the adequacy of the european social survey to measure values in 20 countries

Public Opinion Quarterly, Vol. 72, No. 3, Fall 2008, pp. 420–445

BRINGING VALUES BACK INTHE ADEQUACY OF THE EUROPEAN SOCIAL SURVEYTO MEASURE VALUES IN 20 COUNTRIES

ELDAD DAVIDOVPETER SCHMIDTSHALOM H. SCHWARTZ

Abstract The Schwartz (1992) theory of basic human values haspromoted a revival of empirical research on values. The semi-annual Eu-ropean Social Survey (ESS) includes a new 21-item instrument to mea-sure the importance of the 10 basic values of the theory. Representativenational samples in 20 countries responded to the instrument in 2002–3.We briefly describe the theory and the ESS instrument and assess itsadequacy for measuring values across countries. Using multiple-groupconfirmatory factor analyses, augmented with mean-structure informa-tion, we assess the configural and measurement (metric) invariance ofthe values—necessary conditions for equivalence of the meaning of con-structs and scalar invariance—a precondition for comparing value meansacross countries. Only if such equivalence is established can researchersmake meaningful and clearly interpretable cross-national comparisons ofvalue priorities and their correlates. The ESS values scale demonstratesconfigural and metric invariance, allowing researchers to use it to studyrelationships among values, attitudes, behavior and socio-demographiccharacteristics across countries. Comparing the mean importance of

ELDAD DAVIDOV is with the GESIS-Central Archive for Empirical Social Research, and Universityof Cologne, 50923 Cologne, Germany. PETER SCHMIDT is with the University of Giessen, Faculty ofSocial Sciences, Karl-Gloeckner Str. 22E, 35394, Giessen, Germany. SHALOM H. SCHWARTZ is withthe Department of Psychology, Hebrew University of Jerusalem, 91905 Jerusalem, Israel. Earlierversions of this article were presented at the RC33 conference, Amsterdam 2004, RC28 conference,Oslo 2005, first ESRA conference, Barcelona 2005, and two research seminars of QMSS1 heldby Willem Saris, Lugano 2004 and 2006, with the financial support of the European ScienceFoundation. We would also like to thank Willem Saris and Jaak Billiet for helpful comments onearlier versions of this study. Work on this article was conducted with the help of the GermanIsraeli Foundation (GIF), Project No. I-769–241.4. The work of the third author on this researchwas supported by the Israel Science Foundation Grant No. 921/02–1. Address correspondence toEldad Davidov; e-mail: [email protected], [email protected].

doi:10.1093/poq/nfn035 Advance Access publication August 21, 2008C© The Author 2008. Published by Oxford University Press on behalf of the American Association for Public Opinion Research.All rights reserved. For permissions, please e-mail: [email protected]

Measuring Values in the European Social Survey 421

values across countries is possible only for subsets of countries wherescalar invariance holds.

Introduction

Values are central to public discourse today. Competing groups demand priorityfor the values they hold dear, arguing that conflicting values are unworthy.Theorists have long considered values central to understanding attitudes andbehavior (e.g., Kluckhohn 1951; Allport, Vernon, and Lindsay 1960; Williams1968). They view values as deeply rooted, abstract motivations that guide,justify, and explain attitudes, norms, opinions, and actions (cf. Rokeach 1973;Schwartz 1992; Halman and de Moor 1994; Feldman 2003).

In practice, survey researchers distinguish little between values and attitudes(Halman and de Moor 1994, 22). The confounding of values and attitudesreflects, in part, the absence of a comprehensive theory of the basic motivationsthat are socially expressed as values. It also reflects the lack of reliable, theory-based instruments to measure basic values (Rohan 2000; Hitlin and Piliavin2004).

In 1992, Schwartz introduced a theory of basic human values, building oncommon elements in earlier approaches (e.g., Rokeach 1973). It includes 10motivationally distinct values presumed to encompass the major value orienta-tions recognized across cultures. He also presented a first instrument to measurethese values that he validated cross-culturally (Schwartz 1992). An alternativeinstrument, also validated across cultures, was presented in 2001 (Schwartzet al. 2001). This theory and two instruments have promoted a revival of em-pirical research on relations of values to attitudes and behavior, both withinand across cultures (overviews in Hitlin and Piliavin 2004; Schwartz 2005a,2005b). Recently, the European Social Survey (ESS) incorporated a third, shortinstrument to measure the 10 basic values in its semi-annual studies of atti-tudes and opinions. The 2006 pilot study of the American National ElectionsSurvey also includes a version of the values instrument. This article brieflydescribes the theory of basic values and the ESS instrument, and then as-sesses the adequacy of this instrument for measuring values within and acrosscountries.

Political attitudes and choices are one important domain to which basic valuesare relevant (e.g., Rokeach 1973; Zaller 1992; Knutsen 1995; Miller and Shanks1996). Values may enable people to organize their political evaluations in arelatively consistent manner; they may provide a general structure to politicalattitudes (Feldman 2003). This structuring process is one path through whichvalues may influence political preferences. Converse (1964) likened values to“a sort of glue to bind together many more specific attitudes and beliefs.”Schwartz (1994) argued that systematic variations in the priority individualsgive to different basic values underlie political ideologies. Thus, values mayinfluence political choice through their effects on ideologies.

422 Davidov, Schmidt, and Schwartz

The particular values that structure ideological discourse and underlie po-litical choice depend upon the issues that are central in a given politicalcontext. Consider three examples of studies based on the Schwartz (1992) the-ory and instruments. In the Israeli political arena of 1988, protection of religiouspractice competed with free expression of a secular life style. There, the keyvalues that differentiated party supporters were tradition versus self-direction(Barnea and Schwartz 1998).

In the Italian elections of 2001, the center-right placed particular emphasison entrepreneurship and the market economy, security, and family and nationalvalues. In contrast, the center-left advocated social welfare, concern for socialjustice, equality, and tolerance of diverse, even potentially disruptive groups.The basic values that related significantly to political preferences were securityand power (right) versus universalism and benevolence (left) (Caprara et al.2006).

In a study of political preferences in 14 countries, Barnea (2003) found twomain patterns. Where political competition revolved around issues of nationalsecurity versus equal rights and freedoms for all, the key values that struc-tured voters’ preferences were security and conformity versus universalismand self-direction. Where the focus of political competition revolved aroundthe distribution of material resources, the key values were universalism andbenevolence versus power and achievement.

Other studies have revealed systematic relations of basic values to a widevariety of attitudes and opinions in many countries, using one or another of theSchwartz instruments. For example, basic values exhibit predicted associationswith attitudes toward war, right-wing authoritarianism, and social dominanceorientation (Cohrs et al. 2005); attitudes toward human rights (Spini and Doise1998); interpersonal trust, political activism, and attitudes toward immigration(Schwartz 2007); environmental attitudes (Schultz and Zelezny 1999); andmaterialism (Burroughs and Rindfleisch 2002).

Not only are individual differences in basic value priorities important, butthe prevailing value emphases in societies also relate systematically to na-tional differences in widespread attitudes and public policy (Schwartz 2006b).Hence, basic values have predictive and explanatory potential both at theindividual and societal levels. Moreover, values can reflect major socialchange in societies and across nations. And values may influence the direc-tion of social change and its speed. A simple survey method for measur-ing basic values, such as the ESS instrument, may therefore prove widelyuseful.

The ESS values scale was administered to representative national sam-ples in 20 countries in 2002–3, in 25 countries in 2004–5, and willbe included in future ESS rounds (www.europeansocialsurvey.org). Thefull dataset from the first two ESS rounds, now in the public domain(http://ess.nsd.uib.no), includes a vast amount of information of interestto political scientists, sociologists, and others. Thousands of researchers


have begun to use these data (reported in http://naticent02.uuhost.uk.uu.net/news).

Given the spreading impact of the ESS and the significant role the new humanvalues scale may play in research, it is crucial to examine its validity.1 Thecurrent article assesses how well the ESS human values scale measures the 10basic values in the Schwartz theory. It then examines the equivalence of meaningacross countries of the values that are measured. Only if such equivalence isestablished can researchers make meaningful and clearly interpretable cross-national comparisons of value priorities and their correlates (Billiet 2003). Weemploy multiple-group confirmatory factor analyses (MGCFAs) augmentedwith mean-structure information (see Sorbom 1974; Sorbom 1978) on the datafrom the first round of the ESS to address these issues.

