Professional Documents
Culture Documents
model
Andrei Ion¹, Dragoș Iliescu¹, Said Aldhafri², Neeti Rana³, Kattiya Ratanadilok⁴,
³School of Buddhist Studies and Civilization, Gautam Buddha University, Greater Noida,
(Andrei Ion).
SELF-ARCHIVED PREPRINT
Ion, A., Iliescu, D., Ratanadilok, K., Rana, N., Widyanti, A., & Aldhafri, S. (2017). A Cross-
Cultural Analysis of Personality Structure through the Lens of the HEXACO Model. Journal
Abstract
Across 5 different samples, totaling more than 1,600 participants from India,
Indonesia, Oman, Romania, and Thailand, the authors address the question of cross-cultural
framework, provided by the HEXACO–PI (Lee & Ashton, 2004). The results show that the
HEXACO framework was replicated in some of the investigated cultures. The ESEM data
model
The degree to which human personality has some invariant characteristics in its
structure that could be called universal is a question that has caught the interest of scientists
and also has important practical implications. In the past decades, many large-scale, cross
cultural investigations have aimed at exploring personality trait structure. Most of these large-
scale, cross-cultural investigations (e.g., D. P. Schmitt et al., 2007) have converged toward a
However, a closer look reveals that these findings were based on exploratory data
analytical frameworks, specifically exploratory factor analysis (EFA). The use of more robust
data analytical methods in estimating the structure of personality traits, such as confirmatory
factor analysis, yielded less articulate results (e.g., Borkenau & Ostendorf, 1990). Therefore,
new data analytic methods that can be useful in personality research have been devised, such
as exploratory structural equation modeling (ESEM). These have, to our knowledge, not yet
been used in the analysis of cross-cultural personality research. Encouraged by the recent
developments in data analysis methods and by the need to test whether universality is indeed
This investigation fits into the still vivid debate regarding two of the issues that are
central to our understanding of personality: the universality of personality trait structure and
potentially universal structure of personality traits, have shown findings converging toward a
PERSONALITY HEXACO MODEL 4
five-factor structure. This structure has been labeled the Five-Factor Model (FFM). Some
authors have argued that the five-factor structure is so far the closest match to an
“omnipresent” personality trait structure (De Raad, Perugini, Hrebickova, & Szarota, 1998).
Further research has offered empirical support for the universality of the five-factor structure
(De Fruyt, De Bolle, McCrae, Terraciano, & Costa, 2009; McCrae et al., 2005; D. P. Schmitt
et al., 2007). Specific traits measured by the FFM have been shown to be replicable at the
content level across different cultures (e.g., Nye, Roberts, Saucier, & Zhou, 2008). Gradually,
this model has risen in importance and has become so prominent that currently even the most
severe critics of the FFM acknowledge the fact that it has become central to “contemporary
deviate to a greater or lesser extent from the mainstream FFM reopened debates surrounding
the alleged universal structure that can be retrieved or replicated in most cultures. For
offer support for a narrow, two-factor personality traits structure (e.g., Gurven, von Rueden,
Massenkoff, Kaplan, & Lero Vie, 2013). Even for Western countries, an analysis of factor
structures of personality traits across 12 languages and 14 taxonomies concluded that only
three factors are entirely replicable across all the different cultural groups (De Raad et al.,
2010). Some empirical research advocates in favor of a Big One personality factor (e.g.,
Musek, 2007). In support of a larger number of personality traits, several lexical studies
conducted across Croatian, Turkish, Greek, English, or Dutch groups have revealed a six-
factor personality traits structure (Ashton & Lee, 2007). A study analyzing the personality
traits structure across 11 South African cultural groups and languages reported the emergence
of nine-factor structures (Nel et al., 2012). During the early 1990s a model including seven
PERSONALITY HEXACO MODEL 5
personality dimensions was retrieved (Tellegen & Waller, 1987) and replicated in various
cultures (e.g., Almagor, Tellegen, & Waller, 1995; Benet & Waller, 1995).
The fact that some of the popular personality models are confirmed in different
cultures does not imply that those are the only cross-culturally replicable or “universal”
personality traits structures. We consider that to better understand the universality issue in
personality research, scientific endeavors should focus not only on confirming that the most
popular frameworks are replicable across different cultures, but also on investigating whether
other, less researched frameworks are cross-culturally replicable. To this end we investigated
whether the six-factor model of personality traits, model commonly labeled HEXACO
(Ashton & Lee 2001; Lee & Ashton, 2004) exhibits replicability across different cultural
groups.
One of the more influential projects exploring personality traits structure beyond the
FFM was undertaken by Ashton and Lee (2001; also Lee & Ashton, 2004) resulting in a six-
factor framework of personality traits, the HEXACO model. The instrument attached to this
model and used for measuring the six factors is the HEXACO Personality Inventory
(HEXACO–PI; Lee & Ashton, 2004). The model includes some components that are related,
at least in name, to the FFM, such as Emotionality (E), Extraversion (X), Agreeableness (A),
component that has no counterpart within the FFM, the Honesty-Humility (H) dimension.
