Reverse Coded Issue 1

8/13/2019 Reverse Coded Issue 1

1/20

72

2003 by JOURNAL OF CONSUMER RESEARCH, Inc. Vol. 30 June 2003

All rights reserved. 0093-5301/2004/3001-0006$10.00

Do Reverse-Worded Items Confound Measures

in Cross-Cultural Consumer Research? The

Case of the Material Values Scale

NANCY WONGARIC RINDFLEISCHJAMES E. BURROUGHS*

Most measures of consumer behavior have been developed and employed in theUnited States. Thus, relatively little is known about the cross-cultural applicabilityof these measures. Using Richins and Dawsons (1992) Material Values Scale(MVS) as an exemplar, this article focuses on the problems researchers are likelyto encounter when employing domestic mixed-worded scales (i.e., scales that

contain both positive- and reverse-worded items) in cross-cultural applications.Through an initial study among over 800 adults from the United States, Singapore,Thailand, Japan, and Korea, we show that the cross-cultural measurement equiv-alence and construct validity of the MVS is challenged by its mixed-worded Likertformat. Through a second study among approximately 400 Americans and EastAsians, we find that other mixed-worded scales produce similar problems and thatthe cross-cultural applicability of such scales may be enhanced by replacing itemsposed as statements with items framed as questions.

C onsumer researchers widely recognize the importanceof expanded intellectual boundaries. In particular,there have been several calls to expand the geographic

scope of consumer research to include participants fromcultures outside the United States. For example, in his 1993Association for Consumer Research (ACR) Fellow Speech,Bagozzi (1994) called for exploration of cross-culturaldynamics (p. 9). Similarly, Gorn (1997) recently encour-aged consumer researchers to break out of the NorthAmerican box (p. 6). These calls are increasingly beingheeded, as cross-cultural investigations are published inour fields leading journals with some degree of regularity

*Nancy Wong is assistant professor of marketing at Georgia Institute ofTechnology, DuPree College of Management, Atlanta, Georgia 30332([email protected]). Aric Rindfleisch is assistant professor ofmarketing at the University of WisconsinMadison, School of Business,

Madison, Wisconsin 53706 ([email protected]). James E. Burroughs isassistant professor of commerce at the University of Virginia, McIntireSchool of Commerce, Charlottesville, Virginia 22904 ([email protected]). The authors thank Rick Bagozzi, Hans Baumgartner, RobertFischer, Shankar Ganesan, Gerry Gorn, DawnIacobucci, DavidMick, ChrisMoorman, John Murry, Terry Shimp, seminar participants at the Universityof Hawaii, the University of Pennsylvania, and the University of WesternOntario, and the editors and reviewers for their helpful comments on thisarticle. This research was funded by grants from the Center for InternationalBusiness Education and Research (CIBER) at both Georgia Tech and theUniversity of Hawaii and a grant from the McIntire School at the Universityof Virginia.

(e.g., Aaker 2000; Alden, Steenkamp, and Batra 1999;Steenkamp and Baumgartner 1998; Williams and Aaker2002). As technology provides researchers with increased

global access, this trend should intensify.As noted by a number of scholars, cross-cultural researchpresents a host of challenges not faced domestically (Baum-gartner and Steenkamp 2001; Durvasula et al. 1993; Van deVijver and Leung 1997). Most important, cross-cultural re-searchers need to insure that their concepts and methods areappropriate and valid across all cultures under investigation.This can be a formidable task, as the vast majority of theseconcepts and measures have been designed by Americans andmay not necessarily be cross-culturally applicable (Bagozzi1994; Gorn 1997; Wallendorf and Arnould 1988).

In order to meet these challenges, recent work has providedconsumer researchers with a number of tools and techniquesfor assessing and improving the cross-cultural validity of con-

sumer behavior scales (Baumgartner and Steenkamp 2001;Durvasula et al. 1993; Steenkamp and Baumgartner 1998).Our article seeks to continue this important line of researchby examining a potentially serious threat to the cross-culturalvalidity of consumer behavior measures and offering a pos-sible solution to this problem. In brief, the threat comes fromthe common practice of employing Likert scales that containa mixture of positive-worded items (PWI) andreverse-wordeditems (RWI). Although mixed-worded scales are commonlyused in domestic applications, we contend that they can create


2/20

REVERSE WORDING IN CROSS-CULTURAL RESEARCH 73

a host of problems when measuring consumer behavior inforeign cultures.

The use of mixed-worded scales has a long and contro-versial history. Proponents of mixed wording suggest thatthis practice reduces the dangers of response bias such asacquiescence (Churchill 1979; Nunnally 1978). Critics sug-gest this mixture may lessen a scales internal consistencyand disrupt its dimensionality (Cronbach 1950; Falthzik andJolson 1974). An examination of mixed-worded scales inpractice reveals that while some scales, such as the ChangeSeeking Index (Steenkamp and Baumgartner 1995), appearto work well in American and western European samples,others, such as SERVQUAL (Parasuraman, Berry, and Zei-thaml 1991), do not. A review of the literature suggests thatproblems with mixed-worded scales most often arise formembers of subcultural groups, such as ethnic and racialminorities (e.g., Bachman and OMalley 1984; Schmitz andBaer 2001; Steenkamp and Burgess 2002). This bias appearsto be even more pronounced when domestic scales are ap-

plied in foreign cultures that differ greatly from the domesticculture in terms of values, customs, and language. For ex-ample, a spate of recent studies has found that mixed-wordedmeasures of psychological functioning (e.g., optimism, ag-gression, anxiety) that are unidimensional for Americanssplit into separate positive- versus reverse-wording factorsfor East Asians (e.g., Cheng and Hamid 1997; Lai and Yue2000; Nakano 2001; Suzuki, Tsukamoto, and Abe 2000).

Although prior research suggests that the problems as-sociated with mixed-worded scales are more pronounced incross-cultural settings, the reasons why this occurs are un-clear. Moreover, prior research offers few solutions for howto deal with this problem. Our article seeks to address these

issues by investigating the possible mechanisms behindthese cross-cultural differences and testing a measurement-based solution. In addition, compared to the extant litera-ture, which largely focuses on psychological constructs, em-

ploys student samples, and examines only a single culture(e.g., Cheng and Hamid 1997; Lai and Yue 2000; Meloniand Gana 2001; Nakano 2001), our research focuses onconsumer-related constructs, employs adult subjects, and ex-amines five (national) cultures (i.e., United States, Singa-pore, Thailand, Japan, and Korea).

To accomplish these goals, we present two studies thatfocus on the cross-cultural applicability of mixed-wordedscales, using the Material Values Scale (MVS) as an ex-

emplar. In general, our results show that the cross-culturalapplicability of the MVS is confounded by its mixed-wordedformat. In addition, we find that the use of this format alsoconstrains the cross-cultural applicability of other commonlyused scales in consumer research. Finally, by adapting theMVSs statements into a set of nondirectional questions, wefind the problems associated with mixed-wording scales arelargely alleviated. Before presenting our research, we firstprovide an overview of the practice and potential problemsassociated with mixed-worded scales.

A REVIEW OF THE PRACTICE OFMIXED-WORDED SCALES

Rationale and Employment in Domestic Research

Multi-item Likert scales are a common and recommendedmeans of collecting data on attitudes, beliefs, values, andother latent consumer behavior constructs (Peterson 1994).As a means of avoiding the response biases associated withmulti-item scales that are worded in a single direction (e.g.,acquiescence, straight-line responding, etc.), psychometri-cians often recommend reverse wording (e.g., Baumgartnerand Steenkamp 2001; Churchill 1979; Couch and Kenniston1960; Nunnally 1978). According to this view, multi-itemLikert scales should be balanced by having an equal numberof PWI and RWI to allow any form of systematic responsebias to cancel out. The use of RWI has now become standardpractice in marketing and consumer research. Forty-eight

percent of the scales listed in the Handbook of MarketingScales (Bearden and Netemeyer 1999) use RWI, includingsuch prominent scales as the MVS (Richins and Dawson1992), the Style of Processing scale (SOP; Childers, Hous-ton, and Heckler 1985), and the Exploratory Tendencies inConsumer Behavior scale (ETCB; Raju 1980).

Although mixed-worded scales may reduce the risksof certain types of response biases, the inclusion of RWImay produce unintended problems by reducing a scalesinternal consistency and obscuring its dimensionality (Ben-son and Hocevar 1985; Goldsmith and Desborde 1991;Schriesheim and Hill 1981). Researchers who employmixed-worded scales often find that RWI display some-what lower reliability and weaker item-to-total correlations

than their positive-worded counterparts (Cronbach 1942;Benson and Hocevar 1985; Peabody 1966). Furthermore,when subjected to factor analysis, the RWI often load ona separate factor (Benson and Hocevar 1985; Herche andEngelland 1996; Pilotte and Gable 1990).

Despite the potential dangers posed by RWI, a large num-ber of scales in marketing and consumer behavior use amixed-worded format. Although RWI have been found tobe problematic for some scales (cf. Herche and Engelland1996), most marketing and consumer researchers appear tooperate under the belief that mixed-worded items are rela-tively unproblematic in general. This belief may be buoyedby the fact that many of these mixed-worded scales, suchas the MVS, consistently display good reliability in studies

conducted in the United States (e.g., Mick 1996; Richinsand Dawson 1992; Rindfleisch, Burroughs, and Denton1997). The MVS is not an isolated example. Peter (1981)finds that, on average, mixed-worded marketing and con-sumer behavior scales are just as reliable ( ) as scalesa p .75consisting of only PWI ( ). In sum, domestic con-a p .73sumer behavior researchers appear to have found that manyof the mixed-worded scales they employ do not greatly suf-fer from attenuation of reliability or other psychometricdeficiencies.


