1. Introduction

Cross-cultural comparisons have become standard in the international business literature (Nakata & Hung, 2005; Okazaki& Mueller, 2007; Okazaki, Taylor, & Doh, 2007), with most comparative work being based on surveys. However, survey dataare frequently distorted by certain response patterns, such as acquiescence, disacquiescence, midpoint tendencies andextreme response style (Harzing et al., 2009; Tsui, Nifadkar, & Ou, 2007; van de Vijver & Leung, 2000). The inuence of theseanswering tendencies on study results may be particularly severe when conducting cross-cultural research because culturedetermines how participants respond to rating scales (Baumgartner & Steenkamp, 2001; Harley, 1998; Smith, 2004; van

International Business Review 22 (2013) 814827

Acquiescence

Cross-cultural research

Disacquiescence

Equivalence

Hofstede

Project GLOBE

Response styles

demonstrates that Hofstedes and project GLOBEs cultural dimensions explain variations

in response styles across different countries. Secondly, and most importantly, the paper

explores the impact of culture-dependent response styles on ndings. Remarkably, the

extent of distortion in correlation analysis and mean comparisons is less severe than

expected. Nonetheless, cross-cultural researchers would be well advised to control at least

for (dis-)acquiescence before analyzing and interpreting their data.

2013 Elsevier Ltd. All rights reserved.

Contents lists available at SciVerse ScienceDirect

International Business Review

jo u r nal h o mep age: w ww.els evier .c o m/lo c ate / ib us r evHerk, Poortinga, & Verhallen, 2004). Depending on their cultural background, respondents may be more (or less) likely toanswer in the afrmative or use the extreme points of a scale irrespective of the item content. Consequently, answeringtendencies are a source of unwanted differences in observed measurement scores when comparing data from differentcultures. If culture-specic contamination is large, scholars might not be able to draw valid conclusions in cross-culturalresearch.

While several researchers have already demonstrated in large cross-national survey studies that culture affects thedegree of different response biases (Harzing, 2006; Johnson, Kulesa, Cho, & Shavitt, 2005; Smith, 2004; van Herk et al., 2004),no comprehensive study has yet analyzed the consequences of this methodological issue. In this paper, we examine howDo culture-dependent response styles distort substantial relationships?

Stefan Hoffmann a,*, Robert Mai b,1, Anamaria Cristescu b,1

aChristian-Albrechts-University of Kiel, Faculty of Business, Economics and Social Sciences, Department of Marketing, 24098 Kiel, Germanyb TU Dresden, Faculty of Business and Economics, Chair of Marketing, 01062, Dresden, Germany

A R T I C L E I N F O

Article history:

Received 16 May 2012

Received in revised form 20 November 2012

Accepted 17 January 2013

Keywords:

A B S T R A C T

Survey data are frequently distorted by answering tendencies, such as acquiescence,

disacquiescence, midpoint and extreme response style. Cross-cultural research projects

may be particularly vulnerable to misinterpretations. This paper provides empirical

insights into the manner and degree to which culture systematically distorts ndings from

survey data. The study is based on data from 1027 respondents from Austria, France,

Germany, India, Mexico, Russia, Spain, and the United States. The investigation rstlystrongly culture-dependent response styles distort substantial ndings from cross-cultural research. In doing so, we

* Corresponding author. Tel.: +49 431 880 4737; fax: +49 431 880 3349.

E-mail addresses: stefan.hoffmann@bwl.uni-kiel.de (S. Hoffmann), robert.mai@tu-dresden.de (R. Mai).1 Tel.: +49 351 463 34056; fax: +49 351 463 37176.

0969-5931/$ see front matter 2013 Elsevier Ltd. All rights reserved.http://dx.doi.org/10.1016/j.ibusrev.2013.01.008

S. Hoffmann et al. / International Business Review 22 (2013) 814827 815consider if and when it is necessary to correct for which response styles. Note that we use the term substantial todifferentiate the impact of response bias on important study results (the answers to the research question) from purelymethodological issues. For example, the inuence of a response style on the observed level of a certain scale (e.g.,ethnocentrism) is a methodological issue. A substantial issue would be the relationship between this scale and other relevantconstructs (e.g., cosmopolitanism). For this particular example, we examine whether the inuence of a response style on thescales of ethnocentrism and cosmopolitanism (=methodological issue) affects the relationship between both constructs indifferent cultures (=substantial issue). Expressed more formally, we approach the so far unanswered question of whether thewell-documented methodological issue actually affects substantial relationships in cross-cultural comparison research.

To ll the outlined void, this paper focuses on the four most widely discussed response styles, namely acquiescenceresponse style (ARS), disacquiescence response style (DRS), midpoint response style (MRS) and extreme response style (ERS).Among others, Baumgartner and Steenkamp (2001), as well as Weijters, Geuens, and Schillewaert (2008), suggest methodsfor detecting and correcting the four types of response biases. We apply these methods in a multi-cultural sample in order tomake the following two contributions to the eld. First, we analyze which cultural dimensions vary with response style.While previous studies have already considered some of the possible inuences (Harzing, 2006; Johnson et al., 2005; Smith,2004; van Herk et al., 2004), we extend the literature by replicating ndings on a broader empirical basis and by extendingprevious approaches to a wider set of response styles, countries, and cultural dimensions. The empirical study is based on adata set collected in eight countries (Austria, France, Germany, India, Mexico, Russia, Spain and the United States). In total,the sample consists of 1027 respondents. Our second and main contribution is to demonstrate how strongly culture distortssurvey ndings. To the best of our knowledge, no study has yet systematically conducted this second and most crucial step inanalyzing culture-dependent response style distortions. We analyze contaminations in correlations and mean comparisonsin international business research considering three focal constructs. We choose ethnocentrism, patriotism andcosmopolitanism, because they are often subject to research in international business. On a broad empirical basis, wereveal that although the effect of culture on response styles is signicant, the inuence on substantial relationships is lesssevere than commonly expected.

We organize the remainder of the paper as follows. First, we briey describe correction methods for different types ofresponse styles and review previous research on the relationships between these response styles and the culturaldimensions suggested by Hofstede (2001), as well as project GLOBE (House, Hanges, Javidan, Dorfman, & Gupta, 2004). Wereanalyze and largely conrm these relationships with our empirical data. Building on this robust nding, we then assess theextent to which culture-dependent response styles distort ndings in cross-cultural studies.

2. Conceptual background

2.1. Response styles

The international business literature has identied four main answering tendencies that may distort survey data in cross-cultural research projects (Baumgartner & Steenkamp, 2001; Harzing et al., 2009; Weijters et al., 2008). Firstly, acquiescenceresponse style (ARS) describes the tendency of respondents to agree with items regardless of their content. For example, if therespondent rates a statement on a seven-point scale ranging from 1 = I strongly disagree to 7 = I strongly agree, (s)hechooses 5, 6 or 7 disproportionately often. Secondly, disacquiescence response style (DRS) describes a tendency towarddisagreement. Regardless of the items content, the respondent chooses the categories that disafrm the statement (1, 2and 3). Thirdly, midpoint response style (MRS) describes the tendency of respondents to avoid a clear positioning. Theserespondents tend to choose the middle of the scale (4). Fourthly, extreme response style (ERS) describes the tendency tocheck the most extreme points of the scale (1 and 7). Harzing (2006) additionally identies two subtypes of ERS: positiveextreme response style (7) and negative extreme response style (1).

We follow the procedures suggested by Baumgartner and Steenkamp (2001) and Weijters et al. (2008) to calculateresponse style indicators. Therefore, we counted how often each respondent ticks the different points of seven-pointanswering scales (1 to 7). In order to calculate the indices for different response styles (ARS, DRS, ERS, MRS), we weightedeach of these frequencies differently. The specic weights for each response style and the complete formula for seven-point-rating scales are depicted in Table 1. We calculated the indices in such a way that each index ranges from 0 to 1, with 0indicating no response bias at all and 1 indicating a strong bias. In the next step, the inuence of response styles ispartialled out to estimate unbiased answers (Fischer, 2004; Leung & Bond, 1989; Smith, 2004).