The Theory of Basic Values

Before presenting the ESS values scale and assessing its validity, a briefoverview of the theory from which it derives is necessary. The theory de-fines values as desirable, trans-situational goals, varying in importance, thatserve as guiding principles in people’s lives. It derives 10 motivationally dis-tinct, broad and basic values from three universal requirements of the humancondition: needs of individuals as biological organisms, requisites of coordi-nated social interaction, and survival and welfare needs of groups (Schwartz1992, 2005a). For example, conformity values derive from the prerequisitesof interaction and of group survival. For interaction to proceed smoothlyand for groups to thrive, individuals must restrain impulses and inhibit ac-tions that might hurt others. Self-direction values derive from organismicneeds for mastery and from the interaction requirements of autonomy andindependence.

The 10 basic values cover the distinct content categories found in earlier valuetheories, in value questionnaires from different cultures, and in religious andphilosophical discussions of values. The core motivational goal of each basicvalue, presented in table 1, defines it. The theory also explicates the structureof dynamic relations among the 10 basic values. The main source of the valuestructure is the fact that actions in pursuit of any value have consequences thatconflict with some values and are congruent with other values.2 For example,pursuing achievement values may conflict with pursuing benevolence values.Seeking success for self is likely to obstruct actions aimed at enhancing thewelfare of others who need one’s help. But pursuing both achievement and

1. We discuss validity only in the sense of consistency (Bollen 1989), not in the sense of constructvalidity. The latter would require examining both antecedents and consequences of value priorities,an objective beyond the scope of this paper.2. For a discussion of other sources, see Schwartz (2006a).


Table 1. Definitions of the Motivational Types of Values in Terms of theirCore Goal

POWER: Social status and prestige, control or dominance over people andresourcesACHIEVEMENT: Personal success through demonstrating competence accordingto social standardsHEDONISM: Pleasure and sensuous gratification for oneselfSTIMULATION: Excitement, novelty, and challenge in lifeSELF−DIRECTION: Independent thought and action−choosing, creating,exploringUNIVERSALISM: Understanding, appreciation, tolerance and protection for thewelfare of all people and for natureBENEVOLENCE: Preservation and enhancement of the welfare of people withwhom one is in frequent personal contactTRADITION: Respect, commitment and acceptance of the customs and ideas thattraditional culture or religion provide the selfCONFORMITY: Restraint of actions, inclinations, and impulses likely to upset orharm others and violate social expectations or normsSECURITY: Safety, harmony and stability of society, of relationships, and of self

power values may be compatible. Seeking personal success for oneself tendsto strengthen and to be strengthened by actions aimed at enhancing one’s ownsocial position and authority over others.

The schematic circular structure in figure 1 portrays the pattern of relations ofconflict and congruity among values. The closer any two values in either direc-tion around the circle, the more similar their underlying motivations. The moredistant any two values, the more antagonistic their underlying motivations. Thedivision of the domain of value items into 10 distinct values is an arbitraryconvenience. The circular arrangement of values represents a continuum ofrelated motivations, like the circular continuum of colors, rather than a set ofdiscrete motivations. One could reasonably partition the domain of value itemsinto broader or more fine-tuned distinct value constructs, depending on howfinely one wishes to discriminate among motivations. Reflecting the indeter-minacy of the optimal number of value constructs, items from adjacent valuesoften intermix in empirical studies. Thus, the theory specifies the motivationalorder of value items around the circle and it suggests distinguishing 10 valueconstructs for scientific convenience. The theory leaves the width of the valueconstructs and the absolute distances among them unspecified; hence the widthof the slices in figure 1 is arbitrary.

Two dimensions summarize the structure of relations among the basic values(figure 1): The self-enhancement versus self-transcendence dimension opposespower and achievement values—that emphasize self-interest—to universalismand benevolence values—that entail concern for the welfare and interests of


Universalism

Benevolence

ConformityTradition

Security

Self-Direction

Stimulation

Hedonism

Achievement

Power

Figure 1. Structural Relations Among the 10 Values and the Two Dimensions.

others. The openness to change versus conservation dimension opposes self-direction and stimulation values—that emphasize independent action, thought,and feeling and readiness for new experience—to security, conformity, andtraditional values—that emphasize self-restriction, order, and resistance tochange. The dashed lines around hedonism indicate that it shares elementsof both openness to change and of self-enhancement.

Research with two earlier instruments provides evidence supporting thisstructure in samples from 67 nations (Fontaine et al. (Forthcoming); Schwartz2005a, 2005b; Schwartz and Boehnke 2004). Although individuals differ inthe importance they attribute to various values, the same motivational structureapparently organizes these values across cultures. These studies provide nostrict tests of measurement invariance, however. The current study subjects thenew ESS human values scale to such tests.

The ESS Human Values Scale

The ESS human values scale is derived from the earlier 40-item Portrait ValuesQuestionnaire (PVQ; Schwartz et al. 2001; Schwartz 2005b). Space limitationsin the ESS required reducing the number of items. Some items were droppedand others were revised to encompass additional ideas to preserve coverage ofthe content of the 10 different values. The ESS scale includes verbal portraits


of 21 different people, gender-matched with the respondent. Each portraitdescribes a person’s goals, aspirations, or wishes that point implicitly to theimportance of a value. For example: “Thinking up new ideas and being creativeis important to her. She likes to do things in her own original way” describes aperson for whom self-direction values are important. Respondents’ own valuesare inferred from their self-reported similarity to people described implicitly interms of particular values.3

Regarding each portrait, respondents answer: “How much like you is thisperson?” Six labeled responses range from “not like me at all” to “very muchlike me.” The upper panel of table 2 presents the format of the survey and,below, in the first two columns, the items grouped by type of value. Twoportraits operationalize each value, with three for universalism because of itsvery broad content. The score for the importance of each value is the meanresponse to the items that measure it. Translation into each native languagefollowed procedures explained in Harkness, Van de Vijver, and Mohler (2003,Ch. 3).

Strict probability samples, representing the non-institutionalized populationaged 15 years and older in each of 20 countries, completed the human valuesscale in the first round of the ESS. The value scale was administered followinga face-to-face interview for approximately one hour in respondents’ homes ona wide variety of topics. Eighty-six percent of the respondents completed thesurvey in the presence of the interviewer (66 percent orally, 20 percent written).The interviewer picked up the written values scale later from 2 percent, and12 percent (almost all in Finland and Sweden and half in Norway) returned itby post. The countries, with numbers of respondents who completed the valuesscale and country-level response rates, are: Austria (2,257, RR 60.4), Belgium(1,899, RR 59.2), Czech Republic (1,360, RR 43.3), Denmark (1,506, RR67.6), Finland (2,000, RR 73.2), France (1,503, RR 43.1), Germany (2,919, RR55.7), Great Britain (2,052, RR 55.5), Greece (2,566, 79.9), Hungary (1,685,RR 69.9), Ireland (2,046, RR 64.5), Israel (2,499, RR 70.9), the Netherlands(2,364, RR 67.8), Norway (2,036, RR 65.0), Poland (2,110, RR 73.2), Portugal(1,511, RR 68.8), Slovenia (1,519, RR 70.5), Spain (1,729, RR 53.2), Sweden(1,999, RR 69.4), Switzerland (2,037, RR 33.5), total = 39,596. The first-roundresponse rate of the ESS is calculated according to AAPOR RR5. Detailedinformation on the population, selection procedure, response rates, dates of datacollection and exact question wording in the various languages are available athttp://ess.nsd.uib.no.

3. All the items are double-barreled because each includes two sentences. Schwartz (2003) dis-cusses the rationale for this and presents evidence suggesting that it does not create a problem inthis case.


Tabl

e2.

The

ESS

Hum

anV

alue

sSc

ale:

Form

at,I

tem

Mea

nsan

dSt

anda

rdD

evia

tions

(N=

39,3

99)

Her

ew

ebr

iefly

desc

ribe

som

epe

ople

.Ple

ase

read

each

desc

ript

ion

and

thin

kab

outh

owm

uch

each

pers

onis

oris

notl

ike

you.

Tic

kth

ebo

xto

the

righ

ttha

tsho

ws

how

muc

hth

epe

rson

inth

ede

scri

ptio

nis

like

you.

HO

WM

UC

HL

IKE

YO

UIS

TH

ISPE

RSO

N?

Ver

ym

uch

like

me

Lik

em

eSo

mew

hat

like

me

Alit

tlelik

em

eN

otlik

em

eN

otlik

em

eat

all

1.T

hink

ing

up..

..1

23

45

6V

alue

Item

#(a

ccor

ding

toits

orde

rin

the

ESS

ques

tionn

aire

)an

dw

ordi

ng(M

ale

vers

ion)

Mea

n(S

td.D

ev.)

Self

-dir

ectio

n(S

D)

1.T

hink

ing

upne

wid

eas

and

bein

gcr

eativ

eis

impo

rtan

tto

him

.He

likes

todo

thin

gsin

his

own

orig

inal

way

.2.