The six-factor model of personality traits is not an extension of the FFM, as it has emerged as
a distinct structure obtained via lexical studies conducted in various languages and cultures
(Ashton & Lee, 2007). The dimensions of the HEXACO model have various degrees of
overlap with the dimensions of the FFM. For example, Extraversion, Openness to
Experience, and Conscientiousness are by and large counterparts of the homonymous factors
PERSONALITY HEXACO MODEL 6
of the FFM. On the other hand, the Honesty-Humility component has no corresponding
dimension in the FFM. The HEXACO Agreeableness factor is also not perfectly matched
with the FFM Agreeableness factor. In addition to the traits typically measured in the FFM, it
includes traits of anger and hostility (Lee, Ashton, Morrison, Cordery, & Dunlop, 2008). The
utility of the HEXACO has been investigated in different areas of applied psychology.
Independent findings indicate that the six personality dimensions are predictive for life
outcomes (e.g., Thalmayer, Saucier, & Eigenhuis, 2011), counterproductive work behaviors
(e.g., Marcus, Ashton, & Lee, 2013; O’Neill, Lewis, & Carswell, 2011), job performance
(e.g., Johnson, Rowatt, & Perrini, 2011), academic performance (De Vries, De Vries, &
Born, 2011), and integrity (e.g., Lee et al., 2008). The HEXACO model is comparable to the
FFM in terms of utility in predicting work-related outcomes (e.g., Lee, Ashton, & De Vries,
2005; Lee, Ashton, & Shin, 2005). In some areas, the HEXACO has evidence to support its
incremental validity over and beyond the FFM trait domains; most of the practical advantages
of the HEXACO over the FFM are related to the Honesty-Humility component (Ashton &
Lee, 2007). The six-dimensional model was replicated across more than 10 different cultures
(Aghababaei, Wasserman, & Nannini, 2014; Ashton & Lee, 2007; Wasti, Lee, Ashton, &
Somer, 2008). However, as with other celebrated personality frameworks, all the just
mentioned investigations relied on EFA as the main data analytical framework. This
investigation extends our knowledge of the HEXACO model by examining its replicability
with culturally distinct groups, for which this framework has not been previously studied
structure
EFA is one of the most commonly used data analysis techniques in psychology, and
especially in personality research (e.g., Cudeck & MacCallum, 2007). The results of an EFA
PERSONALITY HEXACO MODEL 7
depend heavily on a set of decisions the researcher makes, mainly regarding the extraction
method, the rotation method, and the number of factors retained in the solution (e.g., Bentler
& Kano, 1990; Conway & Huffcut, 2003; Costello & Osborne, 2005; Fabrigar, Wegener,
MacCallum, & Strahan, 1999). It is due to this dependency that some authors have called the
validity of EFA findings into question, arguing that “EFA can generate a variety of truths
according to the bent or desires or avoidances of the particular factor analyst” (Block, 2010,
p. 6).
Another data-analytic method that gradually became the gold standard in analyzing
the latent structures of different measures is confirmatory factor analysis (CFA). CFA has
often been used for analyses into the structure of personality traits across different cultural
groups (e.g., Cheung & Rensvold, 2000; Little, 1997; Vandenburg & Lance, 2000; Van de
Vijver & Leung, 1997). Motivated by the fact that several straightforward and seemingly
valid personality frameworks were not confirmed via CFA (e.g., Borkenau & Ostendorf,
1990; Church & Burke, 1994; Parker, Bagby, & Summerfeldt, 1993), some authors have
raised concerns about its utility in personality assessment research. Most of the critiques have
predicted factor structure. This imposition is seen by some as inappropriate for analysing
(Aluja, Garcia, Garcia, & Seisdedos, 2005; Marsh & Hau, 2007; McCrae, Zonderman, Costa,
Briefly stated, both EFA and CFA entail a set of advantages and limitations,
especially when deployed in the analysis of broad personality measures. Thus far, regardless
of the personality taxonomy that has been explored, our current understanding regarding the
structure of personality traits has been shaped by research that relied mainly on EFA or CFA.
A new evolution in the area of factor-analytic techniques used in personality research, ESEM,
PERSONALITY HEXACO MODEL 8
builds on the strengths of both these techniques. In this article we compare the utility of
ESEM against that of EFA, having a methodological contribution to the field of personality
research.
ESEM is a data analysis method that builds on the main features of EFA and CFA,
while providing a more flexible analytical framework. ESEM uses the traditional parameters
of structural equation modeling (SEM) that are met in CFA approaches such as fit indexes
and residual correlations or factor regressions. However, it also allows for the rotation of the
initial measurement models (Asparouhov & Muthen, 2009). This way, ESEM tackles some of
the limitations associated with both EFA and CFA, permitting simultaneous estimations of
both structural and measurement models, thus avoiding misspecified models for measures
developed in an EFA framework, and testing a theoretical model without a strict specification
of the measurement structure (Asparouhov & Muthen, 2009). ESEM permits the testing of
personality assessment. This technique has been already successfully pioneered in exploring
the invariance of the FFM across age and gender (e.g., Marsh et al., 2010; Marsh, Nagengast,
& Morin, 2013) or across testing contexts (Furnham, Guenole, Levine, & Chamorro-
Premuzic, 2013).
makes score comparisons across groups impossible and obviates the capacity to analyze how
one or several constructs function across different groups (Little, 1997; van de Vijver &
typically establishes a baseline model by specifying the hypothesized model (Byrne, 2008; N.