2.2. Culture

Hofstedes (1991) software of the mind metaphor is a widely cited concept of culture. A collective level of mentalprogramming is shared by a group of people that distinguishes itself from other groups. In a related manner, the projectGlobal Leadership and Organizational Behavior Effectiveness (GLOBE) emphasizes that culture refers to a set of parametersof collectives that differentiate the collectives from each other in meaningful ways (House, Javidan, Hanges, & Dorfman,2002, p. 5). Many researchers attempt to operationalize the concept of culture through a multidimensional approach.Cultural dimensions are used to describe the differences and similarities between societies. Dimensional concepts have beensuggested, among others, by Hofstede (1991), Triandis (1988), Trompenaars (1994), Schwartz (1994) and House et al. (2004).

S. Hoffmann et al. / International Business Review 22 (2013) 814827816We build on the seminal work of Hofstede (2001), which is the most widespread approach in the international businessliterature, and on project GLOBE (House et al., 2004), which is the most comprehensive and up-to-date approach.

Hofstedes (2001) cultural dimensions are based upon social anthropological assumptions of basic problems with whichevery society is confronted. They reect the different ways people react to problems, as well as the different solutions thatmembers of a society use to solve them. Hofstedes original approach covers the following four cultural dimensions: powerdistance, individualism vs. collectivism, masculinity vs. femininity, uncertainty avoidance. Hofstedes culturalconceptualization remains one of the most cited approaches in the social sciences, and we therefore apply it in thepresent study. It should be noted, however, that the methodological approach underlying Hofstedes work has been subjectto intensive criticism (Dorfman & Howell, 1988; Yoo & Donthu, 2002). Furthermore, since the political and economicsituation has changed in recent decades, some scholars argue that Hofstedes list of cultural proles of different countriesmay even be obsolete (Gooderham & Nordhaug, 2001; McSweeney, 2002). Due to this valid criticism, we additionallyconsider the approach of project GLOBE.

The GLOBE project collected data from 17,300 managers working in 951 organizations in 62 societies (House et al., 2004).GLOBE distinguishes between nine cultural dimensions, namely performance orientation, institutional collectivism, genderegalitarianism, uncertainty avoidance, in-group collectivism, future orientation, humane orientation, assertiveness andpower distance. Project GLOBEs cultural concept has several advantages over other cultural conceptualizations. Mostimportantly, it distinguishes between cultural values and practices (Hanges & Dickson, 2004). Cultural values refer to howmembers of a society wish how things should be done (as it should be). By contrast, cultural practices capture perceptionsof the way things are currently done in a society. They measure how individuals assess present common behaviors andinstitutional practices in their country (as it is). Furthermore, project GLOBE measures culture at both societal andorganizational levels. In this study, we consider only the societal level.

2.3. Previous research on culture-dependent response styles

Only four studies have analyzed the relationships between response styles and cultural dimensions in larger multiple-country studies which enable valid cultural comparisons (Table 2). Smith (2004) focused only on acquiescence responsestyle. His investigation revealed that there is a danger of an acquiescence bias if items cover values, beliefs and attitudes, andif items are personally relevant. This response style bias is positively related to Hofstedes power distance index, to projectGLOBEs future orientation and uncertainty avoidance values, as well as to GLOBEs power distance and in-group collectivismpractices. It is negatively related to Hofstedes individualism index and to GLOBEs gender egalitarianism values.

van Herk et al. (2004) considered acquiescence and extreme response styles in six European countries (Greece, Italy,

Table 1

Formula to calculate indices of answering tendencies.

Response style Formula

Acquiescence response style (ARS) (1)

ARS f 5 1 f 6 2 f 7 3=3kDisacquiescence response style (DRS) (2)

DRS f 1 3 f 2 2 f 3 1=3kMidpoint response style (MRS) (3)

MRS f 4=kExtreme response style (ERS) (4)

ERS f 1 f 7=kNotes. Formula adjusted for seven-point rating scales ranging from 1 = I strongly disagree to 7 = I strongly agree. f(x) = total number of ticks of scale point

x of the seven-point rating scale. k = total number of items.Spain, France, Germany and the United Kingdom). Their study reveals that Greece has the highest acquiescence and extremeresponse style indices, while the United Kingdom scores lowest on these indices. This nding indicates that individualism isnegatively related to acquiescence and extreme response styles.

The research of Johnson et al. (2005) considers how acquiescence and extreme response styles are related to Hofstedesdimensions of power distance, uncertainty avoidance, masculinity and individualism. The study covers 19 countries withrespect to extreme response style measurement and 10 countries for acquiescence measurement. The investigationindicates that all four cultural dimensions negatively inuence acquiescence response style. Power distance and masculinityfoster an extreme response style.

Harzing (2006) collected data from 26 countries to investigate whether acquiescence, positive and negative extremeresponse style and midpoint response style are culture-dependent. She considered the inuence of power distance,individualism and uncertainty avoidance as measured by both Hofstede and project GLOBE. Acquiescence response style isinuenced negatively by Hofstedes individualism index and GLOBEs institutional collectivism practices. It is positivelyaffected by Hofstedes power distance index and GLOBEs in-group collectivism practices, as well as GLOBEs uncertaintyavoidance index practices and values. Negative extreme response style has a negative relationship with GLOBEs powerdistance values and with GLOBEs in-group collectivism practices. A positive extreme response style is positively correlated

S. Hoffmann et al. / International Business Review 22 (2013) 814827 817to GLOBEs in-group collectivism practices and uncertainty avoidance values. A midpoint response style is negatively relatedto Hofstedes indices of individualism and power distance.

To summarize, these studies provide empirical evidence that culture affects response styles. Notably, extant studies haveextensively analyzed cultural inuences on acquiescence response styles, while more research is needed on the otherresponse styles. Additionally, and most importantly, the extent to which culture-dependent response styles affectsubstantial relationships in cross-cultural studies, has not yet been sufciently researched.

3. Design

3.1. Objective of the study

The main aim of the present study is to assess the level of distortion that culture-dependent response styles impose onsubstantial ndings in cross-cultural research. To this end, we conducted a large-scale survey in several countries, capturingvarious cultural proles. We rstly estimate response styles using the formula presented in Table 1. Next, we analyzewhether the mean index of response styles varies across countries, and we assess the relationships between national cultural

Table 2

Overview of prior research on the relationships between culture and response styles.

Cultural dimension Conceptualization Response style

ARS ERS NERS PERS MRS

Individualism Hofstede a,b,c,d b dIn-group collectivism GLOBE practices +a,d d +dInstitutional collectivism GLOBE practices dPower distance Hofstede +a,d/c +c d

GLOBE practices +a

GLOBE values dFuture orientation GLOBE value +a

Masculinity Hofstede c +cGender egalitarianism GLOBE values aUncertainty avoidance Hofstede c

GLOBE practices +d

GLOBE values +a,d +d

Notes. ARS: acquiescence response style; MRS: midpoint response style; ERS: extreme response style; PERS: positive extreme response style; NERS:

negative extreme response style.

Findings: +: signicant positive correlation; : signicant negative correlation.Sources:

a Smith (2004).b van Herk et al. (2004).c Johnson et al. (2005).d Harzing (2006).proles and response styles. Finally, we turn to our main research question and examine the degree to which response stylesaffect ndings in cross-cultural research.

3.2. Sample

The data set for the present research includes samples from eight countries. These countries cover several culturalclusters suggested by project GLOBE (Gupta & Hanges, 2004), namely Anglo (U.S.), Latin America (Mexico), Latin Europe(France, Spain), Germanic Europe (Austria, Germany), Eastern Europe (Russia) and Southern Asia (India).

In contrast to common practice, we did not build on a convenience sample collected in specic rms or in classrooms.Instead, we trained interviewers to recruit respondents in the streets or in city centers. We checked whether the respondentsare natives of the country in which the survey was conducted. If not, they were excluded. The sample size in each country isapproximately 100, ranging from 83 in Mexico to 178 in Russia. In total, we gathered a sample of 1027 respondents. Weapplied quota stops to ensure that the eight national samples were largely equal with respect to age and sex distribution(Table 3). The mean age of the respondents is 33.1 years and 51.3 percent are female.