3/20

74 JOURNAL OF CONSUMER RESEARCH

Problems Posed by Mixed-Worded Scales inCross-Cultural Settings

Although the extent of the problems associated withmixed-worded scales remains a point of contention in do-

mestic settings, the evidence for severe problems in cross-cultural applications is rapidly mounting. For example, al-though mixed-worded materialism scales appear to workreasonably well among Americans, consumer researchersoften find that these measures have questionable reliabilityand validity in cross-cultural applications (e.g., Eastman etal. 1997; Ger and Belk 1996; Wallendorf and Arnould 1988).Similarly, cross-cultural psychologists also find that mixed-worded measures of individual well-being that are psycho-metrically sound in domestic applications are deficient inforeign settings (e.g., Iwata, Roberts, and Kawakami 1995;Iwata, Saito, and Roberts 1994; Meloni and Gana 2001;Nakano 2001).

The reasons why a mixed-worded format creates prob-

lems in cross-cultural contexts are not fully understood.However, there appear to be two possible mechanisms un-derlying diminished internal consistency and disrupted di-mensionality in cross-cultural applications. First, these prob-lems could be because subjects have trouble understandingand responding to RWI. These problems are most pro-nounced for subjects who are young and uneducated (Marsh1996; Pilotte and Gable 1990; Schmitz and Baer 2001).Psychometricians widely agree that it is difficult to designgood RWI and that the negations and contradictions em-ployed as markers of RWI can be confusing (McClendon1991; Rorer 1965; Schuman and Presser 1981). These prob-lems are likely to be exacerbated when domestic mixed-worded measures are translated into languages that employ

different ways of marking negation or contradiction, suchas Japanese or Chinese (Bloom 1981). This view suggeststhat the confusion surrounding RWI introduces unwantednoise, which attenuates a scales psychometric properties(i.e., reducing reliability and obscuring dimensionality).

In short, mixed-worded scales are hampered by the in-consistency in how subjects respond to RWI versus PWI.This appears to be the dominant view shared by most re-searchers who uncover mixed-worded biases (e.g., Marsh1996; Meloni and Gana 2001; Schmitz and Baer 2001).Because they view RWI as confounded and a source ofunwanted noise, the solution typically offered by thesescholars is to remove these offending items. If this view isvalid, analysis of a mixed-worded scales correlations should

reveal that RWI display substantially weaker within-groupintercorrelations than PWI. To illustrate using the MVS asan example, if this is the case we should find that, on av-erage, the correlations between PWI (e.g., My life wouldbe better if I owned certain things I dont have, and Id behappier if I could afford to buy more things) are substantiallyhigher than the correlations between reverse-worded items(e.g., I wouldnt be any happier if I owned nicer things, andI have all the things I really need to enjoy life).

An alternative (and emerging) view is that the problems

associated with mixed-worded scales are because subjectsinterpret and respond to these two sets of items quite dif-ferently (Cheng and Hamid 1997; Javeline 1999; Lai andYue 2000). Rather than representing opposite ends of thesame construct (as assumed by measurement theory), sub-

jects might view PWI and RWI as unrelated. This differ-

ential in how PWI and RWI are interpreted (and respondedto) may be influenced by cultural norms or values. For ex-ample, in some cultures, agreeableness is an important socialnorm, particularly in parts of Asia, where appearing politeis expected. Researchers have suggested that this norm mayprompt subjects to agree to both PWI and RWI (Hui andTriandis 1983; Javeline 1999). If ones dominant norm isto agree with statements irrespective of content, mixed-worded scales should display low reliability and weakerdimensionality. Thus, this view suggests that mixed-wordedscales are hampered by incompatibility in subjects re-sponses to PWI versus RWI. Because of this incompatibilityin the very nature of the items themselves, some proponentsof this view suggest that mixed-worded scales need to be

adapted to formats that are less susceptible to this danger(e.g., Hui and Triandis 1989; Javeline 1999). If this view isvalid, an analysis of a mixed-worded scales correlationsshould reveal that PWI and RWI display weak between-group correlations. To clarify, again using the MVS as anexemplar, if this is the case we should find the correlationsamong items worded in opposite directions (e.g., My lifewould be better if I owned certain things I dont have, andI wouldnt be any happier if I owned nicer things) to besubstantially lower on average (even after recoding) thanthe correlations between items worded in the same direction(e.g., I wouldnt be any happier if I owned nicer things, andI have all the things I really need to enjoy life).

In sum, both views suggest that, when applied cross-

culturally, mixed-worded scales are likely to face difficultyin terms of obtaining adequate internal consistency and di-mensionality. Moreover, in addition to creating difficultiesin terms of establishing measurement equivalence, thesecross-cultural differences may also pose a serious threat tothe construct validity of mixed-worded scales, leading topotentially misleading cross-cultural comparisons. As a re-sult, these cross-cultural applications may lack both mea-surement equivalence and construct validity. In this article,we present two studies that examine these possibilities.

STUDY 1

ObjectiveIn this first study, our objective was to determine the

extent and nature of cross-cultural reverse-wording prob-lems through an examination of the MVS. Introduced in1992, the MVS has become the most widely used and psy-chometrically validated measure of materialism in consumerresearch. The MVS has three dimensions (i.e., happiness,centrality, and success) measured by 10 PWI (e.g., I admirepeople who own expensive homes, cars, and clothes) andeight RWI (e.g., I usually buy only the things I need). This


4/20


scale typically displays good reliability in studies employingAmerican respondents (e.g., Mick 1996; Richins and Daw-son 1992; Rindfleisch et al. 1997).

As recommended by cross-cultural scholars (e.g., Berry1980; Little 1997), our examination focuses on the MVSsmeasurement equivalence and construct validity. As noted

by Hui and Triandis (1983), tests of measurement equiva-lence provide an assessment of a scales internal structure,while tests of construct validity evaluate how a scales un-derlying construct relates to other external constructs; bothare essential in establishing a measures cross-cultural ap-plicability. We assessed measurement equivalence using aseries of nested multiple-group confirmatory factor analysis(CFA) models and construct validity by examining theMVSs nomological relationship to other constructs.

Our cross-cultural examination focused on Americansversus East Asians. We selected Americans as a baselinebecause the MVS was developed in the United States. Weselected East Asians because prior research suggests thatAsian cultural norms of deference and politeness may create

problems for mixed-worded scales in general and the MVSin particular (Hui and Triandis 1983; Suzuki et al. 2000;Wong and Ahuvia 1995). Moreover, linguists suggest thatsome East Asian languages may not easily accommodatecounterfactual statements (Bloom 1981) such as the contra-dictions commonly employed in RWI. For our East Asiansample, we targeted respondents from Singapore, Thailand,Japan, and Korea. The selection of these four Asian culturesalso serves as a natural quasi experiment, as these nationsdiffer systematically in terms of education, economic de-velopment, and Westernization. For example, Japan and Sin-gapore are highly industrialized economies with higher lev-els of income than Korea and Thailand. Moreover, Japanand Thailand are both predominantly Buddhist, whereas Ko-

rea and Singapore have significant representations of otherreligions. Finally, Singapore is the most Westernized sinceit was a British colony for almost 200 years, until 1963.

Respondents and Procedures

Due to the challenges of collecting cross-cultural data,we employed a variety of techniques. We recruited ourAmerican respondents by mailing a survey to 1,000 adultsacross the continental United States. Approximately threeweeks after this initial mailing, nonrespondents were maileda reminder postcard. The surveys for 32 individuals wereundeliverable, reducing the effective sampling frame to 968potential respondents, of which 205 replied, for a 21% re-

sponse rate. Due to severe missing data, five surveys wereunusable. Thus, our final sample in the United States in-cluded 200 respondents. Among our U.S. respondents, 59%were female, 79% were white, their mean age was 43, and47% were college graduates. With the exception of a highnumber of college graduates, our respondents bear closeresemblance to the demographic profile of the United States(U.S. Bureau of the Census 2000).

Due to the questionable reliability of the Thai postal sys-tem, we collected our Thai data using a random selection

door-to-door interview technique (Kau et al. 2000). We hireda Thai survey firm to collect data from 200 adults living inBangkok. This firm randomly selected every third householdin a residential Bangkok neighborhood. We translated oursurvey into Thai using a back-translation procedure (Brislin1970) by employing two native Thai speakers who are fluent

in English. Among our Thai respondents, 50% were female,all were native Thai, their mean age was 35, and 31% werecollege graduates.

We employed the same type of random selection door-to-door interview technique to collect data from 200 adultsin Singapore via a survey research firm. Because Englishis an official language in Singapore and is taught to nearlyall Singaporeans at an early age, we were able to conductthis survey in English. Among our Singaporean respon-dents, 50% were female, 77% were ethnic Chinese, 13%were ethnic Malays, and 10% were ethnic Indians; theirmean age was 34, and 30% were college graduates. Thecomposition of our Thailand and Singapore samplesclosely match the population statistics for both nations in

terms of ethnicity, gender, age, and (for Singapore) edu-cation (http://www.asiandemographics.com).

Finally, we recruited our Japanese and Korean respon-dents by employing a convenience sampling approach at amajor U.S. airport with a large number of flights to EastAsia. This type of sampling procedure has been used inprevious cross-cultural studies (e.g., Aaker 2000; Klein, Et-tenson, and Morris 1998). We had our original English sur-vey back translated into both Japanese and Korean usingtwo native speakers from each culture who are fluent inEnglish. We employed a set of Japanese and Korean grad-uate students to administer the survey. These students wereinstructed to position themselves near the departure gate offlights headed to Tokyo and Seoul and to approach adult

travelers who were sitting or otherwise unhurried. Beforeadministering the survey, students confirmed each respon-dents nationality and residence.