Schmitt & Kuljanin, 2008). A prerequisite of measurement invariance testing consists in the
identification of well-fitting baseline models. It should be noted that several research projects
that are usually referenced in support of the FFM claim for replicability fail to address the
issue of measurement invariance (e.g., McCrae et al., 2005; D. P. Schmitt et al., 2007). Such
a failure could be partly explained by the severity with which CFA sanctions nonzero
deployed for identifying clear baseline models. Establishing an acceptable baseline model is
This article applies the ESEM analytical framework to cross-cultural personality data.
We expected ESEM to yield baseline models that yield acceptable goodness of fit when used
for a complex personality framework (HEXACO), thus enabling the exploration of various
Method
This research investigated the way in which the HEXACO model, as an alternative
groups with different languages: India (Hindi), Indonesia (Indonesian), Oman (Arabic),
Romania (Romanian), and Thailand (Thai). There are two principles that guided the inclusion
of these different cultural groups in this investigation. First, in an attempt to expand our
understanding regarding personality traits structure, we chose countries that do not belong to
the typically overstudied group of Western, educated, industrialized, rich, and democratic
Buddhism in Thailand.
Participants
The data were obtained as part of the cultural adaptation process for the HEXACO
across five different cultures and the corresponding languages: India (Hindi), Indonesia
(Indonesian), Oman (Arabic), Romania (Romanian), and Thailand (Thai). The five samples
thus obtained are each homogeneous from a linguistic, cultural, and religious point of view.
Still, they ensure diversity in the study and a striking difference from the U.S. or Western
European samples that are typically used in personality research: The investigated samples
are different from each other from a linguistic, religious, and cultural point of view: Hindi
language and Hinduism for India, Indonesian language and Islam in Indonesia, Arabic
language and Ibadiyya (a branch of Islam) in Oman, Romanian language and Greek
Thailand. The sample sizes of the five groups included in the analysis ranged from 210 to 482
participants per country. All the investigated groups were assembled via an opportunity
sampling procedure.
The Hindi sample includes a total of 269 participants, with 158 (58.7%) males. The
age range is between 18 and 46 years (M = 22.37, SD = 5.21). The Indonesian sample
includes 482 participants, with 292 (60.5%) males, and ages between 17 and 28 years (M =
20.91, SD = 1.33). The Omani sample includes 255 participants with ages between 16 and 45
years (M = 19.47, SD = 5.07), among them 140 (56%) males. The Romanian sample consists
of 417 participants aged between 18 and 56 years (M = 29.14, SD = 7.86), among them 163
(39%) males. The Thai sample includes 210 participants, with 155 (73%) males, and ages
Measures
Ashton and Lee (2001; also Lee & Ashton, 2004). The HEXACO–PI–R follows a
hierarchical model, including 200 items structured onto 24 facets, and six broad dimensions,
each dimension including four facet scales. The HEXACO–PI–R uses a 5-point Likert
Translation procedure
The HEXACO translation process was based on a similar approach for each of the
This process followed the International Test Commission (ITC, 2010) Guidelines on Test
from a forward translation perspective, as opposed to backward translation, does not work
through the items of a questionnaire in the order in which these items are presented to a test
taker, but begins by grouping the items into their respective scales and facets. The translator
then approaches the translation of each item so that it captures the intent of the respective
scale or facet, as shown by the definition of that construct. This procedure was followed in all
five languages of our study. The resulting draft translation was then submitted to a blind
backtranslation, and supplementary adjustments were made on the items. A review was then
made based on the Item Translation and Adaptation Review Form proposed by Hambleton
and Zenisky (2011). Finally, the results of all these steps were submitted to the original test
authors, and their suggestions were implemented in what was considered a final other
Results
The EFAs were conducted based on facet scores, each facet including four items. The
EFA for all the investigated samples was conducted with a maximum likelihood (ML)
extraction and Oblimin rotation. Due to the normality assumption for ML-based factor
analyses, we also conducted principal axis factor analyses (the latter approach does not have
solutions were retrieved by using the principal axis factoring extraction method. We chose to
approach the factor retention issue by deploying a parallel analysis based on a Monte Carlo
data simulation method (Horn, 1965). In the Monte Carlo simulation, we generated 1,000
The 95th percentile cutoff values for the eigenvalues were included in Table 2.
The EFA-derived six factors displayed eigenvalues greater than the ones obtained via
the simulated results for India, Indonesia, Romania, and Thailand. The Honesty-Humility
component had slightly lower eigenvalues compared to the ones obtained via parallel analysis
(PA) at a 95% confidence interval for India and Oman. Conscientiousness displayed lower
EFA-derived eigenvalues for Oman. The eigenvalues associated with Openness were slightly
lower than those generated via PA for Oman and Thailand. The eigenvalues obtained via
EFA for Emotionality and Extraversion are higher than those generated via PA for all the
investigated groups.
