ccm cross cultural management · 2018-10-01 · bachman & o’malley 1984 usa: african...

The globalization of the world economy andthe increasing importance of multinationalcompanies have made more and moreresearchers realize that theories and conceptsdeveloped in one part of the world (usuallythe USA) might not be applicable across borders. In order to find out which theories

and concepts are universally valid and whichhave to be adapted, cross-national research isnecessary and often this type of research isconducted using surveys. However, cross-national survey research is plagued by manyproblems (for an overview see for instanceSingh, 1995; Usunier, 1998; Van de Vijver

Response Styles in Cross-national Survey ResearchA 26-country Study

Anne-Wil HarzingThe University of Melbourne, Victoria, Australia

ABSTRACT Studies of attitudes across countries generally rely on a comparison ofaggregated mean scores to Likert-scale questions. This presupposes that when peoplecomplete a questionnaire, their answers are based on the substantive meaning of the itemsto which they respond. However, people’s responses are also influenced by their responsestyle. Hence, the studies we conduct might simply reflect differences in the way peoplerespond to surveys, rather than picking up real differences in management phenomenaacross countries. Our 26-country study shows that there are major differences in responsestyles between countries that both confirm and extend earlier research. Country-levelcharacteristics such as power distance, collectivism, uncertainty avoidance and extraversionall significantly influence response styles such as acquiescence and extreme response styles.Further, English-language questionnaires are shown to elicit a higher level of middleresponses, while questionnaires in a respondent’s native language result in more extremeresponse styles. Finally, English-language competence is positively related to extremeresponse styles and negatively related to middle response styles. We close by discussingimplications for cross-national research.

KEY WORDS • acquiescent response style • cross-national survey research • extreme responsestyles • language • research methods

Copyright © 2006 SAGE Publicationswww.sagepublications.com

DOI: 10.1177/1470595806066332

CCM International Journal of

Cross CulturalManagement2006 Vol 6(2): 243–266

http://ccm.sagepub.com/

and Leung, 2000). This article focuses on oneof these problems: differences in responsestyles.

Studies of attitudes across countries havegenerally relied on a comparison of aggre-gated mean scores to Likert-scale questions.This presupposes that when people completea questionnaire, their answers are only basedon the substantive meaning of the items towhich they respond (Baumgartner and Steen-kamp, 2001). However, people’s responsesare also influenced by their response style.‘Response style’ refers to a respondent’s tendency to respond systematically to ques-tionnaire items regardless of their content(Baumgartner and Steenkamp, 2001). Themost commonly cited examples of responsestyles are acquiescence (ARS) or disacquies-cence (DRS); that is, the tendency to agree ordisagree with an item regardless of the con-tent, and extreme response styles (ERS) versus middle response styles (MRS); that is,the tendency to use the extreme or middleresponse categories on ratings scales.1

Previous research, as will be reviewed inthe next section, has shown that there mightbe systematic differences between countrieswith regard to response styles, which wouldmake a comparison of mean scores acrosscountries a hazardous affair. Conclusionsdrawn might simply reflect differences in theway people respond to surveys rather thanpicking up real differences in the manage-ment phenomena across countries. Unfortu-nately, earlier studies looking at responsestyles have focused on comparisons of a limited number of countries only, while thefew available multi-country studies (Baum-gartner and Steenkamp, 2001; Smith, 2004)did not systematically present country differ-ences. In addition, none of the earlier workprovided a clear theoretical rationale forcountry differences in response styles orinvestigated whether the language of thequestionnaire might influence response styles.

This article will offer a systematic com-parison of the response styles identified in 26

countries, covering nearly all major culturalclusters in the world: Northern Europe(Denmark, Finland, Sweden), Western Europe(Austria, Germany, the Netherlands, the UK),Southern Europe (France, Greece, Portugal,Spain, Turkey), Eastern Europe (Bulgaria,Lithuania, Poland, Russia), Latin America(Brazil, Chile, Mexico), North America(USA) and Asia (China, Hong Kong, India,Japan, Malaysia, Taiwan). We will also pro-pose and test several hypotheses with regardto country-level factors influencing responsestyles. Finally, in each country – except forthe UK and the USA – two matched samplesof respondents replied to a questionnaire ineither their native language or English. Wecan therefore also assess the impact of language on response styles. In addition, weexplore the impact of English-language com-petence on response styles.

Literature Review andHypotheses

Previous studies with regard to differences in response styles between countries haveshown fairly consistent results. In the USA,both Hispanics (Clarke III, 2000; Hui andTriandis, 1989; Johnson et al., 1997; Marinet al., 1992; Ross and Mirowsky, 1984) andAfrican Americans (Bachman and O’Malley,1984; Clarke III, 2000; Johnson et al., 1997)showed a greater preference for extremeresponses than European Americans, partic-ularly towards the positive end of theresponse scale, and were also more prone toacquiescence. Comparisons between US/Canadian respondents and Japanese respon-dents showed that the former had higherERS and the latter higher MRS (Chen et al.,1995; Shiomi and Loo, 1999; Takahashi etal., 2002; Zax and Takahashi, 1967). Thesame pattern was found in comparisonsbetween US and Korean respondents (Chunet al., 1974; Lee and Green, 1991).

Other countries are typically covered inonly one or two studies. Bennett (1977)

International Journal of Cross Cultural Management 6(2)244


explained response differences for Chineseand Filipino respondents to different lan-guage questionnaires in terms of switching ofreference groups (Europeans versus locals).However, an overall analysis of their resultsshows that Filipino respondents had a higheracquiescence bias and extreme response biasthan Chinese respondents, who displayed ahigher preference for the middle of the scale.2

Church (1987) also found Filipino respon-dents to have a strong acquiescence bias.Van Herk et al. (2004) found the highest levelof both acquiescence and extreme responsestyles for their Greek respondents, whileSpanish and Italian respondents also hadconsistently higher scores than British,German and French respondents. Moreover,they were able to show that these differenceswere indeed caused by differences inresponse styles only, as there was no relation-ship between higher levels of endorsement onthe questionnaire items and actual behav-iour. The high level of acquiescence forGreek respondents confirmed earlier resultsby Steenkamp and Baumgartner (1998) whofound them to have higher mean scores thanBritish and Belgian respondents. Finally,Brengelmann (1959) found that Germanrespondents showed a higher level of acquiescence than British respondents, whileJaveline (1999) found acquiescence bias to bestronger for Kazakhs than for Russians, andClarke III (2000) found higher levels ofextreme response styles for French respon-dents than for Australians. Table 1 summa-rizes the results of these studies.

Although previous studies have generallyshown consistent results, only a few countrieshave been covered both within individualstudies and across studies. With a limitednumber of exceptions, research has onlyfocused on Hispanics/blacks in the USA orcomparisons of East Asian respondents withUS Americans. Two recent studies (Baum-gartner and Steenkamp, 2001; Smith, 2004)dealing with response styles included a widerrange of countries, but did not report on the

actual differences in response styles betweencountries. The first contribution of this studywill therefore be to provide descriptive resultson response style patterns across a matchedgroup of respondents in 26 countries andcompare these with results from earlier studies.

However, more interesting than thedescriptive results would be an exploration ofwhy response styles differ between countries.Reasons for differences in response style canbe dispositional – that is, related to individualcharacteristics such as age, gender or person-ality – or situational – that is, related to situational characteristics such as the formatof the response scale, the ambiguity of ques-tions, or time pressure (Baumgartner andSteenkamp, 2001). In the context of cross-country differences in response styles, cul-tural differences would be a likely disposi-tional explanation. Here we discuss the likely impact of three dimensions of culturaldifference: power distance, collectivism anduncertainty avoidance.3 The language of thequestionnaire could be an important situa-tional explanation, while the language capa-bility of the respondent in question would bedispositional.

Early studies on differences in responsestyles between countries have typically offeredpost hoc and limited rationales, such as thetendency to be modest for Asian respondents.Yates et al. (1997: 88) even claim that ‘theorigin of response styles themselves [. . .]remain[s] the mystery it has always been’.Javeline (1999) posited that an acquiescencebias might be due to deference or respect for the investigator, and hence predicted and found Kazakhs scoring higher on thisresponse style than Russians as their culturewas seen as more deferential to superiors.She argued that cultures with the same characteristics, such as Central and East-Asian cultures would show a similar tend-ency. Given that our respondents are students, and the investigators are their lecturers, an acquiescence effect might be

Harzing: Response Styles in Cross-national Survey Research 245


present in countries in which deference topeople in positions of higher status is com-mon. This would be likely to be the case incountries characterized by a high score onpower distance as measured in, for instance,Hofstede’s study and the Globe Leadershipstudy (Hofstede, 1980, 2001; House et al.,2004). Smith (2004) offered the same argu-ment when explaining his exploratory results.On the other hand, in countries character-ized by a low score on power distance,respondents would not be afraid to disagreewith the investigator. Hence:

Hypothesis 1: The higher the level of a country’spower distance, the higher its acquiescencebias.