5(1

.2)

11.I

tis

impo

rtan

tto

him

tom

ake

his

own

deci

sion

sab

outw

hath

edo

es.H

elik

esto

befr

eeto

plan

and

notd

epen

don

othe

rs.

2.1

(1.1

)

Uni

vers

alis

m(U

N)

3.H

eth

inks

itis

impo

rtan

ttha

teve

rype

rson

inth

ew

orld

betr

eate

deq

ually

.He

belie

ves

ever

yone

shou

ldha

veeq

ualo

ppor

tuni

ties

inlif

e.2.

1(1

.0)

8.It

isim

port

antt

ohi

mto

liste

nto

peop

lew

hoar

edi

ffer

entf

rom

him

.Eve

nw

hen

hedi

sagr

ees

with

them

,he

still

wan

tsto

unde

rsta

ndth

em.

2.4

(1.1

)

19.H

est

rong

lybe

lieve

sth

atpe

ople

shou

ldca

refo

rna

ture

.Loo

king

afte

rth

een

viro

nmen

tis

impo

rtan

tto

him

.2.

2(1

.0)

Ben

evol

ence

(BE

)12

.It’s

very

impo

rtan

tto

him

tohe

lpth

epe

ople

arou

ndhi

m.H

ew

ants

toca

refo

rth

eir

wel

l-be

ing.

2.3

(1.0

)

18.I

tis

impo

rtan

tto

him

tobe

loya

lto

his

frie

nds.

He

wan

tsto

devo

tehi

mse

lfto

peop

lecl

ose

tohi

m.

2.0

(0.9

)


Tra

ditio

n(T

R)

9.It

isim

port

antt

ohi

mto

behu

mbl

ean

dm

odes

t.H

etr

ies

nott

odr

awat

tent

ion

tohi

mse

lf.

2.8

(1.3

)20

.Tra

ditio

nis

impo

rtan

tto

him

.He

trie

sto

follo

wth

ecu

stom

sha

nded

dow

nby

his

relig

ion

orhi

sfa

mily

.2.

7(1

.4)

Con

form

ity(C

O)

7.H

ebe

lieve

sth

atpe

ople

shou

lddo

wha

tthe

y’re

told

.He

thin

kspe

ople

shou

ldfo

llow

rule

sat

all

times

,eve

nw

hen

no-o

neis

wat

chin

g.3.

1(1

.4)

16.I

tis

impo

rtan

tto

him

alw

ays

tobe

have

prop

erly

.He

wan

tsto

avoi

ddo

ing

anyt

hing

peop

lew

ould

say

isw

rong

.2.

7(1

.3)

Secu

rity

(SE

C)

5.It

isim

port

antt

ohi

mto

live

inse

cure

surr

ound

ings

.He

avoi

dsan

ythi

ngth

atm

ight

enda

nger

his

safe

ty.

2.3

(1.2

)

14.I

tis

impo

rtan

tto

him

that

the

gove

rnm

ente

nsur

ehi

ssa

fety

agai

nsta

llth

reat

s.H

ew

ants

the

stat

eto

best

rong

soit

can

defe

ndits

citiz

ens.

2.4

(1.2

)

Pow

er(P

O)

2.It

isim

port

antt

ohi

mto

beri

ch.H

ew

ants

toha

vea

loto

fm

oney

and

expe

nsiv

eth

ings

.4.

1(1

.3)

17.I

tis

impo

rtan

tto

him

toge

tres

pect

from

othe

rs.H

ew

ants

peop

leto

dow

hath

esa

ys.

3.2

(1.4

)A

chie

vem

ent(

AC

)4.

It’s

impo

rtan

tto

him

tosh

owhi

sab

ilitie

s.H

ew

ants

peop

leto

adm

ire

wha

the

does

.3.

2(1

.4)

13.B

eing

very

succ

essf

ulis

impo

rtan

tto

him

.He

hope

spe

ople

will

reco

gniz

ehi

sac

hiev

emen

ts.

3.2

(1.4

)H

edon

ism

(HE

)10

.Hav

ing

ago

odtim

eis

impo

rtan

tto

him

.He

likes

to“s

poil”

him

self

.2.

9(1

.4)

21.H

ese

eks

ever

ych

ance

heca

nto

have

fun.

Itis

impo

rtan

tto

him

todo

thin

gsth

atgi

vehi

mpl

easu

re.

3.0

(1.4

)

Stim

ulat

ion

(ST

)6.

He

likes

surp

rise

san

dis

alw

ays

look

ing

for

new

thin

gsto

do.H

eth

inks

itis

impo

rtan

tto

dolo

tsof

diff

eren

tthi

ngs

inlif

e.3.

0(1

.4)

15.H

elo

oks

for

adve

ntur

esan

dlik

esto

take

risk

s.H

ew

ants

toha

vean

exci

ting

life.

3.9

(1.5

)


Testing Invariance

When applying a theory and an instrument in different countries or over time,a key concern is to ensure that measurement of the relevant constructs isinvariant cross-nationally or over-time (Cheung and Rensvold 2000, 2002;Harkness, Van de Vijver, and Mohler 2003). The meaning of measurement in-variance is “Whether or not, under different conditions of observing and study-ing phenomena, measurement operations yield measures of the same attribute”(Horn and McArdle 1992, 117). If invariance is not tested, interpretations ofbetween-group comparisons are problematic (Vandenberg and Lance 2000).Absent invariance, observed differences in means or other statistics might re-flect differences in systematic biases of response across countries or differentunderstanding of the concepts, rather than substantive differences (Steenkampand Baumgartner 1998). Equally important, findings of no difference betweencountries do not ensure the absence of “real” differences.

MGCFA (Joreskog 1971) is among the most powerful techniques for testingmeasurement invariance.4 We draw upon the Steenkamp and Baumgartner(1998) procedural guidelines for assessing measurement invariance in cross-national studies with a confirmatory factor analysis (CFA) approach. We followa step-wise procedure from the least to the most demanding form of invariance.

The lowest level of invariance is “configural” invariance, which requiresthat the items in an instrument exhibit the same configuration of loadings ineach of the different countries (Horn and McArdle 1992). That is, the analysisshould confirm that the same items measure each construct in all countries inthe cross-national study.5 Configural invariance is supported if a single modelspecifying which items measure each construct fits the data well in all countries,all item loadings are substantial and significant, and the correlations betweenthe factors are less than one. The latter requirement guarantees discriminantvalidity between the factors (Steenkamp and Baumgartner 1998).

Configural invariance does not ensure that people in different countries un-derstand each of the items the same way. Although the same items form a factorthat represents each construct, the factor loadings may still be different acrosscountries. The next higher level of invariance, “measurement” or “metric” in-variance, assesses a necessary condition for equivalence of meaning. This level

4. Lubke and Muthen (2004) argue that an analysis of Likert data under the assumption ofmultivariate normality may distort the factor structure differently across groups. They proposefitting a model for ordered categorical outcomes. However, De Beuckelaer (2005) demonstratesin simulation studies that using Likert scales and skewed data does not significantly affect theprobability of incorrect conclusions in MGCFA. Moreover, the values scale has six categoriesrather than the usual five in Likert scales. Although the proportional odds model (POM) techniquehas more statistical power (low type II error) than MGCFA, recent studies suggest that POM hasan inflated type I error, and that MGCFA is more flexible (Welkenhuysen-Gybels and Billiet 2002;Welkenhuysen-Gybels 2004).5. The test allows sequentially adding cross-loadings if indicated by the program so long as thesame cross-loadings are allowed across all countries.


requires that the factor loadings between items and constructs are invariantacross countries (Rock, Werts, and Flaugher 1978). It is tested by restrictingthe factor loading of each item on its corresponding construct to be the sameacross countries. Measurement invariance is supported if this model fits thedata well in an MGCFA.

Configural and metric invariance are tested by examining information onlyabout covariation among the items. A third level of invariance is necessary tojustify comparing the means of the underlying constructs across countries. Thisis often a central goal of cross-national research. Such comparisons are mean-ingful only if the items exhibit “scalar” invariance (Meredith 1993; Steenkampand Baumgartner 1998). Scalar invariance signifies that cross-country differ-ences in the means of the observed items result from differences in the meansof their corresponding constructs. To assess scalar invariance, one constrainsthe intercepts of the underlying items to be equal across countries, and tests thefit of the model to the data.