We used target (Procrustes) rotation and the resulting Tucker congruence index to
determine the factor congruence between each of the investigated groups and the original
HEXACO target structure. Tucker’s index is considered an accurate test of replication, and
congruence scores of .90 and above indicate a strong fit to the target structure (McCrae et al.,
PERSONALITY HEXACO MODEL 13
1996). However, congruence indexes are insensitive to the magnitude of loadings. To ensure
that congruence indexes reflect similar rather than opposite loadings, we also computed
Gower’s (1971) similarity coefficient. The similarity coefficient was computed by using a
computation program designed by Barrett (2012). For both analyses, we considered the
HEXACO structure reported by Lee and Ashton (2004) on a Canadian sample of 409
participants as the target structure. The structures obtained via EFA for the five cultures
exhibited a moderate to strong fit with the target structure. Results are presented in Table 2.
The factor solution obtained for the Oman group has the lowest level of congruence with the
target structure (.84), and the solution obtained for the Indian sample has the highest level of
congruence with the target structure (.92). The median congruence coefficient with the six-
factor structure identified by Lee and Ashton (2004) is .88. Gower’s similarity coefficients
ranged between .74 and .91, with the lowest median similarity of .81 (Oman).
For ESEM we used Mplus version 7 (Muthen & Muthen, 2008). The ESEM analyses
were based on the same samples used for the previous analyses and were conducted on
aggregate facet level, each facet including four items. To tackle the effects of potentially
indexes for each model are included in Table 3. We draw attention to the fact that relying on
acceptable or unacceptable goodness of fit (Barrett, 2007). However, the use of fixed
thresholds can lead to labeling misspecified models as acceptable (Beauducel & Wittmann,
CFI of .90 rather than one of .95” (Barrett, 2007, p. 819). The interpretation of fit indexes in
ESEM is based on the same cutoff values as in CFA, values higher than .90 for comparative
fit index (CFI) and lower than .05 for root mean square error of approximation (RMSEA)
indexes. The highest fit indexes were obtained for the Indonesian sample (CFI = .930,
RMSEA = .043) and for the Romanian sample (CFI = .930, RMSEA = .047). The Hindi and
Thai samples exhibited poorer fit indexes: CFI = .906, RMSEA = .053, and CFI = .888,
To accurately interpret the results of an ESEM analysis, the pattern of item and facet
loadings on each factor needed to be examined. Table 4 includes the facet loadings for each
of the five investigated cultures. The geomin loadings across the five cultures reveal two
relationship, presenting the pattern of cross-loadings between facets and all underlying
factors. Given the nonzero relationships between different higher order factors, the
expectation for all the facets to load on a single higher order dimension is less legitimate. Our
results indicate that crossloadings, which decrease the goodness-of-fit indexes in a CFA
For example, even for facets that clearly belong to a specific factor (e.g., the greed
avoidance facet, which loads .65 on the Honesty-Humility dimensions in the Romanian
sample), there are significant secondary loadings (e.g., a –.30 loading on the Emotionality
dimension). In CFA, when not specified, this kind of cross-loading leads to a decrease in
goodness of fit. Instead, in ESEM, such cross-loadings are tolerated, as they could reflect
PERSONALITY HEXACO MODEL 15
between-group differences in the way in which different personality constructs are related to
one another, and they do not artificially decrease the goodness-of-fit indexes. Second, the
examination of loadings across the six factors reveals that there is support for the six-factor
model in India, Indonesia, Romania, and Thailand. The analysis of the loading pattern for
Oman indicates that the Conscientiousness facets tend to “migrate” on other components,
to identify which correlated errors exert a significant impact on goodness of fit. Therefore, we
selected for each analysis those modification indexes that exerted a significant impact on
model fit. As shown in Table 3, the inclusion of facet correlated errors resulted in well-fitting
baseline models for each group. The highest fit indexes were obtained for the Indian sample
(CFI = .965, RMSEA = .032) and for the Indonesian sample (CFI = .953, RMSEA = .035).
For the Romanian and Thailand groups, the fit indexes were slightly below the .95 threshold:
CFI = .943, RMSEA = .043, and CFI = .949, RMSEA = .043, respectively. Even after
specifying three correlated errors, the goodness of fit for the Oman group was lower than
those estimated for the other groups, CFI = .931, RMSEA = .043.
deployment of additional measurement invariance analyses for the Romanian and Indonesian
groups. Because this study is pioneering the use of ESEM in cross-cultural personality
research, we chose to include in the invariance analyses the groups that exhibited slightly
Measurement invariance
To identify whether the HEXACO model functions in a similar way across the five
several ESEM-based models. Table 5 displays the fit indexes for the five group invariance
ESEM analyses. We tested a set of four different models, ranging from configural invariance
to strict factorial invariance. The first model included in the analysis is the least restrictive,
having no imposed invariance constraints. This model exhibited an acceptable goodness of fit
(CFI = .931, RMSEA = .047). The second model tested if the factor loadings for the six-
factor structures are the same across the five investigated groups; this model is also called
decrease of goodness-of-fit indexes (CFI = .852, RMSEA = .055). The third and fourth
models involved imposing additional constraints on factor loadings and item intercepts
(strong measurement invariance) and factor loadings, item intercepts and item uniqueness
(strict measurement invariance), respectively. The ESEM-based invariance analysis for these
two models resulted in an unacceptably poor goodness of fit: CFI = .666, RMSEA = .079 for
strong and CFI = .644, RMSEA = .079 for strict measurement invariance.
