Pty Across Cultures

7/28/2019 Pty Across Cultures

1/21

ALLIKPERSONALITYDIMENSIONSACROSSCULTURES

PERSONALITY DIMENSIONS ACROSS CULTURES

Jri Allik, PhD

In order to generalizethedimensional structureof personalityrelativelyindependent groups of covarying traitsacross languages and cultures,a large number of cultures must be studied. Until recently only a fewworldwide personality datasets have been available. The first largescalestudies indicate that thepattern of covariationbetween personality traitsis universal and is relatively easily generalizable across languages andcultures. In contrast to the structure of personality, the comparison ofthemean trait scoresacrosscultures is much more problematic becausecrossculturaldifferences turnedout tobevery small in their magnitude,

about onethird of the magnitude of individual differences within cul-ture. More integral (e.g., the similarity between personality profiles) orsubtle (e.g., thedisparitybetween positively andnegativelyworded items)measures can reveal more systematic relationships with relevant socio-economic and geographic variables than the mean scores themselves.Relatively modest sizes of crosscultural differences in the mean valuesmay imply that a reasonable scalar equivalence can be achieved, and allindividuals, irrespective of their language and culture, can berepresented in a common metric.

Many popular psychological assessment instruments, originally developed

in English, have been translated into numerous languages and are nowcommonly usedthroughout theworld.Most of these translationswere made

with an explicit or at least tacit assumption that the core psychological con-structs assessed by themeasures substantively transcendhuman languageand culture. The generalizability across languages and cultures, however,

was in most cases presumed, not demonstrated. Therefore it is not surpris-ing that some researchers have expressed concern with this assumption,and especially with practices guided by this assumption (Shweder, 1991).Skeptics have questioned, for instance, whether the uncritical extension of

Western ways of thinking to the rest of the world should serve as standardprocedure in psychological science. From a cultural constructionist point of

view, all personality models are based on conceptions of personhood and

Journal of Personality Disorders, 19(3), 212-232, 2005

2005 The Guilford Press

212

From the Department of Psychology, University of Tartu, Estonia, and The Estonian Center ofBehavioral and Health Sciences.

The author thanks Jeff McCrae and Dave Schmitt for discussions, collaboration, and sharingtheir data.

The preparationof this article was supportedby the EstonianScience Foundation and the Es-tonian Ministry of Science and Education.

Address correspondence to Jri Allik, Department of Psychology, University of Tartu, Tiigi 78,Tartu 50410, Estonia; E-mail: [email protected].


2/21

standards of culturally appropriate behavior that have evolved, as a rule, inAngloAmerican societies. It may be even more so in regard to personalitydisorders which are, as some researchers believe, quintessential cultural

products that owetheirmeaning anddescriptive content toa distinctive cul-tural tradition (Fabrega, 1994). It is also claimed that studies of mental

health are culturally insensitive partly due to the uncritical transfer of toolsand concepts from one language and culture, typically English and theUnited States, to other countries (Rogler, 1999).

Eysenck was one of the pioneers who started to think that the pattern of

covariationbetweenpersonalitytraitsthepersonalitystructuremightbe

universal and rises above the diversity of human cultures (cf. Allik, 2002).

One of the mainreasons for this conclusion was his study of twins thatdem-

onstrateda considerable genetic impact on all personality traits and almost

no influence from the shared environment (Eysenck, 1990). In 1997, Mc-

Crae and Costa made a bold proposal about the crosscultural

generalizability of the FiveFactor Model of Personality. By studying sixtranslations of the Revised NEO Personality Inventory (Costa & McCrae,

1992) into German, Portuguese, Hebrew, Chinese, Korean, and Japanese

they observed that all translations showed similar structures after varimax

rotation of five factors and almost identical factor structure when the

varimax solution was targeted toward the original American factor struc-

ture. Forsome reason, people from different places around the world are in-

clined to think that individuals who, for example, talk a lot are at the same

time optimistic and very happy, or those who often get disgusted with other

people are also inclined to feel inferior to others. Because the samples stud-

ied represented highly diverse cultures with languages from five distinct

language families, McCrae and Costa proposed that personality trait struc-

ture is universal (McCrae & Costa, 1997). The proposal that the personalitystructure is largely or even completely independent from the culture (Mc-