Using this procedure, our graduate assistants collected224 completed surveys (105 Japanese and 119 Koreans).Because of the convenience nature of our sample and useof an international airport as our data collection location,our respondents are high in education: 41% of our Japaneseand 64% of our Korean respondents were college graduates.Our Korean respondents were somewhat older ( )Mp 38than our Japanese respondents ( ). However, wewereMp 29able to target respondents based on gender; thus, close tohalf of our respondents from both countries are female( , ).Japanese p 49% Koreans p 47%

Measures

Our key constructs were assessed using previously pub-lished, multi-item measures using a seven-point Likert for-mat ( , ). We mea-1 p strongly disagree 7 p strongly agreesured materialism using Richins and Dawsons (1992)18-item MVS. With the exception of the MVS, all othermeasures consisted of only PWI. These other measures wereincluded to provide an assessment of the MVSs cross-


5/20


cultural nomological validity. Specifically, we included mea-sures of life satisfaction and individualism. These constructsrepresent two of the most frequently investigated correlatesof materialism. Based on prior research, we expect that ma-terialism should be positively associated with individualismacross our entire sample, as materialism is seen as highly

self-centered in almost all cultures (Eastman et al. 1997;Ger and Belk 1996; Schwartz 1996). In contrast, we expectthat materialism will be negatively associated with life sat-isfaction but only in cultures with high levels of socioec-onomic development. When a culture reaches a high levelof socioeconomic development, material objects are pri-marily used (and marketed) as means of satisfying higher-order needs such as the need for love and belonging (In-glehart 1996; Schudson 1991). Prior research has shownthat objects are inadequate fulfillers of these higher-orderneeds and that an overreliance on object-based need ful-fillment is detrimental to life satisfaction (Burroughs andRindfleisch 2002; Kasser and Ryan 1993). Thus, we predictthat the negative association for materialism and life sat-

isfaction should be present in the United States, Singapore,and Japan (per capita gross domestic product [GDP] allabove $25,000; Central Intelligence Agency 2001) but notKorea (per capita ) or Thailand (per capitaGDP p $16,000

).GDP p $7,000We examined life satisfaction using the Diener et al.

(1985) Satisfaction with Life Scale (SWLS). This is a five-item scale and contains such items as I am satisfied withmy life. This scale is well established in the well-beingliterature and has demonstrated strong psychometric prop-erties in prior empirical investigations (Robinson, Shaver,and Wrightman 1991). Moreover, the SWLS displays ade-quate cross-cultural conceptual equivalence, as the conceptof life satisfaction appears to hold similar meanings across

many cultures (Oishi et al. 1999; Suh et al. 1998). We mea-sured individualism using Triandis and Gelfands (1998)Vertical Individualism scale (VI). This scale contains fouritems such as competition is the law of nature. This mea-sure was developed in a cross-cultural setting and appearsto reflect considerable conceptual equivalence, as the itemsfor this measure are based on an extensive body of cross-cultural research on the meanings of individualism (Markusand Kitayama 1991; Rokeach 1973; Triandis 1995). Thesemeasures appeared in our survey in the following order: (1)life satisfaction, (2) materialism, and (3) individualism.

Test of Measurement Equivalence

We began our examination of the cross-cultural appli-cability of the MVS with an assessment of the measurementequivalence of the MVS across subjects from the UnitedStates, Singapore, Thailand, Japan, and Korea, using thetechniques recently outlined by Steenkamp and Baumgartner(1998). Their approach is similar to methods proposed byother cross-cultural measurement scholars, such as Bensaou,Coyne, and Venkatraman (1999) and Durvasula et al.(1993). We followed this analysis of cross-cultural mea-

surement equivalence with an assessment of the MVSs in-ternal consistency.

As documented by Steenkamp and Baumgartner (1998),there are multiple forms of measurement equivalence (ormeasurement invariance), including configural invariance,metric invariance, and scalar invariance, among others. As

a means of coping with this large selection of equivalencetests, Steenkamp and Baumgartner recommend that thetests employed should correspond to the goals of the re-search. Specifically, for a study such as ours that seeks toexamine the measurement equivalence of a construct,Steenkamp and Baumgartner recommend testing for con-figural invariance. This entails an examination of the de-gree to which observed variables fit latent constructs ineach country through a multicountry confirmatory factoranalysis in which the factor loadings in each country areallowed to vary freely. In addition, because we are inter-ested in understanding the nomological relationship be-tween materialism and other constructs, they recommendtesting for metric invariance, which entails an examinationof the equality of a constructs metrics across all countriesthrough a multicountry CFA in which the factor loadingsin each country are constrained to be equal. As a meansof testing these two forms of equivalence, Steenkamp andBaumgartner recommend the use of hierarchical nestedmodels in which the fit statistics of an unconstrained base-line configural invariance model are examined and thencompared against the fit statistics for a constrained metricinvariance model. In addition to the standard chi-squaredifference test, they also recommend the use of additionalfit indexes, such as the Comparative Fit Index (CFI),Tucker-Lewis Index (TLI), Parsimony Normed Fit Index(PNFI), Consistent version of Akaikes Information Cri-

terion (CAIC), and the root mean square error of approx-imation (RMSEA).Using these recommended procedures, we began with

an examination of the configural invariance of the MVSusing a multicountry CFA model in which the MVS wasspecified as a second-order factor with three underlyingdimensions (happiness, centrality, and success). As shownin table 1, this three-factor model (model A1) produced alow degree of fit ( , , TLI p2x (660) p 2,212.72 CFI p .57.50, , , ). As aPNFI p .43 CAIC p 4,060 RMSEA p .130basis of comparison, Richins and Dawson (1992) specifieda CFA for their MVS across three U.S. samples and foundthat the CFI ranged from .85 to .92 and the RMSEA rangedfrom .056 to .072 (to be reported in Richins 2004). In order

to determine the source of the poor fit for our baseline model,we first examined the item-to-total correlations in each coun-try. This revealed that one item per dimension was weaklyrelated to the other items in each dimension. When we re-visited Richins and Dawsons (1992) original research moreclosely, we found these same items were problematic whenthe scale was first developed. Specifically, they found thatan item for both the centrality (i.e., I put less emphasis onmaterial things than most people I know) and success di-mensions (i.e., I dont pay much attention to the material


6/20


TABLE 1

TESTS OF MEASUREMENT EQUIVALENCE

Model Test x2 df CFI TLI PNFI CAIC RMSEA RMSEAa

A) Study 1: United States,Singapore, Thailand,Japan, and Korea(Likert MVS):

1 Configural invariance(three factor, 18items) 2,212.72 660 .57 .50 .43 4,060 .130 .291

2 Configural invariance(three factor, 15items) 1,577.03 435 .60 .52 .45 3,003 .140 .313

3 Configural invariance(five factor) 743.09 375 .87 .82 .56 2,461 .078 .174

4 Metric invariance (full,five factor) 1,297.36 459 .71 .67 .54 2,391 .110 .246

5 Metric invariance (par-tial, five factor) 770.29 401 .87 .83 .59 2,281 .075 .168

B) Study 2: United States,Japan, and Thailand(Interrogative MVS):

1 Configural invariance(three factor) 499.23 261 .83 .80 .59 1,176 .083 .144

2 Metric invariance (full,three factor) 581.86 285 .79 .77 .60 1,096 .090 .156

3 Metric invariance (par-tial, three factor) 519.24 273 .83 .80 .61 1,113 .083 .144

4 Configural invariance(five factor) 329.47 225 .93 .90 .58 1,264 .060 .104

C) Study 2: United States,Japan, and Thailand(other scales):

1 Configural invariance(one factor, NFC) 894.69 405 .68 .64 .49 1,742 .110 .191

2 Configural invariance(three factor, NFC) 517.45 348 .89 .85 .56 1,659 .064 .111

3 Configural invariance(one factor, NFP) 474.37 195 .60 .52 .41 1,031 .120 .208

4 Configural invariance(three factor, NFP) 226.74 154 .90 .84 .50 1,035 .060 .104

5 Configural invariance(one factor, ETCB) 266.63 81 .73 .63 .50 659 .140 .242

6 Configural invariance(three factor, ETCB) 81.88 51 .95 .90 .42 661 .071 .123

7 Configural invariance(one factor, SOP) 454.75 132 .68 .59 .49 949 .150 .259

8 Configural invariance(three factor, SOP) 146.82 96 .95 .91 .51 845 .061 .106

NOTE. of freedom; Fit Index; -Lewis Index; Normed Fit Index; version ofdfp degrees CFIpComparative TLIpTucker PNFIpParsimony CAICpConsistentAkaikes Information Criterion; and mean square error of approximation.RMSEA p root

a RMSEA adjusted for number of countries (g) by multiplying square root of g.RMSEA# the

objects that other people own) had low factor loadings (Ri-chins and Dawson 1992, table 3). In addition, one of theitems for their happiness dimension (i.e., It sometimes both-ers me quite a bit that I cant afford to buy all the thingsId like) exhibited a high cross-loading on the success di-mension.

In order to determine if these items were the source ofour poor model fit, we eliminated them and reran the analysison the reduced 15-item (nine PWI and six RWI) scale. Asshown in table 1, this reduced model (model A2) resulted in

a slight improvement across most fit indexes ( 2x (435) p, , , ,1,577.03 CFI p .60 TLI p .52 PNFI p .45 CAIC p

, ). Nonetheless, the fit is still far below3,003 RMSEA p .140recommended standards (Byrne 1998), indicating that some-thing more systemic is operative. In particular, we suggestthat the mixed-worded nature of the MVS may be confound-ing its cross-cultural validity.