3.3. Scales

We developed questionnaires in the native language of each country, because Harzing et al. (2009) demonstrated thatsubjects respond to questions in their native language more accurately. In the Indian sample, we used an English version,because English is the ofcial working language in this country. All other versions were created using the translation/back-translation procedure suggested by Brislin (1970).

The interviewers handed out a questionnaire that the participants completed themselves. It took approximately 15 minto answer the questionnaire, which contained roughly 100 items. For the following reasons, we used a relatively long

compute indicators for ARS, DRS, ERS and MRS.

3.4. Validity checks

Since we need identical formats to calculate the inuence of the response tendencies, we used the same question typeand the same scaling format (seven-point rating scales) for all indicators. In order to rule out the possibility of a responseformat bias distorting the ndings, we applied the single-factor test to check for common method variance (PodsakoffMacKenzie, Lee, & Podsakoff, 2003). Given that the target countries for the scales differ across the national samples, we ranthe test for each country separately. An exploratory factor analysis, including all items measured on the rating scalesdemonstrated that one general factor explains between 9.9% and 25.6% of the variance, whereas many factors (2733)would explain between 82.7% and 87.7% of the variance. Hence, there is strong evidence that common method variance didnot inate the results.

As subjects had to answer most questions with regard to three target countries, they responded to more than a hundredindicators in total. Therefore, we explicitly ruled out the possibility that respondent fatigue may bias the results, by testingwhether the variety of answers becomes restricted in the course of answering the questionnaire. In no country, we did ndsubstantial differences in the answering variety. The standard deviation per person of the last twenty-one items (MSD = 1.62

S. Hoffmann et al. / International Business Review 22 (2013) 814827818,

,questionnaire. If a questionnaire contains only a few content-related indicators, it is difcult to discriminate betweensubstantial answers of a respondent (e.g., agreeing with a statement) and a response style (e.g., acquiescence response style).This is particularly true if the questionnaire covers only conceptually related scales for calculating response style. In largeitem sets pertaining to many different areas, by contrast, response bias can more easily be separated from substantialvariation (Harzing et al., 2009). This is due to the fact that substantial answers are more likely to diverge in differentdirections, whereas culturally dependent response styles are constantly driven in one direction across the entirequestionnaire. In order to distinguish response style from substance most effectively, we did not use indicators of culturalscales to compute response style indices, because the indices are related to culture in our analysis. For the same reason, wecalculated the indices with indicators that are independent of our focal constructs in the subsequent analysis(ethnocentrism, patriotism, cosmopolitanism). To be able to use indicators independent of these scales, we applied along questionnaire with indicators belonging to several different scales measuring very diverse aspects. We spread theindicators randomly to calculate response styles across the questionnaire.

The questionnaire covered several scales that are commonly applied in the international business literature, such asthe ethnocentrism, patriotism, and cosmopolitanism of the respondents (indicators taken from Cleveland & Laroche,2007; Kosterman & Feshbach, 1989; Shimp & Sharma, 1987), the general and product-specic country-of-origin imageof different target countries (indicators from Pappu, Quester, & Cooksey, 2007) and consumer animosity (indicatorsfrom Hoffmann, Mai, & Smirnova, 2011; Klein, Ettenson, & Morris, 1998; Rieer & Diamantopoulos, 2007). Most of thesemeasures are country-specic, with each respondent assessing three target countries. The rst target country is the U.S.(except for the U.S. sample), the second is Russia (except for the Russian sample), and the third is Germany (except forthe German and Indian samples). For the U.S., Russia and Germany, the third target country is France, and for India, it isChina.

For all indicators, we applied Likert-type rating scales because this scaling format is most widely applied in internationalbusiness research. In most previous research projects, scholars used ve-point rating scales. In this study, however, we usedseven-point scales following Harzing et al. (2009) who suggested that, on the basis of an extensive empirical examination,seven-point scales are less vulnerable to distortion. Therefore, this scaling format may be a good basis to conservatively testthe validity of cross-national comparisons.

We chose three constructs that are relevant in international business settings to determine the degree to whichsubstantial relationships are distorted by culture-dependent response styles. These constructs are ethnocentrism (5 items),patriotism (4 items), and cosmopolitanism (4 items). From the remaining set of items, we randomly selected 11 items to

Table 3

Description of the national samples.

N Age Sex

M SD Min Max % female

Austria 112 36.7 12.2 18 69 52.7

France 119 29.0 11.5 18 76 50.6

Germany 170 34.8 13.4 18 70 50.6

India 105 29.5 9.6 19 69 41.0

Mexico 83 28.3 12.0 16 57 44.6

Russia 178 35.1 13.5 18 68 58.6

Spain 148 35.9 12.8 18 71 54.7

U.S. 112 31.2 12.0 18 69 58.9

S. Hoffmann et al. / International Business Review 22 (2013) 814827 819ranging from 1.45 [Austria] to 1.99 [Russia]) is almost as high as that for the rst twenty-one answers (MSD = 1.87 rangingfrom 1.76 [India] to 2.02 [France]).

3.5. Discriminant validity and cross-national measurement invariance of focal constructs

In order to ensure the validity of the substantial test, we checked for discriminant validity (Fornell & Larcker, 1981) andcross-national measurement invariance (Steenkamp & Baumgartner, 1998) of the focal constructs ethnocentrism,patriotism, and cosmopolitanism.

A conrmatory factor analysis with AMOS 18.0 (maximum likelihood estimation) across all eight samples conrms themeasurement model (x2 = 1189.21, df = 496, p .001; x2/df = 2.40; CFI = .90; RMSEA = .04). The goodness of t-indices aresimilar for individual analyses of the subsample. The analysis conrms convergent validity. In the entire sample, as well as foreach subsample individually, each indicator signicantly (p .01) loads on the factor to which it is assigned. Furthermore,there is internal consistency, because in the sample as a whole, the average extracted variance (AVE) is well above thegenerally accepted threshold of .5, the construct reliability (CR) exceeds .6, and Cronbachs alpha (a) is higher than .7(ethnocentrism: AVE = .51, CR = .84, a = .83; patriotism: AVE = .67, CR = .88, a = .94; cosmopolitanism: AVE = .66, CR = .91,a = .89). According to the test suggested by Fornell and Larcker (1981), there is discriminant validity between the scales, sincethe squared correlations (ethnocentrism/patriotism: r2 = .20; ethnocentrism/cosmopolitanism: r2 = .09; patriotism/cosmopolitanism: r2 = .03) are lower than the average variances extracted. Analyses for each subsample also conrmedthe internal consistency and discriminant validity.

According to Steenkamp and Baumgartner (1998), this analysis also conrms the cross-cultural congural invariance ofthe constructs. Congural invariance requires that the structures of salient (zero) and non-salient (zero or near zero)loadings in the measurement models are similar across different samples. As outlined above, single-group CFAs for eachsample revealed that the specic measurement models conform to all national samples. Hence, there is full conguralinvariance.

Next, a multi-group conrmatory factor analysis tested the metric invariance of the constructs. This test requires thatfactor loadings do not differ across countries. A comparison of the fully relaxed model of the salient (non-zero) paths and amodel in which these paths are xed as equal across all eight samples, reveals a statistically signicant difference(Dx2 = 271.03; Ddf = 70; p .001). Note that the literature generally conrms that full metric invariance is scienticallyunrealistic (Byrne, Shavelson, & Muthen, 1989; Horn, McArdle, & Mason, 1983; Steenkamp & Baumgartner, 1998). To testhow strongly the measurement model diverges from full invariance, we tested 80 models. In each model, one loading wasallowed to vary freely in one of the eight samples (10 paths 8 countries). In sum, we found statistically signicantdifferences in 21 paths (26%; p .05). Steenkamp and Baumgartner (1998) argue that only highly signicant (p .01)differences should be considered. With regard to this criterion, we found only 12 deviations (15%). Additionally, there is nosystematic pattern. The relaxed paths are randomly spread across samples and indicators. Note that, additionally, all non-salient paths (zero or non-zero) are equal across the samples. Hence, there is partial metric invariance with a relatively lowlevel of variance, which allows cross-cultural comparisons.