Crae& Costa,1996; McCrae & Costa, 1999) was ratherunorthodoxand was

expectedly met with skepticism and denial (Allik & McCrae, 2004a;

Toomela, 2003). It is also important to notice that although the replicability

of the pattern of covariations across cultures turned out to be a rather ro-

bust phenomenon (cf. Rolland, 2002), this fact does not necessarily imply

that all of the finest details are always replicated in every new culture.

Universality by itself is a relative concept implying that something must be

variable as well; it is not an allornothing phenomenon, but a degree of

invariance.

PROBLEMS IN COMPARING PERSONALITY TRAIT SCORES

ACROSS CULTURES

For psychologists seeking to investigate personality traits across cultures,

oneof themore inconvenientproblems hascentered on whether personalitytrait scales possess conceptual equivalence across cultures or not (van de

Vijver & Leung, 1997; van de Vijver & Leung, 2000). It has been particularlytroublesome to establish whether the mean scores across different cultures

PERSONALITY DIMENSIONS ACROSS CULTURES 213


3/21

show metric or scalar equivalence (Byrne & Campbell, 1999; Little, 2000).That is, when comparing the mean scores of different cultures on a person-ality trait scale, any observed differences may be due not only to a real cul-

tural disparity on some personality trait, but also to inappropriatetranslations, biased sampling, or the nonidentical response styles of people

from different cultures. All of these factors can be difficult to control fully,makingsome methodologists extremely skeptical about achieving true met-ric comparability of scores on the same test in different languages or cul-tures (Bijnen& Poortinga,1988; Heine,Lehman,Peng, & Greenholtz, 2002;

van de Vijver & Leung, 1997). Although much of this skepticism is certainlywarranted, a recent, largescale intercultural comparison provided evi-dence that personality scales may be more robust than methodologists hadinitially presumed (McCrae, 2001; McCrae, 2002). Perhaps because the

many sources of error cancel out, it is possible to obtain meaningful resultswhen scores are compared across cultures (Allik & McCrae, 2002).

Among the more common methods for establishing the crossculturalcomparability of personality trait measures is first to show that the trait

scales contained in the measures are internally reliable across all targeted

languages and cultures. A second frequently employed technique is to dem-

onstrate a high degree of factorial structure invariance across different lin-

guistic and cultural contexts (Barrett, Petrides, Eysenck, & Eysenck, 1998;

Caprara, Barbaranelli, Bermudez, Maslach, & Ruch, 2000; Rolland, 2002).

Metric equivalence can also be established through differential item analy-

sis and bilingual administrations. The number of bilingual studies is sur-

prisingly small, however, and theses studies rarely demonstrate consistent

differences between languages (e.g.,Konstabel, 1999).However, if the ques-

tionnaire format is replaced with autobiographical narratives, then indeed

language may, to some extent, dictate how individuals describe themselves(Marian & Kaushanskaya, 2004). Historically, if trait scales from a person-

ality measure showed high internal reliability, invariant factor structure,

and item equivalence across different languages and cultures, comparing

the mean scores across cultures was often deemed a reasonable next step

(Steel & Ones, 2002; van de Vijver & Leung, 2001). However, even with evi-

dence of reliability, factor invariance, and item equivalence, problems can

remain in how to metrically interpret meanlevel differences in personality

traits across cultures. For example, it is possible that people from different

countries have different internal standardsandmake their judgments in the

relation to these standards (cf. Heine et al., 2002).