Discussion
This investigation is centered on two core issues. First, we investigated whether the
six -factor structure proposed by the HEXACO framework is replicable across five culturally
distinct groups. Second, we explored the utility of ESEM as a data analysis technique in the
cross-cultural analysis of personality and compared it with more traditional approaches based
The degree to which different personality trait frameworks are replicable across
diverse social, religious, or cultural groups represents a substantive issue, having important
practical and theoretical implications for personality research. In this study we used two
different data-analytic frameworks, EFA and ESEM, in an effort to estimate the extent to
PERSONALITY HEXACO MODEL 17
which the six-factor personality framework can be considered replicable across five diverse
cultural groups and across different analytical frameworks. The results point toward the
When using EFA, the six-factor solution predicted by the HEXACO framework was
extracted across all the five investigated cultures. The percentage of variance explained by
the six-factor solution prior to factor rotation ranged between 37.8% (Indonesia) and 40.9%
(India), and the average percentage of explained variance across the five samples was 38.9%.
We then showed in the Procrustes analysis and Gower’s similarity analysis that the structures
emerging in the five different countries are similar to the one obtained on a Canadian sample,
as part of the first published article addressing the psychometric characteristics of the
HEXACO (Lee & Ashton, 2004). EFA results obtained in this study are strikingly similar to
those commonly reported for the FFM (Aluja et al., 2005; McCrae et al., 1996).
The ESEM solutions displayed an acceptable fit to the data. This tendency was
observed across all the investigated groups. The optimal goodness of fit obtained for each
group can be partly explained by the less restrictive assumptions on which ESEM rests, each
facet or observed variable being allowed to load other factors beside the target factor to
which it theoretically belongs. Even if the six-factor personality model is far from being
equivalent across the five investigated cultures, the invariance analyses suggest that the
Raad et al., 2010) also emerged in our investigation, as the Honesty-Humility component was
not retrieved in all the cultures as a stand-alone dimension. Overall, the results indicate that
the pattern of primary and secondary factor loadings does not vary to a great extent across the
The inclusion of correlated errors resulted in improvements in model fit statistics for
each cultural group. For the Indian sample, specifying correlated errors between the modesty
A significant yet insufficient improvement in goodness of fit was obtained for the Oman
Romanian sample, the inclusion of correlated errors between fairness (Honesty-Humility) and
similar approach was used to obtain a better fitting model for the Thailand group. For this
group the inclusion of three correlated errors generated a significant gain in goodness of fit.
The three correlated errors are sincerity (Honesty-Humility) and fairness (Honesty-Humility),
and diligence (Conscientiousness). Even though different facets, some from the same domain
and some from different domains, were thus correlated for each culture, correlated errors
group. These correlated errors suggest that within the sample of reference there is insufficient
factor models
PERSONALITY HEXACO MODEL 19
The EFA solutions described earlier are very similar to the solutions usually reported
for the FFM (e.g., Aluja et al., 2005; Costa & McCrae, 1992; Hendriks et al., 2003). To
illustrate the similarity of the EFA results for the two competing models, we compare our
results with the ones obtained by Hendriks et al. (2003) in a study that investigated the
functioning of the FiveFactor Personality Inventory (FFPI) across 13 cultures. The average
percentage of explained variance prior to applying rotation obtained in this research is similar
to that reported by Hendriks et al.: The five-factor solution retrieved by Hendriks et al.
reported an average proportion of explained variance equaling 38.1%, the lowest percentage
of variance being encountered in Slovakia (32.1%) and the highest in Germany (49.7%). Our
results indicate that for the five diverse cultures, the proportion of explained variance
obtained prior to factor rotation ranged between 37.0% and 40.9%. In the Hendriks et al.
study, the average congruence between the solutions obtained in 13 countries and the U.S.
solution was .86, and in our study the average congruence is .88. The analyses conducted by
Hendriks et al. led to the rightful conclusion that the FFPI can be considered “a valid and
(Hendriks et al., 2003, p. 362). We consider that the results presented in this research support
As ESEM is a relatively new data analytical framework, there are relatively few
available research papers reporting ESEM results for the FFM. However, the studies
conducted so far show results comparable to those obtained in this research (e.g., Furnham et
al., 2013; Marsh et al., 2010). For example, using the NEO–FFI and data collected from a
large German sample, Marsh et al. (2010) obtained fit indexes that are similar to the ones
In line with the results and their comparison with other reported findings in the
investigated groups.
of fit for the six-factor framework across all the investigated cultural groups.
3. Most of the correlated errors that exerted a negative influence over the goodness
4. Deployment of EFA and ESEM on the same data set leads to convergent
groups.
5. The six-factor personality traits framework yielded results comparable to the ones
typically exhibited by the other celebrated frameworks across the two data
analytical methods.
Thus, the findings of this research provide evidence in favour of the replicability of an
alternative, non-FFM personality traits framework, the HEXACO, both by EFA and ESEM
standards across five different cultural groups. Although this framework does not keep up
with the popularity of the FFM, the findings reported in this research indicate that both the
five- and six-factor personality traits frameworks exhibit a similar functioning across
cultures. Although the usage of fit indexes derived either via ESEM or via CFA is on its own
not sufficient evidence to advocate the universality of different personality traits models, this
research demonstrates that there are other, non-FFM personality traits frameworks that
exhibit an acceptable goodness of fit across religiously and linguistically diverse groups.