Collectivist countries are characterizedby harmony, avoidance of confrontationsand more conformity behaviour. Individualinitiatives and opinions tend to be discour-aged and opinions are predetermined by thein-group (Hofstede, 2001). We expect thatthis would lead respondents from collectivistcountries to give either middle or slightly positive responses (e.g. 3 or 4 on a 5-pointscale) as these are most likely to avoid con-


Table 1 Results of earlier studies on differences in response styles between countries

Study Year Countries/groups included Result

Ross & Mirowsky 1984 USA: Hispanics versus ARS & ERS higher for Hui & Triandis 1989 European Americans HispanicsMarin et al. 1992Johnson et al. 1997Clarke III 2000

Bachman & O’Malley 1984 USA: African Americans ARS & ERS higher for Johnson et al. 1997 versus European Americans African AmericansClarke III 2000

Zax & Takahashi 1967 USA/Canada & Japan ERS higher for USA/CanadaChen et al. 1995 MRS higher for JapanShiomi & Loo 1999Takahashi et al. 2002

Chun et al. 1974 USA & Korea ERS higher for USALee & Green 1991 MRS higher for Korea

Bennett 1977 China & Philippines ARS & ERS higherChurch 1987 for Philippines

MRS higher for ChinaSteenkamp & Baumgartner 1998 Greece, UK, Belgium ARS highest for Greece

Van Herk et al. 2004 Greece, Spain, Italy, UK, ARS & ERS highest for Germany, France Greece

ARS &ERS lowest for UK,Germany, France

Brengelmann 1959 Germany & UK ARS higher for Germany

Javeline 1999 Kazakhstan & Russia ARS higher for Kazakhstan

Clarke III 2000 France & Australia ERS higher for France

ARS = acquiescent response style, ERS = extreme response style, MRS = middle response style.


frontation and preserve harmony. In explain-ing his exploratory results, Smith (2004) alsoclaimed that in-group harmony in collectivistcultures might lead respondents to giveacquiescent answers. In contrast, individual-ist countries are characterized by an accep-tance of confrontations and lower emphasison conformity and harmony. Individual ini-tiative is expected and speaking one’s mind is appreciated. In individualist countries wewould therefore expect a higher willingnessto disagree. Hence:

Hypothesis 2: The higher the level of a country’scollectivism, the higher its middle response andacquiescence bias.

In countries with a high level of uncer-tainty avoidance, people experience higherlevels of stress and anxiety, and have a needfor clarity and structure. Change and inno-vation are generally resisted and diversity lessvalued than in countries with a low level ofuncertainty avoidance. Truth is seen asabsolute and students prefer structured learn-ing situations and seek the ‘right answers’(Hofstede, 2001). Tolerance for ambiguity isalso negatively related to uncertainty avoid-ance (House et al., 2004). We expect that thelower tolerance for ambiguity and diversityof respondents in high uncertainty avoidancecountries will lead to a preference for affirm-ative answers (acquiescence) over disagree-ment (disacquiescence). The preference forthe absolute truth and the right answerswould reinforce this tendency. Respondentsare likely to agree with what they think theinvestigator sees as the ‘right answer’, ratherthan questioning this by disagreeing. Hence:

Hypothesis 3: The higher the level of a country’suncertainty avoidance, the higher its acqui-escence bias.

As Smith (2004) indicated, response biascan be seen not only as the manifestation ofnation-level intergroup relations (such aspower distance, individualism/collectivismand uncertainty avoidance), but also as anation-level reflection of individual commu-

nication styles. Communication styles mightbe a particularly useful explanatory factor ofextreme versus middle response styles. Asimilar argument was put forward byBachman and O’Malley (1984) to explain thehigher level of extreme responses by blackAmericans in comparison to white Ameri-cans. They argued these response stylesmight be related to differences in linguisticsstyles. Blacks might be more willing toexpress their opinions in unqualified terms,while whites show a greater caution or inhi-bition to do so. Hui and Triandis (1989) alsoseemed to refer to this distinction when theyclaimed that in low-ERS countries indi-viduals prefer to appear modest and non-judgmental. In high-ERS countries on theother hand extreme responses would be seenas a demonstration of sincerity, convictionand individual expressiveness.

In order to test this assumption, ideally wewould need a measure of the level of restraintand modesty versus expressiveness and exag-geration in communication styles. Gudykunstet al.’s (1988) concept of succinct versus elab-orate communication styles comes very closeto this. However, no country scores are avail-able for these communication style differ-ences and the same examples (Middle Easterncultures having an elaborate style and manyAsian cultures having a succinct style) arerepeated over and over again. Hall’s (1976)distinction between direct and indirect (orlow/high context) communication styleswould seem to be related to this as well, butthe parallel isn’t complete. Although an indi-rect communication style is often related tounderstatement, succinctness and an exten-sive use of silence (e.g. as for Japan and many other Asian countries), Latin American and Mediterranean countries are generally characterized to have high context communi-cation, but would seem to have a moreexpressive communication style than Asians.Furthermore, some low context countrieshave rather restrained communication stylesas well (Scandinavian countries, Germanic



countries). Finally, Trompenaars’ (1997) distinction between affective and neutral cultures might be expected to show some relationship with communication styles. Un-fortunately, published empirical data on thisdimension are limited to the results of onequestion (‘Would you show emotions openlyif you felt upset at work?’). In addition,Trompenaars’ results seem counterintuitivefor many countries, such as Denmark, Fin-land and Germany, are classified on the affective side.

Recently, however, there has been increasing interest in comparing the big-fivepersonality characteristics (neuroticism,extraversion, openness, agreeableness andconscientiousness) across cultures. Hofstedeand McCrae (2004) and Van Hemert et al.(2002) have shown that its factor structuresare replicated not just at an individual levelwithin countries, but also at a country level,and hence meaningful comparisons can bemade across countries. One of the big-five per-sonality characteristics, extraversion, wouldseem to bear a strong positive relationship tothe level of expressiveness and exaggerationin communication styles. Hence:

Hypothesis 4: The higher the country-levelextraversion, the higher the extreme responsebias.

As indicated, the language of the ques-tionnaire might be an important situationaldeterminant of response styles. Unfortu-nately, there has been no previous researchthat has systematically investigated this.However, Church et al.’s (1988) study foundthat concepts are more refined in the mindsof the respondents when responding in theirnative language. This might lead respondentsto prefer more neutral answers when pre-sented with a questionnaire in a non-nativelanguage. Gibbons et al. (1999) found thatfor one of the questionnaires they tested witha bilingual sample, items were found to bemore meaningful and elicited more extremeresponses when they were presented in the

respondent’s native language. Finally, McCrae(2002) found standard deviations for theNEO-PI-R to be higher in the Filipino subsamples when the questionnaire wasadministered in Filipino than when it wasadministered in English. This sparse evi-dence seems to indicate that respondentsmight be more likely to use the full range intheir native language, while being moreprone to neutral (middle) answers in a foreignlanguage. Hence:

Hypothesis 5a: Respondents will be more likelyto choose extreme responses when replying toa questionnaire in their native language thanwhen replying to the same questionnaire in anon-native language.

And:

Hypothesis 5b: Respondents will be more likelyto choose middle responses when replying to aquestionnaire in a non-native language thanwhen replying to the same questionnaire intheir native language.

An important reason for the preferencefor neutral responses when replying to aquestionnaire in a non-native languagemight be a lack of understanding of the language in question. We argue that whenrespondents feel they might not understand aquestion properly, they would be more likely to choose a safe (middle) response. Ifthis were true, the level of competence in aforeign language would be related to differ-ences in response style. Therefore:

Hypothesis 6: When responding to a question-naire in a non-native language, respondents’level of non-native language competence willbe positively related to extreme response stylesand negatively related to middle responsestyles.

Methodology

Sample and Data CollectionProcedures

The project coordinator recruited countrycollaborators through personal contacts andnetworking at professional conferences such



as the Academy of Management. Once theproject had started, several researchers contacted the project coordinator directly,offering to collect data in their country. Allcountry collaborators received a 15-pagedocument containing very detailed instruc-tions about the aim of the study; items andconstructs; results of the pilot study; trans-lation, data collection and data entry proce-dures; as well as agreements about co-authorship. All collaborators received accessto the final data set. A document with per-sonal introductions of all collaborators wasprepared to promote group cohesion andfacilitate networking among collaborators.

