Ai

Ha

1b

a

KHBBMEC

1

daodgvcStcFcrctBB

lA

P

c

0h

International Journal of Hospitality Management 36 (2014) 156–166

Contents lists available at ScienceDirect

International Journal of Hospitality Management

jou rn al hom ep age: www.elsev ier .com/ locate / i jhosman

ssessing equivalence of hotel brand equity measuresn cross-cultural contexts�

aemoon Oha,∗, Cathy H.C. Hsub,1

Department of Hospitality and Tourism Management, Isenberg School of Management, University of Massachusetts-Amherst, 90 Campus Center Way Flint07, Amherst, MA 01003, United StatesSchool of Hotel and Tourism Management, The Hong Kong Polytechnic University, 17 Science Museum Rd., TST East, Kowloon, Hong Kong

r t i c l e i n f o

eywords:otel brand

a b s t r a c t

The authors synthesize the measurement equivalence or invariance literature and illustrate how to con-

rand equityrand choiceeasurement invariance

quivalenceross-cultural

duct equivalence analyses by using a hotel brand equity model as an example. The illustration focuseson how to assess the model’s generalizability across three selected cultural or cross-country factors: thehotel’s brand identity (domestic vs. foreign), the customer’s first language (Mandarin vs. English), andthe customer’s country of residence (Asia vs. Europe vs. North America). Results support the model’sconfigural and metric generalizability across the three cross-cultural contexts. The authors show how to

uival

interpret the results of eq

. Introduction

Measuring brand equity (BE) in the hotel industry oftenemands cross-national or cross-cultural research design andnalyses. Not only is the nature of the hotel business global inperations, but the business also constantly, and increasingly,eals with customers from diverse national or cultural back-rounds. Such diversity in background becomes frequent sources ofariance in customer perceptions and behaviors, also causing con-erns in customer-based measurement of hotel BE (Motameni andhahrokhi, 1998). Consequently, researchers face numerous ques-ions associated with whether a model developed in one culturalontext will work in another (Steenkamp and Baumgartner, 1998).or example, would a model structure remain consistent acrossultural groups or segments of customers? Is the strength of theo-etical relationships among the model constructs equivalent acrossultural contexts? Understanding variances attributable to contex-ual differences like these will advance theoretical knowledge onE measurement as well as managerial decisions on building globalE strategies (Kish et al., 2001; Yoo and Donthu, 2002).

The BE literature, especially of hospitality and tourism, generallyacks research efforts to address potential cross-cultural variations.lthough researchers have proposed models for measuring hotel BE

� The work described in this paper was supported by a grant from The Hong Kongolytechnic University (Project no. 8-ZH74).∗ Corresponding author. Tel.: +1 413 545 2061; fax: +1 413 545 3765.

E-mail addresses: oh@isenberg.umass.edu (H. Oh),athy.hsu@polyu.edu.hk (C.H.C. Hsu).

1 Tel.: +852 3400 2323; fax: +852 2362 9362.

278-4319/$ – see front matter © 2013 Elsevier Ltd. All rights reserved.ttp://dx.doi.org/10.1016/j.ijhm.2013.09.002

ence analyses and discuss a few related methodological issues.© 2013 Elsevier Ltd. All rights reserved.

(e.g., Bailey and Ball, 2006; Hsu et al., 2012; Kayaman and Arasli,2007; Prasad and Dev, 2000; So and King, 2010; Xu and Chan, 2010),few have assessed their models for equivalence or generalizabilityacross the cultural backgrounds of customers. While measuring andtracking hotel BE has a number of significant reasons, such as under-standing customer feedback, the hotel’s competitive position, andthe impact of marketing mix (Prasad and Dev, 2000), relying on aBE model that is robust to likely cultural influences will make thesereasons more valid.