We began our investigation of this suspected confoundby examining the correlation between the MVSs PWI andRWI in each culture. An inspection of these correlations


7/20


FIGURE 1

CONFIRMATORY FACTOR ANALYSIS MODEL (STUDY 1)

revealed that the MVSs nine PWI and six RWI itemsare strongly correlated for Americans ( , ),rp .57 p !.001weakly correlated for Singaporeans ( , ) andrp .16 p !.05Koreans ( , ), and statistically unrelated forrp .17 p !.10both Thai ( , NS) and Japanese ( , NS). Arp .11 rp .09Fishers z-test reveals that the differences between Amer-icans and each East Asian culture are significant at p !

. This suggests that RWI are being responded to quite.001differently among East Asians compared to Americans. Inorder to assess their relative consistency, we also exam-ined the within-item correlations of RWI and PWI foreach country. This examination shows that, with the ex-ception of Japan, the average interitem correlations for the

RWI ( , ,United States p .62 Singapore p .67 Thailand p, , ) are similar to the PWI.66 Japan p .31 Korea p .78

( , , ,United States p .82 Singapore p .77 Thailand p .69, ). None of the differences (withJapan p .80 Korea p .78

the exception of Japan) between RWI versus PWI corre-lations are statistically significant. This suggests the lackof measurement equivalence associated with the MVS ismainly due to incompatibility between PWI versus RWI.

In order to control for such method-related divisions, mea-surement scholars recommend employing a CFA that in-

cludes separate method factors in addition to any conceptualfactors (Little 1997; Watson 1992). Following this approach,we respecified our three-factor baseline model as a second-order, five-factor multitrait-multimethod model (MTMM)(see Bagozzi and Yi 1991 for an elaboration of this pro-cedure). This model, which is portrayed in figure 1, specifiesmaterialism as a second-order factor composed of three traitfactors (i.e., centrality, happiness, success) and two methodfactors (i.e., PWI, RWI). Thus, each item is specified asloading on both a trait and method factor in accordance withits conceptualization and direction of item wording. Unlikethe three trait factors, the two method factors were not spec-ified to load on the second-order factor (i.e., materialism)

but were specified to be correlated with each other.As shown in table 1, this five-factor model (model A3)

results in a considerable improvement in model fit over ourprevious model ( , ,2x (375) p 743.09 CFI p .87 TLI p

, , , ) and is.82 PNFI p .56 CAIC p 2,461 RMSEA p .078comparable to the fit indexes originally achieved by Richinsand Dawson (1992) in the United States. Although stillslightly below recommended standards (Byrne 1998), thisis likely due to the large number of items in the scale. Asnoted by several measurement scholars (Bagozzi and Heath-


8/20


TABLE 2

PARAMETER ESTIMATES FOR STUDY 1

Country/scale

Meantrait-factorloadingsa

Meantrait

t-values

Meanmethod-factor

loadingsa

Meanmethodt-values

Traitvariance

Methodvariance

Errorvariance

PWI-RWIfactor

correlation

PWI-RWIbivariate

correlation Alpha

United States:MVS1b (.56)

.38(5.53)2.76 .49 5.46 .17 .28 .55 .61 .57 .83

MVS2c (.82).72

(6.18)2.59 NA NA NA NA NA NA NA NA

Singapore:MVS1 (.38)

.26(2.81)1.85 .53 6.30 .13 .30 .57 .07 .16 .72

MVS2 (.82).75


Thailand:MVS1 (.37)

.27(3.10)2.18 .40 3.30 .12 .18 .70 .94 .11 .49

MVS2 (1.00).93


Japan:MVS1 (.43)

.32

(2.57)

2.04 .37 2.85 .13 .24 .63

.11

.09 .68MVS2 (.96)

.95(3.93)3.24 NA NA NA NA NA NA NA NA

Korea:MVS1 (.47)

.26(3.33)1.35 .60 5.43 .12 .37 .51 .52 .17 .82

MVS2 (1.00).85


NOTE. -worded items; -worded items; Values Scale; applicable.PWIppositive RWIp reverse MVSpMaterial NAp notaWithin-group completely standardized factor loadings for five-factor configural invariance model; ( ) p three-factor model.bFirst-order factors.cSecond-order factors.

erton 1994; Netemeyer, Pullig, and Bearden 2002; Podsakoffand MacKenzie 1994), latent constructs such as the MVSthat contain a large number of items often have great dif-ficulty attaining satisfactory fit indexes in totally disaggre-gated models such as ours. More central to our focus thanabsolute fit is the relative improvement in fit once item-

wording direction is controlled. These findings complementour earlier analysis of the correlations between PWI andRWI by further establishing that the poor initial performanceof the MVS can be traced to a confound produced by item-wording direction.

To probe this issue further, we examined the five-factormodels parameter estimates. This information is providedin table 2, which shows the trait and method loadings; par-titioning of variance among trait, method, and error; cor-relations (both bivariate and factor) between PWI and RWI;and the overall coefficient alpha in each culture. As shownin this table, although Americans have somewhat higher traitloadings (and comparable method loadings) than EastAsians, the main cross-cultural difference appears to lie in

the correlation between the MVSs PWI and RWI factors.Although this correlation is actually negative for most ofour Asian respondents ( , ,r p .07 r p .94Singapore Thailand

, ; M: ), it is stronglyr p .11 r p .52 rp .38Japan Koreapositive for our American respondents ( , ).rp .61 p !.001

A test of differences in factor correlations reveals that thismethod correlation is significantly different between theUnited States and each Asian culture at . The factp !.001that these cross-cultural differences are expressed primarilythrough differences in method correlation rather than traitand method loadings (and associated variances) suggests the

biases associated with mixed-worded scales have more todo with RWI incompatibility than RWI inconsistency. Inessence, the inability of the MVS to achieve adequate cross-cultural configural invariance appears mainly due to the factthat East Asians respond to (and possibly interpret) RWIvery differently than PWI.

Having achieved configural invariance, we proceeded totest for metric invariance of the MVS by constraining thefactor loadings to be equal across all five countries (Steen-kamp and Baumgartner 1998). As shown in table 1 (modelA4), although several of the constrained MVS models fitstatistics are within expected parameters, the addition ofthese constraints results in a significant increase in chi-square ( , ). As a result, the MVS2Dx (84) p 554.27 p !.001

does not display full metric equivalence.As recognized by Steenkamp and Baumgartner (1998)

and other cross-cultural researchers, few scales achieve fullcross-cultural metric equivalence. Consequently, they rec-ommend that researchers conduct tests of partial metric


9/20


equivalence by examining which factor loadings are invar-iant and relaxing the constraints on noninvariant loadingsbetween two or more countries as a means of improvingmodel fit. This procedure resulted in 26 (out of 84) pairsof invariant factor loadings across the set of five countries.As specified by Steenkamp and Baumgartner (1998), this

partial metric invariance model has at least two item load-ings per factor that are invariant across two or more coun-tries. As shown in model A5 of table 1, the fit statistics forthis partially constrained model are not statistically differentfrom the unconstrained configural model, indicating that theMVS, after accounting for item wording differences, dis-plays partial metric equivalence across the United States,Singapore, Thailand, Japan, and Korea. Our other twoscales, individualism and life satisfaction, also achieved par-tial metric equivalence. As noted by Steenkamp and Baum-gartner (1998), partial metric invariance provides an ac-ceptable level of cross-cultural equivalence in order toconduct cross-cultural analytical comparisons such as testsof nomological validity.

Test of Construct Validity

Having established the measurement equivalence of theMVS as well as our other key measures, we turned to anexamination of its construct validity. Following Cronbachand Meehl (1955), our examination of construct validityfocuses on an assessment of the nomological validity of theMVS (also see Churchill 1979 and Peter 1981). Nomologicalvalidity is the extent to which a constructs hypothesizedrelations are supported by its measurement data. In essence,nomological validity is confirmed when a measure behavesas expected in relation to other constructs (Churchill 1979,p. 72). Again, material values are expected to be positively

associated with individualism in all five cultures and neg-atively associated with life satisfaction in the United States,Singapore, and Japan, but not in Thailand or Korea.

We began by examining materialisms relationship toboth individualism and life satisfaction without makingany correction for the effects of item-wording direction. Asa means of correcting for attenuation due to measurementerror, we examined the MVSs cross-cultural nomologicalvalidity through an analysis of the factor correlations pro-duced by a structural model (using LISREL 8.51) that in-cluded material values (three-factor partial metric invari-ance model), life satisfaction, and vertical individualism( , , ,2x (1,235) p 3,163.67 CFI p .69 TLI p .65 PNFI p

, , ). In this model, ma-.52 CAIC p 5,407 RMSEA p .110

terialism (specified as a second-order factor) is positivelyrelated (all ) to individualism in all five culturesps !.01( , , , ,r p .74 r p .54 r p .83 r p .59U.S. Singapore Thailand Japan

). Materialism is negatively related to life sat-r p .66Koreaisfaction in both the United States ( , ) andrp .21 p !.05Japan ( , ), unrelated to life satisfaction inrp .31 p !.05Singapore ( , NS) and Korea ( , NS), and pos-rp .01 rp .17itively related to life satisfaction in Thailand ( ,rp .48

). Although the factor correlations for materialismp !.01and individualism are consistent with expectations, the cor-

relations for materialism and life satisfaction are somewhatdivergent, as prior theory and research suggests that ma-terialism should be negatively associated with life satisfac-tion among Singaporeans. Moreover, the strong positive as-sociation between materialism and life satisfaction for Thairespondents is incongruent with nearly all extant theories of

materialism, which do not view materialism as enhancinglife satisfaction for any culture.