Respondents were final year universitystudents following a course in BusinessAdministration, Business and Management,Commerce or a similar subject. They weregenerally between 21 and 22 years old. Thegender distribution varied from 27% femalein India to 77% female in Hong Kong.International students were excluded fromour sample, so that our comparisons onlyincluded students who could be assumed tobe representative of the country they studiedin. The resulting sample sizes ranged from 85for Russia to 210 for the Netherlands, but formost countries were around 100. Data werecollected in class between March 2001 andApril 2003. Although data were collected ona voluntary basis, response rates were high –generally between 80% and 100%. The useof a student sample poses limitations in termsof representativeness; especially in develop-ing countries students might be differentfrom the population as a whole and might be more westernized than non-students.However, this does mean that any cross-country differences in response styles mightbe attenuated, so in fact our study provides amore stringent test of these differences (Allikand McCrae, 2004).

The project was part of a large-scalestudy investigating the impact of the lan-guage of the questionnaire on students’responses. Yang and Bond (1980) suggest

that when learning a second language, indi-viduals might be subconsciously influencedby the culture of that language and acquiresome of the cultural attitudes and valuesassociated with it, a process called culturalaccommodation. In our study responses wereshown to be significantly different betweenthe English-language questionnaire and thenative-language questionnaire (see Harzingand Maznesvki, 2002; Harzing et al., 2005),showing a pattern of cultural accommoda-tion. This means that responses in the nativelanguage are likely to be closer to the ‘true’responses and response styles than responsesin English. Hence in the first part of thisstudy we only used the sample of studentsthat responded to the questionnaire in theirnative language. In the second part of the study, which focused on the impact oflanguage and English-language competenceon response styles, we compared the two language versions and only used the sampleof students that responded to the question-naire in English.

The original questionnaire was designedin English. It was pilot tested in the UK inOctober 2000. The pilot study coincidedwith a discussion among the first eight coun-try collaborators about translatability ofitems. Several items that proved to be diffi-cult to translate were replaced. Subsequently,bilingual country collaborators were respon-sible for the translation of the originalEnglish questionnaire. Translations wereconducted using translation–back-translationprocedures. The translator and back-transla-tor were separate individuals who did notenter into a discussion until after they had finished their translations. Discussionsbetween translator and back-translator usu-ally resulted in the change of some of thetranslations. Where difficulties remained, athird bilingual person was consulted. Theback-translated versions were verified by theproject coordinator for consistency acrosslanguages, which usually resulted in furtherchanges and discussions between translator



and back-translator. For several of the Euro-pean languages, the project coordinator provided independent verification of thetranslated versions.

Questionnaires were completed in eitherEnglish or the native language of the countryin question. Collaborators were instructed to make sure that the different language versions were randomly distributed. In mostcountries English and native language ques-tionnaires were distributed in the same class.In the remaining countries, different classesof the same module or related module wereused to separate English and native languagequestionnaires. Respondents were not allowedto choose which language version they completed. An equal number of English-language and native-language questionnaireswere distributed.

To verify whether collaborators had succeeded in the randomization process, wetested whether the two language groups differed systematically on the question: ‘Howtypical do you consider your view to be ofpeople who live in the country in which youwere born?’ None of the 25 country samplesshowed a significant difference between the language versions on the ‘typical view’question, which shows that there were no systematic differences between the two language samples. However, in some of thecountries there was a difference in age andgender distribution between the different language versions. We therefore included ageand gender as control variables in our statis-tical analysis.

Measures

Dependent variables Measures of thevarious response styles were constructedusing the responses to all attitudinal 5-pointLikert-scale questions in the questionnaire.These questions dealt with four differenttopic areas: cultural norms and values withregard to activity (e.g. ‘Sitting around withoutdoing something is a waste of time’), culturalnorms and values with regard to relationships

(e.g. ‘Good team members subordinate theirown interests to those of the team’), reasonsfor choosing electives (e.g. ‘Because I think Ican get a high mark for it’), and characteris-tics of the ideal type of job after graduation(e.g. ‘Have an opportunity for high earnings’).The first three sets of questions had scaleanchors running from ‘strongly disagree’ to‘strongly agree’, while scale anchors for thelast set of questions ran from ‘of very little orno importance’ to ‘of utmost importance’.The response format was identical for allquestions; that is, ‘strongly agree’ and ‘ofutmost importance’ were always on the right.A total of 69 questions were used to create thedependent variables.4

The level of acquiescence was calculatedby dividing the number of questions thatreceived a 4 or a 5 (agree/strongly agree, veryimportant/of utmost importance) responseby the total number of questions for eachrespondent. The resulting scores ranged from0.00 to 1.00. Disacquiescence was calculatedin a similar way, using the number of ques-tions that received a 1 or a 2 (strongly dis-agree/disagree, of very little importance/oflittle importance) response. We also calcu-lated the acquiescence balance by subtractingdisacquiescence from acquiescence, resultingin a score from –1.00 to 1.00. Following VanHerk et al. (2004), the acquiescence balancewas used as the final measure of acquiescentresponse style as it measures the tendency toagree more than disagree. Middle responsestyle was calculated as the proportion of ques-tions that received a middle (3) response foreach respondent. Extreme response style wasdivided into positive extreme response style(proportion of 5 responses) and negativeextreme response style (proportion of 1responses). For the analyses relating to language the two measures of ERS were combined.

As indicated, we used only the responsesto the native-language questionnaire for ourhypotheses relating to the impact of nationalcultural dimensions and the national level



of extraversion on response styles. As thisanalysis is conducted at country level, indi-vidual response styles were aggregated to thecountry level. For the hypotheses relating tothe impact of language on response styles we used questionnaires in both languages(hypothesis 5) or English-language question-naires only (hypothesis 6). Analyses were con-ducted at the individual level.

Independent variables Power distance,collectivism/individualism and uncertaintyavoidance were measured using Hofstede’s5

(1980, 2001) and the Globe study’s (House et al., 2004) country-level scores for thesedimensions. As Hofstede did not includecountry scores for Lithuania these were takenfrom a Lithuanian study (Mockaitis, 2002).Data for Lithuania, Bulgaria and Chile weremissing in the Globe study. The personalitycharacteristic of extraversion is part of twowell-established personality measurementinstruments: the Eysenck Personality Ques-tionnaire (EPQ) and the revised NEOPersonality Inventory (NEO-PI-R). Wedecided to use the scale from the EPQ since– tapping into aspects such as expressivenessand liveliness – this would seem to capturemuch of what was defined earlier as anexpressive/elaborate communication style.In contrast, the extraversion scale of theNEO-PI-R taps not only into friendliness,gregariousness and cheerfulness, but also intoassertiveness, activity and excitement seek-ing, which would not seem to be related asdirectly to an expressive communicationstyle. Moreover Hofstede and McCrae(2004) show that the extraversion scale of theNEO-PI-R showed very strong (0.57–0.64)correlations with the two culture variablesincluded in our study (individualism andpower distance), making it less useful as a distinct predictor of response styles. TheEPQ was not significantly related to any ofHofstede’s culture dimensions (correlationsvaried from 0.07 to 0.11), nor to most of theeight Globe study culture dimensions (corre-

lations varied from 0.00 to 0.32, only the correlation with in-group collectivism wassignificant, 0.50, p = .018), making it moreuseful as a distinct explanatory factor.

Country mean scores for extraversionwere taken from Van Hemert et al. (2002)who summarized the results of a range ofcountry studies that included extraversion asone of their concepts. However, if a particu-lar country was not reported in Van Hemertet al. but was included in Lynn and Martin(1995) – who provided a similar overview –we used the latter data. In three other cases(Japan, the Netherlands and France) we alsoused the Lynn and Martin data, because forthese countries the studies that were reportedby Van Hemert et al. included data for fewerrespondents. For Denmark, Austria, Turkey,Malaysia and Taiwan no extraversion scoreswere available. For the first four countries we used the score of their closest culturalequivalent in our sample (based on Hof-stede’s dimensions) that also shared histori-cal, geographical or linguistic links: Sweden,Germany, Greece and India, respectively.Taiwan didn’t have close cultural equivalentsand its nearest equivalent (Brazil) did notshare any historical, geographical or linguis-tic links. Hence we excluded Taiwan fromour analysis.

For our analysis of the impact of the language of the questionnaire on responsestyles, we compared response styles betweenthe native-language questionnaires and theEnglish-language questionnaires. In order tomeasure the impact of English-languagecompetency on response styles, we asked students to assess their capability to under-stand written English on an 8-point scale(very weak to fully bilingual). As for the latterquestion some of the categories had very few observations, we collapsed the eight categories into three (very weak–average,good/very good, excellent/bilingual).

Control variables There are several demo-graphic variables that have been shown to



influence response styles in earlier studies.ERS has been shown to increase with age(Greenleaf, 1992; Ross and Mirowsky, 1984)and males have been shown to have a higherlevel of acquiescence than females (Johnsonet al. 1997; Ross and Mirowsky, 1984). Bothvariables were therefore included as controlvariables in the individual-level analysis.Level of education (Greenleaf, 1992; Johnsonet al., 1997; Landsberger and Saavedra,1967; Marin et al., 1992) and socioeconomicstatus (Ross and Mirowsky, 1984) have alsobeen shown to impact on response styles, butas our respondents are reasonably wellmatched on this characteristic, it is notincluded as a control variable.