Both interests in and needs for cross-cultural studies seemto have emerged boldly enough to necessitate a methodologicalillustration for future applications broadly in general hospitalityresearch, needless to say hospitality BE research. In their com-prehensive review of hospitality marketing research, Line andRunyan (2012) summed (p. 485): “The methodological goal of mostdomains has recently moved toward the examination of the cross-cultural validity of scales commonly used within the domain. . ..Utilizing such scales cross-culturally is important, but ensuringthat the scales are cross-culturally valid is an equally importantmethodology issue, often ignored in [hospitality] research. . .. Theproper method of establishing cross-cultural validity is to testfor measurement invariance (Steenkamp and Baumgartner, 1998).Unfortunately, such tests are rarely enacted. Indeed, although pop-ular measurement scales of hospitality phenomena are employedcross-culturally, invariance research is absent in top hospitalityjournals. As such, we suggest that future studies address invari-ance as it relates to the cross-cultural employment of hospitality

marketing constructs.”

This study responds to Line and Runyan’s (2012) call and aimsto introduce and illustrate a methodological procedure of assessingand validating measurement invariance of a research model,

H. Oh, C.H.C. Hsu / International Journal of Hospitality Management 36 (2014) 156–166 157

Perceived Quality

Brand Aw areness

Brand Image

Management Tr ust

Brand Reliability

Brand Loyal ty

Brand ChoiceIntention

er-bA

eiiayMautHaptprraagcfm

2

2

oDnb(t(isdc(

fchsbaap

Fig. 1. A model of customdapted from Hsu et al. (2012).

specially of hotel BE as an example, in cross-cultural contexts. Thellustration follows the methodological procedure of measurementnvariance or equivalence analysis in application of multi-samplenalysis with structural equation modeling (SEM), a procedure notet formally introduced in the hospitality literature (e.g., Horn andcArdle, 1992; Steenkamp and Baumgartner, 1998; Vandenberg

nd Lance, 2000). Although some researchers have recently begunsing multigroup invariance analysis as part of their hypothesisests (e.g., Boo et al., 2009; Chung et al., 2011; Hallak et al., 2012;an et al., 2010), the analysis procedure, its logic, and its generalpplications still remain largely unexplained. For an illustrativeurpose, therefore, this study examines three cross-cultural fac-ors: hotel brand identity (domestic vs. foreign), the customer’srimary language spoken (Mandarin vs. English), and the traveler’segion of residence (Asia vs. Europe vs. North America). The keyesearch question is whether a model’s measurement structurend its theoretical expositions (i.e., structural coefficients) are ten-ble, and hence generalizable, among cross-cultural hotel customerroups. The multinational nature of the hotel business suits suchross-cultural examinations. As the world gets smaller especiallyor the hotel business, cross-cultural generalization of research

odels is no longer a goal; it is a requirement.

. Brand equity and culture

.1. The hotel BE model

BE research is largely rooted in the seminary conceptual worksf Aaker and Keller (Aaker, 1991, 1996; Keller, 1993, 2003a).efined as “a set of brand assets and liabilities linked to a brand, itsame and symbol that add to or subtract from the value providedy a product or service to a firm and/or to that firm’s customers”Aaker, 1991, p. 15), BE serves as a comprehensive index estimatinghe value exchanged between a brand and its customers. To Keller2003a), BE was the added value resulting in different market-ng outcomes, a common denominator for interpreting marketingtrategies, and the value of a brand that could be created in manyifferent ways. As implied in these definitions, capturing BE in pre-ision is a challenge and requires a multi-dimensional approachKeller, 2003b).

Researchers have proposed customer-based measurementrameworks of hotel BE in application of Aaker’s and Keller’s con-eptualization. Bailey and Ball (2006) explored the meaning ofotel BE, while Kayaman and Arasli (2007) examined relation-hips among selected BE sub-constructs such as perceived quality,

rand loyalty, and brand image. Prasad and Dev (2000) proposed

hotel BE index consisted of top-of-mind brand recall, brandwareness, satisfaction, return intent, price-value relationship, andreference. Kim and colleagues followed Aaker’s (1991) proposed

ased hotel brand equity.

model more closely to measure BE of both the luxury/mid-scalehotels and restaurants (Kim and Kim, 2005; Kim et al., 2008).Better-performing casino hotels were found to perform better oncustomer-based brand equity measures, say, brand loyalty, brandimage, and brand awareness (Tsai et al., 2010). More recently, Hsuet al. (2012) proposed a customer-based hotel BE measurementmodel based on a series of qualitative and quantitative studies.While these models either focused on or extended different aspectsof Aaker’s (1991, 1996) and Keller’s (1993, 2003a) works, they com-monly recognized four essential components of hotel BE: brandloyalty, brand awareness, perceived quality, and brand awareness.