However, the Honesty-Humility component was the most problematic component, presenting
In our invariance analyses, the 24 facets tended to load on the same factors across the
constraints on facet loadings (weak invariance), on facet loadings and intercepts (strong
significant decay in goodness of fit. This indicates that the functioning of the HEXACO
across the five investigated groups differs with respect to the parameters presented earlier.
This failure to meet the standards imposed in the invariance analyses can be explained by the
both. Regardless of the cause, these issues are signaled by inequivalent parameters (loadings,
successfully retrieved across the five groups, it differed significantly with respect to facet
loadings, intercepts, and uniqueness. Further research is needed to understand the source of
these differences. In this research, the samples were uneven in both composition and size. In
the absence of matched samples, we cannot assert whether failure to meet the invariance
standard stems from true construct and construct measurement differences or sample-related
confounding variables.
ESEM proves to be a versatile data analytical framework, and its application in the
CFA. As shown in this research, ESEM effectively overcomes the measurement invariance
ESEM permits the comparison of competing personality traits models across the same data
set, in less restrictive conditions. ESEM can be used as an alternative to CFA in testing
different levels of measurement invariance. This is an important advantage that ESEM has
over the standard CFA approach. We consider that ESEM could be used as an extension to
the analyses conducted with different personality traits measures across large, cross-national
studies.
PERSONALITY HEXACO MODEL 22
Results are encouraging especially because the HEXACO model is more complex
than the FFM. If more complex models can now be addressed in their cross-cultural
invariance, this could help to address the replicability issue in a more robust manner than was
previously possible.
However, obtaining acceptable fit indexes across different cultures does not fully
consider that the universality issue should be tackled both from a methodological perspective
that typically focuses on the replicability of personality traits framework, but also, and more
important, from the perspective of the theoretical foundations and antecedents of those
personality traits frameworks that prove to be replicable. We consider that more research into
the foundations of hypothesized traits is needed in the case of such models as the HEXACO.
Limitations
Several limitations of this research should be noted. First, most of the limitations of
this research stem from the structure and sizes of the samples. The volume and structure of
the samples differ from one culture to another. To rule out these potential sources of error,
across gender, age, or other covariates that can induce a systematic bias. Although the
samples are sufficiently large to permit the computation of EFA or ESEM, between-group
analyses are likely to be biased by the differences in sample size and structure. One way of
tackling this issue is to create matched samples. However, due to uneven sample sizes and
structures, creating matched samples would yield unacceptably small sample sizes.
Moreover, the samples are far too small to be considered nationally representative, thus
hindering the generalization of this study’s findings. To reliably tap into country-level
equivalent psychometric instruments, but also on groups that can be considered nationally
representative. Although tempting from an exploratory perspective, we did not compare the
constructs across the five groups via mean difference tests. Before deploying such
comparison tests, several issues must be considered: (a) the comparability of the investigated
groups (across age, gender, and educational level), (b) invariance of the measurement
instrument across the investigated groups, and (c) nationally representative samples must be
personality data need to carefully consider these constraints before comparing personality
Second, this investigation was not based on item-level analyses, but on four-item facet
scores. Although not uncommon in personality research (e.g., Marsh, Trautwein, Ludtke, K €
€oller, & Baumert, 2006), using aggregate scores instead of individual items does not permit
the investigation of potential differential item functioning. Moreover, when not conducting
drawbacks entailed by aggregate-level analyses are that they do not permit a clear
identification of the sources of nonequivalence and they can yield potentially inflated factor
personality traits structures was not and will not be solved via cross-sectional research
designs, regardless of the magnitude of the samples or number of cultures included in the
analysis. Even if a certain personality traits structure would show replicable results across
different cultures, asserting the universality of that specific structure would not be
assumption. It implies that there are some factors that determine the structure of personality
traits in all humans, regardless of social and cultural environments. Future research should
attempt to tap more profoundly into the biological foundations of those universally replicable
personality traits structures via research designs that permit the investigation of causality
assumptions.
Acknowledgment
This article is the result of an Advanced Research and Training Seminar organized at
the 27th International Congress of Applied Psychology (July 11–16, 2010, Melbourne,
Applied Psychology, and International Association for Cross-Cultural Psychology, and led by
References
Aghababaei, N., Wasserman, J. A., & Nannini, D. (2014). The religious person revisited:
Almagor, M., Tellegen, A., & Waller, N. G. (1995). The big seven model: A cross-cultural
replication and further exploration of the basic dimensions of natural language trait
Aluja, A., Garcia, O., Garcia, L. F., & Seisdedos, N. (2005). Invariance of the “NEO–PI–R”
Asparouhov, T., & Muthen, B. (2009). Exploratory structural equation modeling. Structural
Ashton, M. C., & Lee, K. (2001). A theoretical basis for the major dimensions of personality.
Ashton, M. C., & Lee, K. (2007). Empirical, theoretical, and practical advantages of the
11, 150–166.