Results

Descriptive Results

Table 2 provides the descriptive results for the various response styles in the 26 coun-tries included in our survey. Our results con-form very closely to previous cross-countrystudies, many of which were conducted several decades ago. Students from Spanish-speaking countries show higher ERS andhigh acquiescence, while East Asian (Japan-ese and Chinese) respondents show a relative-ly high level of MRS. German respondentsshowed higher acquiescence than Britishrespondents did, and within Europe theGreeks stand out as having the highest level ofacquiescence and ERS.6

Table 2 also shows that while someregions – e.g. Northern and Western Europe– show fairly similar response patterns, otherregions are much less homogeneous. WithinEastern Europe two clear patterns are visible,with Russia and Poland showing high dis-acquiescence, low MRS and low positiveERS, while Bulgaria and Lithuania show thereverse pattern. In Southern Europe, Greeceand Turkey and Spain and Portugal formrather similar pairs, while some aspects of theFrench response style are more similar toNorthern/Western Europe, confirming its

geographical and cultural position as abridgehead between Northern/WesternEurope and Southern/Latin Europe. WithinLatin America, Mexico and Chile are rathersimilar, showing the typical Hispanicresponse style. Brazil, however, shows a mixof Hispanic and Northern/Western Euro-pean response styles. Brazil has been identi-fied as a country with particularly strong evidence of distinct subcultures (Lenartowiczand Roth, 2001), which might make ourresults difficult to generalize.

The largest differences, however, arefound in the Asian cluster that shows threevery different patterns. Respondents inChina and Hong Kong show medium acqui-escence, low disacquiescence, low positiveand negative ERS and high MRS, whileMalaysia and India show the reverse patternon nearly all of these indicators. Taiwantakes up a middle position between theseextremes. A third and very distinct pattern isshown by Japan, which has the lowest acqui-escence, the highest disacquiescence and thehighest MRS of all 26 countries. The resultsfor Malaysia show that ethnic backgroundalso influences response styles: Malaysian-born Chinese respondents had responsestyles that differed significantly from Malayrespondents and were generally situatedbetween Malay and Chinese (mainlandChina and Hong Kong) response styles.7

Malay respondents had a significantly higherARS (t = 2.727, p = .008) and positive ERS (t = 2.209, p = .031) than Malayisan-bornChinese respondents.

The Impact of CulturalDimensions and Extraversionon Response Styles

Table 3 summarizes the intercorrelations ofall variables used in our study. As would beexpected, the different types of response stylesshow strong intercorrelations. However, aswe will see, they do differ in terms of the factors that influence them. When we com-pare the correlation pattern for acquiescence




Table 2 Overview of response style differences across 26 countries, % of answers in particular categories, native language questionnaires only

1. Acqui- 2. Disacqui 3. Acqui- ERS ERS Middle escence escence escence positive negative response

% of 4/5 % of 1/2 balance % of 5 % of 1 % of 3 Country answers answers (1–2) answers answer answers

USA (n = 61) .54 .24 .30 .17 .03 .22

Northern EuropeDenmark (n = 44) .44 .27 .17 .10 .05 .28Finland (n = 87) .48 .32 .15 .13 .09 .20Sweden (n = 62) .45 .31 .13 .14 .10 .24

Western EuropeAustria (n = 53) .48 .26 .21 .15 .07 .26Germany (n = 50) .52 .21 .31 .15 .05 .26Netherlands (n = 109) .48 .31 .17 .08 .05 .21UK (n = 46) .48 .27 .21 .11 .04 .24

Eastern EuropeBulgaria (n = 78) .54 .20 .34 .18 .04 .26Lithuania (n = 57) .53 .20 .32 .16 .04 .26Poland (n = 54) .56 .27 .29 .12 .03 .16Russia (n = 44) .54 .29 .25 .14 .04 .16

Southern/Latin EuropeFrance (n = 42) .54 .31 .23 .17 .07 .14Portugal (n = 76) .53 .26 .28 .17 .06 .21Spain (n = 83) .54 .22 .32 .12 .04 .22Greece (n = 58) .58 .20 .38 .22 .05 .13Turkey (n = 78) .57 .21 .36 .22 .05 .21

Latin AmericaBrazil (n = 72) .51 .30 .22 .20 .09 .19Chile (n = 53) .58 .19 .39 .19 .04 .23Mexico (n = 50) .60 .21 .39 .28 .06 .19

AsiaChina (n = 50) .51 .19 .33 .12 .02 .26Hong Kong (n = 54) .54 .18 .36 .12 .03 .26India (n = 50) .60 .17 .44 .28 .05 .16Japan (n = 45) .39 .32 .07 .10 .08 .28Malaysia (n = 65) .61 .17 .44 .19 .03 .21

• Malay (n = 38) .63 .15 .48 .22 .02 .21• Chinese (n = 27) .58 .20 .38 .13 .03 .22

Taiwan (n = 60) .61 .18 .43 .17 .03 .19

Overall average (n = 1581) .52 .25 .27 .15 .05 .22

ERS = Extreme response style.


International Journal of C

ross Cultural M

anagement6(2)

254

Table 3 Correlation matrix

Variable 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

1. Acquiescence balance 1 –.30 –.61*** .66*** .72*** .14 –.44* .54** –.60*** .48* .25 .58** –.27* .63*** .11 .36

2. Middle response style 1 –.11 –.51** –.40 –.25 .44* –.34 .10 –.47* .20 –.46* .62** –.45* –.47* –.15

3. Extreme response style negative 1 .11 –.40 .000 .14 –.05 .26 –.04 –.54** –.37 .20 –.36 .11 .05

4. Extreme response style positive 1 .56** .24 –.44* .55** –.46* .55** –.12 .46* –.32 .49* .30 .53**

5. in-group collectivism practices 1 .14 –.26 .47* –.82*** .70*** .17 .77*** –.61** .86*** .39 –.10

6. in-group collectivism values 1 .05 –.06 .05 .24 –.06 .19 –.23 .15 .19 .50*

7. Institutional collectivism practices 1 –.72** .09 –.45* .40 –.13 .41 –.19 –.46* –.32

8. Institutional collectivism values 1 –.46* .46* –.37 .26 –.28 .40 .43* .28

9. Individualism (Hofstede) 1 –.47* –.23 –.66*** .35 –.79*** –.32 .08

10. Power distance practices 1 –.16 .57** –.60** –.58** .56** .05

11. Power distance values 1 .31 .10 .31 –.39 –.21

12. Power distance (Hofstede) 1 –.58** .80*** .26 –.11

13. Uncertainty avoidance practices 1 –.62*** –.68*** .00

14. Uncertainty avoidance values 1 .32 –.15

15. Uncertainty avoidance (Hofstede) 1 .07

16. Extraversion 1

Notes: N = 1581. *** p <0.001, ** p < 0.01, * p < 0.05 = p <0.1, all two-tailed.

at The U

niversity of Melbourne Libraries on A

ugust 29, 2013ccm

.sagepub.comD

ownloaded from


in our study with that reported for six differ-ent studies by Smith (2004), we find nearlyidentical patterns. All cultural dimensionsthat showed a significant correlation withacquiescence in our study also did so inSmith’s study, often for all five or six studiesthat Smith analysed. In most cases even themagnitude and level of significance of thecorrelations is very similar. This furtherstrengthens Smith’s argument that bias willbe consistently predicted by the same valueprofile. It also reinforces our earlier argu-ment of consistency of response styles overtime, as our data were collected after thestudies that were reported in Smith (2004).Finally, it shows that even though we used astudent sample, our results are very similar to those of studies that used managerial samples.

Table 4 shows the regression results foreach set of culture dimensions (Hofstede,Globe practices, Globe values) and extraver-sion. It is apparent that whereas acquies-cence and a positive extreme response styleare well explained by the variables includedin our study, this is less so for a negativeextreme response and middle response style.We will see that language is a better predictorfor ERS and MRS. Hypothesis 1 predicted apositive relationship between power distanceand acquiescence. Table 4 shows partial support for this hypothesis as Hofstede’spower distance dimension shows a positivecorrelation with acquiescence. However, neither of the Globe measures of power dis-tance shows the same relationship. Hof-stede’s power distance measure also showeda significant negative correlation with middleresponse styles, while Globe power distancepractices (values) are weakly positively (nega-tively) related to negative ERS. Hypothesis 2 predicted a positive relationship betweencollectivism and acquiescence and middleresponse style. Again there is partial supportfor this hypothesis. Hofstede’s individualismmeasure (the opposite of collectivism) showsa significant negative relationship with both

an acquiescent and a middle response style.Globe’s in-group collectivism practices alsoshow the predicted positive relationship withacquiescence, but are not related to middleresponse style. They do have a significantpositive (negative) relationship with positive(negative) ERS. Institutional collectivismpractices on the other hand are significantlynegatively related with acquiescence, thoughagain no relationship is found with middleresponse style. Hypothesis 3 predicted a posi-tive relationship between uncertainty avoid-ance and acquiescence. Again partial supportfor this hypothesis is found in that bothGlobe uncertainty avoidance practices andvalues show the expected relationships, whileuncertainty avoidance values also have astrong positive relationship with ERS. Theresults for Hofstede’s uncertainty avoidancedimension are not significant, although thereis a weakly significant negative relationshipwith middle response styles.