As indicated in figure 1, we reexamined the relationshipbetween materialism (second-order factor) and both indi-vidualism and life satisfaction by rerunning our analysisusing the five-factor MTMM model in which the effects ofitem-wording direction are accounted for ( 2x (1,207) p

, , , ,2,406.94 CFI p .81 TLI p .78 PNFI p .60 CAIC p, ). This revised analysis reveals that4,545 RMSEA p .076

materialism is still positively related to individualism in theUnited States ( , ), Thailand ( ,rp .76 p !.01 rp .47 p !

), Japan ( , ), and Korea ( ,.01 rp .59 p !.01 rp .82 p !), but not in Singapore ( , NS). Moreover, al-.01 rp .11

though materialism is still negatively related to life satis-

faction in both the United States ( , ) andrp

.18 p !.05Japan ( , ), it is now unrelated to life sat-rp .31 p !.05isfaction in Thailand ( , NS) and positively relatedrp .10to life satisfaction in Singapore ( , ) and Korearp .48 p !.01( , ). Thus, although the use of method factorsrp .36 p !.01to control for item-wording direction enhances measurementequivalence, it does not appear to improve construct validity,as several of these findings still do not conform to our the-oretical expectations.

Discussion

This first study suggests that the cross-cultural applica-bility of the MVS is constrained by its mixed-worded Likert

scale format. Tests of measurement equivalence reveal thatthe dimensional structure of the MVS lacked configural in-variance until we added method-based factors that accountedfor the scales two different types of item wording. In ad-dition to obscuring the MVSs factor structure, the mixed-worded nature of this scale also partially clouds its cross-cultural construct validity (even after controlling for thesemethod effects).

The problems associated with the MVSs mixed-wordedformat suggest that East Asians interpret (or at least respondto) PWI and RWI quite differently than Americans. Thus,the direction of item wording appears to have a powerfulinfluence on the responses of East Asians to Likert scales.Moreover, this influence has the potential to affect the sub-

stantive interpretations of cross-cultural research. In sum,study 1 suggests that the mixed-worded format of scalessuch as the MVS represent a substantial threat to the validityof measures in cultures where differences in values, lan-guage, or customs lead individuals to interpret RWI differ-ently from PWI.

Although this initial study has identified the presence ofa measurement confound, the nature of this confound isunclear. We suggest that the MVSs lack of cross-culturalmeasurement equivalence and nomological validity is pri-


10/20


marily due to the mixed-worded nature of its Likert format,which results in a measurement-based division between theScales PWI and RWI among East Asians. However, thisdivision could also be due to conceptual-based differencesin the how the construct of materialism is interpreted byAmericans versus East Asians (Wallendorf and Arnould

1988). As a means of clarifying the source of this cross-cultural division, we conducted a follow-up study.

STUDY 2

Objective

Based on the results of our first study, we believe themixed-worded nature of the MVSs Likert format limits itscross-cultural applicability. As seen in study 1, the MVSsPWI and RWI are strongly correlated for Americans butweakly correlated (at both the bivariate and factor levels)among East Asians. Thus, although these two types of itemsseem to be relatively isomorphic for domestic respondents,

their integrity appears to be compromised when appliedcross-culturally. We proposed that this measurement dividedisrupts the cross-cultural measurement equivalence andnomological validity of mixed-worded scales. If our prop-osition is correct, we should find that the conceptual mean-ing of materialism is cross-culturally robust. Moreover, othermixed-worded Likert scales should display similar cross-cultural divisions. Finally, alternative scale formats thatavoid declarative statements should display stronger cross-cultural applicability. Thus, we conducted a second studythat (1) investigated the conceptual equivalence of the mean-ing of materialism, (2) examined the cross-cultural appli-cability of a set of additional mixed-worded Likert-scaledmeasures among Americans versus East Asians, and (3)

explored the cross-cultural applicability of an alternative(i.e., interrogative) format of the MVS.

Respondents and Procedures

As in study 1, our respondents include both Americansand East Asians. Recent world events precluded interna-tional airports as a source of respondents. Therefore, werecruited respondents using an intercept approach at shop-ping malls located in major cities in the United States,Thailand, and Japan. Natives of each country were em-ployed to collect this data. This type of sampling approachhas been used in previous cross-cultural studies (e.g., Kleinet al. 1998; Webster and Beatty 1997). The survey instru-

ment was designed in English and then converted into bothThai and Japanese using the same back-translation pro-cedures used in study 1. This approach yielded 392 com-pleted surveys (126 Americans, 150 Thai, and 116 Jap-anese). Demographically, our respondents were approxi-mately similar to study 1 in terms of gender (female:

, , ),United States p 54% Thailand p 50% Japan p 65%age (M: , , ),United States p 35 Thailand p 37 Japan p 36and education (college graduate: ,United States p 36%

, ).Thailand p 20% Japan p 51%

Measures

As a means of assessing the conceptual equivalence ofthe MVS, we created an adjective checklist that asked re-spondents to check all the words that they felt would de-scribe someone who is very materialistic (i.e., someone who

places high importance on worldly possessions in theirlives). This checklist contained 20 adjectives. Based onprior materialism research (e.g., Belk 1985; Fournier andRichins 1991; Richins and Dawson 1992), we chose 10adjectives that should typify a materialistic individual (i.e.,acquisitive, competitive, egotistical, envious, greedy, inse-cure, possessive, selfish, shallow, and snobbish) and 10 ad-

jectives that are uncharacteristic of someone high in ma-terialism (i.e., altruistic, considerate, friendly, humble,introverted, reflective, scholarly, sincere, thoughtful, and un-derstanding). This checklist was presented at the end of oursurvey instrument.

In order to identify other important consumer behaviorscales that use a mixed-worded Likert format, we consulted

theHandbook of Marketing Scales (Bearden and Netemeyer1999). From this handbook, we selected (1) the short formof the Need for Cognition Scale (NFC) by Cacioppo, Petty,and Chuan (1984), (2) the Need for Precision Scale (NFP)by Viswanathan (1997), (3) the innovativeness dimensionof the ETCB by Raju (1980), and (4) the verbal componentof the SOP by Childers et al. (1985). All of these measureswere published in leading journals, display acceptable psy-chometric properties in domestic applications, are well es-tablished, and contain a mixture of PWI and RWI (NFC:nine PWI and nine RWI; NFP: five PWI and eight RWI;ETCB: four PWI and five RWI; SOP: seven PWI and fourRWI). Each was measured on a seven-point Likert scale inwhich disagree and agree.1 p strongly 7 p strongly

As a means of testing the MVSs cross-cultural appli-cability using an alternative format, we followed the rec-ommendations of Schuman and Presser (1981) by replacingthe MVSs mixed-worded Likert format with a 15-item in-terrogative question format. Thus, rather than forcing re-spondents to agree or disagree with statements (e.g., I admirepeople who own expensive cars, homes, and clothes), weasked respondents to react to questions (e.g., How do youfeel about people who own expensive cars, homes, andclothes?) using a set of specific response options (e.g., donot admire vs. greatly admire). We altered the direction(i.e., left or right side) of these anchors to mirror the orig-inal direction of the Likert-style MVS (MVSLikert). Thisnew interrogative-formatted MVS (MVSinterrogative) is dis-

played in the appendix.By changing the MVSs mixed-worded statements into

a series of nondirectional questions, we avoid the dangercreated when promaterialism and antimaterialism state-ments are not at opposite ends of a single conceptual con-tinuum. Also, by being more inquisitive and using responseoptions customized for each question (rather than a uni-form disagreeagree scale), this format focuses respondentattention more intently on the content of each question.Finally, simply inquiring about a respondents position on


11/20


an issue (as opposed to telling the respondent they are aparticular way, to which they are then forced to agree ordisagree) will hopefully minimize agreement for the sakeof being polite. Unlike some alternative formats, such asa semantic-differential scale, the interrogative format al-lowed us to retain the essence and richness of the MVSs

complex Likert statements. This interrogative question for-mat is conceptually similar to the forced-choice formatadvocated by several psychometricians (e.g., Bentler, Jack-son, and Messick 1971; Javeline 1999) but avoids the mea-surement limitation of forced-choice questions by allowingresearchers to collect interval-scaled data. In addition, thealternating directionality of the interrogative scales an-chors allowed us to avoid the dangers of acquiescenceresponse bias that the mixed-worded Likert format wasdesigned to solve.

Finally, our survey included the SWLS and VI measuresemployed in study 1. The multi-item measures appeared inour survey instrument in the following order: (1) SWLS,(2) MVSinterrogative, (3) VI, (4) NFC, (5) NFP, (6) ETCB, and

(7) SOP.

Results and Discussion

Thus far, our findings suggest the MVSs mixed-wordedformat presents a cross-cultural measurement confound.However, it is also possible that cultural distinctions betweenAmericans versus East Asians present a conceptual con-found. In other words, the very essence and meaning of theconcept of materialism may differ between East and West.The results of our adjective checklist do not, however, showthis to be the case, as the conceptualization of materialismexhibits a considerable degree of cross-cultural robustness.The 10 materialistic adjectives obtained much higher levels

of endorsement for both Americans and East Asians (M:United , , )States p 47% Thailand p 42% Japan p 32%than the nonmaterialistic adjectives (M: United States p

, , ). A few of the 10 ma-3% Thailand p 9% Japan p 2%terialistic adjectives received low rates of endorsement (i.e.,!20%) in Thailand and Japan. Specifically, insecurity re-ceived only an 11% endorsement in both nations, whileenvious and snobbish each received a 13% endorsement inJapan, and greedy received an 11% endorsement in Thai-land. Thus, materialism appears to be viewed by East Asiansas an indicant of strong self-absorption (e.g., being egotis-tical, possessive, and selfish) rather than a marker of weakself-image (e.g., being envious and insecure). However,cross-cultural differences in terms of the conceptualization

of materialism appear minimal, and the content of the MVSseems conceptually meaningful across all three cultures.Thus, it appears unlikely that the findings from study 1 aredue to a lack of conceptual equivalence.

