PERSONALITY PROCESSES AND INDIVIDUAL DIFFERENCES
Universal Features of Personality Traits From the Observer’s Perspective:
Data From 50 Cultures
Robert R, McCrae and Antonio Terracciano
National Institute on Aging, Netional Inettutes of Health,
Department of Health and Human Services
78 Members of the Personality Profiles
of Cultures Project,
“Totest hypotheses about the miversaiy of personality tras, college stents in 5O cultures identified
an adult ce colege aged an of woman whom they knew well and rated the
985 targets using he
3d: person version of the Revised NEO Personality Inventory. Factor analyses within eultores showed
that the normative Amorcan selfrepor sirctare wae clearly replicated in most coltares ard was
recogalzable in all Sex differences replicated eater selfrepet resus, with the mest pronoticed
Aifecoces in Western cules. Cross-sectional age eferences for 3 iciors followed the pater
‘etfed in slf-epons, with moderate rates of change during college age and Slower changes sr age
40, With few exceptions, hese dala support the hypothese that features of personality teste ere
cormon tall burn groupe
Strong claims have recently been made sbout the universality of|
personality traits. McCrae and Coste (1997) argued that the five-
factor model (FFM) of personality is found in all cultures! =
hhypotbesis subsequently supported in a wider range of cultures
Rolland, 2002), McCrae etal. (1999) reported that cross-sectional
age differences were similar in different cultures whose cohorts
hhad experienced very different life histories, and Costa, Terrac-
Robert R. McCrae and Antonio Terracciano, Gerontology Research
‘Centr, National Tastiaze on Aging; the 78 contbutng members of the
Personality Profiles of Cultura Psject.
"The 78 members ofthe Personality Profs of Cultures Project are sted
{in onde of dita subisson in the Appendix.
Robert R. MeCrae receives royals from te Revised NEO Personality
Inventory (NO.PLR). Portion ofthese data were prosanted athe Second
World Coagess on Women's Moni) Health, Washington, DC, March
2004, Coech paticipation was supported by Grant 40/DU/ISD7 from the
‘Grant Ageney of the Czech Republic and is related to Research Plan AV
(027025918 ofthe Insite of Psychology, Academy of Sciences ofthe
(Cuach Republic Seri Gulg’s participation wa supported by the Turkish
Acideniy of Sciences, Buskinahd and French Swiss paiipation was
supposed by a grat ffom the Swiss National Science Foundaton to J.
Rossier. The data collection in Hong Kong was supprted by RC Diret
Allocation Grants DAGO2I03 HSSI4 and DAGORO4HSS14 awarded to
M. Vik Data collection in Malaysia wae suppored by UKM Fandanentl
Research Grant 1ID/015/2003
‘We thank Dave Schmit for contact information on potential collabora-
toss, Bob Smith fr authorizing use of tbe NEO.PI-R, and the tansators
\whose eller work on Form Sof the NEO-PLR me this projet feasb
or asistance oa this project we thank Herbert Biggs, Luciana de Almoids,
Hiaéson W. Carvalho, Marco Montastoyos Calgaro, Andis da Siva Be,
“Zheng Li, Ana Butkovi, Ole Dreyer, Susy Ball, Anna Gramberg, Ho:
nathan Hlavow, V. S. Bose, Suguna Kennan, K. Sarita, K. Mixa
ser
iano, and McCrae (2001) reported pancultural pattems of gender
\ifferences, MeCrae, Costa, Marin et al. (2004) provided data on
‘cross-observer agreement suggesting that even in collectivistic
cultures, where there is purportedly a greater emphasis on rele
tionships than on traits, people accurately perceive their own and
others” traits, These recurring regularities —despte differences in
Tanguage, history, religion, and eulture—soggest that personality
Lidwina Dominica R, Vinx Bucyamin, Hiromi Inte, Kenji Sugiyama,
Midori Takayama, Rezita Kemi, Rosin! Ismail, Anna Nedtwig, Zack
ary Smith, Aaron Wolen, Mays Tamir, Chiste Naps Seollon, Valery
(Oyl, Ivan G. Seni J.C. Munene, ilve Kozel, Manca Jaki, Simona
‘Zistaik, Nadia Messoulan, Facando Abul, Fermenda Melina, Daitna
Bon, Sebastén Mosquera, Ludnila Fispo, Lorena Fichoverry, Femando
Vera, Catherine Curl, Richard Chan, Chistopher Pak, Herbert
roudenthalr, Andes Fink, and Cornelia Hohesbeller.
