresses the need of formulating a theory for the
career development of women (Sidiropoulou-Dimakakou, 1996). However, the
structure of RIASEC types in Greece needs to be further researched since gender
differences in Holland types do not seem to hold true according to a meta-analysis
by Rounds and Tracey (1993). The specific findings of the present study might also
be due to the fact that the samples are not matched-pairs ones (Anderson et al.,
1997), and that the sample size is relatively small and might yield some instability
in the results. Along with this limitation, some other restrictions should be noted
such as the limited range of the age of the participants, as the sample is drawn only
from a university students population. Furthermore, a more heterogeneous sample
(in terms of fields of study) might reveal different structures and should be
considered for future research.
Despite the above limitations, the current results support the continuing use of
Hollands theory as a conceptual framework for vocational and educational
counseling in Greece. Rayman and Atanasoff (1999) argue that the theory is simple,
has face validity, provides an understandable organizational framework, and is
transferable into practice. The results of this study have implications for vocational
interest theories. Exploring Hollands theory in Greece and other countries in the
framework of a cultural system, may initially provide support for Hollands
personality types application to these cultures, assuming that we can accept some
limited alteration to the theorys statistical properties because of cultural
differences. Further cross-cultural research of vocational interest structure may
open up new paths in understanding social, cultural and contextual factors that
influence vocational interests and choice.

Acknowledgements The authors would sincerely like to thank the four anonymous reviewers and the
Journal Editor for their directions and highly constructive comments during the revision procedure.


Anastasi, A., & Urbina, S. (1997). Psychological testing (7th ed.). NJ: Prentice-Hall.
Anderson, M. Z., Tracey, T. J. G., & Rounds, J. (1997). Examining the invariance of Hollands vocational
interest model across gender. Journal of Vocational Behavior, 50, 349364.