We next turn to the generalizability of our findings. If ourpremise about the incongruity of mixed-worded scales isgeneralizable, we should find that the association betweenthe PWI and RWI for these four additional mixed-wordedscales is weaker among East Asians than Americans. More-over, this item-wording difference should also undermine

the cross-cultural measurement equivalence of these scales.This is largely the pattern of effects we find.

Our analysis began with an inspection of the bivariatecorrelation between the PWI and RWI for each of thesefour mixed-worded scales. In congruence with the resultsfrom study 1, the PWI and RWI of these scales display

a higher degree of correlation among our American re-spondents than our Thai or Japanese respondents (NFC:

, , ; NFP: ,r p .61 r p .09 r p .41 r p .32U.S. Thailand Japan U.S., ; ETCB: ,r p.27 r p .11 r p .54 r pThailand Japan U.S. Thailand

, ; SOP: , ,.31 r p .46 r p .60 r p.03 r pJapan U.S. Thailand Japan). Tests of differences between correlations reveal that.42

the United States has significantly higher PWI-RWI cor-relations than Thailand for all four constructs (NFC: p !

, NFP: , ETCB: , SOP: ) and high-.01 p !.01 p !.05 p !.01er than Japan for three of these constructs (NFC: ,p !.05NFP: , SOP: ). As before, we also examinedp !.05 p !.05the within-item type correlations for all four scales. Acrossthese four scales, the average within PWI ( ,r p .69U.S.

, ) and within-RWI ( ,r p .62 r p .57 r p .75Thailand Japan U.S.

, ) correlations are quite similar.r p .54 r p .69Thailand JapanThese findings are congruent with our results for study 1and strengthen the notion that the problems associated withmixed-worded scales lie primarily in the incompatibility be-tween RWI versus PWI rather than in the inconsistency ofRWI.

Next, we examined the factor structures (i.e., configuralinvariance) of these scales using a multicountry CFA anal-ysis. Because these measures are conceptualized as uni-dimensional, we specified the items of each scale to loadon a single factor. The results are shown in table 1 andreveal that the single-factor models display relatively lowfit indexes. For example, the CFIs associated with these

multicountry models are far below recommended standards( , , , ).NFC p .68 NFP p .60 ETCB p .73 SOP p .68

If our conceptualization about the cross-cultural limi-tations of mixed-worded scales is valid, we should find thefactor structure of these scales improves when their mixed-worded nature is statistically controlled. Thus, we reana-lyzed their factor structure by specifying a series of three-factor models that included a trait-based factor and twoseparate method-based factors. These models are similarto the five-factor model used in study 1. Essentially, theitems for each scale were specified to load on a single traitfactor and either a positive- or reverse-worded method fac-tor. As shown in table 1, though not always above rec-ommended standards, the fit indexes for these three-factor

models represent a significant improvement for all fourscales (CFIs: , , ,NFC p .89 N FP p .90 ETCB p .95

). In sum, these results affirm and extend ourSOP p .95earlier findings by suggesting that the confounding effectsof mixed-worded items generalize beyond just the MVS.

Next, we examined the parameter estimates associatedwitheach of these models. This information is provided in table3, which shows the trait and method loadings; partitioning ofvariance among trait, method, and error; correlations (bothbivariate and factor) between PWI and RWI; and overall


12/20


TABLE 3

PARAMETER ESTIMATES FOR STUDY 2

Country/scaleMean trait

factor loadingsa

Meantrait

t-values

Meanmethod

factor loadingsa

Meanmethodt-values

Traitvariance

Methodvariance

Errorvariance

PWI-RWIfactor

correlation

PWI-RWIbivariate

correlation Alpha

United States:NFC .53 2.69 .26 2.31 .30 .14 .56 .42 .61 .89NFP .37 2.71 .41 2.06 .16 .20 .64 .28 .32 .76ETCB .45 2.51 .34 1.08 .26 .23 .51 .68 .54 .76SOP .47 3.29 .35 2.73 .27 .25 .48 .37 .60 .81MVS1b .51 3.87 .35 2.44 .31 .14 .55 .70 .62 .88MVS2c .79 7.05 NA NA NA NA NA NA NA NA

Thailand:NFC .26 2.07 .28 1.95 .09 .11 .80 .28 .09 .57NFP .26 2.27 .39 2.57 .12 .22 .66 .25 .27 .31ETCB .37 2.71 .42 2.18 .19 .25 .56 .08 .31 .70SOP .22 1.28 .43 1.99 .08 .27 .65 .05 .03 .51MVS1 .35 2.21 .32 2.07 .16 .15 .69 .56 .44 .70MVS2 1.00 3.98 NA NA NA NA NA NA NA NA

Japan:NFC .39 2.56 .34 2.42 .20 .18 .62 .30 .41 .82NFP .34 2.52 .33 1.74 .17 .18 .65 .13 .11 .61ETCB .37 2.23 .41 1.72 .19 .25 .56 .43 .46 .72SOP .40 2.81 .37 2.52 .20 .23 .57 .04 .42 .69MVS1 .38 1.79 .38 4.20 .17 .20 .63 .90 .64 .82MVS2 .97 3.01 NA NA NA NA NA NA NA NA

NOTE. -worded items; -worded items; for Cognition scale; for Precision scale;PWIppositive RWIp reverse NFCpNeed NFPpNeed ETCBpExploratoryTendencies in Consumer Behavior scale; of Pr ocessing scale; Values Scale; applicable.SOPpStyle MVSpMaterial NAp not

aWithin-group completely standardized factor loadings for three-factor configural invariance model.bFirst-order factors.cSecond-order factors.

coefficient alpha in each culture. Once again, Americans havesomewhat higher trait loadings (and comparable method load-ings) than East Asians. As before, the cross-cultural differencelies mainly in the correlation between each scales PWI and

RWI method factors. Across these four scales, this correlationis generally higher among Americans than among Thai orJapanese (M: , , ). A testr p .44 r p .17 r p .21U.S. Thailand Japanof differences in factor correlations reveals that the method-factor correlations are higher in the United States than in bothThailand and Japan for NFC, ETCB, and SOP (at p !.05

). The fact that these cross-cultural differences are ex-.10pressed primarily through differences in method-factor cor-relations rather than trait and method loadings (and associatedvariances) lends further support to the thesis that the inabilityof the MVS to achieve adequate measurement equivalence ismainly because East Asians respond to (and possibly inter-pret) RWI very differently than PWI.

Our next analysis explores a possible solution to this prob-

lem by examining the measurement equivalence of our newinterrogative-format MVS. If our suspicion about item-wording effects is correct, we should find that the MVSinter-rogative displays a higher degree of configural invariance thanthe MVSLikert. To conduct this assessment, we specified a three-factor trait-based multicountry CFA model (i.e., the initialbaseline model used in study 1). As shown in table 1, the fitindexes associated with this three-factor model (model B1;

, , , ,2x (261) p 499.23 CFI p .83 TLI p .80 PNFI p .59, ) are stronger than the fitCAIC p 1,176 RMSEA p .083

indexes displayed for the MVSLikert three-factor model andcomparable to the five-factor model in study 1.

In order to provide a direct comparison to the five-factormodel in study 1, we specified a similar model here by

creating two correlated method factors (i.e., PWI and RWI).Although our new scale no longer employs a mixed-wordedformat, the directions of our anchors were mixed so thathigher levels of materialism are indicated by moving rightfor what were formerly PWI and by moving left for whatwere formerly RWI. The overall fit statistics for this newmodel are reported in table 1 (model B4), and the parameterestimates are reported in table 3. As shown in table 1, al-though this five-factor configural invariance model shows amoderate improvement in some fit statistics over the three-factor model, two measures that account for parsimony (i.e.,PNFI and CAIC) do not.

An examination of parameter estimates reveals that thefive-factor MVSinterrogative model has higher trait loadings

(United , , ) in allStates p .51 Thailand p .35 Japan p .38three countries than the five-factor MVSLikertmodel (United

, , ) used in studyStates p .38 Thailand p .27 Japan p .321. In addition, the interrogative format model has a lowerpercentage of method variance (United ,States p 14%

, ) in all three countriesThailand p 15% Japan p 20%than the Likert format model (United ,States p 28%

, ). However, these differ-Thailand p 18% Japan p 24%ences are relatively modest. Most important, although onlythe United States had a significant positive correlation be-


13/20


tween PWI and RWI factors in the Likert model, the factorcorrelations between PWI and RWI for the interrogativeformat is positive and significant in the United States( , ), Thailand ( , ), and Ja-rp .70 p !.001 rp .56 p !.001pan ( , ). These factor correlations are con-rp .90 p !.001gruent with the bivariate correlations between what were

previously PWI and RWI, which are significant amongAmericans ( ), Thai ( , ), andrp .62, p !.01 rp .44 p !.01Japanese ( , ) despite the fact that higher lev-rp .64 p !.01els of materialism are still indicated by moving right forsome items and left for others. Collectively, these findingslend evidence that the interrogative format can provide aproactive remedy to the item-wording differences thatweaken the measurement equivalence of mixed-worded Lik-ert scales in cross-cultural applications.