German, Russian, and Czech daa were taken fam ever ste (Me-
‘Cra, Costs, Bibitkovs, ee, 2004; Ostendort & Angleliner, 2008, ard
prone of the Bralian, Lebanese, and That data aro also reprted in
chapters in MeCree (in pres), McCrae, Tesraccian, and Khouty (i
‘ress and Costa and MeCrae in press, respectively.
Correspondence concerning thistle should be addressed to Robert.
‘MeCeee, Gerontology Research Center, National Intute on Aging, Ne
Vina! Instiues of Health, Departneat of Health and Human Sevies, Box
#03, $600 Nathan Shack Diive, Bakimoro, MID 2224-5825, E-mail:
mecraej@groniasiger
"Throughout this ale, we use the tem cures looely 10 seer 1
‘ations or subgroups within naons, We ae aware tht rations donot eve
monolithic cules and that our samples do not necessary eet te full
calla avery seen within nations,
iaanennessteae sh548 McCRAB BT AL
‘oats are basic features of the human species (AMik & MeCae,
2002), The present study offers new tests of these hypotheses of
sniversality
‘One obvious Imitation to prior claims is that cultures and
subcultures heve not been exhaustively studied. Only 2 few Afti-
‘cen—and no Arsbio—coltues. have been included in previous
studies using the Revised NEO Personality Inventory (NEO-PIR;
Costa & McCrae, 1992s) or other measures of the FFM (eg.,
Heaven, Connors, & Stones, 1994). No preliterate cultures have
been examined; in fact, most studies have used college student
samples, whose members may be relatively Westernized.
‘A second limitation is that most studies have relied exclusively
conselfrepor methods, leaving the possibility that method artifacts
may be responsible for some or all of the findings. Observer
stings fonm an alternative method of personality measurement,
{known to be convergent but not wholly redundant with self-reports
(MoCrae, Costa, Martin, etal, 2004). In American studies, ob-
server ratings typically yield similar conclusions about structure
‘and about age and gender differences (eg, Costa & McCrae,
19926), but this is not invariably the case im eross-caltaral re-
search. For example, Bxtraversion and Openness to Experience
both appeared to decline cros-sectionally in Gorman adults when
self-reports were analyzed but not when observer ratings were
analyzed (MeCrae et al, 2000), In a Czech sample, age associa.
tions found in self-reports wese replicated in ebserver ratings for
Extraversion and Openness bat not for Nevroticism or Agreeable-
ness (McCrae, Costa, Hiebfékov4, etal, 2004).
“The NEO-PER offers two versions: a self-report Form S and an
‘observer rating Focm R, with the same items rephrased inthe third
person, The factor structure of Form R in Americen semples
closely resembles that of Form $ (e.g, Piedmont, 1994), and the
same is true in German, Russian, and Czech-language versions
(McCrae, Coste, Martin, et a, 2004; Ostendorf & Angleitne,
2004). However, there appear to be no published stodies of the
factor strctare of observer rating measures of the FFM in non-
Western cultures. The present article includes data from move than
a dozen,
Past and Present Designs
‘Most previous cross-cultural studies of the FFM were based on
secondary analyses of data collected for a varisty of purposes
(Costa et a.,2001; MeCree, 2002; Rolland, 2002). Samples varied
in size and composition (althoogh only normal volunteer data were
sed) and in the time period of data collection. Jn many cases, only
summary statistics were available, and demographic data were
generally not available. No atempt was made fo assess or control
the quality of the data. It is testimony to the robustness of the
underlying effects that clear regularities emerged despite these
limitaions
In the present study, we collected data from college stodents
‘who were asked to identify an individual from one of four target
groups—-college-aged men, college-aged women, adult men, and
adult women—and provide ratings ofthat target on Form R ef the
NEO-PL-R. Because a uniform approach was taken to data collee-
tion, results are more likely to Be comparable across cultures (ef.