4. Culture-dependent response styles

4.1. National differences

We rst identied answering tendencies for each study participant individually. We then calculated the means for thenational samples. As expected, the answering tendencies varied across the eight countries (Fig. 1). ANOVA reveals that thereare signicant differences across countries for the acquiescence response style (F(7, 1017) = 28.253, p .001),disacquiescence response style (F(7, 1017) = 14.584, p .001) and midpoint response style (F(7, 1017) = 8.545, p .001).The effects of ARS (h2 = 163) and DRS (h2 = .091) are moderate. The magnitude of the MRS effect is weak (MRS: h2 = .056).There are no differences in extreme response styles (F(7, 1017) = .351, n.s.). These ndings are highly consistent with priorwork on response bias. Note that we used seven-point scales, which provide more options than ve-point scales. With seven-point scales, ERS occurs less often, because respondents are able to state mild disagreement (2) or agreement (6). MRS is lesslikely, because respondents tend to use 3 and 5, instead of choosing the midpoint. In fact, previous research found astrong decrease in ERS (Clarke, 2001; Harzing et al., 2009; Hui & Triandis, 1989) and a slight decrease in MRS when largerscales are used (Harzing et al., 2009).

Next, we performed the post hoc StudentNewmanKeuls test (SNK) to identify patterns of countries that do notdiffer in their degree of response style. SNK groups country samples that do not differ statistically signicantly withrespect to a specic response tendency. In this way, we detect countries that do not need to be singled out with respectto a specic answering tendency. Survey data of countries in the same group are biased in a similar manner, so thatcomparisons of countries in the same group are less distorted. Cross-cultural research could therefore be conducted inthese countries without correcting for the specic answering tendency. It is also important that the SNK can assign acountry to multiple groups. Consequently, this country would not diverge from all other countries in the multiplegroups to which it is assigned. However, the remaining countries in these groups differ from each other, if not groupedtogether.

ARS DRS

.29 (.12)India .42 (.16)U.S.

S. Hoffmann et al. / International Business Review 22 (2013) 814827820Because the analysis detects no overall differences for ERS, we did not perform post hoc tests regarding ERS. The SNK with

ERS MRS

.15 (.08)

.15 (.07)

.18 (.08)

.20 (.09)

.23 (.08)

.23 (.13)

.24 (.11)

.10 .20 .30 .40

Germany

U.S.

Austria

Spain

France

Mexico

Russia

.26 (.11)

.27 (.15)

.28 (.18)

.31 (.17)

.32 (.16)

.34 (.12)

.37 (.18)

.10 .20 .30 .40

India

France

Mexico

Austria

Spain

Russia

Germany

.27 (.20)

.27 (.22)

.27 (.22)

.27 (.24)

.28 (.18)

.28 (.21)

.29 (.23)

.29 (.22)

.10 .20 .30 .40

India

Austria

Spain

Mexico

France

Germany

U.S.

Russia

.15 (.17)

.17 (.16)

.18 (.16)

.21 (.20)

.22 (.21)

.25 (.23)

.28 (.20)

.28 (.20)

.10 .20 .30 .40

Russia

India

U.S.

Mexico

Germany

Spain

France

Austria

Fig. 1. Response style means by countries. Notes. Standard deviation in brackets.a signicance level of a = .05 conrms that there are groups of countries that do not vary in their intensity of ARS, DRS, andMRS. With regard to ARS, there are four groups (1: Germany, U.S., Austria, p = .116; 2: Austria, Spain, p = .209; 3: France,Mexico, Russia, p = .493; 4: India).2 There are ve groups for DRS (1: India, France, Mexico, p = .479; 2: France, Mexico,Austria, Spain, p = .109; 3: Austria, Spain, Russia, p = .085; 4: Russia, Germany, p = .085; 5: U.S.). Regarding MRS, the analysisdetects three homogeneous groups (1: Russia, India, U.S., p = .445; 2: India, U.S., Mexico, Germany, p = .160; 3: Mexico,Germany, Spain, France, Austria, p = .055). While individualism/collectivism is the most relevant cultural dimension in mostcross-cultural studies, the groupings for MRS demonstrate that individualistic (e.g., U.S.) and collectivistic (e.g., Russia, India)countries are mixed up. This indicates that cultural dimensions other than individualism/collectivism are probably morerelevant to the degree of MRS. In the next section, we analyze the cultural inuence on the response styles.

4.2. Cultural imprint of response styles

We now examine the relationship between response styles and cultural dimensions. Again, we use the aggregatedresponse styles at a national level. Additionally, we apply cultural indices at a national level, according to Hofstede (2001)and GLOBE (House et al., 2004). The GLOBE indices are measured at societal, as well as organizational levels. Since ourprimary data was collected in terms of general national samples and not within specic rms, we only consider the GLOBEsocietal scores. We used GLOBE cultural values and practices. The cultural indices from Hofstede and GLOBE are based onaggregated answers. We checked the inuence of these cultural proles on response styles at a national level (aggregatedmeans). Hence, the analysis of relationships is on both sides (culture and response style) at the same level of aggregation,ensuring that there is no ecological fallacy bias. Since data stem from different sources (cultural value scores of large-scaleresearch projects vs. respondents of our survey), it is impossible that common method variance biases the analysis.

Due to the small number of countries, we performed a nonparametric bootstrapping procedure to generate relativelyrobust ndings (Davison & Hinkley, 1997; Efron & Tibshirani, 1993). This procedure does not require any assumptions aboutthe distribution of the population. We drew 1000 repeated samples from the original sample to assess condence intervalsfor the estimates of all parameters. To avoid misinterpretations due to the small sample, we interpret only relationshipswhich are statistically signicant in the bootstrapping analysis with at least a medium effect (r .50). These correlations arebold in Table 4.

2 Note that the groups of the SNK should not be interpreted as distinct clusters. Some countries may therefore belong to different groups.

S. Hoffmann et al. / International Business Review 22 (2013) 814827 821In line with extant studies, this examination reveals several cultural inuences on ARS. Additionally, the analysisdemonstrates some inuences on DRS and MRS. Obviously, the cultural background of a respondent may stimulate apreference to answer in the afrmative, to disagree, or to favor the midpoint of a scale. There is, however, no culturalinuence on the tendency to choose extreme points (ERS) independently of their valence. Given that we considered seven-point rating scales, this result is in line with previous ndings (Clarke, 2001; Harzing et al., 2009; Hui & Triandis, 1989).

As the following discussion of the most relevant results illustrates, the relationships found are plausible from a theoreticalperspective. Hofstedes power distance correlates positively with ARS and negatively with MRS. Members of societies thatexpect and accept unbalanced power tend to agree to statements, regardless of their content (Hofstede, 2001). Presumably, theyconsciously or unconsciously fear that their responses may have negative consequences for them. To express submissiveness to

Table 4

Correlations between response styles and cultural dimensions at a national level.

Concept Cultural dimension ARS DRS ERS MRS

GLOBE practices Assertiveness .877 .523 .078 .609Institutional collectivism .333 .184 .265 .060In-group collectivism .810 .631 .372 .445Future orientation .489 .258 .216 .409Gender egalitarism .130 .006 .593 .012Humane orientation .508 .348 .141 .287Performance orientation .549 .179 .156 .699Power distance .274 .009 .483 .236Uncertainty avoidance .550 .292 .030 .559

GLOBE values Assertiveness .249 .270 .429 .167Institutional collectivism .122 .169 .307 .493In-group collectivism .061 .029 .289 .033Future orientation .522 .452 .291 .351Gender egalitarism .934 .674 .119 .409Humane orientation .398 .259 .426 .476Performance orientation .686 .433 .089 .525Power distance .009 .045 .469 .154

Uncertainty avoidance .716 .578 .129 .325

Hofstede Power distance .784 .458 .116 .544Individualism .612 .514 .392 .296Masculinity .402 .190 .289 .291Uncertainty avoidance .019 .169 .005 .332

Notes. Marked bold if meaningful relationship according to the following criteria: r .5 and statistically signicant (p .05) on basis of bootstrapping with1000 samples.their superiors, they avoid contradicting them, and thus display acquiescence. By contrast, assertive societies encourage theirmembers to be competitive and to demonstrate tough behavior. Hence, such people are more likely to disagree withstatements in surveys. Accordingly, GLOBE assertiveness practices correlate negatively with ARS and positively with DRS.