Finally, we examined the metric equivalence and nomo-logical validity of the MVSinterrogative. Due to its lack of item-wording differences, the MVSinterrogative should displaystronger metric equivalence compared to its Likert coun-terpart employed in study 1. In order to assess metric equiv-alence, we constrained the three-factor MVSinterrogativefactorloadings to be equal across each country. As shown in table1, although all of the constrained models fit statistics aresimilar to the baseline model, the addition of these con-straints does result in a significant increase in chi-square( , ). Thus, the interrogative format2Dx (24) p 82.63 p !.01of the MVS does not display full metric equivalence acrossour American, Thai, and Japanese respondents. After in-specting the modification indexes, we relaxed 12 (out of 24)factor loadings, and the MVSinterrogativeachieved partial metricequivalence across countries. Compared to the MVSLikertinstudy 1, the MVSinterrogativeobtained a higher percentage (50%vs. 31%) of invariant factor loadings. Again, this suggests

that the mixed-worded nature of the MVSLikertis hamperingits cross-cultural applicability.Next, we explored the nomological validity of the

MVSinterrogative by examining its (second-order) factor cor-relations with vertical individualism and life satisfactionfor the United States, Thailand, and Japan. We specified astructural model that included material values (three-factorpartial metric invariance model), individualism, and life sat-isfaction. Because the MVSinterrogative is not confounded bymixed-worded items, we should find that its pattern of nom-ological relations broadly matches our predicted relation-ships. Recall that these predictions suggested that materi-alism should be positively associated with individualism inall three cultures and negatively related to life satisfaction

for only wealthy cultures (i.e., United States and Japan).This pattern of effects is evident for our MVSinterrogative. Whenmeasured using this format, the MVS exhibits a positiveassociation with individualism for all three countries (UnitedStates: , ; Japan: , ; Thailand:rp .49 p !.01 rp .21 p !.05

, ) and a negative association with life sat-rp .41 p !.01isfaction for both Americans ( , ) and Jap-rp .32 p !.05anese ( , ) but not for Thai ( , NS).rp .38 p !.01 rp .09

In sum, the results of this follow-up study provide ad-ditional evidence that the MVSs lack of equivalence is

largely a measurement confound due to the mixed-wordednature of its Likert format. This evidence comes in threeforms. First, the meaning of materialism displays consid-erable cross-cultural convergence, eliminating lack of con-ceptual equivalence as an alternative explanation; second,the low degree of association between the PWI and RWI

among East Asian respondents is not isolated to just theMVS; and third, when the Likert format of the MVS isreplaced with an interrogative format, it displays strongcross-cultural measurement equivalence and nomologicalvalidity among Americans, Thai, and Japanese.

GENERAL DISCUSSION

In recent years, a growing number of consumer research-ers have embarked on cross-cultural research in order tounderstand, explain, and predict the behavior of participantsof our global consumer culture (e.g., Aaker 2000; Alden etal. 1999; Williams and Aaker 2002). Unfortunately, the gen-eralizability of the conceptualizations and measures em-

ployed in these studies remains unclear, as few studies haveexamined their cross-cultural applicability. Thus, we believeour inquiry into the cross-cultural measurement equivalenceand construct validity of mixed-worded scales provides animportant and much needed measurement foundation forfuture cross-cultural consumer research. Our inquiry indi-cates that the cross-cultural applicability of domestic scalessuch as the MVS is confounded by their mixed-wordedLikert format. In this final section, we discuss our findingsand their implications.

The Cross-Cultural Applicability of Domestic-Based Mixed-Worded Scales

Collectively, our findings suggest that the structure ofmixed-worded Likert scales such as the MVS, NFC, andSOP scales is culturally biased. As noted by Bagozzi (1994)in his ACR Fellow Speech, the potential for this type ofcultural bias is widespread due to consumer researcherspenchant for applying U.S.-made scales to foreign cultureswithout considering their cultural appropriateness. As henotes, Something is to be learned by taking existing the-ories and measures from one culture to another. But moreprogress will be made when we identify theoretical differ-ences and develop measures that are valid across cultures(p. 9). To date, this call remains largely unheeded, as themajority of cross-cultural consumer behavior research is car-ried out using measures developed in the United States with

few, if any, adjustments.Our two studies directly examine the appropriateness of

such practices by focusing on how item-wording directioninfluences cross-cultural validity. In aggregate, we find thatmixed-worded measures are quite problematic when appliedcross-culturally. Using confirmatory factor analysis to con-duct tests of measurement equivalence, we find that theMVS displays a low degree of fit to a data set that combinesthe responses of Americans, Singaporeans, Thai, Japanese,and Koreans. However, this fit improves substantially after


14/20


controlling for the difference between the scales PWI andRWI. More specifically, while both types of items are largelyequivalent for American respondents, they appear to rep-resent somewhat different constructs for East Asian re-spondents. For example, the MVSs (15-item scale) PWIand RWI demonstrate a high degree of correlation for Amer-

ican respondents ( ) but an unacceptably low degreerp .57of correlation for East Asian respondents ( ,r p .16Singapore

, , ; M: ). Ar p .11 r p .09 r p .17 rp .09Thailand Japan Koreatest of differences reveals that these correlations are signif-icantly higher among Americans than each of the four EastAsian samples at . Similar results can be seen forp !.001the other four mixed-worded scales in study 2. Likewise,the pattern of factor correlations for PWI and RWI shownin tables 1 and 3 reveals a similar set of results. Thus, forour East Asian respondents, direction of item wording ap-pears to have a pronounced effect on the way these indi-viduals respond to consumer behavior scales. In short, itappears that mixed-worded Likert scales such as the MVShave difficulty attaining cross-cultural measurement equiv-alence due to this method division.

In addition to its cautionary measurement equivalence,we also find that the MVS demonstrates inconsistent cross-cultural construct validity, as this scales nomological re-lations are somewhat at odds with extant theory. Moreover,controlling for method divisions between PWI and RWIthrough the specification of a five-factor model does littleto enhance the scales nomological validity. This inconsis-tent nomological validity provides additional evidence thatthe mixed-worded nature of the Likert format serves as animpediment to cross-cultural generalizability and that thesimple addition of method factors is not an ideal solution.

Although our U.S. data have substantially stronger cor-

relations between PWI and RWI than any of our four Asiancultures, the difference in trait loadings between Americansand East Asians was much less pronounced. On average,the trait loadings associated with our five mixed-wordedconstructs across studies 1 and 2 was .44 in the UnitedStates, .36 in Japan, and .28 in Thailand. Moreover, themethod loadings associated with these five constructs aresomewhat similar across all countries. Thus, mixed-wordedbias appears to be evidenced primarily by method correlationattenuations rather than by lowered trait loadings (or height-ened method loadings). This lends support to the assertionthat this bias is primarily attributed to the manner in whichPWI and RWI are viewed or responded to, as the noise (i.e.,method loadings and variance) associated with these two

different sets of items is similar across all cultures. Thus,of the two possible mechanisms that we identified earlier,it appears that reverse-wording incompatibility is more cul-pable than reverse-wording inconsistency.

If incompatibility between PWI and RWI is the mainculprit, the recommended solution should be to modify scaleconstruction rather than to eliminate items. We employ thissolution in study 2 through the use of an interrogative formatof the MVS. By most accounts, this solution appears to workwell, as it produces stronger measurement equivalence, log-

ical nomological validity, improved correlations betweenPWI and RWI (at both the bivariate and the factor level),higher trait loadings, and a lower percentage of methodvariance across all cultures than the standard Likert format.Despite these enhancements, our interrogative MVS stillcontains a substantial degree of method variance. Thus, an

opportunity exists for future researchers to enhance our ef-forts by developing valid cross-cultural scales that avoid theproblems of mixed-worded items while also capturinghigher levels of trait variance. However, this is likely to bea daunting task, as measurement reviews consistently revealthat, for most scales, method variance exceeds trait variance(Cote and Buckley 1987; Fiske and Campbell 1992).

Possible Factors Limiting the Cross-CulturalApplicability of Mixed-Worded Scales

Our results suggest that among East Asians, the PWI ofLikert scales are empirically distinct from RWI. However,the cause of this division is uncertain. As outlined by Little(1997), cross-cultural measurement problems can be tracedto three possible factors: (1) translation errors, (2) cross-cultural response biases, or (3) substantive cultural differ-ences. We explore each of these potential explanations inthe context of our data.

Translation Errors. The translation of measurementitems from one language to another may result in error andmisspecification. Thus, the translation of the MVS may havecreated sufficient confusion to attenuate the association be-tween PWI and RWI for our East Asian respondents. How-ever, we believe this possibility is remote. First, we carefullyemployed recommended back-translation procedures (Brislin1970) using native speakers from each culture. Second, ourSingaporean data also reveal an empirical division betweenthe MVSs PWI and RWI even though this data was collectedusing a survey written in English. Finally, the consistent pat-tern of weak correlations between these two sets of itemsamong respondents from all four East Asian cultures in study1, as well as the four different scales employed in study 2,implies that this pattern is unlikely due to stochastic trans-lation errors. In sum, we believe that the observed divisionbetween the PWI and RWI among our East Asian respondentsis due to something more substantial than translation errors(i.e., substantive cultural differences).