Schwvartz, 1992). Samples are similar in size, age and sex of
targets, and time period in which data were collected. In addition,
item-level data and basic demographics are available for each
sample,
‘The use of college student raters also offers advantages. College
students are not, in general, representative oftheir natiousl popu-
lations, and this is particularly ikely to be true in less affluent
cultures. However, tis fact is less problematic in observer rating
studies than in self-report studies: Raters could choose anyone they
new well as a target, yielding a wider age and educational range
than would normally be abtcined in self-report studies, For exam-
ple, sbout 119% ofthe targets in the present study had fewer than
9 years of education.
In alton, college students may be more familiar and comfort
able with questionnaire methods than members of the general
population, especially in non-Western cultures (cf. Marsclla, Do-
‘banoski, Hamada, & Morse, 2000), yielding more meaningful dats,
However, even college students may differ across cultures in
testsking experience and attiudes; in panicalsr, cultural difer-
fences in acquiescence have been noted (Smith, 2004). In the
present stady, we attempted to assess the quality of the dat in each
sample to compare cultures on data quality, and to take quality
{nto account in interpreting results, t must be stresed that quality
4g primarily a fonction of the fit between an imported Western
personality measure and the experiences and attitudes of each
calture; poorer data quality should not be seen as evidence of
problems with ether the instrument or the respondents but rather
of their mismatch,
Finally the use of observer ratings permits an analysis of certain
aspects of person perception and assessment, When self-reports are
examined, target and rater are completely confounded, making it
impossible to know whether ratings are a function ofthe person
being rated or the person making the ratings. It is possible, for
example, that women everywhere score higher on Neuroicism not
because they are less emotionally stable but merely because they
‘are betier able to perceive negative affect (ef. Feldman Barrett,
Lane, Sechrest, & Schwartz, 2000; Teracciano, Merit, Zender-
‘man, & Evans, 2003) or more willing to abuse it toa target than
‘men are. In the present design both mien and women rato men and
‘women, soit is possible to estimate sex differences in rating styles
cor biases,
Method :
Cultures
‘We resritedeolaboriors fromm a wide rango of colts, subject to the
segusement that prospective participants would be feat in Poglish or one
‘ofthe other langages for which an authorized NEO-PL translation was
available. Collabacstors wore pxmarlyIndivdoals who bed previously
ved the NEO-PLR in their ova research or who had been rembes of
another multinational study (Schmit ete, 2008. To increase repaseas:
sion of Ain and Arabic cultures, we seasched the Inert an Paye-
INFO for personally psychologists from those area, Data gathered so far
are from 50 coltures representing six continents, using anslaiont Into
Indo Horopesn, amito-Semitc, SinoTbetan, Daic, Usalic, Maley
Polynesian, Dravidian, and Alas Tangunges. American end Brazilian dota
were gueied from several sites. Genman, Rosin and Czech data were
‘obtained by selecting targets of the Inended ages from existing observer
rating data (Mere, Coss, Mann ts, 2004; Ostendor de Angleter,
2004). Cates andthe language in which the NEO-PLR was administered
are given inte first two columns of Table 1. As noted in the table, 22 of
these cultures have not previously been studied wsing the NEOPIR,UNIVERSAL FEATURES OF TRAITS 549
Table 1
Characteristics of the Samples
Raters ‘Tegets
Mean age Mean age
Cute Language N Smale yeas) mle gears) —_ducationt
Agent ‘spanish 204 49 216 510 353 24
‘Austalid® lish 206 aL 212 510. 342 24
Austia German’ 158 132 23 505 a3 7
Belgium Femi 29 182 ter Sia 337 20
Botan 16, 312 212 su 460 22