GLOBE in-group collectivism practices are positively related to ARS and negatively related to DRS. In-group collectivismdescribes the extent to which individuals express their loyalty toward other members of their family and peers. The behaviorof these societies members is more likely to preserve harmony and not to contradict (Gelfand, Bhawuk, Nishii, & Bechtold,2004). The norms of the group are more important than individual opinions. Hence, they are prone to an acquiescenceresponse style. Conversely, the more people are independent, the less they feel a need to conform and to afrm statements.Likewise, individualism measured by Hofstede (=which is the inverse concept of collectivism) is negatively correlated withARS and positively related to DRS.

Performance orientation is negatively related to ARS and positively to MRS. Respondents from societies that stronglyencourage innovation and performance (e.g., Germany, the U.S. and Austria) do not tend to answer in the afrmative,regardless of their content. They prefer to think themselves rather than simply agreeing with a given statement. GLOBEfuture orientation values are positively correlated with ARS. Future orientation describes how members of societies think inthe future. Presumably, living mentally in the future allows people to be more afrmative in the present. Additionally,GLOBE gender egalitarianism values are related to response styles. Societies that do not accept male dominance tend to adopta disacquiescence response style and avoid acquiescence. Members of these societies are more active, vocal and open tochange (Emrich, Denmark, & Den Hartog, 2004). They are skeptical and question the status quo. Accordingly, they rejectgiven statements more easily than individuals from countries with lower gender egalitarianism scores.

GLOBE uncertainty avoidance values are positively related to ARS and negatively related to DRS. Societies with highuncertainty avoidance scores are characterized by a high level of anxiety that reects the individual needs to avoiduncertainty through explicit rules (Sully de Luque & Javidan, 2004). Hence, members of these societies conrm rather thancontradict statements. Since most questionnaires based on rating scales provide no space to explain contradictory responses,this creates an ambiguous situation which these individuals tend to avoid. Remarkably, GLOBE uncertainty avoidancepractices are negatively correlated with ARS. Extant research also provides evidence of a positive (Harzing, 2006; Smith,2004), as well as a negative relationship (Johnson et al., 2005) between uncertainty avoidance and ARS. Generally, the GLOBE

study also revealed several negative relationships between values and practices (House et al., 2004). From a theoreticalperspective, the seemingly contradictory ndings can be solved by Javidan, Dorfman, Sully de Luque, and Houses (2006)deprivation hypothesis. In societies that are already performing the desired action (high practice score), the increment issmaller than that of societies with lower practice scores. Since the latter strive strongly toward improvement (very highvalue score), values and practices are negatively related.

5. Impact of response styles on substantial relationships

So far, we have shown that the culture-dependent response styles are largely consistent with previous ndings and in linewith theoretical reasoning. On this robust basis, we now answer our main research question, namely the degree to whichresponse styles distort substantial ndings of cross-cultural studies. We explore the impact on the ndings of correlation

S. Hoffmann et al. / International Business Review 22 (2013) 814827822studies, as well as mean comparisons across countries. To this end, we consider three pivotal constructs of cross-culturalresearch: ethnocentrism, patriotism, and cosmopolitanism.

To control for response styles, we partial out the inuence of the response tendencies from these constructs. Therefore,we rst run OLS regression analysis that regressed ethnocentrism, patriotism, and cosmopolitanism (=dependent variables)respectively on response style indices (=independent variables), yielding the following coefcients: ethnocentrism (singleregressions: ARS: b = .048, n.s.; DRS: b = .171, p .001; MRS: b = .085, p .01; multiple regression: R = .173), patriotism(single regressions: ARS: b = .062, p .05; DRS: b = .059, p .10; MRS: b = .063, p .05; multiple regression: R = .143), andcosmopolitanism (single regressions: ARS: b = .210, p .001; DRS: b = .003, n.s.; MRS: b = .014, n.s.; multiple regression:R = .291). We use the residuals of the single linear regressions to estimate scales corrected by ARS, DRS, and MRS,respectively. We also run a multiple regression that includes ARS, DRS, and MRS simultaneously to produce data correctedjointly for the three response styles.

5.1. Correlation analysis for the whole sample

First, we consider the effects of response styles for the entire sample without distinguishing between different nations.The ndings of this analysis are relevant for researchers who gather data in several countries and then treat them as one dataset. As a baseline, we examine the rst-order correlations between ethnocentrism, patriotism, and cosmopolitanism withoutpartialling out response styles. As expected, ethnocentrism and patriotism are positively correlated. By contrast, there is anegative relationship between cosmopolitanism and the other two constructs (Table 5).

Rerunning the correlation analysis with the corrected scales reveals no relevant inuence of the response-styles. Thevalence of the correlations never changes and the magnitude of the difference is minimal (ranging von Dr = .000 to Dr = .017).In conclusion, our analysis of the whole sample highlights that response styles though dependent on culture do notdramatically affect correlations between focal constructs.

5.2. Correlation analysis at the national level

Next, we explore the impact of response styles on substantial relationships separately for each nation. Hence, we consideran issue that is relevant when country serves as a moderating variable and the strength of relationships is compared acrosscountries. Table 6 displays the differences between unadjusted and adjusted correlations for each pair of variables, for eachcountry, and for each type of correction. Additionally, Table 7 displays the mean differences between unadjusted andadjusted correlations for each country, and for each type of correction. Obviously, ARS and DRS cause the most pronounceddistortions, whereas the inuence of MRS is generally very low. Hence, we advise scholars to control at least for (dis-)acquiescence response styles before running substantial analyses.

Response style corrections are vital for survey research, if the values of adjusted and unadjusted parameters differmarkedly. In the following sections, we consider a change of Dr .025 as relevant. This is a rather arbitrary threshold whichwe selected as a conservative threshold that classies even small changes as important. Distortions are even more evident ifthe valance of the relationship changes (positive vs. negative sign). We describe and give examples of the most problematicresponse styles for each nation separately.

Table 5

Correlation analysis of unadjusted and adjusted constructs.

Not adjusted Adjusted

ARS DRS MRS ARS/DRS/MRS

r Dr r Dr r Dr r Dr

Ethnocentrism/patriotism .393 .392 .001 .390 .003 .401 .008 .390 .003

Ethnocentrism/cosmopolitanism .245 .261 .016 .254 .009 .247 .002 .258 .013Patriotism/cosmopolitanism .126 .143 .017 .122 .004 .126 .000 .120 .006

Notes. r = Pearson-Product-Moment-Correlation, Dr = difference to unadjusted r.

Table 6

Correlation analysis at a national level.

Not adjusted Adjusted

ARS DRS MRS ARS/DRS/MRS

r Dr r Dr r Dr r Dr

Austria Ethno/Patr .315** .323** .008 .334** .019 .306* .009 .327** .012

Ethno/Cosmo .192* .196* .004 .248** .056 .201* .009 .228* .036Patr/Cosmo .149 .150 .001 .174 .025 .141 .008 .155 .006

France Ethno/Patr .156 .154 .002 .151 .005 .171 .015 .162 .006

Ethno/Cosmo .137 .131 .006 .185* .048 .146 .009 .183* .046Patr/Cosmo .083 .131 .048 .052 .031 .076 .007 .077 .006

Germany Ethno/Patr .099 .097 .002 .096 .003 .112 .013 .091 .008

Ethno/Cosmo .224** .240** .016 .208** .016 .222* .002 .213** .011Patr/Cosmo .135 .180* .045 .146 .011 .140 .005 .156* .021

India Ethno/Patr .271** .248* .023 .262** .009 .263* .008 .253** .018

Ethno/Cosmo .035 .111 .076 .062 .027 .029 .006 .104 .069Patr/Cosmo .381** .359** .022 .378** .003 .372* .009 .353** .028

Mexico Ethno/Patr .548** .555** .007 .543** .005 .522* .026 .545** .003

Ethno/Cosmo .024 .025 .001 .048 .072 .017 .007 .029 .053Patr/Cosmo .207 .252* .045 .205 .002 .209 .002 .243* .036