Barrett, P. T. (2007). Structural equation modelling: Adjudging model fit. Personality and
http://www.pbarrett.net/Gower/Gower.html
Beauducel, A., & Wittmann, W. W. (2005). Simulation study on fit indices in confirmatory
factor analysis based on data with slightly distorted simple structure. Structural
Benet, V., & Waller, N. G. (1995). The “Big Seven” model of personality description:
Bentler, P. M., & Kano, Y. (1990). On the equivalence of factors and components.
Block, J. (2010). The five factor framing of personality and beyond: Some ruminations.
Borkenau, P., & Ostendorf, F. (1990). Comparing exploratory and confirmatory factor
analysis: A study on the five factor model of personality. Personality and Individual
Cheung, G. W., & Rensvold, R. B. (2000). Assessing extreme and acquiescence response sets
Church, T. A., & Burke, P. J. (1994). Exploratory and confirmatory tests of the big five and
Conway, J. M., & Huffcut, A. I. (2003). A review and evaluation of exploratory factor
6,147–168.
Costa, P. T., & McCrae, R. R. (1992). Revised NEO Personality Inventory (NEO PI–R) and
Costello, A. B., & Osborne, J. W. (2005). Best practices in exploratory factor analysis: Four
recommendations for getting the most from your analysis. Practical Assessment,
Cudeck, R., & MacCallum, R. C. (Eds.). (2007). Factor analysis at 100: Historical
De Fruyt, F., De Bolle, M., McCrae, R. R., Terraciano, A., & Costa, P. T. (2009). Assessing
the universal structure of personality in early adolescence: The NEO–PI–R and NEO–
De Raad, B., Barelds, D. P. H., Levert, E., Ostendorf, F., Mlacic, B., Di Blas, L., …
De Raad, B., Perugini, M., Hrebickova, M., & Szarota, P. (1998). Lingua franca of
De Vries, A., De Vries, R. E., & Born, M. P. (2011). Broad versus narrow traits:
Fabrigar, L. R., Wegener, D. T., MacCallum, R. C., & Strahan, E. J. (1999). Evaluating the
Methods, 4, 272–299.
Furnham, A., Guenole, N., Levine, S. J., & Chamorro-Premuzic, T. (2013). The NEO–PI–R:
Factor structure and gender invariance from exploratory structural equation modeling
Gower, G. C. (1971). A general coefficient for similarity and some of its properties.
Gurven, M., von Rueden, C., Massenkoff, M., Kaplan, H., & Lero Vie, M. (2013). How
universal is the Big Five? Testing the Five-Factor Model of personality variation
Hambleton, R. K. (2005). Issues, designs, and technical guidelines for adapting tests in
Hambleton, R. K., & Zenisky, A. L. (2011). Translating and adapting tests for cross-cultural
Hendriks, A. A. J., Perugini, M., Angleitner, A., Ostendorf, F., Johnson, J. A., De Fruyt, F.,
Horn, J. L. (1965). A rationale and test for the number of factors in factor analysis.
International Test Commission (2010). Guidelines for translating and adapting tests.
Johnson, M. K., Rowatt, W. C., & Perrini, L. (2011). A new trait on the market: Honesty
Lee, K., & Ashton, M. C. (2004). Psychometric properties of the HEXACO personality
Lee, K., Ashton, M. C., & De Vries, R. E. (2005). Predicting workplace delinquency and
integrity with the HEXACO and five-factor models of personality structure. Human
Lee, K., Ashton, M. C., Morrison, D. L., Cordery, D., & Dunlop, P. D. (2008). Predicting
integrity with the HEXACO personality model: Use of self- and observer reports.
Lee, K., Ashton, M. C., & Shin, K. H. (2005). Personality correlates of workplace anti-social
Little, T. D. (1997). Mean and covariance structures (MACS) analyses of cross-cultural data:
Marcus, B., Ashton, M. C., & Lee, K. (2013). A note on the incremental validity of integrity
Marsh, H. W., & Hau, K.-T. (2007). Applications of latent-variable models in educational
Marsh, H. W., Hau, K. T., & Wen, Z. (2004). In search of golden rules: Comment on
Marsh, H. W., Ludtke, O., Muth € en, B., Asparouhov, T., Morin, A. J. S., Trautwein, U., &
Nagengast, B. (2010). A new look at the big-five factor structure through exploratory
Marsh, H. W., Nagengast, B., & Morin, A. J. S. (2013). Measurement invariance of Big-Five
factor structure over the life span: Exploratory structural equation modeling tests of
PERSONALITY HEXACO MODEL 30
gender, age, plasticity, maturity and la dolce vita effects. Developmental Psychology,
49, 1194–1218.
Marsh, H. W., Trautwein, U., Ludtke, O., K € €oller, O., & Baumert, J. (2006). Integration of
McCrae, R. R., Terracciano, A., & Personality Profiles of Cultures Project. (2005).
McCrae, R. R., Zonderman, A. B., Costa, P. T., Bond, M. H., & Paunonen, S. V. (1996).
Musek, J. (2007). A general factor of personality: Evidence for the big one in the five-factor
Muthen, L. K., & Muthen, B. (2008). Mplus user’s guide. Los Angeles, CA: Muthen &
Muthen.