For the purpose of illustrating cross-cultural equivalence anal-ysis procedures, this study chose Hsu et al.’s (2012) model forseveral reasons. First, the model was comprehensive extend-ing Aaker’s (1991) model by including seven BE sub-constructs,namely, perceived quality, brand awareness, brand image, man-agement trust, brand reliability, brand loyalty, and brand choiceintention (see Fig. 1). The first five constructs directly affected brandloyalty which in turn determined brand choice intention. Second,the model was developed in China, yet it has not been testedfor cross-cultural generalizability with hotel customers originatingfrom other countries. Third, the model was one of the most recentlyproposed hotel BE models providing potentially more accumulatedthoughts on BE measurement issues. Hsu et al. provided conceptualbackgrounds with necessary explanations about each theoreticalrelationship for the model; this study uses the model as a caseexample to illustrate methodologically cross-cultural equivalenceassessment procedures and, as a result, aims to contribute to futurehotel BE research and theory development in global scale.

2.2. Cross-cultural factors

To illustrate cross-cultural equivalence analysis, this studychose three cultural factors: hotel brand stayed (domestic vs.foreign), the traveler’s primary language spoken (Mandarin vs.English), and the traveler’s region of current residence (Asia vs.Europe vs. North America). First, research on the effects of coun-try of origin and consumer ethnocentrism supported an inferenceon potential differences between the domestic and foreign hotelbrands the customers chose and stayed. Defined as the countryof manufacture or assembly (Han and Terpstra, 1988), country oforigin serves as an extrinsic cue affecting broadly the consumer’sproduct evaluation by positively associating the product’s qualitywith the characteristics of the country the product originates from(Ahmed et al., 2002). Consumer ethnocentrism is the belief held

by consumers about the appropriateness and morality of buyingforeign-made products (Shimp and Sharma, 1987). Motivated byethnocentrism, consumers tend to choose domestic or local prod-ucts even if they are lower in quality than foreign-made products

1 f Hosp

(aah(

2tsosefAmvs

ttisrott2tt

3

3

ieFroattsm2srnp

rasc2Ltiivvv�i

58 H. Oh, C.H.C. Hsu / International Journal o

Wall and Heslop, 1986). Thus, as suggested by country of originnd consumer ethnocentrism research, travelers choosing to stayt a domestic hotel, as compared to a foreign hotel brand, couldave a different brand choice process as well as evaluation schemaShanahan and Hyman, 2007).

Second, language is a way of marking cultural identity (Gao,006). Gao argued that the meanings of a particular language pointo the culture of a particular social group and that the analy-is of those meanings involves the analysis and comprehensionf that culture. Chinese-speaking vs. foreign language travelershowed significant differences in tour guide evaluations (Huangt al., 2010). Evaluations of service performance were also dif-erent between Asians (mainly Chinese) and foreigners (mainlymericans) (Kim et al., 2010; Mattila, 2000). Therefore, the pri-ary language used by the traveler may be an important cultural

ariable determining how the traveler evaluates products and, con-equently, judges the equity of a hotel brand.

Finally, place of residence is a proxy variable demarcating cul-ural differences. For example, Asian (vs. Western) consumersended to exhibit a stronger (weaker) “concern for face” and “beliefn fate,” which motivated them to feel more dissatisfied with aervice failure in social attributes (Chan et al., 2009). Country ofesidence and nationality of origin also caused differential effectsn the traveler’s product evaluation, satisfaction with a destina-ion, perception of service quality, value judgment, and likelihoodo return to a destination or destination loyalty (Forgas-Coll et al.,012; McCleary et al., 2006). Viewed similar to product or destina-ion evaluations, the evaluation of hotel BE may differ across theraveler groups defined by the place of their residence.