Russia Ethno/Patr .494** .499** .005 .495** .001 .497* .003 .499** .005

Ethno/Cosmo .170* .162* .008 .150* .020 .168* .002 .137 .033Patr/Cosmo .128 .155 .027 .092 .036 .168* .040 .069 .059

Spain Ethno/Patr .398** .407** .009 .423** .025 .412* .014 .433** .035

Ethno/Cosmo .552** .546** .006 .561** .009 .552* .000 .567** .015Patr/Cosmo .302** .265** .037 .314** .012 .309* .007 .285** .017

U.S. Ethno/Patr .250** .254** .004 .257** .007 .263* .013 .243** .007

Ethno/Cosmo .176 .161 .015 .124 .052 .175 .001 .087 .089Patr/Cosmo .167 .177 .010 .142 .025 .166 .001 .125 .042

Difference to not adjusted Min .001 .001 .000 .003

Max .076 .072 .040 .089

Mean .018 .022 .009 .028

SD .019 .019 .009 .023

Notes. Ethno = ethnocentrism, Patr = patriotism, Cosmo = cosmopolitanism, r = Pearson-Product-Moment-Correlation, Dr = difference to unadjusted r.Level of signicance:

* p .05.** p .01.

S. H

offm

ann

et al.

/ In

ternatio

nal

Busin

ess Review

22

(2013)

814827

823

S. Hoffmann et al. / International Business Review 22 (2013) 814827824Within the U.S. sample, there is the strongest deviation between unadjusted and the adjusted correlations. Aftercorrecting for ARS, DRS, and MRS conjointly, there is a mean difference of Dr = .046. The strongest inuence is due to DRS(mean Dr = .028). Although not statistically signicant, the magnitude of the correlation between ethnocentrism andcosmopolitanism changes by Dr = .089.

In the Indian sample, there are also deviations between the unadjusted and adjusted parameters (mean Dr = .038 whencorrecting for ARS, DRS, and MRS conjointly). In particular, ARS distorts the ndings (mean Dr = .040). Again, the strongestdeviation is in the relationship between ethnocentrism and cosmopolitanism, which changes by Dr = .069.

In the Russian data, there is a weak inuence of response styles, if DRS, ARS, and MRS are considered jointly (meanDr = .033). The most relevant effects (ARS-corrected: Dr = .027; DRS-corrected: Dr = .036; MRS-corrected: Dr = .040) are onthe relationship between patriotism and cosmopolitanism. Apparently, response bias causes an overestimation ofcorrelation coefcients for the Russian respondents (particularly for the construct cosmopolitism). For example, thesignicant relationship between cosmopolitanism and ethnocentrism (r = .170) becomes insignicant when correcting forall response styles (r = .137).

The Mexican survey data are also distorted by the joint inuence of ARS, DRS, and MRS (mean Dr = .031), with DRS exertingthe most important inuence (mean Dr = .026). The Mexican subsample provides evidence that neglecting response stylecorrection can also lead to an underestimation of important relationships. For instance, the unadjusted correlation betweenpatriotism and cosmopolitanism is insignicant. This relationship becomes signicant when adjusting for ARS (r = .252) andfor all response styles (r = .243).

In Spain, there is a relatively low inuence of the corrections (mean Dr = .022). Yet, some relationships are affected byresponse styles. For example, the unadjusted correlation between patriotism and cosmopolitanism (r = .302) differs fromthe correlations which are adjusted by ARS (r = .265, Dr = .037).

In France, the most severe bias is DRS (mean Dr = .028). The unadjusted relationship between ethnocentrism andcosmopolitanism is not signicant (r = .137). However, it becomes signicant when adjusting for DRS (r = .185, Dr = .048)and for all response styles (r = .183, Dr = .046).

In the Austrian sample as well, DRS causes the most severe bias (mean Dr = .033). For example, the unadjusted correlationof ethnocentrism and cosmopolitanism is r = .192, while the DRS-corrected correlation is markedly stronger (r = .248).Hence, researchers need to correct data collected in Austria for a disacquiescence response style, before running substantialanalysis.

Finally, response styles barely distort ndings that are based on the German data set. Nonetheless, in some cases, ARSmight cause bias. The analysis reveals that some correlations change after having controlled for ARS. The unadjustedrelationship between patriotism and cosmopolitanism (r = .135) is not statistically signicant, while the ARS-adjusted one

Table 7

Mean differences between adjusted and unadjusted correlations at a national level.

ARS Dr DRS Dr MRS Dr ARS/DRS/MRS Dr

U.S. .010 .028 .005 .046India .040 .013 .008 .038Russia .014 .019 .015 .033Mexico .018 .026 .012 .031Spain .017 .015 .007 .022

France .019 .028 .010 .019Austria .004 .033 .009 .018Germany .021 .010 .007 .013

Note. The most relevant differences are displayed in bold gures (Dr .25).is both stronger and signicant (r = .180, Dr = .045).

5.3. Impact on mean comparisons

Finally, we check whether response styles affect the results of mean comparisons across countries. We consider pair-wise mean differences (Table 8). We rst look at the mean difference based on the original data and consider whether thisdifference is signicant. We then look at the mean difference of the corrected scales. Due to space limitations, we reportonly the scales corrected simultaneously for ARS, DRS and MRS. However, the results are similar when taking into accountone response style indicator individually. We ran 74 comparisons including all combinations of the three focal constructsand all combinations of pair-wise country comparisons. In total, we found that the analyses turned from statisticallyinsignicant to signicant (a < .05) in only three comparisons when correcting for the response styles. A signicantcomparison never became insignicant when controlling for answering tendencies. In only four cases did the sign of themean comparison change meaning. However, in all of these cases, the difference was not statistically signicant for eitherthe original or for the corrected scales. All in all, we found that response styles have only a small inuence on the results ofmean comparisons.

Table 8

Inuence of culture-dependent response styles of mean comparisons.

Country Country Ethnocentrism Patriotism Cosmopolitanism

#1 #2 Not adjusted Adjusted Change Not adjusted Adjusted Change Not adjusted Adjusted Change

DM p DM p val p DM p DM p val p DM p DM p val p

Austria France 1.006 .000 1.004 .000 .976 .000 1.008 .000 1.121 .000 1.004 .000 Austria Germany .981 .000 .946 .000 1.084 .000 1.078 .000 .249 .029 .327 .003 Austria India .097 .581 .031 .861 .676 .000 .505 .005 .475 .000 .349 .004 Austria Mexico .008 .965 .032 .866 Yes .444 .021 .338 .079 .434 .001 .470 .000 Austria Russia 1.071 .000 .993 .000 .555 .001 .647 .000 .217 .055 .140 .190 Austria Spain .818 .000 .797 .000 .370 .026 .397 .017 .648 .000 .634 .000 Austria U.S. 1.493 .000 1.433 .000 .670 .000 .644 .000 .958 .000 1.019 .000 France Germany .024 .875 .059 .704 .107 .499 .070 .659 .872 .000 .677 .000 France India .908 .000 .974 .000 1.652 .000 1.514 .000 .646 .000 .655 .000 France Mexico .997 .000 1.036 .000 1.421 .000 1.346 .000 1.555 .000 1.474 .000 France Russia .066 .668 .011 .942 Yes .421 .008 .362 .022 .904 .000 .864 .000 France Spain .188 .238 .208 .192 .606 .000 .611 .000 .473 .000 .370 .001 France U.S. .487 .004 .428 .012 .307 .080 .364 .038 Yes .163 .185 .015 .898 Yes Germany India .884 .000 .915 .000 1.760 .000 1.584 .000 .226 .052 .022 .840 Germany Mexico .973 .000 .977 .000 1.528 .000 1.416 .000 .683 .000 .797 .000 Germany Russia .090 .516 .048 .731 .529 .000 .432 .002 .032 .752 .187 .050 YesGermany Spain .164 .261 .149 .305 .714 .000 .681 .000 .399 .000 .307 .002 Germany U.S. .512 .001 .487 .002 .414 .010 .434 .007 .709 .000 .692 .000 India Mexico .089 .640 .062 .743 .231 .236 .167 .391 .909 .000 .820 .000 India Russia .974 .000 .962 .000 1.231 .000 1.152 .000 .258 .025 .209 .056