In sum, meaningful comparison of construct means across countries requiresthree levels of invariance, i.e., configural, metric and scalar. Only if all theselevels of invariance are supported, can we assume that scores are not biased andconfidently carry out mean comparisons. We adopt a “bottom-up” test strategy,starting with the weakest constraints and proceeding to the most severe. Thisenables us to establish first whether even weak forms of invariance are absent.We test all countries together because theory implies that the measurementmodel should hold for them all.6

Data Analysis

The data analysis starts with 20 separate CFAs. This is followed by a simul-taneous MGCFA in order to assess configural invariance. We then modify themodel slightly, following the modification indices that are compatible withtheory, in order to specify an acceptable model that is invariant across the20 countries. Next, we conduct metric and scalar invariance tests. Finally, wetest for invariance of the structural covariances across countries (which is not anecessary condition for the comparison of means).

First, we computed a Pearson (product moment) intercorrelation matrix plusstandard deviations for each country. We then converted these into covariancematrices as input for estimating CFAs. We used pairwise deletion for miss-ing values because, with fewer than five percent missing values as observedhere, pairwise deletion is better than listwise and is adequate (see Brown1994; Schafer and Graham 2002).7 Correlations among the items ranged from

6. Several sources provide details on measuring invariance (e.g., Vandenberg and Lance 2000;Steenkamp and Baumgartner 1998; Cheung and Rensvold 2002; Vandenberg 2002; De Beuckelaer2005).7. Simulations demonstrate that results of pairwise deletion are robust when there are fewer than 5percent missing values and that the improvement of newer methods is minimal (Schafer and Graham


negative for indicators of value constructs that are theoretically opposed tohighly positive for indicators of adjacent value constructs or of the same con-struct (see figure 1).

Our first question is: How well does the ESS human values scale measure the10 basic values in the Schwartz theory across the 20 countries? Byrne (2001,175–6) notes the importance of conducting single-group analyses prior to testsof multi-group invariance. We therefore began by performing separate CFAs ineach country on the a priori model of 10 values with their respective indicators.We used the Amos 6.0 software package in all of the analyses (Arbuckle 2005).The analyses revealed that all of the items had a factor loading of .4 or higheron their respective factors in each of the countries. However, in every country,at least two pairs of constructs were dependent on each other.

In order to solve the problem of the non-positive definite covariance matrices8

of the constructs, we unified the pairs of strongly associated values. Table 3reports the number of distinct values found in each country and lists the pairs ofvalues that were unified into one construct. After unifying strongly associatedvalue constructs, the formal test of dimensionality suggested that there werebetween five and eight distinct values in the different countries. Of the 71 pairsof value constructs that were unified across the 20 countries, 69 are pairs ofvalues that are adjacent in the value circle of the Schwartz theory (figure 1).Thus, they represent closely linked motivations and do not violate the theorizedcircular motivational structure. They do, however, suggest that the ESS valuesscale may not capture all of the fine-tuned distinctions in the theory.

We next tested for configural invariance across the 20 countries by computingthe CFA for all the countries simultaneously. We expected to find approximatelythe average number of distinguishable values found in the 20 separate countryanalyses. We started with the a priori model of 10 values with their respec-tive indicators. Three pairs of values were highly intercorrelated, power withachievement, conformity with tradition, and universalism with benevolence.This produced a problem of non-positive definite matrices of the constructs.The high intercorrelations indicated that the pairs of values were too close tobe modeled separately. We therefore unified each pair into a single construct toform seven distinct values that we tested in all subsequent analyses.

Figure 2 shows the model with the seven value constructs and the relationsbetween indicators and constructs. Note that the value constructs that wereunified represent values that are adjacent in the value circle of the Schwartztheory; each unified construct shares a broad basic motivational goal. Thus,these findings support the theorized circular motivational structure. They again

2002). Therefore, the gain from using full information maximum likelihood for the problem ofmissing values is minimal here. Comparing our results with the FIML procedure produced nodifferences in the unstandardized regression coefficients until the third place after the comma.8. The problem of non-positive definite covariance matrix of the constructs means that someconstructs are associated with each other so strongly that they cannot be separated.


Table 3. Number of Values Found in Each Country after Unifying Valuesto solve the Problem of Non-Positive Definite Matrices of the Constructs inSingle-Country CFAs

Country Number of values Unified valuesa

Austria 8 POAC, COTRBelgium 6 POAC, COTR, UNBE, STSDCzech Republic 7 POAC, UNBE, COTRDenmark 8 COTR, POACGermany 7 POAC, UNBE, COTRFinland 8 COTR, POACFrance 7 COTR, POAC, UNBEGreat Britain 8 COTR, POACGreece 5 POAC, COTR, UNBE, HEST, STSDHungary 5 UNBE, COTR, POAC, SECUN, HESDIreland 6 POAC, COTR, UNBE, HESTIsrael 7 UNBE, POAC, STSDNetherlands 8 COTR, POACNorway 8 POAC, COTRPoland 6 UNBE, COTR, HEST, POACPortugal 7 COTR, UNBE, HESTSlovenia 5 COTR, UNBE, HEST, POAC, STSDSpain 8 COTR, POACSweden 8 COTR, POACSwitzerland 7 COTR, POAC, UNBE

aFor abbreviations of values, see table 2.

suggest, however, that the ESS values scale may not capture all of the fine-tuneddistinctions in the theory.

We proceeded to improve the model by adding paths suggested by the mod-ification indices. These indices revealed that four items intended to measureparticular value constructs also had significant, secondary loadings on othervalue constructs. Table 4 presents the chi-square values, degrees of freedom,and fit statistics for the basic model of seven values (1) and the models aftersequentially adding paths for the items with an additional loading on anothervalue construct (2, 3, and 4). Each model produced some improvement inseveral of the fit indices.9

Model 2 allowed a positive path from power item #2 to the combinedconformity-tradition value construct, a negative path to the combined univer-salism/benevolence value construct, and a negative path from tradition item #9to the combined power/achievement value construct. Model 3 allowed a nega-tive path from stimulation item #15 to the combined universalism/benevolencevalue construct. Finally, model 4 allowed a positive path from power item #17

9. Because of the large number of cases, the chi-square values are very high.


Tabl

e4.

Fit

mea

sure

sfo

ra

Mul

ti-G

roup

Con

firm

ator

yFa

ctor

Ana

lysi

sof

Seve

nV

alue

s,C

onst

rain

ing

Con

figur

alIn

vari

ance

acro

ss20

Cou

ntri

es,w

ithM

odifi

catio

ns∗

Mod

elR

MR

NFI

CFI

RM

SEA

PCL

OSE

AIC

BC

CC

hisq

uare

df

1.B

asic

mod

el0.

090

0.86

60.

879

0.01

41.

000

32,0

7932

,109

29,5

593,

360

2.A

ddin

gpa

ths

for

item

s#2

,#9

0.08

10.

879

.892

0.01

31.

000

29,1

6429

,195

26,5

243,

300

3.A

ddin

ga

path

for

item

#15

0.07

20.

892

0.90

50.

013

1.00

026

,371

26,4

0323

,691

3,28

04.

Add

ing

apa

thfo

rite

m#1

70.

064

0.90

10.

914

0.01

21.

000

24,5

8924

,622

21,8

693,

260

∗ RM

R=

root

mea

nsq

uare

resi

dual

;N

FI=

the

Ben

tler-

Bon

ett

norm

edfit

inde

x;C

FI=

the

com

para

tive

fitin

dex;

RM

SEA

=ro

otm

ean

squa

reer

ror

ofap

prox

imat

ion;

PCL

OSE

=pr

obab

ility

ofcl

ose

fit;A

IC=

the

Aka

ike

info

rmat

ion

crite

rion

;BC

C=

the

Bro

wne

-Cud

eck

crite

rion

;df=

the

num

ber

ofde

gree

sof

free

dom

.For

deta

ilsse

efo

rex

ampl

eA

rbuc

kle

(200

5).


HE

ST

SD

UNBECOTR

SEC

POAC

ac1

d3

1

ac2

d4

1

he1

d5

1

1

he2

d6

1

st1 d71 1

st2 d81po1d1

11po2d2

1

sec1d2011

sec2d211

co1d1811

co2d191

tr1

d16

1tr2

d17

1be1

d14

1be2

d15

1

un1 d1111

un2 d121

sd1 d91

sd2 d101

un3 d131

1

Figure 2. A Confirmatory Factor Analysis with Seven Value Constructs.NOTE.—Broken lines indicate paths added subsequently based on the modifi-cation indices (see text). In order to set the metric for a factor, it is necessary tofix the factor loading of one of the indicators to one.

to the combined conformity/tradition value construct. These paths appear asbroken lines in figure 2.