As predicted in Hypothesis 4 extraversionhas a significant positive relationship withextreme response styles. This result is con-sistent, regardless of which other variablesare included in the analysis. However, thisresult is limited to positive extreme responsestyles; that is, the tendency to strongly agree.This is not unexpected as, similar other studies, the major differences between coun-tries were found in positive ERS and henceresults for ERS and acquiescence are related.Country-level extraversion was also found tobe a significant positive determinant of anacquiescent response style.

Overall, respondents in countries withhigh power distance values seem to preferpositive extreme response styles to middleresponse styles and negative extremeresponse styles, and collectivism appears tolead to a preference for acquiescence andmiddle response styles. Uncertainty avoid-ance is associated with a higher level ofacquiescence, and a preference for increaseduncertainty avoidance is very strongly associ-ated with extreme positive answers. Extra-



version shows its strongest impact on positiveERS (p < .001), but also shows secondaryexplanatory power for acquiescence (posi-tive, p < .01). Overall, in-group collectivismpractices and extraversion seem to be thecharacteristics that most consistently influ-ence response styles, while uncertainty avoid-ance values and Hofstede’s power distanceand individualism measures are quite influ-ential too. However, we should note thatthese four cultural dimensions show verystrong intercorrelations – typically around.60 to .80.

The Impact of Language onResponse Styles

Hypotheses 4 and 5 referred to the impact oflanguage of the questionnaire and English-language competency on response styles. Totest the relative impact of language versuscountry, SPSS’s General Linear Model(GLM) Factorial procedure was used. TheGLM procedure is a technique that providesregression analysis for one dependent vari-able by one or more factors and/or variables.In contrast to linear regression analysis, thistechnique allows a combination of categori-cal and continuous independent variables,


Table 4 Regression analysis of the impact (standardized beta coefficients) of power distance,individualism/-collectivism, uncertainty avoidance and extraversion on response styles

Acquiescence% of 4/5 answers – Negative ERS Positive ERS MRS

Response style % of 1/2 answers % of 1 answer % of 5 answers % of 3 answers

Hofstede values (n = 25)Power distance .372* –.363 .342 –.800**Individualism –.445* .095 –.265 –.567**Uncertainty avoidance –.166 .231 .089 –.338=Extraversion .446** –.016 .579*** –.101Adjusted R-square .550 .018 .498 .399

Globe practices (n = 23)Power distance –.165 .585= .176 –.151In-group collectivism 1.029*** –.762* .564* –.008Institutional collectivism –.289* .221 –.111 .064Uncertainty avoidance .381* –.014 .177 .456Extraversion .371** .015 .541*** –.093Adjusted R-square .761 .096 .593 .138

Globe values (n = 23)Power distance .150 –.581* –.210 .306in-group collectivism –.161 –.015 –.240 –.150Institutional collectivism .272 –.202 –.022 .038Uncertainty avoidance .533* –.105 .712** –.515Extraversion .501* –.027 .722*** –.066

Adjusted R-square .615 .159 .596 .042

Notes: *** p <0.001; ** p < 0.01; * p < 0.05; = p < 0.1, all two-tailed. ERS = Extreme response style, MRS =Middle response style.


without the necessity to recode categoricaldata into individual dummy variables.Because the sample sizes for English andnative-language questionnaires varied slight-ly in the different countries, we included theinteraction effect between language of thequestionnaire and country into the firstmodel. Further, since the 26 countries in oursurvey might differ in their average English-language competence, the interaction effectbetween country and English-language com-petence was included in the second model. Inaddition, our literature survey indicated thatdemographic variables such as age and gender might be related to response stylesand hence they were included as control variables. As Table 5 shows, even when con-trolling for country, demographic variablesand the interaction between country and language, language remains a very importantdeterminant of response styles. In fact, in fiveof the six analyses it is the most importantdeterminant. Confirming our analyses, therespondent’s country also has an importantimpact on response styles. Age does notappear to have a systematic or consistentimpact, which is not surprising given therestricted age range in our student sample.Gender does appear to have some impact,

with male students generally showing higherERS and female students showing higherMRS, which confirms earlier studies (John-son et al. 1997; Ross and Mirowsky, 1984).8

In terms of our specific hypotheses, Table6a shows strong confirmation of bothhypothesis 5a and hypothesis 5b. Extremeresponses are more likely when a respondentis responding in his or her native language,while middle responses are more likely whenEnglish language questionnaires are used. Asa result the standard deviation is significantlyhigher for native-language questionnairesthan for English-language questionnaires. AsTable 6b shows, higher English-languagecompetence is significantly positively relatedto extreme response styles and significantlynegatively related to middle response styles,thus confirming hypothesis 6. Standard deviation differs accordingly: the higher theEnglish-language competence, the higher thestandard deviation.

Discussion and Conclusion

Our study has shown that there are substan-tial differences in response styles across countries that, without exception, confirmedpatterns found in earlier studies. These


Table 5 Impact of language versus country and control variables on response styles, F-valuesand 2-tailed significance levels in GLM analysis

Language EnglishResponse questionnaire competence Country* Adjustedstyle n = 2940 n = 1402 Country language Age Gender R-square

ERS 34.952*** – 15.520*** 2.078** 2.226 9.151** .128MRS 43.993*** – 11.897*** 4.042*** 8.294** 0.984 .114SD 48.483*** – 12.615*** 2.172*** .888 3.514= .114ERS – 6.226*** 5.220*** 1.458* .160 .282 .137MRS – 2.667= 3.661*** 1.003 1.654 4.511* .092SD – 8.178*** 3.245*** 1.336= .870 .313 .113

*** p <0.001; ** p < 0.01; * p < 0.05; = p <0.1, all two-tailed.ERS = Extreme response style, MRS = Middle response style, SD = Standard deviation.


strongly consistent results point to responsestyle differences between countries that arevery stable across time. One of the earlieststudies on response style differences (Zax andTakahashi, 1967) linked the high MRS andlow positive ERS in Japan to child-rearingpractices that promote restraint and suppressimpulsive displays. The authors suggestedthat these results might change since theirrespondents were born just after World WarII and the response styles of their childrenmight draw closer to the western groups theyare striving to emulate (Zax and Takahashi,1967). Our study shows that respondentswho were born some 40 years later still dis-play very similar patterns.

Given that no less than 26 countries wereincluded in our study, we now know muchmore about response styles for a wide rangeof countries. Our results for Malaysia also

showed that ethnic background could have apersistent effect on response styles, withMalay respondents showing response stylesdifferent from those of Chinese respondents.Over 35 years ago, Mitchell (1968) observeda similar difference between Chinese andIndian respondents in Malaysia. Hence our results are relevant not only for cross-national surveys, but also for cross-culturalsurveys within nations. A second contribu-tion of our study was to test the impact ofvarious cultural dimensions and one person-ality characteristic on a variety of responsestyles. The results generally confirmedSmith’s (2004) exploratory results, butextended them to a wider range of responsestyles, including extreme response style andmiddle response style. We also showed thatextraversion was one of the most importantand consistent determinants of response


Table 6a Impact of language of the questionnaire on response styles

Language of the questionnaire SignificantResponse 1. English language 2. Native language Significance differences,style n = 1443 n =1529 t-value 2-tailed p < 0.05

ERS .1817 .2111 5.826 .000 2 > 1MRS .2399 .2177 6.152 .000 1 > 2SD 1.0296 1.0770 6.695 .000 2 > 1

ERS = Extreme response style, MRS = Middle response style, SD = Standard deviation.

Table 6b Impact of English language competence on response styles, English-language questionnaires only

Ability to understand written English

1. Very weak– 2. Good/ 3. Excellent/ Significant Response Average, Very Good, Bilingual, Significance differences, style n = 315 n = 696 n = 417 F-value 2-tailed p < 0.05

ERS .1529 .1742 .2176 25.782 .000 3 > 2 > 1MRS .2648 .2356 .2277 13.826 .000 1 > 2&3SD .9602 1.0238 1.0942 51.335 .000 3 > 2 > 1

ERS = Extreme response style, MRS = Middle response style, SD = Standard deviation.


styles. Future researchers might want to focuson other determinants more directly relatedto communication style rather than culturaldimensions in explaining differences inresponse styles. A third contribution of ourstudy was to show that the language of the questionnaire and the English-languagecompetency of the respondents influenceextreme and middle response styles as well asstandard deviation. Responses in the nativelanguage showed higher ERS and SD andlower MRS, while the level of English-language competency was related to higherERS and SD and lower MRS in respondingto English-language questionnaires.