. Model equivalence and hypotheses

.1. Models of equivalence

Assuring measurement equivalence is a prerequisite for mean-ngful comparisons across cultural groups and an important step instablishing the generalizability of theories or models (Milfont andischer, 2010; Steenkamp and Baumgartner, 1998). In substantiveesearch involving multinational businesses like hotels, researchersften rely on a measurement instrument that was developed innother country, culture, or customer segment. In such cases, twoypical assumptions are that the instrument will operate exactlyhe same way and that the underlying construct(s) being mea-ured will have the same theoretical structure and psychologicaleaning across groups, cultures, or countries of interest (Byrne,

008). These two critical assumptions are, however, rarely testedtatistically and this has been the case in hospitality and tourismesearch. Conclusions and cross-cultural comparisons drawn from aon-equivalent measurement instrument can misguide theoreticalrogress as well as managerial decisions.

Integrated procedures to test equivalences of multivariateesearch models are available in the SEM literature (see Jöreskognd Sörbom, 2006). A number of researchers have endeavored totreamline and visualize the procedure for applications in cross-ultural psychological and marketing research (e.g., Byrne, 2001,006, 2008; Steenkamp and Baumgartner, 1998; Vandenberg andance, 2000). In essence, the procedure tests across groups gshe equivalence of five sample-implied parameter matrices, thats, in the LISREL convention, �x, Ф, �ı, �x, and �, where �x

s the factor loading matrix of measured variables xs, Ф is theariance–covariance matrix of the latent variables �s, �ı is the

ariance–covariance matrix of measurement errors ıs, �x is theector of intercepts for xs, and � is the vector of means fors. By constraining the parameters of these five matrices to benvariant across groups, one may test measurement equivalences

itality Management 36 (2014) 156–166

pertaining to three equations: the population variances and cova-riances, which derive from ˙xx = �xФ�′

x + �ı, the measurementmodel (x = �x + �x� + ı), and the mean structure (�x = �x + �x�).Specifically, the procedure generally suggests testing the followingseries of equivalences in sequence.

3.1.1. Equivalence of the covariance matrices and mean vectorsSteenkamp and Baumgartner (1998) recommend that one starts

out testing the invariance of the variance–covariance matrices(˙xx) and mean vectors (�x) across groups, both jointly and sep-arately. In these invariance models, each parameter value of thetwo matrices (i.e., ˙xx and �x) is constrained to be invariant acrossgroups. The results will provide useful information about whetherthe (co)variances or means are primarily responsible for the overalllack of invariance, if any. If the covariances and means are invariantacross groups, the data can be pooled and group-specific modelingis unnecessary.

3.1.2. Establishing the baseline configural modelThe configural validity (i.e., the same structure of subscales and

their covariances) of the hypothesized model needs to be tested ineach group separately for its goodness of fit to the data from theperspectives of both parsimony and substantive meaningfulness(Byrne, 2008; Ployhart and Oswald, 2004). The same model is fitto each group independently in an effort to identify a theoreticallyjustified, statistically acceptable model for each group. Desirablefor the sake of generalizability and efficiency of remaining cross-cultural equivalence analyses is the same model fitting equally wellin each group. Fitting acceptably across groups, the model servesas the baseline model for subsequent equivalence analyses (Byrne,2008). This step is necessary before proceeding to test configuralequivalence in order to understand whether the lack of fit, if any,comes from any particular group(s). Any group producing an ill fitdictates modifications to the model that may require group-specificmodeling hindering further equivalence analyses.

3.1.3. Configural equivalence.In this model, one specifies the baseline model with the ref-

erence group and then imposes the same model structure inthe remaining groups. This pattern equivalence model tests theassumption that the same model structure is applicable across thecultural groups under study. The model(s) fitting well in each groupin the previous stage indicate(s) that the factor structure is similaracross groups, but it is “not necessarily equivalent across groups asequivalence of the factors and their related items has not yet beenput to the test” (Byrne, 2008, p. 873, emphasis original). Towardgeneralizability of the model across the groups, this pattern equiv-alence test should produce an acceptable fit whose value thenbecomes a baseline statistic to which the fit of the subsequentlynested equivalence models can be compared.