Another way to increase confidence in the crosscultural comparability of

personalitymeasures is to show that themean levels of differentassessment

instruments that intend to measure the same construct, or approximately

the same construct, are highly correlated across multiple languages or cul-

tures. For example, if two conceptually similar personality trait scales are

used in a large number of different cultures, a positive association between

the mean levels of those trait scales across the broad set of cultures would

provide evidence that both measures are tapping the same underlying con-

struct (Campbell& Fiske,1959).Of course, inorder toanalyze the compara-

214 ALLIK


4/21

bility of personality measures using this crosscultural convergent

validation strategy, large numbers of cultures must be studied using

conceptually similar measures of personality.

LARGESCALE CROSSCULTURAL STUDIES

There have been many studies in which pairs of cultures have been com-pared on personality measures (cf. Katigbak, Church, & Akamine, 1996; cf.Katigbak, Church, GuanzonLapena, Carlota, & del Pilar, 2002), but there

have only been only a few in which a sufficiently broad sample of cultureswas examined. Although many personality instruments like the Tempera-ment and Character Inventory (Cloninger, 2004) have successfully trans-lated into different languages, only few worldwide personality datasets have

been available for statistically meaningful comparisons to be made.

One of the first comprehensive personality trait measures to enjoy world-

wide popularity and a fairly large number of translations into different lan-guages was Eysencks Personality Questionnaire (EPQ; Eysenck &

Eysenck, 1975). In 1984, meanlevel trait scores from 25 countries were

made available (Barrett & Eysenck, 1984). Ten years later, the number of

countries in which three broad personality traitsNeuroticism,

Extraversion, andPsychoticismwere measured by theEPQwasexpanded

to 37 (Lynn & Martin, 1995). Soon it was found that the internal reliability

and factorial structure of the EPQ across languages and cultures appeared

to be replicable (Barrett et al., 1998). Although evidence of the crosscul-

tural generalizability of the EPQseemed overwhelming, there was still some

room for reservations. Some of these doubts were met by van Hemert and

colleagues (van Hemert, vande Vijver, Poortinga, & Georgas, 2002) whocrit-

ically reanalyzed available EPQ data both at the individual and the countrylevel. Many previous studies were dropped because of insufficient informa-

tion (e.g., the number of women and men was not specified). The final set

contained studies in 38 countries with a total of 68,374 respondents. Using

the multilevel factor analysis, they found that the EPQ has a different facto-

rial structure at both levels. In some countries (China, India, Japan, and

Uganda) the pattern of correlation between EPQ scales, differed from those

found elsewhere. From the exploration of the meaning of the EPQ scales,

countrylevel correlations were found with a variety of country characteris-

tics, such as Hofstedes measures of cultural differences and Dieners sub-

jective wellbeing. A disappointing result of this analysis was that Lynn and

Martins (1995) findings were not replicated, as no expected correlations

were found, for example, between extraversion and the death rate.

However, because no other large personality datasets were available for

comparison, it remained unclear as to whether meanlevel differences in

EPQ scores across cultures converged with other similar measures. Again,

such crosscultural construct validity evidence would have made it more

likely that national differences in personality, as measured by the three

broad trait scales of the EPQ, were due to real cultural disparities, and not

some other biasing factors.

PERSONALITY DIMENSIONS ACROSS CULTURES 215


5/21

The most comprehensive instrument thus far designed to measure the Big

Five or FFM is the Revised NEO Personality Inventory (NEOPIR; Costa &

McCrae, 1992). Recently, the NEOPIR was translated into many different

languages and administered to samples from over two dozen countries. In2001, NEOPIR data from 26 countries or cultural regions became avail-

able to the research community (McCrae, 2001) and the database was soon

expanded by 10 additional cultures covering 5 major language families:

IndoEuropean, Uralic, Altaic, Dravidian, and SinoTibetian (McCrae,

2002). In every culture and language that has been studied, the trait scales

of the NEOPIR have displayed adequate levels of internal reliability, and

the factorial structure of the NEOPIR has been considered robust

(McCrae, 2001; McCrae, 2002).