Cross-Cultural Response Biases. In addition to re-flecting respondents opinions and beliefs, responses to mea-

surement items may also reflect a systematic bias due to anindividuals particular response style (i.e., responding toitems in a consistent manner that is unrelated to itemcontent). As outlined by Baumgartner and Steenkamp(2001), there are several types of response styles, such asacquiescence, disacquiescence, extreme response style, andmidpoint responding. Of all these potential response styles,acquiescence and disacquiescence are the ones most likelyto produce a low degree of correlation between PWI andRWI. Acquiescence (disacquiescence) is defined as the ten-


15/20


dency to agree (disagree) with items irrespective of content(Couch and Kenniston 1960). Both types of response biasmay produce a lack of association between mixed-wordeditems. Specifically, acquiescence leads respondents to agreewith both PWI as well as a RWI, while disacquiescence hasthe same type of effect, in reverse. This pattern of systematic

agreement (disagreement) to items regardless of contentwould severely attenuate the correlation between a scalesPWI and RWI after recoding for analysis.

As a means of assessing the potential impact of acquies-cence and disacquiescence on the relationship between theMVSs PWI and RWI, we follow the techniques used byBaumgartner and Steenkamp (2001). Specifically, we esti-mated (dis)acquiescence by examining respondents degreeof (dis)agreement to both positive- and reverse-worded MVSitems. We selected one PWI and RWI from each of the MVSsthree dimensions (i.e., centrality, happiness, and success) tocreate a six-item scale for computational purposes. Extremeagreement (i.e., response of seven) to both a positive- anda reverse-worded statement was assigned an acquiescence

score of three. Extreme agreement to both a positive-wordedstatement and strong agreement (i.e., response of six) toa reverse-worded statement (or vice versa) was assignedan acquiescence score of two. Strong agreement to both apositive-worded statement and a reverse-worded statementwas assigned an acquiescence score of one. We used theinverse of this procedure to calculate disacquiescence. Thus,the maximum possible (dis)acquiescence score was nine,and the minimum possible score was zero.

In general, respondents from all five cultures displayed lowlevels of (dis)acquiescence (mean acquiescence: United

, , ,States p .19 Singapore p .14 Thailand p .77 Japan p, ; mean disacquiescence:.29 Korea p .25 United States p, , , ,.28 Singapore p .01 Thailand p .29 Japan p 1.65

). As seen from the mean scores, theKorea p .16(dis)acquiescence response bias appears nominal with theexception of Thailand, which exhibited a modest level ofacquiescence, and Japan, which exhibited a moderate levelof disacquiescence (although scores of .77 and 1.65 out ofa possible nine are still notably low). Thus, it appears un-likely that the lack of correlation between the MVSs PWIand RWI among our East Asian respondents is due to thistype of systematic response bias.

Substantive Cultural Differences. Thus far, we havefound that our East Asian respondents were unlikely to havebeen confused by translation issues and that the influenceof systematic response biases is fairly modest. Therefore, it

appears that the observed distinction between PWI and RWImay be largely due to substantive cultural differences be-tween our American and East Asian respondents. This cul-tural difference is not surprising considering that all of thefocal scales in our two studies were conceptualized, devel-oped, and purified by Americans using sets of respondentsfrom the United States. Thus, although PWI, such as I ad-mire people who own expensive cars, homes, and clothes,and RWI, such as I like to keep my life simple as far aspossessions are concerned, may be at opposite ends of a

conceptual continuum for Americans, the meanings of theseitems may not be so black and white among East Asians.Moreover, this distinction is consistent across respondentsfrom different Asian cultural backgrounds. Although Sin-gaporeans, Thai, Japanese, and Koreans all reside in EastAsia, this is an expansive region, and the members of these

four cultures speak very different languages, possess dif-ferent levels of economic resources, and follow differentcustoms and traditions. Thus, locating a set of commoncultural forces that might be responsible for the pattern ofeffects we observed for our MVS among our East Asianrespondents is a challenging task.

One cultural commonality that Singaporeans, Thai, Jap-anese, and Koreans share is the influence of Buddhist teach-ings and practices. For much of the past 1,500 years, Bud-dhism has been a major religion in each culture. Althoughthe religious underpinnings of American culture draw asharp divide between materialism and antimaterialism(Schudson 1991), the Buddhist foundations of East Asianculture focuses on bridging the gap between these two values(Zimmer 1993). Thus, although materialism and antimater-ialism appear to be oppositional for Americans, this op-position may not hold among East Asians.

Buddhisms ontology and epistemology appear to makeEast Asians relatively comfortable with apparent contradic-tions. As part of a focus on what is often referred to as themiddle way (Sopa and Jones 2000), Buddhism (as well asseveral other East Asian belief systems such as Confucian-ism) seeks to minimize dualities and encourage more holisticand dialectical thinking. As noted by Peng and Nisbett(1999), the Western philosophical traditions of Americanculture view contradictory propositions as unacceptable. Incontrast, they suggest that Chinese (as well as other East

Asians) deal with contradictions through a compromise ap-proach, by showing tolerance of contradictions by findinga middle way by which truth can be found in each of twocompeting hypotheses (p. 742). To test these propositions,Peng and Nisbett conduct a series of experiments, whichreveal that Chinese students are more accepting of contra-dictory proverbs than their American counterparts.

Peng and Nisbetts findings are congruent with recent re-search by Bagozzi, Wong, and Yi (1999), which finds thatalthough Americans appear to experience positive and neg-ative emotions in an oppositional, bipolar manner, East Asiansappear to experience positive and negative emotions in a moredialectical, holistic manner. In explaining these results, Ba-gozzi et al. (1999) suggest that Americans tend to polarize

contradictions while East Asians seem to accept contradic-tions as part of the natural order (p. 667). These results haverecently been extended by Williams and Aaker (2002), whofind that East Asians are more accepting of conflicting emo-tions than Americans. Thus, when faced with the apparentcontradictions embodied in a mixed-worded scale, Americansmay have a predisposition to view PWI and RWI as polaropposites while East Asians have a predisposition to viewthese items as related parts of a larger order.

As a means of investigating whether Buddhism might


16/20


provide at least part of the explanation for our findings, wereturned to our Singapore data from study 1. Singaporestands at a religious crossroad, as it is one of the few placesstrongly represented by three of the worlds major religions(Christianity, Buddhism, and Islam), all of which have beenestablished in Singapore for more than 500 years (Tamney

1996). Thus, Singapore provides a quasi-experimental set-ting. Recognizing this religious diversity, we initially in-cluded a question on religious affiliation as part of ourSingaporean survey. Among Singaporean non-Buddhists( ), the MVSs PWI and RWI are positivelycorrelatedn p 59( , ). However, among Singaporean Buddhistsrp .29 p !.05( ), there is no relation between these two types ofn p 81items ( , NS). Thus, it appears that the differencesrp .02between the way East Asians and Americans respond tomixed-worded scales in general, and the MVS in particular,may be at least partially due to substantive differences inthe religious belief systems underlying eastern versus West-ern cultures (see Gorsuch et al. 1997 for a related finding).We conclude with a discussion of the implications of these

cultural differences.

Implications for Future Cross-Cultural ConsumerResearch

Our examination of conceptual equivalence suggests thatRichins and Dawsons (1992) conceptualization of materialvalues appears to be cross-culturally applicable. However,their measure of this construct (i.e., the MVS) is not. Inparticular, our findings suggest that the MVS mixed-wordedformat limits the scales cross-cultural measurement equiv-alence and may hamper its nomological validity. Althoughit poses some problems for Americans, this mixture is par-

ticularly volatile for East Asians, who do not appear to viewthese two types of items as being in opposition. As we showin study 2, other mixed-worded scales pose similar prob-lems. Thus, our inquiry provides modern day verificationto the historic warnings from such measurement scholars asCronbach (1950) and Jordan (1965) that mixed-worded Lik-ert scales may present severe measurement difficulties. De-spite these warnings, the problems associated with mixed-worded scales have received little attention amongcontemporary consumer researchers (see Falthzik and Jolson1974 and Herche and Engelland 1996 for exceptions), andeven less scrutiny in the case of cross-cultural consumerresearch. Our research suggests this issue deserves renewedattention, as mixed-worded scales that perform acceptably

in the United States may be problematic elsewhere.As an example of these problems, consumer researchers

should consider the case of cross-cultural applications ofUnited States mixed-worded materialism measures (i.e.,Belk 1985; Richins and Dawson 1992) in Asian cultures.Typically, these studies have faced difficulty obtaining sat-isfactory levels of reliability for materialism (e.g., Eastmanet al. 1997; Sirgy et al. 1995). As a result, the findings ofsuch studies may be confounded by the mixed-worded na-ture of these scales. For example, in contrast to both prior

theory as well and our current findings, Sirgy et al. (1995)find that materialism is negatively associated with subjec-tive well-being among Asians in less developed nations.We believe that this type of divergent result may be largelydriven by the fact that prior studies have not accountedfor the biases posed by the mixed-worded nature of Likert-

formatted materialism scales. Without accounting for (andcorrecting) this bias, research on the correlates of mate-rialism among Asians may lead to misleading conclusions.

Although our findings are localized to East Asia, previousresearch suggests that problems with mixed-worded Likertscales may not be limited to this region. For example, Marin,Gamba, and Marin (1992) find that respondents from LatinAmerican cultures display a high level of acquiescence andoften agree with both PWI and RWI. Likewise, Steenkampand Burgess (2002) find nonwhite South Africans responddifferently to RWI than to PWI (see also Meloni and Gana2001). In addition, although our focus is on consumer be-havior constructs, prior research indicates that manageriallyfocused mixed-worded scales may also be problematic in

foreign applications (e.g., Grewal and Tansuhaj 2001; Spec-tor et al. 1997). Because of the apparent widespread natureof these problems, we advise future cross-cultural research-ers to employ mixed-worded scales with great caution andto examine the fit between PWI and RWI within each culture(and subculture) they investigate.

Recognizing the difficulties posed by mixed-worded Lik-ert scales, some psychometricians (e.g., Schriesheim

Reverse Coded Issue 1

Documents