An important explanation for responsestyle differences across languages might bedifferential interpretation of equivalent scaleanchors in different languages. Even thoughscale anchors might translate into appropri-ate local equivalents, the intensity associatedwith these equivalents might be differentfrom the original language. Voss et al. (1996)show that while the magnitude estimates forgood and very good were 74 and 87 inEnglish, they were 91 and 101 in the equiva-lent Japanese translation. As far as we areaware no equivalent research has been donefor disagree–agree scales or unimportant–important scales, but if results would be similar, this could explain the very low acquiescence of our Japanese respondents.

Our analysis of response style differencesfocused on aggregate response styles for thequestionnaire as a whole, as we reasoned thatresponse styles would occur irrespective ofthe type of question concerned. However, wemight ask ourselves whether differences inresponse styles are more likely for questionsrelating to cultural dimensions than for moreneutral questions such as elective choice andto a lesser extent ideal job characteristics. We therefore re-ran all the analyses usingresponse styles for each group of questions(culture activity dimension, culture relation-ship dimension, elective choice and ideal jobcharacteristics) as a dependent variable.

While F-values differed slightly, overall dif-ferences in response styles between countriesturned out to be highly significant (all p <.000) regardless of the type of questions.There was no distinguishable pattern thatcould indicate that response style differenceswere more important for some questionsthan for others.

A second question in this respect wouldbe whether Hofstede’s/Globe’s cultural dimen-sions and extraversion would have a moresignificant explanatory power for some sets ofquestions than for others. Given that some ofour questions dealt with cultural dimensions(albeit slightly or even considerably differentfrom the Hofstede/Globe dimensions thatwere included as explanatory variables),some of our results might be due to ‘true’content-related correlation between the inde-pendent and dependent variables rather thancorrelation between the Hofstede/Globe cultural dimensions and response styles. Inparticular our culture relationship dimensioncould be expected to have a content-orientedlink with the individualism/collectivism/power distance dimensions. However, oursplit-group analysis showed that relationshipsbetween the independent variables andresponse styles were generally upheld foreach group of questions, hence lendingstrength to the idea that the underlying rea-son for the relationship is indeed differencesin response styles. What did become appar-ent though is that the relationship betweenextraversion and response styles wasstrongest for the ideal job characteristics. Theideal job questions were measured on a different scale (from ‘of very little or noimportance’ to ‘of utmost importance’) thanthe three other sets of questions that used astrongly disagree to strongly agree format.This might indicate that scale anchors refer-ring to the level of importance might even bemore susceptible to an acquiescent responsebias than scale anchors referring to the levelof agreement.

A third main topic in our article was the



impact of the language of the questionnaireon response styles. Differentiating our analy-sis by group of questions did not show anymajor differences between the groups in thisrespect. In all four groups both language andcountry, as well as the interaction betweenthe two, were highly significant determinantsof ERS, MRS and standard deviation. Asreflected in the overall analysis, the languageof the questionnaire was generally (in 9 of the12 comparisons) a more important determi-nant of ERS, MRS and SD than the countryin which the data were collected. Withregard to English-language competence, ourdetailed results are slightly different. Eventhough in the overall analysis English-language competence was a significant factorinfluencing response styles, in some of theindividual analyses this was not the case. It isclear from both the overall analysis and thedetailed analysis that the language of thequestionnaire is a more significant factorinfluencing response styles than English-language competence as such. Finally, adetailed comparison of ERS, MRS and SDfor each of the four groups of questions generally confirmed the overall analysis, withERS and SD being higher and MRS beinglower in the native language. Differenceswere highly significant (p = .000) for each ofthe four groups of questions, except for ERSelectives (p = .003) and SD electives (p =.037). English-language competence had thesame impact for each of the four groups ofquestions, with higher level competenceresulting in higher ERS and SD and lowerMRS. Except for MRS culture activity (p =.16) and MRS culture relationship (p = .011),results were significant at p = .000. Overall,our detailed analyses showed that resultswere very similar across all four groups ofquestions. This reinforces our argument thatdifferences in response styles are a major factor to take into account in any inter-national comparisons.

An important question that we have notanswered yet is how to eliminate or attenuate

the impact of response styles. An establishedprocedure for removing bias associated withscale response is standardization (Leung andBond, 1989). This procedure has becomeincreasingly popular in cross cultural studies(Fischer, 2004). However, standardizationmight also remove some of the true differ-ences in responses. It remains difficult toassess what part of, for instance, a high meanscore is caused by acquiescence bias and whatpart truly reflects a strong opinion about thesubject in question. In addition, for question-naires that cover different topical areas, stan-dardization over the questionnaire as a wholemight cause a strong response bias for onepart of the questionnaire to unduly impact onthe scores of another part of the question-naire. This would reduce the validity of cross-country comparisons at the level of differentaspects of the questionnaire (Maznevski et al., 2002). Fischer (2004) reviews differentmethods of standardization and provides an excellent overview of the problems andlimitations associated with them.

Rather than trying to eliminate responsebias retrospectively through standardization,researchers could attempt to avoid it by care-ful questionnaire design. Several options areavailable. First, Smith (2003) suggests thatthe use of a mixture of positive and negativestatements will mitigate both acquiescenceand disacquiescence, because it might leadrespondents to consider the exact meaning ofthe question more closely and as a result givemore meaningful responses. But even if thiseffect does not materialize, at least responseswill cancel each other out, so that the averagefor the respondent in question represents amiddle position, which would be a betterreflection of his or her true opinion than oneextreme or the other (Smith, 2003). Theproblem, however, is that questionnaireitems containing negations are difficult totranslate into some languages.

A solution to mediate the impact ofextreme response styles is to use Likert scaleswith a larger number of categories, which



allows respondents with a relatively strongopinion to voice a more nuanced position,rather than being forced to choose the mostextreme answer. Hui and Triandis (1989)found that ERS for Hispanics disappearedwhen 10-point Likert scales were used.

Most studies that show response bias usedLikert scales with ordered scale anchors suchas ‘strongly disagree’ to ‘strongly agree’, or‘of little or no importance’ to ‘extremelyimportant’. These anchors might be vulner-able to acquiescence bias as respondents arekeen to agree, whether this is caused by high levels of power distance, collectivism,uncertainty avoidance or extraversion. Ourdetailed analysis showed that scale anchorsreferring to the level of importance are evenmore problematic in this respect than scaleanchors referring to the level of agreement. Arelated problem in this respect – discussedearlier with special reference to Japan – isthat scale anchors are often difficult to trans-late and translations might not result in metric equivalence. An alternative would beto use scale anchors as part of the questionand let them reflect opposites rather thanlevel of agreement. This would make the‘right answer’ less obvious and would alsoforce respondents to carefully consider eachquestion as most scale anchors would be different. Of course this technique wouldincrease the level of cognitive involvementrequired and might lead to lower responserates. In addition, careful translation andpilot testing would become even more crucialas a respondent’s interpretation of the ques-tions would be framed by single words,whereby words that are seen as opposites insome countries might not be opposites inother countries. However, if translationproblems can be solved, responses might bemore meaningful. The Globe study (House etal., 2004) used many items that were con-structed in this way – such as, ‘In this society,people are generally: tough/tender’ or ‘Inthis society, people place more emphasis on:solving current problems/planning for the

future’. It is probably not coincidental thatresponse bias was found to be modest in thisstudy.

A final remedy might be to ask respon-dents to rank statements rather than usingLikert scales. Of course, this is only possible ifthe subject area is such that a hierarchicalordering of statements can be expected. Thecharacteristics of ideal types of jobs, as dis-cussed earlier, would be one subject area inwhich this would have been possible. On theother hand, asking respondents to rank morethan a handful of statements puts a very high demand on their cognitive abilities andmight lead them to discard the questionnairealtogether. In addition, statistical analysesthat can be performed with rank-orderedscores are more limited than those that can be used for interval or quasi-intervalscales (Alwin and Krosnick, 1985). An inter-esting alternative to ranking is suggested byLenartowicz and Roth (2001) in their studyof cultural values. They first asked respon-dents to indicate the most and least impor-tant value and rate its importance using a 10-point scale. Then respondents were asked torate the next most and least important valueswithin the range of the previously rated values, and so on. For the analysis of final ratings all subject scores were then trans-formed to bring all the respondents’ ratingsto the same range of values, hence eliminat-ing response styles. This solution preservesthe hierarchical measurement, but alsoincludes individual ratings. Again, though, itputs heavy demands on the respondent’stime and cognitive capabilities.