Direct comparisons of the NEOPIR with the EPQ have suggested that

translations of both instruments provide reasonably comparable estimates

of mean levels of extraversion and neuroticism across cultures. For exam-

ple, the meanlevel scores of neuroticism and extraversion as measured bytheNEOPIR andtheEPQwere significantly correlated across 18nations,r= .80 and r= .51, respectively (McCrae, 2002). Thus, if a nation scored rela-

tively high on the EPQ Neuroticsm and Extraversion scales, it was likely to

score high on the NEOPIR Neuroticism and Extraversion scales as well.

For this result, however, it was necessary to omit EPQ data from India as a

suspected outlier. These empirical findings, though limited to 18 cultural

regions, can be taken as supportive evidence that at least two of the Big Five

dimensions, neuroticism and extraversion, can be comparably measured

across human languages and cultures.

Recently McCrae and Terracciano initiated a large crosscultural project

to extend findings on the universality of personality traits in several impor-

tant ways (McCrae et al., 2005). Most of crosscultural studies have reliedexclusively on selfreport methods, leaving thepossibility that, forexample,

the selfserving bias distorts findings. For this reason the thirdperson ver-

sion of the NEOPIR was used instead. College students in 50 cultures

identified an adult or collegeage man or woman who they knew well and

rated the11,985self-nominated targetpersons. Factoranalyses withincul-

tures showed that the normative American selfreport structure wasclearly

replicated in most cultures, and was at least recognizable in all 50 cultures.

The otherreport findings replicated some regularities that were identified

on the basis of selfreports. For example, like earlier selfreport results, sex

differences were the most pronounced in Western and less in Eastern and

African cultures (cf. Costa, Terracciano, & McCrae, 2001). Crosssectional

age differences for three factors followed the pattern identified in selfre-

ports, with moderate rates of change during college age and very slow

changes afterage 40. The authors conclude thatwitha few exceptions, their

data support the hypothesis that features of personality traits are common

to all human groups (McCrae et al., 2005).

Perhaps David Schmitt conducted the largest crosscultural study of per-

sonality so far as a part of the International Sexuality Description Project, a

collaborative effort of over 100 social, behavioral, and biological scientists

216 ALLIK


6/21

from 56 countries (Schmitt et al., 2003). All studied samples were adminis-

tered the Big Five Inventory (BFI) of personality traits (BenetMartinez &

John, 1998). The BFI, a short version of the Big Five family of question-

naires, was translated from English into 29 languages and administered to17,837 individuals from 56 nations. Results indicated that the fivedimen-

sional structure of the BFI was highly robust across major regions of the

world (Schmitt, Allik, McCrae, BenetMartnez et al., submitted). However,

the correlationbetweenBFI scales and their counterparts fromthe EPQ and

NEOPIR was disappointingly low. Although the EPQ and the BFI do not

conceptualize Neuroticism and Extraversion in a completely identical way,

it is reasonable to expect at least modest convergent correlations. There

were 26 countries for which the mean scores of Neuroticism and

Extraversion were measured by both the EPQ (Lynn & Martin, 1995; van

Hemert et al., 2002) and the BFI. As expected, the Neuroticism scales of the

two instruments were significantly correlated,r= 0.49,p= .01, but the cor-

relation between the BFI Extraversion scale and its EPQ counterpart wasdisappointingly low and did not reach statistical significance, r= 0.18. The

problems with convergent correlations were not only due to conceptual dif-

ferences between instruments of measure. In some cases, the convergent

correlation was absent when two conceptually similar the Big Five instru-

ments, the BFI and NEOPIR, were compared. There was an overlapping

set of 27 countries (i.e.,Austria, Belgium, Canada, Croatia, Czech Republic,

Estonia, France, Switzerland,Germany, Hong Kong, India, Indonesia, Italy,

Japan, Malaysia, Netherlands, Peru, Philippines, Portugal, South Africa,

South Korea,Spain,Taiwan, Turkey, United States,Serbia, andZimbabwe)

in which the correlations between the BFI and NEOPIR scales were possi-

ble. Although the convergent correlations of BFI and NEOPIR nationlevel

scores on Neuroticism, r= .45, p

Pty Across Cultures

Documents

Transcript of Pty Across Cultures