Following these modifications, the various fit indices indicate a fit betweenthe model and the data that is satisfactory for not rejecting a model accordingto Hu and Bentler (1999); Marsh, Hau, and Wen (2004) (RMR = 0.06, NFI =0.90, CFI = 0.91, RMSEA = 0.01, PCLOSE = 1 and AIC and BCC lower thanin previous models). Hence, the configural invariance of the 7-factor modelcannot be rejected. That is, we can treat the specification of the items that indexeach of the seven factors as invariant across the 20 countries.

Interestingly, all the modifications introduced in models 2–4 entail addingpaths between single indicators and motivationally opposed latent factorsthat were formed by combining two value constructs. The negative paths in-dicate that the association between the opposing latent value constructs didnot capture all of the opposition for three items. The positive paths indicatethat these associations overestimated the opposition for two items. The needfor these modifications may be due to the reduction from 10 original factors


to 7. For example, the unified power-achievement latent construct yielded anoverall association with the conformity-tradition construct that required addingpositive paths for the two power items. In the theorized motivational circle ofvalues, power is closer to conformity-tradition than achievement is. We wouldtherefore expect the power items to correlate less negatively with conformity-tradition than the achievement items with which they were combined. Perhaps,had the number of indicators in the ESS scale not been reduced to two perconstruct (three for universalism), permitting retention of all 10 factors, theadded paths would not have been necessary.

As noted, the ESS values scale captures the order of motivations postulatedby the value theory, but the CFAs do not support all of the fine-tuned dis-tinctions in the theory. On the other hand, multi-dimensional scaling analysesof the same values data in each of the countries do support these distinctions(Schwartz 2007). These analyses also yielded structures that correspond withthe theorized circular motivational structure of values. Moreover, in 15 of the 20countries, it was possible to partition the two-dimensional MDS (Multidimen-sional Scaling) space into distinct regions for each of the 10 values. Partitioningin the remaining countries yielded distinct regions for eight values and a jointregion for two values adjacent in the value circle. Whereas MDS focuses on thedifferences among items, CFA focuses on what they have in common. This mayexplain why CFA supports fewer distinctions. It is apparently more sensitiveto the very limited number of items per construct, especially because items in-tended to measure theoretically adjacent constructs are necessarily substantiallycorrelated.

We next turn to our second question: To what extent do the values thatare measured have equivalent meanings across countries? We address thisquestion by testing measurement or metric invariance in a second step ofthe analysis with MGCFA. This step constrains the factor loadings betweenthe indicators and the constructs in the model to be the same in all of thecountries. If the factor loadings are invariant, we can conclude that the meaningof the values, as measured by the indicators in the ESS, may be identical inthe 20 countries. Because configural invariance is a prerequisite for metricinvariance, we conducted this test on the model of seven value factors includingthe modifications made in the test of configural invariance (see broken lines infigure 2).

Table 5 summarizes the fit indices for sequentially more constrained models.The first row repeats the indices for the unconstrained, configural invariancemodel. The second row provides the indices for the metric invariance modelthat constrains the loadings of each item on its respective factor to be thesame in each of the 20 countries. These indices suggest a reasonable fit forthe metric invariance model too, a fit sufficient not to reject the model. Theincrease in AIC and BCC shows that one does better allowing variation inthe loadings, but the constrained model is quite good itself (Hu and Bentler1999; Marsh, Hau, and Wen 2004) (RMR = 0.08, NFI = 0.89, CFI = 0.91,


Tabl

e5.

Fit

Mea

sure

sfo

ra

Mul

ti-G

roup

Con

firm

ator

yFa

ctor

Ana

lysi

sof

Seve

nV

alue

s,C

onst

rain

ing

Met

ric

(Mea

sure

men

t),

Scal

ar,a

ndSt

ruct

ural

Cov

aria

nces

Inva

rian

ceac

ross

20C

ount

ries

∗

Mod

elty

peR

MR

NFI

CFI

RM

SEA

PCL

OSE

AIC

BC

CC

hisq

uare

df

1.U

ncon

stra

ined

:con

figur

alin

vari

ance

only

0.06

40.

901

0.91

40.

012

1.00

024

,589

24,6

2221

,869

3,26

02.

Met

ric

(mea

sure

men

t)in

vari

ance

0.07

50.

890

0.90

50.

012

1.00

026

,227

26,2

5124

,229

3,62

13.

Scal

arin

vari

ance

–∗∗0.

802

0.81

60.

016

1.00

045

,894

45,9

2143

,588

3,88

74.

Stru

ctur

alco

vari

ance

sin

vari

ance

0.11

50.

860

0.87

60.

013

1.00

031

,805

31,8

1630

,871

4,15

3

∗ RM

R=

root

mea

nsq

uare

resi

dual

;N

FI=

the

Ben

tler-

Bon

ett

norm

edfit

inde

x;C

FI=

the

com

para

tive

fitin

dex;

RM

SEA

=ro

otm

ean

squa

reer

ror

ofap

prox

imat

ion;

PCL

OSE

=pr

obab

ility

ofcl

ose

fit;A

IC=

the

Aka

ike

info

rmat

ion

crite

rion

;BC

C=

the

Bro

wne

-Cud

eck

crite

rion

;df=

the

num

ber

ofde

gree

sof

free

dom

.For

deta

ilsse

efo

rex

ampl

eA

rbuc

kle

(200

5).

∗∗T

heR

MR

inde

xis

notp

rovi

ded

byth

epr

ogra

mA

mos

whe

nm

eans

and

inte

rcep

tsar

ees

timat

ed.


Table 6. Unstandardized Regression Coefficients for 20 Countries in the Mea-surement Invariant Model (All Coefficients are Significant, p < 0.01)

SD UNBE COTR SEC POAC HE ST

SD1 1.00SD11 0.87UN3 1.00UN8 1.14UN19 1.11BE12 1.03BE18 1.19TR9 0.87 −0.29TR20 0.85CO7 1.00CO16 1.11SEC4 1.00SEC15 0.99PO2 −0.50 0.03 1.00PO17 0.37 0.78AC4 1.19AC13 1.28HE10 1.00HE21 1.01ST6 1.00ST15 −0.99 1.35

The path coefficients added to the original model in the modification process are in bold. Emptycells represent no direct relation between the values and the indicators.

RMSEA = 0.01 and PCLOSE = 1.0).10 All factor loadings in the configurallyinvariant model are high, including those that are statistically differentacross countries. The statistical differences and similarities may supportpartial measurement invariance of some of the constructs in some countries.However, the constrained (metrically invariant) model is substantially moreparsimonious and meaningful than the model with only configural invariance,and it is reasonable to accept it (Little 1997).

Table 6 summarizes the invariant factor loadings between indicators and val-ues for the 20 countries in the metric invariance model. All the item loadings forthe metric invariance model are significant. Appendix reports the standardizedfactor loadings for each country separately. The items with added paths basedon the modification indices are at the bottom of the table.

The third step of the MGCFA tests for scalar invariance, a necessary condi-tion for comparing value means across countries. This step is augmented with

10. As the sample size is very large, we do not apply the chi-square difference test (Cheung andRensvold 2002).


mean-structure information (see Sorbom 1974; Sorbom 1978).11 It constrainsthe intercepts of the indicators in the model, in addition to the factor loadingsbetween the indicators and the constructs, to be the same in all of the 20 coun-tries. If the factor loadings and the intercepts are invariant, one can legitimatelycompare value means. Where scalar invariance holds, value means should becomputed as parameters of the Structural Equation Modeling (SEM) modeland not from composite scores calculated from the observed variables. Thisis because SEM controls for measurement errors of the observed indicators(Sorbom 1974).

The fit indices for the scalar invariance model suggest that one should rejectthis more restrictive model. The increase in AIC and BCC is very large, and bothCFI and NFI decrease substantially. Only PCLOSE and RMSEA suggest a goodfit (NFI = 0.80, CFI = 0.82, RMSEA = 0.02 and PCLOSE = 1.0). We thereforeconclude that the scale does not meet the requirements of scalar invariance. Inother words, one cannot use these data to compare priorities on all seven valuesacross the full set of 20 countries. However, partial scalar invariance may holdfor some or all of the values across subsets of countries.12 This would allowmean comparisons of these values across the countries in the subset (Byrne,Shavelson, and Muthen 1989; Steenkamp and Baumgartner 1998).

To illustrate, consider the subset of Denmark and Spain. To assess partialscalar invariance, we first set all intercepts to be equal in both countries(i.e., we constrain full scalar invariance). Then, following the modificationindices suggested by the program, we released the equality constraints forthe intercepts of the indicators TR9, CO16, UN3, BE18, and AC13. Thisresulted in an acceptable model fit (CFI = 0.92, NFI = 0.90, RMSEA =0.04, Pclose = 1.00), permitting mean comparisons of all seven values.13

The comparisons reveal that people in Spain attribute significantly moreimportance to power/achievement (mean difference 0.42), security (0.91),conformity/tradition (0.10), and universalism/benevolence (0.12) values. Bycontrast, Danish people attribute significantly more importance to hedonism(0.40) and self-direction (0.11) values. People in the two countries do notdiffer in their mean level of stimulation values.