Brazil a7 23 244 313 3 23
Burkina Faso m3 248 3 Bt 20
Canada Enlist 13 253 200 263 299 26
Chit Spanish? 194 237 204 435 343 28
Chine Chinese m a5 2a S18 3s 24
Croatia Cromian 191 35 203 503, 345 25
eech Republic 0 BD an 45 455 22
Denmark 153 138 249 a 385 24
Batosia 23 m9 98 537 337 26
Bihiopia® i772 28 503 345 1s
France m 83 m 493. 373 23
Gemany Gernact 3300 MT 401 263 419 =
Hong Kong Chinese 207 3 20.1 sor 344 23
Ieeland? Teele 199 503 2569 497 356 25
India 185 492 210 503, 33 22
Tdowesi 388 197 495 353 21
Italy 9s 87 234 492 359 24
Japan 191 92 195 503, 34 26
Kawai? 468s 201 466 352 25
Lebanon 2 3RS 89 410 313 23
Malaysia 2 ma 20 239 370 23
Malt 22 267 209 500 336 26
esi? m 80 239 538 aa 26
Moreeeo™ m aid 24 33 330 23
‘New Zealand 2m ons 9 505 345 23
Nigeis" i467 283 49 342 25
Pers 286 213 85 352 26
Prlipines 197 253 153 503, 233 27
Poland 7 97 21 47 349 26
Portugal 8207 247 310 MB 23
Presto Rica BL — Sa 339 28
Rusia 30000 iB 361 481 367 o
Serbia 200 205 2s 505 BAT 26
Slovaks Slovakian 505 201 495 335 23
Slovenia? Slovene 29 7 28 414 356 23
Scat Keres Korean® 196 0s 229 Sis 240. 26
Spain Spanish 200 180 210 500 359 23
Stiverlant Geeman® 2a A 204 ass 373 26
‘Sowkzerland repci® 26s a7 213 «83 365 23
‘Tiland™ Ths 2 BS 36 507 347 27
‘Turkey ‘Tutine 208 442 3 500 341 26
Uganda English 1 446 238 500 353. 23
Unied Kingdom: Englane™ Engst we 7 Ba 490 372 27
Unled Kingdom: Nother Heland™ ——ngih® 10s 3 213 491 333 26
Unites Sates English vis a9 205 196 34 26
‘Note, Dashes in cells indicate data were nt avaiable, NEO-P-R = Revised NEO
Mean lev, where 0 = no edveation, I primary, 2 = secondary, 3 = college. Not inched in previous cross culture! comparisons using the
NEO-PL-R, “Used a ptlished version ofthe NEO-P-R in pateipant native language
Instrument
‘The NEO-PLR is e 240.tom measuse of the FFM. It conains 20
clehtitem fcc sales, sx for each ofthe five Basie personality fcr,
[Neurotiism (8), Extavesion (8), Openness to Expetienes (0), Agree
ableness (A), and Consienionsnes (C) Responses are made on 5-poit
Likert scale, from srongly disagree fo strongly agree. The fctors can be
timated by domain scores, which sum the relevant 6 facet, or more
precisely by factor scores, which ace » weighted combination of al 30,
facets (Costa & McCrae, 19925, Table 2). Two parallel forms have beea
‘developed: Form $ for selfreports and Form R for observer ratings, in
hich the tems have been rephrased in the thed person. Evidence onthe550 McCRAB BY AL.
reliability nd obidity‘of the English version are presented inthe NEO-
PLR manual (Costa & MeCra, 1922). Although the NEO-PLR dees not
Include a socal desiailiy scale Piedmon,, MoCrae, Riemann, &
Angleiner, 2000), it does provide some checks for peeteal validity, ana
protocols deemed invalid have substanlly lower sees sabi (Canter et
a, 2000),
Form S ofthe NBO-PLR hos been translated into over 30 languages
1m almost all cases, wanslaions were done by Uilingual psychologists
aatve tothe culture. An independent back-translation was reviewed by
the test authors, and modifications were made as needed. In some cass,
the translations have been extensively validated and published (eg
Hoekstra, Onnel, & De Pruyt, 1996; Shimonaka, Nakazato, Gondo, &
Takayam, 1999}; In other cases, the translations ean be considered
research instroments. For the present study, collaborators modified the
fire-pereon version to crest third-person version. They elo tans
Iated the instructions, which were reviewed in back-transation by
Robet R. MeCrte and Antonio Terraceiane, Revisions were made
based on these reviews.
Participants, Targets, and Procedures
Parcipnts were ootege stdens? who volimtcere to participate anon
_yoously in a sudy of personality across cultures. The composition of the
“amples by sex and the mean age ofthe raters are given in the thd and
‘ovr colamns of Table {. "The gsea majoiy of raters were native-born
chizens of thee counties and generally reflected the clinic make-up of,
their counses?