3.1.4. Metric equivalence.Also called measurement (unit) equivalence (van de Vijver and

Leung, 1997), the metric invariance model tests whether thestrengths of relations between specific scale items and their respec-tive latent constructs are identical across groups (Milfont andFischer, 2010). In this model, one freely estimates the factor load-ings in the reference group, and then constrains them to beinvariant in the remaining groups (i.e., �1 = �2 = · · · = �g). Resultsprovide evidence of whether the respondents in each group inter-preted and responded to each measurement item in the same

way. This equivalence is a more stringent test than the configu-ral equivalence as it imposes the assumption of equal metrics orscale intervals across groups (Rock et al., 1978). For meaningfulcomparisons and thus cross-group generalizability of the model,

f Hosp

tp

3

gto1fla2otr

3

eeitips

3

godb

�

jciUrs

3

ugeTieB

3

tmailccuoi

H. Oh, C.H.C. Hsu / International Journal o

his equivalence should be based on the configural equivalence androduce a good fit.

.1.5. Scalar equivalenceThe scalar, or intercept (�x), invariance model tests whether

roup differences in the means of the observed items are consis-ently related to the corresponding group differences in the meansf the underlying construct(s) (Steenkamp and Baumgartner,998). Establishing scalar invariance (i.e., �1 = �2 = · · · = �g), there-ore, indicates that individuals who have the same score on theatent construct would obtain the same score on the observed vari-ble regardless of their group membership (Milfont and Fischer,010). Scalar invariance is necessary for meaningful comparisonsf latent factor means. One freely estimates the intercept terms inhe reference group while constraining them to be invariant in theemaining groups.

.1.6. Error variance equivalenceOne may want to test whether the amount of measurement

rror for each item is equal across groups. The errors in the ref-rence group are freely estimated, but they are constrained to benvariant in the other groups (i.e., �1 = �2 = · · · = �g). Note thathis equivalence is overly strict and is likely of least interest andmportance to researchers in most situations (Bentler, 2005) and, inractice, rarely holds. For these reasons, this model is not a requiredtep in equivalence analyses.

.1.7. Factor variance/covariance equivalenceEquivalence in both factor variances and covariances across

roups (i.e., Ф1 = Ф2 = · · · = Фg) indicates that the range of scoresn the latent factors and the correlations among the latent factorso not vary across groups (Milfont and Fischer, 2010). The test maye separated into factor variance equivalence (i.e., �1

jj= �2

jj= · · · =

gjj) and factor covariance equivalence (i.e., �1

jk= �2

jk= · · · = �g

jk;

= 1, . . ., m and k = 1, . . ., [j − 1]). Because the latter has direct impli-ations for theoretical relationships among the latent constructs,t is usually of more interest and research focus (Byrne, 2008).sing a structural model, one can achieve the same test results by

e-specifying factor covariance equivalences into equivalences oftructural parameters across groups.

.1.8. Factor mean equivalenceThis model tests whether groups differ in the means of the

nderlying constructs. One fixes the factor means in the referenceroup to zero and constrains the factor mean between the refer-nce and remaining groups to be invariant (i.e., �1 = �2 = · · · = �g).his model must include the equivalence of means for the observedtems (i.e., scalar invariance) because analysis for factor meanquivalence assumes item means are equivalent (Byrne, 2008;yrne and Watkins, 2003).

.2. Model assessment

In line with the two-step approach to testing covariance struc-ure models (Anderson and Gerbing, 1988), models of equivalences

ay also comprise measurement equivalence and structural equiv-lence (Byrne, 2008; Vandenberg and Lance, 2000). The formerncludes configural, metric, scalar, and error variance equiva-ence models, while the latter pertains to factor variance, factorovariance, and factor mean equivalence models. To satisfy basic

onditions for generalizability of the target model, both the config-ral and metric equivalence models should fit the data acceptably inrder to assure that regardless of group membership respondentsnterpret and respond to the instrument in the same way. Models

itality Management 36 (2014) 156–166 159

of other equivalences are tested for a variety of other substantiveinterests.