Following similar procedures, one can find the items whose equality con-straints should be released to obtain partial scalar invariance across varioussubsets of countries. We found, for example, that it is legitimate to comparethe importance to Danes and to Belgians of all seven values. Partial scalar

11. This type of MGCFA is often referred to in the literature as Mean and Covariance Structure(MACS) modeling.12. In partial scalar invariance, the intercepts of at least two indicators per construct must beequal.13. In MGCFA, mean comparison of latent variables is done by setting the mean values of theconstructs to zero in one group and then estimating the mean difference in the other groups.Mean differences are on the original scale net of measurement error. Arbuckle (2005) providescomputation details. Alternatives are proposed by Little, Slegers and Card (2006).


invariance also allows mean comparisons between Belgium and Spain for fivevalues (security, self-direction, power/achievement, conformity/tradition, anduniversalism/benevolence). Researchers interested in comparisons among par-ticular countries can test which values show full or partial scalar invariance forthose countries and may be compared. In sum, although one should not com-pare the seven values simultaneously across the 20 countries, one can comparemeans for some values across subsets of countries where scalar invariance orpartial scalar invariance hold.

As noted above, the value theory does not specify the distances of the valueconstructs from one another around the circle. As a further question, we askedwhether the pattern of distances is the same across countries. We tested thispossibility by constraining the covariances among the factors to be the same.The fourth row in table 5 reports the fit statistics for this model. The RMR andNFI fit indices indicate a relatively poor fit (RMR = 0.12, NFI = 0.86, CFI =0.88), and substantial increases in AIC and BCC also indicate a significant lossof fit. We reject therefore a model which postulates equivalence of structuralcovariances across countries. The relations among the seven values are notinvariant. The pattern of distances between the values is not the same in allcountries. This finding suggests that societal factors may influence the strengthof opposition and/or congruence among the different values.14

Discussion and Conclusion

The semi-annual ESS is a major source of data for researchers across the socialsciences. Its 21-item human values scale, designed to measure the full range ofdistinct, basic values in population surveys, has quickly attracted interest. Sixpresentations at the 1st European Association for Survey Research Conferencediscussed data using the scale. Given the interest in this new scale and itspotential use in surveys around the world, it is crucial to assess its validity.This article examined the fit of the ESS values instrument to the theory onwhich it was based. It then investigated the appropriateness of this instrumentfor studying values across 20 countries, using data from the first round of theESS. We addressed three questions:

(1)How well does the ESS human values scale measure the 10 basic valuesin the Schwartz theory?

(2)To what extent do the values that are measured have equivalent meaningsacross countries?

(3)To what extent can the values scale be used to compare latent means acrossthe 20 countries?

14. For example, using MDS to assess the value structure, Fontaine et al. (2008) find that theopposition of self-direction and universalism values to power and security values is greater insocieties with a more autonomous culture and with higher socio-economic development.


Separate CFAs in each country supported the adequacy of the ESS scalefor measuring the 10 values of the Schwartz theory. However, very high cor-relations between some pairs of values in these analyses and in the MGCFAacross countries caused a problem of non-positive definite covariance matrices.To solve this problem, we unified these pairs into single value constructs.This yielded seven distinct values in the MGCFA: security, self-direction,stimulation, hedonism and combined tradition/conformity, universalism/benevolence, and power/achievement values. We further modified the multi-group model by adding paths from four items to other constructs. This modelsupported the configural invariance of the seven value factors across the20 countries.

The three combined value constructs were formed by unifying values thatare adjacent in the value structure. Hence, the ESS human values scale cap-tures the motivational circle of values in the theory on which it is based. Theemergence of only seven distinguishable values may be a consequence of thefact that the ESS included only 21 items to measure all 10 value constructs,constructs that are correlated according to the theory. This compares with 40items in the full PVQ and 45 in the Schwartz Value Survey, both of whichdiscriminated the 10 values successfully (Schmidt et al. 2007); Schwartz andBoehnke 2004). As noted, MDS analyses of the ESS scale in 20 countries, anapproach less sensitive to high correlations among items, supported discrim-ination of the 10 value constructs (Schwartz 2007). Nonetheless, the currentanalyses suggest one can use the scale to measure only seven distinct values withconfidence.

The critical question for use of the values scale in different countries andfor comparing value relations in one country to those in another is whether thevalues it measures have similar meanings in each country. The test of mea-surement (metric) invariance addressed this question. It led to the conclusionthat the meaning of the values, as measured by the indicators of the ESS, isprobably the same in the 20 countries. Thus, the ESS human values scale, whenused to measure seven distinct values, meets the tests of configural and metricinvariance of the latent factors across countries. In spite of cultural differences,people in Europe appear to understand the meaning given to the values by theirindicators in a similar manner.15

The final question concerned the comparability of value means acrosscountries. The test of scalar invariance addressed this question. The scalefailed to exhibit scalar invariance across the 20 countries. Hence, one shouldnot compare the mean importance of the values across all 20 countries

15. The current research reveals the levels of configural, metric, and scalar invariance of theESS values instrument only in the countries studied here. One cannot generalize from this tofuture studies in these or other countries. Such studies should repeat the current analyses toreassess invariance. The statistical method presented here can establish necessary conditions forequivalence of meaning. Cognitive interviews offer a supplementary tool to assess the equivalenceof meaning of the value instrument in various countries.


simultaneously. However, as illustrated for Denmark and Spain, one can com-pare means for values across subsets of countries where scalar invariance orpartial scalar invariance are found.

The findings of the current research justify employing the human value scalein survey research for numerous purposes. One can examine change in valuescores across time as an indicator of fundamental societal change in responseto historical, demographic, and social structural developments (e.g., impactson peoples’ basic values of wars, dropping birthrates, and inflation). Values,measured with the ESS scale may also be used to understand differences amongnational populations in their responses to government policies (e.g., towardimmigration) and to major events (e.g., terror attacks). Future studies mayalso address the way different values mediate the effects of individuals’ socio-demographic characteristics (e.g., age, education, occupation) on their attitudesand opinions (e.g., prejudice against out-groups, trust in institutions).

Because the human value scale demonstrates meaning equivalence acrosscountries, researchers can also legitimately assess and draw conclusions aboutsimilarities and differences in the relations of value priorities to other vari-ables across countries (Van de Vijver and Leung 1997; De Beuckelaer 2005,Ch. 5). Within country studies have shown many meaningful associations be-tween individuals’ value priorities and their attitudes (e.g., political preferences,left-right orientation, views on abortion, marriage, and religion) and behavior(e.g., voting, political activism, membership in voluntary organizations) (seesummaries in Schwartz 2005b, 2006a, 2007). Researchers can now test whetherthese and other patterns of value/attitude and value/behavior relations general-ize across countries. If they find differences between countries in relations ofvalue priorities to attitudes and behavior, the evidence for meaning equivalencemakes it legitimate to seek explanations for such country differences.16

Without establishing configural and metric invariance, as done in the currentresearch, none of the above applications of the ESS human values scale couldbe undertaken with confidence. This study provides the critical legitimacyfor such comparative work—evidence for equivalence of the meanings of thevalues across countries. As interdependence among nations increases, suchcomparative work becomes more and more important. Indeed, the problem ofmeaning equivalence applies to within country studies of diverse ethnic groupsas well. The current research can serve as an example of what needs to be doneto assess equivalence of meaning, rather than to assume it, as has been typicalin comparative survey research.

16. In studies that relate individuals’ values to other variables, it is crucial to correct in-dividual differences in use of the response scale, as explained in Schwartz (2003, 275)and Schwartz (2006a). The ESS website describes procedures for making this correction(http://www.europeansocialsurvey.org/).


App

endi

x:S

tand

ardi

zed

Fact

or

Lo

adin

gsan

dC

ross

Lo

adin

gsfo

rea

chC

oun

try

(Bas

edo

nth

eM

etri

cIn

vari

ance

Mo

del)

Item

Val

uefa

ctor

AT

BE

CH

CZ

DE

DK

ES

FIFR

GB

GR

HU

IEIL

NL

NO

PLPT

SESL

HE

10H

E0.

710.

680.

750.

760.

800.

690.

800.

820.

750.

770.

590.

600.

730.

760.

610.

760.

810.

800.

730.