[Roters were randomly stsigned to one of four target conditions and
‘were asked for ratings of collegeaged women, cllege-aged men, adult
(over 40) men, or adelt weren, For the colege-aged targets, rites were
ashe the flowing:
Please think of a woman {man aged 18-PF whom you know well
She fhe] should te somsoxe who is 2 native-born elizen of yout
‘country, She he] can be aelative ora fiend orneghbor—romeone
you Uke, ar someano you do not lke. She he) can be a college
‘seat, but she fe] need not be,
Inthe adult conditions the age specified was over age 4010 form a cleat
conta to the collegeaged ares, Raters were then asked to estimate the
age and years of forual education Coane, IB yeas, 9-12 years, over 12
yeas) af he target and to provide demographic information on themselves
before completing the NEO-PLR. Data onthe compositions of the target
samples by fe, their mean age, hd thoi degre af education aro giver ia
the ast thes columns of Table 1
Data Quatity Assessment
‘When insiuments and methods developed and validated in one culture
sre exported to sather, their psyehomesrc propenies may be affect
“That might be due to el diferences in psychological functioning, but it
right alo be die to cultare elated aifets, bless of meaning ray be
lost in translation; response sgles may vary aross cules the tsk of
‘conletng a questionnaire may be unfaniiar and coafusing. Healy, an
fsexsment ofthe quality ofthe dats should be made before substantive
ress reconsidered Deviations from srct epliaten ein Be dtcouned
iF there ae independent indicators that the-instament itl is les than
‘optimal in some ell contents,
"The NEO-PER manual (Costa & McCrae, 19928) specifies tat pote
cols with more than 40 missing sesponses ae considered fnvalid. In
setten, repetitive responses (eg. more thon 9 consecutive disagree
respons of 10 consecutive netral sponses), which ae rare in volunteer
‘arpls, se considoced evidence of random responding. Cases considered
invalid by citer of these esteia were eliminated. However, we also
considered that the fequency of valid responses in a sample probably
reflec’ the quali of data in that administration in genera, and we wed
the pereeatage of valid protocols in the unsereened snp (anging fom
85.1% to 100%) as 2 fst indeator of data quay.
‘Acqpiescence can be estimated by counting the numer of agree and
strongly agree responses to all tems. Because NBOLPI- scales ae bal
anced in keying, the net effect of acqucscent responding is limited and
soyuiecence does not invalidate a prtocol, However, iti a possible
Indicator of poorer quality dt Using de cutoff scores in the mana, we
calcolted the frequency of quiescent (150 agree or strongly agree
responses) or nay-saying (250 agree or strongly agre responses) prote-
cal
ineach unscrened sample (fom 0% to 21.5%) as second (reversed)
ex of data quay.
Where fewer than 40 tems ave missing, missing dita are tested by
sofsttming a vevtral response. Before making that substittion, we
‘counted the numberof missing responses and used the sample mean (from
‘to 11 items) as third (revered indicator of date quality.
‘We considered ir likely that fewer problems would gccue when tater
‘completed the quesonnare inter native languaye o when the samples
tsa whole were jadged by oar cllhoratees a being ent in the second
Tangvage that wat used, Our fourth indicator of data quality Was scored 2
for native language, 1 for very Huent inthe second language, and 0 for
somewhat unt in the second langvage. Although many ofthe uapl
lished NEO-FLR translations ave excellent, its probably fir assume
that polished versione ae farther along in paychometic development han
‘mest unpublished versions. All samples that were tesied in a second
Tanguage used a published version (English or French); for samples in
ich the naive language was sed our fifth ineator was seored 1 for
‘publised and 0 for unpublished tslatons (ee Table 1 now).