Regardless of what remedy is used toeliminate or alleviate response bias, the firststep towards finding a solution is acknow-ledging that response bias can be a seriousthreat to valid comparisons across countries.We hope this article has provided a step inthat direction and that in future response biaswill receive the attention it deserves fromresearchers in the area of international andcross cultural management.



Notes1 In this article we only deal with response

styles that are independent of item content.Socially desirable response styles that varywith item content are not discussed.

2 Culpepper et al. (2002) propose aninteresting distinction in response styles forChinese respondents. They review twodifferent streams of literature: one claimingthat Chinese respondents have a highertendency for extreme response styles and onemaintaining that Chinese respondents have amid-point response bias. Both are supportedby empirical studies. They resolve thisapparent contradiction by looking at the typeof questions concerned. For simple questionsrelated to knowledge that is ‘time-tested andwidely accepted’, Confucian influence leadsChinese respondents to see debate betweenopposing viewpoints as undesirable. They areless likely to weigh up pros and cons andinstead have a tendency to prefer extremeresponses that reflect the time-tested andwidely accepted wisdom. However, for itemsrequiring ‘ideographic characterizations’ thatinvolve judgements that are more complexand analytical, more modest responses arelikely, based on the Confucian philosophythat does not value assertiveness and thedisplay of strong independent opinions. Theirempirical study that included both types ofquestions fully supported this distinction. Asour study includes questions of the latter type(ideographic characterization), we wouldexpect Chinese respondents to be displayinga rather modest response style.

3 Smith (2004) found some of the other Globedimensions to be linked to acquiescence aswell. However, a theoretical rationale forthese links was not obvious and in most casesthese dimensions only predicted acquiescencefor one of studies included in his review,while the predictors that we did include inour study showed a consistent effect across allseven studies covered by Smith.

4 As we are interested in response stylepatterns, not in the scoring on individualquestions or constructs, we did not constructscales. Response styles were calculated usingall 69 questions, so that each item had anequal contribution to the composite responsestyle variables.

5 Although Hofstede’s work has elicited somecriticism, it is largely accepted as a helpful,although crude, way to quantify cultural

differences (see Harzing and Hofstede, 1996,for a discussion of the various critiques andthe extensive use of Hofstede’s dimensions inother studies; and see Søndergaard, 1994, fora summary of reviews, replications andcitations).

6 We conducted formal statistical tests for allcountries that were included in both previousstudies and our study. Confirming studies byRoss and Mirowsky (1984), Hui and Triandis(1989), Marin et al. (1992), Johnson et al.(1997) and Clarke III (2000), Mexico – themost likely home country of the Hispanics inthese studies – and the USA show verysignificant differences in both ARS (t =2.944, p = .004) and ERS (t = 5.866, p =.000). Confirming Zax and Takahashi (1967),Chen et al. (1995), Shiomi and Loo (1999)and Takahashi et al. (2002), Japaneserespondents showed significantly higher (t =2.726, p = .008) levels of MRS than USrespondents. They also showed significantlylower (t = 3.636, p = .000) levels of positiveERS, but higher levels of negative ERS (t =3.806, p = .000). Differences in ARS betweenGerman and British respondents weresignificant (t =3.765, p = .000) and parallelthose found in Brengelmann (1959). Inconformance with Steenkamp andBaumgartner (1998) and Van Herk et al.(2004) Greek respondents showedsignificantly higher levels of ARS andpositive ERS than British (t = 6.128/6.216, p= .000/.000), German (t = 2.491/2.705, p =.014/0.008), French (t = 4.565/1.380, p =.000/.194) and Spanish (t = 2.558/5.343, p =.012/0.000) respondents. These consistentresults for Greece are difficult to reconcilewith the Globe Leadership project that foundGreece to have one of the lowest levels ofacquiescence. Mean scores for Greece fallwithin the lowest bands for most of the Globeculture dimensions for practices and/orvalues. The only exception is GenderEgalitarianism where Greece falls within thehighest band for both practices and values. Anoticeable feature of the Globe questionnaireis that many items were reverse-scored.Judging from the sample items listed inHouse et al. (2004) Gender Egalitarianismwas the only dimension where items were notreverse-scored. It is possible that reversescoring has influenced the typical Greekacquiescent response pattern.

7 It is possible that the language of thequestionnaire (Malay) resulted in Chinese



response styles that were more similar toMalay respondents than they would havebeen if the questionnaire had been inChinese. This, however, only reinforces theethnic background argument.

8 These studies looked at acquiescence ratherthan ERS and MRS, but the three responsestyles are obviously interrelated. A GLManalysis with acquiescence as a dependentvariable confirmed the significant impact ofgender.

ReferencesAllik, J. and McCrae, R.R. (2004) ‘Toward a

Geography of Personality Traits: Patterns ofProfiles across 36 Cultures’, Journal of Cross-Cultural Psychology 35(1): 13–28.

Alwin, D.F. and Krosnick, J.A. (1985) ‘TheMeasurement of Values in Surveys: AComparison of Ratings and Rankings’, PublicOpinion Quarterly 49: 535–52.

Bachman, J.G. and O’Malley, P.M. (1984) ‘Yea-saying, Nay-saying, and Going to Extremes:Black–White Differences in Response Styles’,Public Opinion Quarterly 48: 491–509.

Baumgartner, Hans and Steenkamp, Jan-Benedict E.M. (2001) ‘Response Styles inMarketing Research: A Cross-nationalInvestigation’, Journal of Marketing Research 38:143–56.

Bennett, M. (1977) ‘Response Characteristics ofBilingual Managers to OrganisationalQuestionnaires’, Personnel Psychology 30: 29–36.

Brengelmann, J.C. (1959) ‘Differences inQuestionnaire Responses between Englishand German Nationals’, Acta Psychologica 16:339–55.

Chen, C., Lee, S.-Y. and Stevenson, H.W. (1995)‘Response Style and Cross-culturalComparisons of Rating Scales among EastAsian and North American Students’,Psychological Science 6(3): 170–5.

Chun, K.-T., Campbell, J.B. and Yoo, J.H.(1974) ‘Extreme response style in cross-cultural research’, Journal of Cross-CulturalPsychology 5(4): 465–80.

Church, A.T. (1987) ‘Personality Research in aNon-Western Culture: The Phillipines’,Psychological Bulletin 102(2): 272–92.

Church, A.T., Katigbak, M.S. and Castaneda, I.(1988) ‘The Effects of Language of DataCollection on Derived Conceptions ofHealthy Personality with Filipino Bilinguals’,Journal of Cross-Cultural Psychology 19: 178–92.

Clarke III, I. (2000) ‘Extreme Response Style inCross-cultural Research’, Journal of SocialBehavior and Personality 15(1): 137–52.

Culpepper, R.A., Zhao, L. and Lowery, C.(2002) ‘Survey Response Bias among ChineseManagers’, Academy of Management Proceedings,Denver, 9–14 August 2002.

Fischer, R. (2004) ‘Standardization to Accountfor Cross-cultural Response Bias: AClassification of Score Adjustment Proceduresand Review of Research in JCCP’, Journal ofCross-Cultural Psychology 35(3): 263–82.

Gibbons, J.L., Zellner, J.A. and Rudek, D.J.(1999) ‘Effects of Language andMeaningfulness on the Use of ExtremeResponse Style by Spanish–EnglishBilinguals’, Cross-Cultural Research 33(4):369–81.

Greenleaf, E.A. (1992) ‘Measuring ExtremeResponse Style’, Public Opinion Quarterly 56(3):328–51.

Gudykunst, W.B., Ting-Toomey, S. and Chua,E. (1988) Culture and Interpersonal Communication.Newbury Park, CA: Sage Publications.

Hall, E. (1976) Beyond Culture. New York: AnchorPress.

Harzing, A.W.K. and Hofstede, G. (1996)‘Planned Change in Organizations: TheInfluence of National Culture’, Special Issueof the Research in the Sociology of Organizationsseries, pp. 297–340. Greenwich, CT: JAIPress.

Harzing, A.W.K., Maznevski M. (with countrycollaborators) (2002) ‘The Interactionbetween Language and Culture: A Test of theCultural Accommodation Hypothesis inSeven Countries’, Language and InterculturalCommunication 2(2): 120–39.

Harzing, A.W.K. and 32 country collaborators(2005) ‘The Use of English Questionnaires inCross-national Research: Does CulturalAccommodation Obscure NationalDifferences?’, International Journal of CrossCultural Management 5(2): 213–24.

Hofstede, G. (1980) Culture’s Consequences.International Differences in Work-related Values.London: Sage Publications.

Hofstede, G. (2001) Culture’s Consequences,Comparing Values, Behaviors, Institutions andOrganizations across Nations, 2nd edn. ThousandOaks, CA: Sage Publications.