Nel, J. A., Valchev, V. H., Rothmann, S., Van de Vijver, F. J. R., Meiring, D., & De Bruin,
Nye, C. D, Roberts, B. V., Saucier, G., & Zhou, X. (2008). Testing the measurement
O’Neill, T. A., Lewis, R. J., & Carswell, J. J. (2011). Employee personality, justice
Parker, J. D. A., Bagby, R. M., & Summerfeldt, L. J. (1993). Confirmatory factor analysis of
the revised NEO Personality Inventory. Personality and Individual Differences, 15,
463–466.
Schmitt, D. P., Allik, J., McCrae, R. R., Benet-Martınez, V., Alcalay, L., & Ault, L. (2007).
The geographic distribution of Big Five personality traits: Patterns and profiles of
173–212.
Schmitt, N., & Kuljanin, G. (2008). Measurement invariance: Review of practice and
Tellegen, A., & Walker, N. G. (1987, August). Reexamining basic concepts of personality
and personality disorder. Paper presented at the 95th annual meeting of the American
Thalmayer, A. G., Saucier, G., & Eigenhuis, A. (2011). Comparative validity of brief to
medium length big five and big six personality questionnaires. Psychological
Van de Vijver, F. J. R., & Leung, K. (1997). Methods and data analysis for cross-cultural
Vandenburg, R. J., & Lance, C. E. (2000). A review and synthesis of the measurement
Wasti, S. A., Lee, K., Ashton, M. C., & Somer, O. (2008). The Turkish personality lexicon
665–684.
Yuan, K.-H. (2005). Fit indices versus test statistics. Multivariate Behavioral Research, 40,
115–148.
PERSONALITY HEXACO MODEL 32
Honesty-Humility 1.21 (1.29) .91 .88 1.21 (1.19) .91 .91 1.13 (1.30) .86 .78 1.54 (1.23) .85 .84 1.32 (1.33) .85 .88
Emotionality 2.83 (1.55) .92 .90 2.08 (1.35) .94 .87 1.95 (1.49) .87 .82 2.17 (1.37) .91 .89 1.86 (1.54) .91 .87
Extraversion 3.53 (1.67) .94 .90 3.41 (1.49) .91 .89 4.59 (1.70) .82 .74 3.45 (1.53) .88 .90 4.18 (1.78) .81 .86
Agreeableness 2.83 (1.55) .92 .89 2.89 (1.40) .92 .86 2.29 (1.58) .90 .89 2.33 (1.44) .82 .84 2.22 (1.63) .91 .88
Conscientiousness 2.01 (1.40) .92 .89 1.80 (1.30) .90 .89 1.36 (1.42) .78 .73 1.95 (1.32) .92 .91 1.49 (1.46) .79 .80
Openness 1.57 (1.34) .91 .89 1.25 (1.24) .93 .88 1.24 (1.35) .88 .86 1.64 (1.27) .88 .76 1.38 (1.39) .86 .87
*The results of the parallel analysis are presented in brackets. TCI = Tucker’s Congruence Index; GSI = Gower’s Similarity Coefficient.
PERSONALITY HEXACO MODEL 34
.02 .05 .07 -.01 .10 .09 .02 -.02 -.02 .00 .06 .04 .23 .00 .00 -.10 -.09 -.01 .01 .00 .34 .43 -.02 .43 .30 .01 .00 .18 -.01 .20
Diligence
-.01 -.03 -.04 .01 -.03 .07 .01 .01 .01 .01 -.01 -.22 .27 -.01 .00 -.01 .00 -.01 -.01 .00 .41 .48 .12 .38 .37 .11 .01 .30 .00 .11
Perfectionism
.02 .10 -.01 -.02 .05 -.10 -.09 .03 -.02 .00 .00 .02 .24 .03 .11 .14 .03 .14 -.01 .10 .40 .34 .01 .44 .31 -.00 .07 .02 .01 .00
Prudence
.01 .07 .01 -.01 .01 .05 .02 .16 -.03 .00 -.07 -.03 -.02 -.01 .00 .08 .02 .11 .01 .17 .02 -.02 -.03 -.01 .00 .38 .46 .37 .73 .41
Aesthetic App.*
.02 .01 -.01 .01 -.01 -.18 -.16 .00 -.01 -.20 -.02 -.04 .03 -.05 .00 .02 .04 .09 -.01 .04 .14 .09 -.11 -.01 .15 .30 .30 .32 .32 .12
Inquisitiveness
-.00 -.06 .05 .02 -.05 .03 .05 .06 .05 .00 .18 .18 .06 .04 .12 -.00 .00 -.04 .04 .05 -.01 .00 -.02 .06 .03 .58 .39 .48 .37 .39
Creativity
.18 -.05 .00 .01 -.03 -.02 -.15 -.21 .03 .00 .01 .10 -.12 .03 .01 -.16 -.07 -.01 -.06 .00 -.09 .03 .03 -.02 .00 .25 .23 .40 .28 .20
Unconventionality
Note: The different countries are noted as 1 (India) , 2 (Indonesia), 3 (Oman), 4 (Romania) and 5 (Thailand). *Aesthetic Appreciation.
PERSONALITY HEXACO MODEL 37