‘inl, we asked collsboratrsdzely i thee were any problems. The
‘most comuron problem mentioned war the length ofthe qvesionnsires end
the time required to complete them, Presence or absence of x problema was
‘our sinth indicator of data quality,
Results and Discussion
Data Quality and Internal Consistency
‘The six indicators of data quality were modestly intercorrelated
(rank-order rs ~ 09-66; coefficient a ~ 76; all indices were
significantly reisted to at least two other indices), so we expressed
each as a rank score and used the mean of the six indicators a an
overall measure of data quality. This value is reported in the
second columa of Table 2, and the entries are listed in descending
drder. By and large, the entzies atthe top of the list are from
affluent, mostly Western nations, whereas those at the botiom are
from underdeveloped nations. In part, this is probably due to the
availability of translations in most Buropean languages bat rela-
Lively Few Afiiean languages, In pat, it probably also reflects the
fact thatthe NEO-PI-R was developed within the Wester tradition
of psychological measurement, and completing it is perhaps
more meaningful task for Westerners
In item analyses, we examined the comected item/domain cor
relations inthe full sample and in each culture, Ia te fll sarmple,
* In Gennany, Russa, and dhe Czech Republi, where existing data were
‘wed, raters were usually spouses or same-age pees of the targets. Non-
stent ravers were also rarely included i the new data collection.
2 Byceptions were Russia and Malaysia, where samples were almost
centely composed of ele Russians and Malays, respectively
+ Because of « misunderstanding, partcpents in Uganda end France
‘ere asked to compete al four veesions, which many raters found bur
ensome.UNIVERSAL FEATURES OF TRAITS 551
Table 2
Quality, Reliability, and Factor Replicability of the Semples
Internal consistency Congruenc eooticients*
itare Quy ON BO Ac ON FO AC tol
Germany 39 kM) KGS
Spain ss 23st 85
French Switzer 30 k's ets sg
Denar sso 33) 2 ss tga
Farce 359383 T9586 sts
German Switresand m7 33 sts 586
Chile 42ST 8384989395
New Zestend m3 92 tk 8s T3888
Belgiom B32 tks 5H
Portugal 290 wd 85 ss
‘Tunkey m3 390 3083985558585
Poland 370k Ss
Seria 31830 mks mee
Male 36 tO 9884 Sg
(Czech Repatiie L090 8 BSG 8586
ston gor 83 otk 695979
United Kingdom: Nontern Leland = 3059280908558]
Slovakia 5od 9093 8969s
Teeland 2s 908883 85985
‘Ausiie wi 930k 93k bss
United Kingdom: England ms ts 8s tw
Canada m9 8s 07 zo 833 Oss
Austalia 73 20S 8S 95s 86
Japan 9 0738s
South Korea 276 gs ik ws
Hong Kong 3 32 83g 8585
Brea 26390 0B kr 58k
Ialy 28 97th 86s
United Sates 7 3837 e566 7
Thailand 20008887 3 3 ts
Indonesia 2s 8 B 9 3 eG
Argentina 2s Mw Si 95 ese
Borkina Faso 216 8h ks st 3 35 ts
Kuwait B37 Bt Bo sss
Mexica 987 B92 9583S Ss
Piippines a3 BL BH 83 7 hs
(Croatia 7300 2 95 3565s 86
Russa 166380 9 tbs
China 37s 7 3s 058g
Indie fel 3770 8 88 3
Pera BE a 1 95 83
Slovenia Be 90 St 8 55s 8686
Malaysia Bs 30 8 ot nO 80
Bossivana Bs os So 92 RSD
Nigeria Boa 5 rT
Poerto Rico Bo 308s sc 9s e858
Ethiopia, om m7 a ee a a |
Lebaron oo ess 6558s ss
Ugande co 3 7 1 9 3 ee BM os 50
Mecoseo Ss 8 SS
ote. Alpha les than 70 an congruence coelficens les than 85 are given in boldface ype. N= Neazticiny = Bstraversion O = Openness, A =
‘Agreesbeness; = Conseienioustss
These ae factor and ttl eangrence coefficients comparing fve Procriste-rotaed principal components in each ample wth the American normative
selteepon stnctre (Costa & McCrae, 1923),
these correlations were postive for 239 tems, tem 17, “Chav &
leisurely style in work and play” (reversed), was the exception; in
some evltues it was 2 good indicator of Extraversion, but in most
it appeared to assess Introversion, Within cultures, 394 of the
12,000 (i. 240 X 50) comreeted itewvtomain correlations (3.3%)
were negative, These tended ta cocur forthe same items across
cultures (such as Tem 17) and for the same subset of cultures,
Jeading to lower internal consistency. Alihough poorly performing,
items might be tated as missing, we retained them in the present
study. The third through seventh columns of Table 2 report coef:
ficient alpha for the 48-item domain scales. In general, these are
4uite high, with median values of 90, 80, 88, 92, and 94 for N,992 MCRAE
, O, A, and C, respectively, Nevertheless, there are some in-
stances of low alphas (12 of 250, oF 4.8%, ze lower than .70),
especially for O—a domain that hes also shown problematic
reliability in self report data in Malaysia and Zimbabwe (Mastoe,
Jin, & Cooper, 2000; Piesimont, Bain, McCrae, & Costa, 2002),
‘The value of 25 in Nigesia is particularly notable, suggesting the
possibilty that O is not a meaningful dimension In that culture
However, an altemative interpretation is that low alphas reflect
‘only poor data quality. That hypothesis is supported by rank-order
corelations of column 2, Quality, with columns 3-1 (rs = 63-82,
P< 001), Careless or soquiescent responding, fatigue, or failure
to understand the nuances of language cen have serious conse
quences for item-level analyses, When aggregated into facet
scales, however, some of this error may be reduce.