India Spain .720 .000 .766 .000 1.046 .000 .902 .000 .173 .150 .284 .012 YesIndia U.S. 1.396 .000 1.402 .000 1.345 .000 1.150 .000 .482 .000 .670 .000 Mexico Russia 1.063 .000 1.025 .000 .999 .000 .985 .000 .651 .000 .610 .000 Mexico Spain .809 .000 .828 .000 .814 .000 .735 .000 1.082 .000 1.104 .000 Mexico U.S. 1.485 .000 1.464 .000 1.114 .000 .982 .000 1.391 .000 1.489 .000 Russia Spain .254 .079 .197 .172 .185 .211 .250 .092 .431 .000 .494 .000 Russia U.S. .421 .007 .439 .005 .115 .474 .002 .989 Yes .741 .000 .879 .000 Spain U.S. .675 .000 .636 .000 .300 .072 .247 .138 .310 .008 .385 .001

Notes. DM = differences of means between the country named rst and the country named second. Adjusted = inuences of the response styles ARS, DRS, and MRS are partialled out. val = valence! sign (positive/negative) of DM between adjusted and unadjusted scores. p = signicance of DM between adjusted and unadjusted scores (threshold of p = .05).

S. H

offm

ann

et al.

/ In

ternatio

nal

Busin

ess Review

22

(2013)

814827

825

S. Hoffmann et al. / International Business Review 22 (2013) 8148278266. Conclusions

Many cross-cultural business studies still do not control for response styles. However, there are repeated claims in theliterature that researchers should do so. Since extant studies stress that culture affects response styles, it is reasonable toconclude that study ndings might be distorted by culture-dependent response styles. Our ndings contradict thisconventional wisdom. On the basis of broad empirical ndings, we reveal that even though the effect of culture on responsestyles is signicant, the inuence on substantial relationships is less severe than commonly expected. Although there areculture-dependent response styles, the effects on mean comparisons and correlation analyses are relatively small. Therefore,the present paper makes a methodological contribution to the literature, with compelling implications for research practiceand the empirical testing of theories.

The present study demonstrates that there are remarkable differences in response styles across different nations. Thesedifferences are at least partly due to the cultural orientation of the various national groups. Our ndings about the culturaldependency of response styles are largely consistent with theory and empirical results of previous studies. Building on thesestable ndings, the present paper primarily strives to answer the question of whether country-specic differences inresponse styles affect the substantial ndings of cross-national research. Notably, the extent of distortion in correlativestudies is less severe than expected, and our analysis suggests that response styles are only a minor methodological threat.Distortions in mean comparisons are even less severe. Nonetheless, some examinations reveal differences between theunadjusted and the adjusted scales. Hence, although in most circumstances, culture-dependent response styles seem tocause no major distortion, scholars would still be well advised to control for response-style biases before performing furtherstatistical analysis. This investigation demonstrates that, in some cases, omitting to correct answering tendencies can lead tothe overestimation of substantial relationships in some cultures, whereas they are underestimated in others. The degree towhich survey data are distorted depends on the country sample and its specic tendency toward response bias. Hence,although the likelihood of severe distortion is generally low, there is no guarantee of drawing valid cross-culturalconclusions, without controlling for response styles. In line with previous research, the most important response style bias is(dis-)acquiescence. We therefore recommend checking whether or not ARS and DRS distort the ndings in any national andcross-national study. ARS is particularly strong in collectivistic countries with a high level of power distance and low level ofassertiveness. If sample countries t this prole, a correction for ARS is essential. DRS is strong in cultures with the oppositecultural prole. In these cultures, we suggest controlling for DRS.

In order to be able to control for response styles, we propose that international business researchers should include itemsin their questionnaires that are conceptually unrelated to the object of investigation. These items may be used to calculateresponse style indicators that are independent of their focal construct. In this way, they can disentangle substantial effectsand response bias.

7. Further research

Like all empirical research, our study is limited in some respects. The nations we surveyed cover six cultural clusterssuggested by project GLOBE, including the Anglo countries, Latin America, Latin Europe, Germanic Europe, Eastern Europe,and the Southern Asia cluster. Although our approach is wider than that of previous studies, not all cultural clusters areincluded. Note however, that the objective of this study is not to assess the characteristic response styles of certain regions,but to relate response styles to cultural dimensions. Even if our study does not cover a specic region, international businessresearchers and practitioners can conclude, on basis of the reported relationships (Table 4) and the cultural prole of thefocal region (provided by Hofstede or GLOBE), what prevalent response styles are to be expected in a given region.Nonetheless, to provide more valid conclusions and to establish external validity, we call for replications and extensions ofour ndings in the cultural clusters not yet considered.

Furthermore, we used seven-point scales, because they provide more options than those with less differentiation. Extantstudies reveal that the number of scaling points affects the magnitude of various response styles (Clarke, 2001; Harzing et al.,2009; Hui & Triandis, 1989). Hence, we recommend future research that replicates our study with other scaling formats. Inparticular, analysis with ve-point-scales is needed since they are often applied in survey research. ERS and MRS might bemore relevant in this context.

Finally, this study analyzed the relationship between culture and response style at a national level, using aggregatedresponse styles (means) for the national subsamples. However, within any given culture, there might be intra-cultural andinter-individual variations. For this reason, we suggest replicating our ndings using scales for individual culture, such as theCVSCALE (Yoo, Donthu, & Lenartowicz, 2011) or independent and interdependent self-construals (Singelis, 1994). Moreover,the interaction effect of national culture at an aggregated level with individual characteristics (e.g., personality) on responsestyles could be analyzed using multi-level approaches (e.g., Raudenbush & Bryk, 2001).

References

Baumgartner, H., & Steenkamp, J.-B.E.M. (2001). Response styles in marketing research: A cross-national investigation. Journal of Marketing Research, 38(2), 143156.

Brislin, R. W. (1970). Back-translation for cross-cultural research. Journal of Cross-Cultural Psychology, 1(3), 185216.

S. Hoffmann et al. / International Business Review 22 (2013) 814827 827Byrne, B. M., Shavelson, R. J., & Muthen, B. (1989). Testing for the equivalence of factor covariance and mean structures: The issue of partial measurementinvariance. Psychological Bulletin, 105, 456466.

Clarke, I., III (2001). Extreme response style in cross-cultural research. International Marketing Review, 18(3), 301324.Cleveland, M., & Laroche, M. (2007). Acculturation to the global consumer culture: Scale development and research paradigm. Journal of Business Research, 60(3),

249259.Davison, A. C., & Hinkley, D. V. (1997). Bootstrap methods and their application. Cambridge: Cambridge University Press.Dorfman, P. W., & Howell, J. P. (1988). Dimensions of national culture and effective leadership patterns: Hofstede revisited. Advances in International Comparative

Management, 3, 127150.Efron, B., & Tibshirani, R. J. (1993). An introduction to the bootstrap. New York: Chapman Hall.Emrich, C. G., Denmark, F. L., & Den Hartog, D. N. (2004). Cross-cultural differences in gender egalitarianism: Implications for societies, organizations, and leaders.

In R. J. House, P. J. Hanges, M. Javidan, P. W. Dorfman, & V. Gupta (Eds.), Culture, leadership, and organizations: The GLOBE Study of 62 Societies (pp. 343394).Thousand Oaks: Sage.

Fischer, R. (2004). Standardization to account for cross-cultural response bias: A classication of score adjustment procedures and review of research in JCCP.Journal of Cross-Cultural Psychology, 35(3), 263282.

Fornell, C., & Larcker, D. F. (1981). Evaluating structural equation models with unobservable variables and measurement error. Journal of Marketing Research, 18(1),3950.

Gelfand, M. J., Bhawuk, D. P. S., Nishii, L. H., & Bechtold, D. J. (2004). Individualism and collectivism. In R. J. House, P. J. Hanges, M. Javidan, P. W. Dorfman, & V. Gupta(Eds.), Culture, leadership, and organizations: The GLOBE Study of 62 Societies (pp. 437512). Thousand Oaks: Sage.