Hofstede, G. and McCrae, R.R. (2004)‘Personality and Culture Revisited: LinkingTraits and Dimensions of Culture’, Cross-Cultural Research 38(1): 52–88.

House, R.J., Hanges, P.J., Javidan, M., Dorfman,



P.W. and Gupta, V. (2004) Culture, Leadership,and Organizations, The Globe Study of 62 Societies.Thousand Oaks, CA: Sage Publications.

Hui, C.H. and Triandis, H.C. (1989) ‘Effects ofCulture and Response Format on ExtremeResponse Style’, Journal of Cross-CulturalPsychology 20(3): 296–309.

Javeline, D. (1999) ‘Response Effects in PoliteCultures. A Test of Acquiescence inKazakhstan’, Public Opinion Quarterly 63(1):1–28.

Johnson, T., O’Rourke, D., Chavez, N.,Sudman, S., Warnecke, R., Lacey, L. andHorm, J. (1997) ‘Social Cognition andResponses to Survey Questions amongCulturally Diverse Populations’, in L. Lyberg,P. Biemer, M. Collins, E. de Leeuw, C.Dippo, N. Schwarz and D. Trewin (eds) SurveyMeasurement and Process Quality, pp. 87–113.New York: John Wiley & Sons

Landsberger, H.A. and Saavedra, A. (1967)‘Response Set in Developing Countries’,Public Opinion Quarterly 14: 214–29.

Lee, C. and Green, R.T. (1991) ‘Cross-culturalExamination of the Fishbein BehavioralIntentions Model’, Journal of InternationalBusiness Studies 25(2): 289–305.

Lenartowicz, T. and Roth, K. (2001) ‘DoesSubculture within a Country Matter? ACross-cultural Study of Motivational Domainsand Business Performance in Brazil’, Journal ofInternational Business Studies 32(2): 305–25.

Leung, K. and Bond, M. H. (1989) ‘On theEmpirical Investigation of Dimensions forCross-cultural Comparisons’, Journal of Cross-Cultural Psychology 20(2): 133–51.

Lynn, R. and Martin, T. (1995) ‘NationalDifferences for Thirty-seven Nations inExtraversion, Neuroticism, Psychoticism andEconomic, Demographic and OtherCorrelates’, Personality and Individual Differences19(3): 403–6.

McCrae, R.R. (2002) ‘NEO-PI-R Data from 36Cultures’ in R.R. McCrae and J. Allik TheFive-factor Model of Personality across Cultures, pp. 105–25. New York: Kluwer AcademicPublishers.

Marin, G., Gamba, R.J. and Marin, B.V. (1992)‘Extreme Response Style and Acquiescenceamong Hispanics. The Role of Acculturationand Education’, Journal of Cross-CulturalPsychology 23(4): 498–509.

Maznevski, M.L., DiStefano, J.J., Gomez, C.B.,Noorderhaven, N.G. and Wu, P.-C. (2002)‘Cultural Dimensions at the Individual Levelof Analysis: The Cultural Orientations

Framework’, The International Journal of CrossCultural Management 2(3): 275–98.

Mitchell, R.E. (1968) ‘Survey Materials Collectedin the Developing Countries: Obstacles toComparisons’, in S. Rokkan (ed.) ComparativeResearch across Cultures and Nations. The Hague:Mouton.

Mockaitis, A.I. (2002) ‘The National CulturalDimensions of Lithuania’ Ekonomika, 59:67–77.

Ross, C.E. and Mirowsky, J. (1984) ‘Socially-desirable Response and Acquiescence in aCross-cultural Survey of Mental Health’,Journal of Health and Social Behavior 25(2):189–97.

Shiomi, K. and Loo, R. (1999) ‘Cross-culturalResponse Styles on the KirtonAdaptation–Innovation Inventory’, SocialBehavior and Personality 27(4): 413–20.

Singh, J. (1995) ‘Measurement Issues in Cross-national Research’ Journal of InternationalBusiness Studies 26: 597–620.

Smith, P.B. (2004) ‘Acquiescent Response Bias asan Aspect of Cultural Communication Style’,Journal of Cross-Cultural Psychology 35(1): 50–61.

Smith, T.W. (2003) ‘Developing ComparableQuestions in Cross-national Surveys’, in J.A.Harkness, F.J.R. Van de Vijver and P.P.Mohler (eds) Cross-cultural Survey Methods, pp. 69–92. New York: John Wiley & Sons.

Søndergaard, M. (1994) ‘Research Note:Hofstede’s Consequences: A Study ofReviews, Citations and Replications’,Organization Studies 15(3): 447–56.

Steenkamp, J.B.E.M. and Baumgartner, H.(1998) ‘Assessing Measurement Invariance inCross-national Consumer Research’, Journal ofConsumer Research 25(June): 78–90.

Takahashi, K., Ohara, N., Antonucci, T.C. andAkiyama, H. (2002) ‘Commonalities andDifferences in Close Relationships among theAmericans and Japanese: A Comparison bythe Individualism/Collectivism Concept’,International Journal of Behavioral Development26(5): 453–65.

Trompenaars, F. (1997) Riding the Waves of Culture.Understanding Cultural Diversity in Business.London: Nicholas Brealey Publishing.

Usunier, J.-C. (1998) International and Cross-culturalManagement Research. London: SagePublications.

Van de Vijver, F.J.R. and Leung, K. (2000)‘Methodological Issues in PsychologicalResearch on Culture’, Journal of Cross-CulturalPsychology 31: 33–51.

Van Hemert, D.A., Van de Vijver, F., Poortinga,



Y.H and Georgas, J. (2002) ‘Structural andFunctional Equivalence of the EysenckPersonality Questionnaire Within andBetween Countries’, Personality and IndividualDifferences 33: 1229–49.

Van Herk, H., Poortinga, Y.H. and Verhallen,T.M.M. (2004) ‘Response Styles in RatingScales: Evidence of Method Bias in Datafrom 6 EU Countries’, Journal of Cross-CulturalPsychology 35(3): 346–60.

Voss, K.E., Stem, D.E., Johnson, L.W. and Arce,C. (1996) ‘An Exploration of theComparability of Semantic Adjectives inThree Languages. A Magnitude EstimationApproach’, International Marketing Review 13(5):44–58.

Yang, K.S. and Bond, M.H. (1980) ‘EthnicAffirmation by Chinese Bilinguals’, Journal ofCross-Cultural Psychology 11: 411–25.

Yates, J.F., Lee, J.-W. and Bush, J.G. (1997)‘General Knowledge Overconfidence: Cross-national Variations, Response Style, and“Reality”’, Organizational Behavior and HumanDecision Processes 70(2): 87–94.

Zax, M. and Takahashi, S. (1967) ‘CulturalInfluences on Response Style: Comparisons ofJapanese and American College Students’,Journal of Social Psychology 71: 3–10.

ANNE-WIL HARZING is Associate Professor inInternational Management and Director ofStudies for the PhD Programme in theDepartment of Management of the University ofMelbourne, Parkville Campus, Victoria 3010,Australia.[email: [email protected]]


Résumé

Styles de réponse dans les enquêtes de recherche transnationale : une étudesur 26 pays (Anne-Wil Harzing)Les études d’attitudes dans plusieurs pays les pays reposent généralement sur une comparai-son des moyennes agrégées des scores obtenus aux questions de l’échelle de Likert. Ceci présuppose que, quand les gens remplissent un questionnaire, leurs réponses sont basées surla signification substantielle des questions auxquels ils répondent. Cependant, les réponses despersonnes sont également influencées par leur propre style de réponse. Par conséquent, lesétudes que nous entreprenons pourraient simplement ne refléter que les différences dans lafaçon qu’ont les gens de répondre aux enquêtes, plutôt que de relever les vraies différencesdans les perceptions du management à travers les pays. Notre étude sur 26 pays montre qu’ily a des différences majeures dans les styles de réponse entre les pays qui à la fois confirmentet pousse plus loin la recherche antérieure. Les caractéristiques au niveau de chaque paystelles que la distance au pouvoir, le collectivisme, la fuite de l’incertitude et l’extroversioninfluencent toutes de manière significative les types de réponse tels que l’assentiment et lesmodes de réponses extrêmes. De plus, les questionnaires en langue anglaise s’avèrent obtenirde plus forts taux de réponses intermédiaires, alors que les questionnaires dans la languematernelle des personnes interrogées donnent lieu à des types de réponse plus extrêmes. Enconclusion, le niveau de maîtrise de la langue anglaise est lié positivement aux types deréponses extrêmes et négativement aux taux de réponses intermédiaires. Nous terminons endiscutant les implications en recherche transnationale.



26

Anne-Wil Harzing

26

Anne-Wil Harzing


ccm cross cultural management · 2018-10-01 · bachman & o’malley 1984 usa: african...

Documents