Factor Structure
‘The first substantive question addressed here isthe universality
of the FFM in observer ratings. An analysis combining raw data
from each sample would confound the covariation of tits aoross
individuals with covariation across cultures (Bond, 2001). We
therefore standardized data within each culture Go that the means
of all facets in each culture were transformed to 0, the standard
deviations to 1.0), and factored the 30 facet scales The first six
igenvalues were 6.67, 4.40, 351, 2.43, 1.46, and 0.84, unmistake
ably suggesting x five-fector solution. After varimax rotation, the
‘expected structure of the FEM was clealy replicated, with factor
congruence coefficients ranging from .96 10 .98. The principal
ifference between this matrix ané the normative sel-report mi
‘tix is that the Form R factors account for more total variance than
Form § factors (61.6% vs. 56.9%), and the A and C factors account
for a larger percentage of the common variance (23.9% and
26.4%) in observer ratings than in self-reports (19.8% and 22.2%).
‘That phenomenon had already been noted in American Form R
ata Costa & McCrae, 19928)
‘Although the varimax-rotted strctore i almost identical to the
American self report normative structure, comparisons of factor
structures are most direct when orthogonal Procruses rotation is
‘used to align factors maximally with tho target (McCrae, Zonder.
man, Costa, Bond, & Paunonen, 1996). Table 3 reports the factor
stracture forthe total sample and gives variable, Factor, and total
‘congruence coefficients. In this study, E3: Assertiveness has a
‘somewhat stronger (negative) loading on A than on F, but all other
facets have their primary loading on the intended factor, and
secondary loadings (uch as N2: Angry Hostility on A. and 03:
Feelings on B) are also replicated, as attested to by the large
varisble congruence coefficients, The FFM structure was also
replicated within each ofthe four age-and-sex target groups, with
factor congruence coefficients after Procrustes rotalion ranging
from 96 10 98°
Although itis clear from Table 3 that the FEM does in fact
represent the stuctare of observer-ated personality traits across
cultures, its possible tht there is 2 minority of cultures in which
the structure is not found. Factor analyses with Procrustes rotation
were therefore conducted in each sample separately; results are
ssammarized as foctor and total congruence coefficients in the lest
six columns of Table 2. By the 85 esiterion of factor repicability
(Haven & ten Berge, 1977), 94.4% of the factors are replications
fof the American normative Farm $ structure Statistically, there is
AL,
‘Table 3
Factor Loodlings for Observer-Rated NEO-PLR Facet Scales in
the Combined Sample
Factor
NEOPRiee NFO A vee
NI: Anxiety SL 07) -05 06 12 98
1N2: Angry Hostty 2 05 12-37 0 oe
3: Depression my M18 02 “og —li7 ow
NéSeliConciouness 68 23-518 Ot 9
NS: Impisiveness| 33g I 36 oP
Ne: Vainerablty I -be 1 06 23 (9
EI: Warmth “Mom se
BD Gregaroasness, = =12, 76 Os 5 ow"
EX Asetvencet <2 ao 10 at 38 oF
Bat Activity = 8 os = 25 aw
FS Excitement Seeking 07 (S431 -=25 29 “a
FG Positive Emotions” = —.14 703217 a9
01: Fansy 13 27 6) -32 9
(02: Aextetes 16 1¢ 75 i410 9
2s as it ots ot