68H

E21

HE

0.62

0.66

0.65

0.73

0.75

0.71

0.80

0.83

0.66

0.78

0.64

0.68

0.75

0.75

0.64

0.77

0.80

0.67

0.75

0.69

ST6

ST0.

720.

660.

610.

640.

660.

720.

680.

740.

630 .

700.

710.

600.

690.

690.

680.

740.

670.

730.

720.

64ST

15ST

0.92

0.79

0.79

0.86

0.84

0.96

0.91

0.89

0.83

0.92

0.80

0.74

0.91

0.84

0.89

0.98

0.87

0.93

0.95

0.74

PO2

POA

C0.

600.

590.

600.

660.

630.

560.

580.

670.

620.

660.

550.

590.

630.

490.

630.

660.

550.

590.

650.

52PO

17PO

AC

0.47

0.45

0.41

0.51

0.45

0.43

0.45

0.54

0.42

0.48

0.45

0.44

0.47

0.40

0.49

0.50

0.40

0.44

0.50

0.44

SEC

5SE

C0.

720.

660.

660.

610.

670.

600.

750.

650.

740.

610.

720.

700.

630.

560.

640.

620.

580.

680.

640.

61SE

C14

SEC

0.70

0.64

0.67

0.57

0.65

0.57

0.69

0.60

0.73

0.59

0.71

0.66

0.61

0.59

0.68

0.60

0.60

0.67

0.58

0.62

CO

7C

OT

R0.

610.

510.

530.

570.

590.

600.

540.

630.

550.

630.

460.

460.

600.

560.

590.

620.

600.

580.

590.

54C

O16

CO

TR

0.74

0.64

0.59

0.69

0.71

0.65

0.74

0.70

0.70

0.75

0.66

0.66

0.73

0.71

0.72

0.67

0.71

0.70

0.64

0.69

UN

3U

NB

E0.

600.

450.

420.

500.

470.

370.

620.

520.

510.

470.

580.

460.

560.

490.

490.

460.

550.

630.

540.

49U

N8

UN

BE

0.63

0.51

0.54

0.57

0.54

0.48

0.66

0.58

0.59

0.58

0.60

0.56

0.60

0.51

0.57

0.57

0.49

0.65

0.59

0.51

SD1

SD0.

680.

540.

470.

630.

600.

570.

680.

520.

510.

600.

650.

530.

540.

460.

530.

620.

530.

670.

580.

54SD

11SD

0.69

0.54

0.52

0.60

0.60

0.54

0.67

0.47

0.44

0.58

0.64

0.56

0.56

0.50

0.55

0.52

0.60

0.68

0.52

0.55

UN

19U

NB

E0.

670.

490.

520.

640.

550.

490.

660.

560.

580.

550.

660.

670.

590.

470.

560.

500.

620.

680.

540.

57A

C4

POA

C0.

710.

650.

610.

740.

690.

620.

670.

760.

660.

720.

670.

680.

720.

650.

700.

710.

640.

670.

710.

64A

C13

POA

C0.

770.

710.

690.

770.

750.

700.

760.

820.

700.

770.

720.

750.

770.

740.

790.

780.

740.

770.

790.

74T

R20

CO

TR

0.56

0.46

0.47

0.45

0.51

0.49

0.51

0.49

0.47

0.50

0.58

0.48

0.57

0.47

0.51

0.48

0.57

0.59

0.47

0.54

TR

9C

OT

R0.

570.

480.

520.

490.

550.

470.

580.

500.

560.

540.

480.

540.

540.

560.

470.

490.

460.

590.

520.

62B

E18

UN

BE

0.69

0.56

0.58

0.61

0.60

0.62

0.67

0.56

0.60

0.57

0.61

0.60

0.61

0.58

0.59

0.54

0.62

0.70

0.62

0.49

BE

12U

NB

E0.

680.

580.

560.

590.

590.

580.

720.

590.

610.

620.

650.

630.

640.

570.

610.

600.

620.

770.

640.

60T

R9

POA

C−0

.17

−0.1

8−0

.17

−0.2

0−0

.18

−0.1

5−0

.21

−0.1

9−0

.18

−0.1

8−0

.20

−0.2

0−0

.19

−0.1

6−0

.17

−0.1

7−0

.18

−0.1

9−0

.18

−0.1

9PO

2U

NB

E−0

.23

−0.1

6−0

.18

−.21

−0.2

0−0

.17

−.22

−0.2

0−0

.22

−0.2

1−0

.19

−0.2

2−0

.21

−0.1

7−0

.20

−.21

−.18

−.25

−0.2

3−0

.18

PO2

CO

TR

0.02

0.02

0.02

0.02

0.02

0.02

0.02

0.02

0.02

0.02

0.01

0.02

0.02

0.02

0.02

0.02

0.02

0.02

0.02

0.02

ST15

UN

BE

−0.4

2−0

.28

−0.3

0−0

.38

−0.3

6−0

.31

−0.4

0−0

.34

−0.4

0−0

.39

−0.3

2−0

.38

−0.3

8−0

.31

−0.3

7−0

.36

−0.3

2−.

43−0

.41

−0.3

3PO

17C

OT

R0.

250.

200.

210.

200.

230.

230.

200.

230.

210.

240.

180.

190.

220.

230.

220.

230.

170.

230.

230.

23

The

last

5ro

ws

indi

cate

the

path

coef

ficie

nts

adde

dto

the

orig

inal

mod

elin

the

mod

ifica

tion

proc

ess.


References

Allport, Gordon W., Philip E. Vernon, and Gardner A. Lindzey. 1960. A Study of Values. Boston:Houghton Mifflin.

Arbuckle, James L. 2005. Amos 6.0 User’s Guide. Chicago, IL: SPSS.Barnea, Marina. 2003. Personal Values and Party Orientations in Different Cultures. Unpublished

doctoral dissertation, Israel: The Hebrew University of Jerusalem.Barnea, Marina, and Shalom H. Schwartz. 1998. “Values and Voting.” Political Psychology 19:17–

40.Billiet, Jaak. 2003. “Cross-Cultural Equivalence with Structural Equation Modeling.” In Cross-

Cultural Survey Methods, eds. J. A. Harkness, F. J. R. Van de Vijver, and P. Ph. Mohler, pp.247–64. New York: John Wiley.

Bollen, Kenneth A. 1989. Structural Equations with Latent Variables. New York: Wiley.Brown, Roger L. 1994. “Efficacy of the Indirect Approach for Estimating Structural Equation

Models with Missing Data: A Comparison of Five Methods.” Structural Equation Modeling1:287–316.

Burroughs, James E., and Aric Rindfleisch. 2002. “Materialism and Well-Being: A ConflictingValues Perspective.” Journal of Consumer Research 29:348–70.

Byrne, Barbara M. 2001. Structural Equation Modeling with AMOS. Basic Concepts, Application,and Programming. London: Lawrence Erlbaum Associates.

Byrne, Barbara, Richard J. Shavelson, and Bengt Muthen. 1989. “Testing for the Equivalenceof Factor Covariance and Mean Structures: The Issue of Partial Measurement Invariance.”Psychological Bulletin 105:456–66.

Caprara, Gian Vittorio, Shalom H. Schwartz, Cristina Capanna, Michele Vecchione, and ClaudioBarbaranelli. 2006. “Personality and Politics: Values, Traits, and Political Choice.” PoliticalPsychology 27:1–28.

Cheung, Gordon W., and Roger B. Rensvold. 2000. “Assessing Extreme and Acquiescence Re-sponse Sets in Cross-Cultural Research Using Structural Equations Modeling.” Journal ofCross-Cultural Psychology 31:187–212.

Cheung, Gordon W., and Roger B. Rensvold. 2002. “Evaluating Goodness-of-Fit Indexes forTesting Measurement Invariance.” Structural Equation Modeling 9:233–55.

Cohrs, J. Christopher, Barbara Moschner, Jurgen Maes, and Sven Kielmann. 2005. “Personal Valuesand Attitudes Toward War.” Peace and Conflict: Journal of Peace Psychology 11:293–312.

Converse, Philip E. 1964. “The Nature of Belief Systems in Mass Publics.” In Ideology andDiscontent, ed. D. Apter, pp. 206–61. New York: Free Press.

De Beuckelaer, Alain. 2005. Measurement Invariance Issues in International Management Re-search. PhD dissertation. Limburg: Limburgs Universitair Centrum.

Feldman, Stanley. 2003. “Values, Ideology, and Structure of Political Attitudes.” In Oxford Hand-book of Political Psychology, eds. D. O. Sears, L. Huddy, and R. Jervis, pp. 477–508. New York:Oxford University Press.