124 Int J Educ Vocat Guidance (2008) 8:111125

Christensen, L. B. (1988). Experimental methodology (4th ed.). Boston: Allyn & Bacon Inc.
Einarsdottir, S., Rounds, J., Egisdottir, S., & Gerstein, L. H. (2002). The structure of vocational interests
in Iceland: Examining Hollands and Gatis RIASEC models. European Journal of Psychological
Assessment, 18(1), 8595.
Fouad, N. L., & Dancer, L. S. (1992). Cross-cultural structure of interests: Mexico and the United States.
Journal of Vocational Behavior, 40(2), 129143.
Furnham, A. (2001). Vocational preference and P-O Fit: Reflections on Hollands theory of vocational
choice. Applied Psychology: An International Review, 50(1), 529.
Haverkamp, B. E., Collins, R. C., & Hansen, J-I. C. (1994). Structure of interests of Asian-American
college students. Journal of Counseling Psychology, 41, 256264.
Herr, E. L., Cramer, S. H., & Niles, S. G. (2004). Career guidance and counseling through the life span
(6th ed.). USA: Pearson.
Hesketh, B., & Rounds, J. (1995). International cross-cultural approaches to career development. In
W. B. Walsh, & S. H. Osipow (Eds.), Handbook of vocational psychology: Theory research and
practice (2nd ed., pp. 367390). Mahwah, NJ: Erlbaum.
Hofstee, W. K. B., de Raad, B., & Goldberg, L. R. (1992). Integration of the big five and circumplex
approaches to trait structure. Journal of Personality and Social Psychology, 63, 146163.
Holland, J. L. (1985). Making vocational choices: A theory of vocational personalities and work
environments (2nd ed.). Englewood Cliffs, NJ: Prentice-Hall.
Holland, J. L. (1992). Making vocational choices (2nd ed.). Odessa, FL: Psychological Assessment-
Holland, J. L. (1997). Making vocational choices (3rd ed.). Englewood Cliffs, NJ: Prentice-Hall.
Kline, P. (1993). The handbook of psychological testing. London: Routledge.
Myors, B. (1996). A simple, exact test for the Holland hexagon. Journal of Vocational Behavior, 48, 339351.
Panagiotou, I. (2003). Dierevnisi ton typon prosopikotitas kai epilogis epaggelmatos (mia efarmogi tis
typologikis theorias tou Holland): I periptosi ton mathiton Technikis ekpaidevsis tis Milou
(in Greek) [Examination of personality types and occupational choice -an application of Hollands
theory: the case of the Technical education students in Milos island]. Unpublished M.Sc. thesis,
University of Athens.
Prediger, D. J. (1982). Dimensions underlying Hollands hexagon: Missing link between interests and
occupations? Journal of Vocational Behavior, 21, 259287.
Rayman, J., & Atanasoff, L. (1999). Hollands theory of career intervention: The power of the hexagon.
Journal of Vocational Behavior, 55, 114126.
Rounds, J., & Tracey, T. J. G. (1993). Predigers dimensional representation of Hollands RIASEC
circumplex. Journal of Applied Psychology, 78, 875890.
Rounds, J., & Tracey, T. J. G. (1996). Cross-cultural structural equivalence of RIASEC models and
measures. Journal of Counseling Psychology, 43, 310329.
Ryan, J., Tracey, T., & Rounds, J. (1996). Generalizability of Hollands structure of vocational interests
across ethnicity, gender, and socioeconomic status. Journal of Counseling Psychology, 43, 330337.
Sidiropoulou-Dimakakou, D. (1996). I anagaiotita diatyposis mias theorias epaggelmatikis anaptyxis ton
gynaikon (in Greek) [The necessity of formulating a theory for women career development]. Leschi
ton Ekpedeftikon [Educators forum], 14, 3334.
Sidiropoulou-Dimakakou, D. (2002). Epaggelmatiki axiologisi kai symvouleftiki ton gynaikon (in Greek)
[Career assessment and counseling for women]. In M. Kassotakis (Ed.), Symvouleftiki kai
Epaggelmatikos Prosanatolismos: Theoria kai Praxi [Counseling and Career Guidance: Theory and
Practic] (pp. 623634). Athens: Tipothito.
Sidiropoulou-Dimakakou, D. (2003). Epaggelmatiki symvouleftiki kai politismiki diaforetikotita
(in Greek) [Career counseling and cultural diversity]. Psychologia, 10(2 & 3), 399413.
Sidiropoulou-Dimakakou, D., Touloumakou, A., & Papadakou (2004). Dislitourgikes pepithiseis stin
epilisi provlimaton stadiodromias kai sti lipsi epaggelmatikon apofaseon: H sxesi tous me ta
epaggelmatika endiaferonda ke tous tipous prosopikotitas kata Holland (in Greek) [Disfunctional
perceptions in problem solving and career decision making: Their relation with career interests and
Hollands personality types]. In M. Makris & D. Desli (Eds.), I gnostiki psychologia simera: Gefires
gia ti meleti tis noisis [The cognitive psychology today: Bridges for the study of mind] (pp. 198200).
Athens: Tipothito.
Tetradakou, N. (1997). Epaggelmatika endiaferonta teliofiton lykeiou symfona me tin typologiki theoria
tou J. Holland kai epilogi spoudon tritovathmias ekpaidevsis (in Greek) [Vocational interests of

Int J Educ Vocat Guidance (2008) 8:111125 125

senior high school students according to J. Hollands theory, and choice of university fields].
Unpublished M.Sc. thesis, University of Athens.
Touloumakou, A. (2005). The Self-Directed Search. Stathmisi tou ergaleiou Dokimion autoanaforas tou
John, L. Holland. (in Greek) [Standardization procedures for Hollands Self-Directed Search].
Unpublished M.Sc. thesis, University of Athens.
Tracey, T. J. G. (2000). Analysis of circumplex models. In H. E. A. Tinsley & S. D. Brown (Eds.),
Handbook of applied multivariate statistics and mathematical modeling (pp. 641664). San Diego:
Tracey, T. J. G., & Rounds, J. (1993). Evaluating Hollands and Gatis vocational interest models:
A structural meta-analysis. Psychological Bulletin, 113, 229246.
Tracey, T. J. G., & Rounds, J. (1996). A spherical representation of vocational interests. Journal of
Vocational Behavioral, 48, 341.
Van de Vijver, F., & Leung, K. (1997). Methods and data analysis for cross-cultural research. Thousand
Oaks, CA: Sage.
Veligekas, P., Mylonas, K., & Zervas, Y. (2007). Goal orientation and beliefs about the causes of success
among Greek track and field athletes. Perceptual and Motor Skills, 105, 927938.

Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.