Gooderham, P., & Nordhaug, O. (2001). Are cultural differences in Europe on the decline? European Business Forum, 8, 4853.Gupta, V., & Hanges, P. J. (2004). Regional and climate clustering of societal cultures. In R. J. House, P. J. Hanges, M. Javidan, P. W. Dorfman, & V. Gupta (Eds.), Culture,

leadership, and organizations: The GLOBE Study of 62 Societies (pp. 178218). Thousand Oaks: Sage.Hanges, P. J., & Dickson, M. W. (2004). The development and validation of the GLOBE culture and leadership scales. In R. J. House, P. J. Hanges, M. Javidan, P. W.

Dorfman, & V. Gupta (Eds.), Culture, leadership, and organizations: The GLOBE Study of 62 Societies (pp. 122151). Thousand Oaks: Sage.Harley, J. R. (1998). Timidity as a response style to psychological questionnaire. Journal of Psychology: Interdisciplinary and Applied, 32(2), 201210.Harzing, A.-W. (2006). Response style in cross-national survey research: A 26-country study. International Journal of Cross-Cultural Management, 6(2), 243266.Harzing, A.-W., Baldueza, J., Barner-Rasmussen, W., Barzantny, C., Canabal, A., Davila, A., et al. (2009). Rating versus ranking: What is the best way to reduce

response and language bias in cross-national research? International Business Review, 18, 417432.Hoffmann, S., Mai, R., & Smirnova, M. (2011). Development and validation of a cross-nationally stable scale of consumer animosity. Journal of Marketing Theory and

Practice, 19(2), 235252.Hofstede, G. (1991). Cultures and organisations. Software of the mind. London: McGraw-Hill.Hofstede, G. (2001). Cultures consequences: Comparing values, behaviors, institutions and organizations across nations (2nd ed.). Thousand Oaks: Sage.Horn, J. L., McArdle, J., & Mason, R. (1983). When is invariance not invariant: A practical scientists look at the ethereal concept of factor invariance. The Southern

Psychologist, 1, 179188.House, R. J., Hanges, P. J., Javidan, M., Dorfman, P. W., & Gupta, V. (2004). Culture, leadership, and organizations: The Globe Study of 62 Societies. Thousand Oaks: Sage.House, R. J., Javidan, M., Hanges, P. J., & Dorfman, P. Q. (2002). Understanding cultures and implicit leadership theories across the globe: An introduction to project

GLOBE. Journal of World Business, 37, 310.Hui, C. H., & Triandis, H. C. (1989). Effects of culture and response format on extreme response style. Journal of Cross-cultural Psychology, 20(3), 296309.Javidan, M., Dorfman, P., Sully de Luque, M., & House, R. J. (2006). In the eye of the beholder: Cross cultural lessons in leadership from project GLOBE. Academy of

Management Perspective, 20, 6790.Johnson, T., Kulesa, P., Cho, Y. I., & Shavitt, S. (2005). The relation between culture and response styles: Evidence from 19 countries. Journal of Cross-Cultural

Psychology, 36(2), 264277.Klein, J. G., Ettenson, R., & Morris, M. D. (1998). The animosity model of foreign product purchase: An empirical test in the Peoples Republic of China. Journal of

Marketing, 62(1), 89100.Kosterman, R., & Feshbach, S. (1989). Toward a measure of patriotic and nationalistic attitudes. Political Psychology, 10(2), 257274.Leung, K., & Bond, M. H. (1989). On the empirical investigation of dimensions for cross-cultural comparisons. Journal of Cross-Cultural Psychology, 20(2), 296309.McSweeney, B. (2002). Hofstedes model of national differences and their consequences: A triumph of faith a failure of analysis. Human Relations, 55(1), 89118.Nakata, C., & Hung, Y. (2005). Progress and promise: The last decade of international marketing research. Journal of Business Research, 58(7), 611618.Okazaki, S., & Mueller, B. (2007). Cross-cultural advertising research: Where we have been and where we need to go. International Marketing Review, 24(5), 499518.Okazaki, S., Taylor, C. R., & Doh, J. P. (2007). Market convergence and advertising standardization in the European Union. Journal of World Business, 42(4), 384400.Pappu, R., Quester, P. G., & Cooksey, R. W. (2007). Country image and consumer-based brand equity: Relationships and implications for international marketing.

Journal of International Business Studies, 38(5), 726745.Podsakoff, P. M., MacKenzie, S. B., Lee, J.-Y., & Podsakoff, N. P. (2003). Common method biases in behavioral research: A critical review of the literature and

recommended remedies. Journal of Applied Psychology, 88(5), 879903.Raudenbush, S. W., & Bryk, A. S. (2001). Hierarchical linear models: Applications and data analysis methods. Newbury Park: Sage.Rieer, P., & Diamantopoulos, A. (2007). Consumer animosity: A literature review and a reconsideration of its measurement. International Marketing Review, 24(1),

87119.Schwartz, S. H. (1994). Beyond individualism/collectivism: New cultural dimensions of value. In U. Kim, H. C. Triandies, C. Kagitcibasi, S. C. Choi, & G. Yoon (Eds.),

Individualism and collectivism: Theory, method, and applications (pp. 85115). Thousand Oaks: Sage.Shimp, T. A., & Sharma, S. (1987). Consumer ethnocentrism: Construction and validation of the CETSCALE. Journal of Marketing Research, 24(3), 280289.Singelis, T. M. (1994). The measurement of independent and interdependent self-construals. Personality and Social Psychology Bulletin, 20(5), 580591.Smith, P. B. (2004). Acquiescent response bias as an aspect of cultural communication style. Journal of Cross-Cultural Psychology, 35(1), 5061.Steenkamp, J.-B.E.M., & Baumgartner, H. (1998). Assessing measurement invariance in cross-national consumer research. Journal of Consumer Research, 25(1), 7890.Sully de Luque, M., & Javidan, M. (2004). Uncertainty avoidance. In R. J. House, P. J. Hanges, M. Javidan, P. W. Dorfman, & V. Gupta (Eds.), Culture, leadership, and

organizations: The GLOBE Study of 62 Societies (pp. 602653). Thousand Oaks: Sage.Triandis, H. C. (1988). Collectivism vs. individualism: A reconceptualization of the basic concept in cross-cultural psychology. In G. Verma & C. Bagley (Eds.), Cross-

cultural studies of personality, attitudes and cognition (pp. 6095). London: MacMillan.Trompenaars, F. (1994). Riding the waves of culture: Understanding cultural diversity in business. London: Economist Books.Tsui, A. S., Nifadkar, S. S., & Ou, A. Y. (2007). Cross-national, cross-cultural organizational behavior research: Advances, gap and recommendations. Journal of

Management, 33, 426478.van de Vijver, F. J. R., & Leung, K. (2000). Methodological issues in psychological research on culture. Journal of Cross-Cultural Psychology, 31, 3351.van Herk, H., Poortinga, Y. P., & Verhallen, T. M. M. (2004). Response style in rating scales. Evidence of method bias in data from six EU countries. Journal of Cross-

Cultural Psychology, 35(3), 346360.Weijters, P., Geuens, M., & Schillewaert, N. (2008). Assessing response styles across modes of data collection. Journal of the Academy of Marketing Science, 36(3),

409422.Yoo, B., & Donthu, N. (2002). A review on Geert Hofstedes book titled cultures consequences: Comparing values, behaviors, institutions and organization across

nations, 2nd ed.. Journal of Marketing Research, 39(3), 388389.Yoo, B., Donthu, N., & Lenartowicz, T. (2011). Measuring Hofstedes ve dimensions of cultural values at the individual level: Development and validation of

CVSCALE. Journal of International Consumer Marketing, 23(34), 193210.

Do culture-dependent response styles distort substantial relationships?IntroductionConceptual backgroundResponse stylesCulturePrevious research on culture-dependent response styles

DesignObjective of the studySampleScalesValidity checksDiscriminant validity and cross-national measurement invariance of focal constructs

Culture-dependent response stylesNational differencesCultural imprint of response styles

Impact of response styles on substantial relationshipsCorrelation analysis for the whole sampleCorrelation analysis at the national levelImpact on mean comparisons

ConclusionsFurther researchReferences