Fontaine, Johnny R. J., Ype H. Poortinga, Luc Delbeke, and Shalom H. Schwartz. 2008. “StructuralEquivalence of the Values Domain Across Cultures: Distinguishing Sampling Fluctuations fromMeaningful Variation.” Journal of Cross-Cultural Psychology 39: 345–65.

Halman, Loek, and Ruud deMoor. 1994. “Value Shift in Western Societies.” In The IndividualizingSociety: Value Change in Europe and North America, eds. P. Ester, L. Halman, and Ruud deMoor, pp. 1–20. Tilburg, the Netherlands: Tilburg University Press.

Harkness, Janet A., Fons J. R. Van de Vijver, and Peter Ph. Mohler (eds). 2003. Cross-CulturalSurvey Methods. New York: John Wiley.

Hitlin, Steven, and Jane Allyn Piliavin. 2004. “Values: Reviving a Dormant Concept.” AnnualReview of Sociology 30:359–93.

Horn, John L., and J. J. McArdle. 1992. “A Practical and Theoretical Guide to MeasurementInvariance in Aging Research.” Experimental Aging Research 18:117–44.


Hu, Litze, and Peter M. Bentler. 1999. “Cutoff Criteria for Fit Indexes in Covariance StructureAnalysis: Conventional Criteria Versus New Alternatives.” Structural Equation Modeling 6:1–55.

Joreskog, Karl G. 1971. “Simultaneous Factor Analysis in Several Populations.” Psychometrika36:409–26.

Kluckhohn, Clyde K. M. 1951. “Values and Value Orientations in the Theory of Action.” In Towarda General Theory of Action, eds. Talcott Parsons, and Edward Shils. Cambridge, MA: HarvardUniversity Press.

Knutsen, Oddbjørn. 1995. “Party Choice.” In The Impact of Values, eds. J. W. van Deth, and E.Scarborough, pp. 460–91. New York: Oxford University Press.

Little, Todd D. 1997. “Mean and Covariance Structures (MACS) Analyses of Cross-Cultural Data:Practical and Theoretical Issues.” Multivariate Behavioral Research 32:53–76.

Little, Todd D., David W. Slegers, and Noel A. Card. 2006. “A Non-arbitrary Method of Identifyingand Scaling Latent Variables in SEM and MACS Models.” Structural Equation Modeling 13:59–72.

Lubke, Gitta H., and Bengt O. Muthen. 2004. “Applying Multigroup Confirmatory Factor Modelsfor Continuous Outcomes to Likert Scale Data Complicates Meaningful Group Comparisons.”Structural Equation Modeling 11:514–34.

Marsh, Herbert W., Kit-Tai Hau, and Zhonglin Wen. 2004. “In Search of Golden Rules: Commenton Hypothesis-Testing Approaches to Setting Cutoff Values for Fit Indexes and Dangers inOvergeneralizing Hu and Bentler’s (1999) Findings.” Structural Equation Modeling 11:320–41.

Meredith, William. 1993. “Measurement Invariance, Factor Analysis and Factorial Invariance.”Psychometrika 58:525–43.

Miller, Warren E., and J. Merrill Shanks. 1996. The New American Voter. Cambridge, MA: HarvardUniversity Press.

Rock, Donaid A., Charles E. Werts, and Ronald L. Flaugher. 1978. “The Use of Analysis ofCovariance Structures for Comparing the Psychometric Properties of Multiple Variables acrossPopulations.” Multivariate Behavioral Research 13:403–18.

Rohan, Meg J. 2000. “A Rose by Any Name? The Values Construct.” Personality and SocialPsychology Review 4:255–77.

Rokeach, Milton. 1973. The Nature of Human Values. New York: Free Press.Schafer, Joseph L., and John W. Graham. 2002. “Missing Data: Our View of the State of the Art.”

Psychological Methods 7:147–77.Schmidt, Peter, Bamberg, Sebastian, Davidov, Eldad, Herrmann, Johannes, and H. Schwartz

Shalom. 2007. “Wie lassen sich Werte messen? Validierung des neuen Messinstruments ‘Por-traits Value Questionnaire’.” Zeitschrift fur Sozialpsychologie 38(4):249–63.

Schultz, P. Wesley, and Lynette Zelezny. 1999. “Values as Predictors of Environmental Attitudes:Evidence for Consistency across 14 Countries.” Journal of Environmental Psychology 19:255–65.

Schwartz, Shalom H. 1992. “Universals in the Content and Structure of Values: Theoretical Ad-vances and Empirical Tests in 20 Countries.” Advances in Experimental Social Psychology25:1–65.

Schwartz, Shalom H. 1994. “Are there Universal Aspects in the Content and Structure of Values?”Journal of Social Issues 50:19–45.

Schwartz, Shalom H. 2003. A Proposal for Measuring Value Orientations across Nations. Chapter7 in the ESS Questionnaire Development Report. http://www.europeansocialsurvey.org.

Schwartz, Shalom H. 2005a. “Basic Human Values: Their Content and Structure Across Countries.”In Valores e Comportamento nas Organizacoes [Values and Behavior in Organizations], eds.A. Tamayo, and J. B. Porto, pp. 21–55. Petropolis, Brazil: Vozes.

Schwartz, Shalom H. 2005b. “Robustness and Fruitfulness of a Theory of Universals in Individ-ual Human Values.” In Valores e comportamento nas organizacoes [Values and Behavior inOrganizations], eds. A. Tamayo, and J. B. Porto, pp. 56–95. Petropolis, Brazil: Vozes.


Schwartz, Shalom H. 2006a. “Basic Human Values: Theory, Measurement, and Applications.”Revue Francaise de Sociologie 47:249–88.

Schwartz, Shalom H. 2006b. “A Theory of Cultural Value Orientations: Explication and Applica-tions.” Comparative Sociology 5:137–82.

Schwartz, Shalom H. 2007. “Value Orientations: Measurement, Antecedents and Consequencesacross Nations.” In Measuring Attitudes Cross-Nationally – Lessons from the European SocialSurvey, eds. Roger Jowell, Caroline Roberts, and Rory Fitzgerald, pp. 161–93. London: Sage.

Schwartz, Shalom H., Gila Melech, Arielle Lehmann, Steven Burgess, Mari Harris, and VickiOwens. 2001. “Extending the Cross-Cultural Validity of the Theory of Basic Human Valueswith a Different Method of Measurement.” Journal of Cross Cultural Psychology 32:519–42.

Schwartz, Shalom H., and Klaus Boehnke. 2004. “Evaluating the Structure of Human Values withConfirmatory Factor Analysis.” Journal of Research in Personality 38:230–55.

Sorbom, Dag. 1974. “A General Method for Studying Differences in Factor Means and FactorStructure between Groups.” British Journal of Mathematical and Statistical Psychology 27:229–39.

Sorbom, Dag. 1978. “An Alternative to the Methodology for Analysis of Covariance.” Psychome-trika 43:381–96.

Spini, Dario, and Willem Doise. 1998. “Organizing Principles of Involvement in Human Rights andtheir Social Anchoring in Value Priorities.” European Journal of Social Psychology 28:603–22.

Steenkamp, Jan-Benedict E. M., and Hans Baumgartner. 1998. “Assessing Measurement Invariancein Cross National Consumer Research.” Journal of Consumer Research 25:78–90.

Vandenberg, Robert J. 2002. “Towards a Further Understanding of and Improvement in Measure-ment Invariance Methods and Procedures.” Organizational Research Methods 5:139–58.

Vandenberg, Robert J., and Charles E. Lance. 2000. “A Review and Synthesis of the Measure-ment Invariance Literature: Suggestions, Practices and Recommendations for OrganizationalResearch.” Organizational Research Methods 3:4–69.

Van de Vijver, Fons J. R., and Kwok Leung. 1997. “Methods and Data Analysis of ComparativeResearch.” In Handbook of Cross-Cultural Psychology, eds. J. W. Berry, Y. H. Poortinga, andJ. Pandey, pp. 257–300. Chicago, IL: Allyn and Bacon.

Welkenhuysen-Gybels, Jerry. 2004. “The Performance of Some Observed and Unobserved Con-ditional Invariance Techniques for the Detection of Differential Item Functioning.” Quality &Quantity 38:681–207.

Welkenhuysen-Gybels, Jerry, and Jaak Billiet. 2002. “A Comparison of Techniques for DetectingCross-Cultural Inequivalence at the Item Level.” Quality & Quantity 36:197–218.

Williams, Robin M., Jr. 1968. “Values.” In International Encyclopedia of the Social Sciences, ed. Edward Sills. New York: Macmillan.

Zaller, John R. 1992. The Nature and Origins of Mass Opinion. Cambridge, UK: CambridgeUniversity Press.

the adequacy of the european social survey to measure values in 20 countries

Documents