Use of multiple fit indices is advisable when assessing the good-ness of equivalence and nested models. No single, ideal index isavailable to determine goodness of equivalence models. One mayevaluate overall model fit by using the chi-square to degrees-of-freedom ratio (2/df), with the ratio of 3:1 or less as an indicatorof good fit (Wheaton et al., 1977; Carmines and McIver, 1981) andRMSEA (root mean square error of approximation) with the valueof .06 or smaller as good fit and the value up to .09 as acceptable (Huand Bentler, 1999). In addition, some incremental fit indices mayassess overall goodness of models. With a caution given to its well-known sensitivity to sample size (Anderson and Gerbing, 1988), thechi-square difference test is usable in comparing nested models.Both the comparative fit index (CFI) and the nonnormed fit index(NNFI) attest to good fit when their values are .95 or higher (Huand Bentler, 1999). Lower values of a consistent version of Akaike(1987) information criterion (CAIC) qualify a better fit when com-paring models. Steenkamp and Baumgartner (1998) particularlysupported using RMSEA, NNFI, and CAIC as most effective indices indistinguishing between correctly and incorrectly specified models.

3.3. Partial equivalence

The concept of partial equivalence often becomes a questionin equivalence analysis. In multivariate measurement situations,all equivalence models may be further analyzed for chosen sub-sets of parameter equivalences. In practice, researchers may facesituations where they find only a subset of parameters to be invari-ant across all or a subset of groups because the assumption thatall parameters are equivalent across all groups is often unrealisticor too stringent. This is true particularly where a priori theoreti-cal knowledge or research design provides justifiable reasons forsuch partial equivalence. For example, one may apply equivalenceanalysis to experimental data in which group identity is typicallypredetermined and treatment effects are theoretically predictable(see Ployhart and Oswald, 2004). In such a case one may specifyonly necessary parameters such as selected means to be equal ordifferent across groups, depending on the goals of the experiment.

Where a priori knowledge is unavailable, partial equiva-lence analysis is of limited use. Several researchers proposedsystematically incorporating partial equivalence analysis whencomparing groups, especially when data conditions meet such pur-poses (Byrne, 2008; Milfont and Fischer, 2010; Steenkamp andBaumgartner, 1998). In the absence of a priori knowledge aboutdata conditions, however, the question is how one determines thesubset of equivalent or nonequivalent parameters. Steenkamp andBaumgartner advised a conservative use of modification indices tothis end. In this study, partial equivalence models are attempted,with reference to modification indices, only for the scalar and factormean equivalence models for the purpose of illustration, becausereliance on modification indices for model specification tends tocapitalize on chance.

3.4. Equivalence and hypotheses

Depending on goals and availability of source theories, one mayposition a cross-cultural study in one or more of the four cate-gories: generalizability, psychological differences, theory testing,or external validation (van de Vijver and Leung, 1997). The firsttwo are typical when the researcher does not consider contextualfactors other than mere cross-cultural comparisons. The latter two

serve studies with specific contextual factors such as demographicand psychological variables. More hypothesis-driven are general-izability and theory testing studies, while the other two are ratherexploratory in nature. Generalizability studies attempt to establish

1 f Hosp

torig

iBabfttmcslg(pDca

4

4

(cornvhctasftiq2

i1tap

60 H. Oh, C.H.C. Hsu / International Journal o

he generalizability of research findings obtained in one group withthers, which in reality is an overriding goal in most cross-culturalesearch as well as in this illustrative study. Theory-driven stud-es are conducted when sufficient previous research is available forenerating specific hypotheses.

In equivalence studies, the scope and conclusions depend crit-cally on study goals (Ployhart and Oswald, 2004; Steenkamp andaumgartner, 1998). The goal of this study is to illustrate how tossess whether a hotel BE model would generalize across culturaloundaries. For the illustrative model chosen (Hsu et al., 2012),ew theories are available to predict exact differences among thehree sets of cross-cultural groups defined by hotel brand identity,he first language spoken, and the tourist’s place of residence. The

odel’s recency also makes it difficult to precisely predict its appli-ability across the cultural traveler groups. As such, although moretringent forms of equivalence are generally preferable, estab-ishing configural and metric equivalences seems a reasonableoal in assessing the model toward cross-cultural generalizabilitySteenkamp and Baumgartner, 1998). In other contexts, researchersroposed and tested equivalence hypotheses in a null form (e.g.,eng et al., 2005; Doll et al., 1998). Hence, toward the model’s cross-ultural generalizability, which is defined as satisfactory configuralnd metric equivalence, this study proposes:

H1: The hotel BE model (Hsu et al., 2012) is configurally equivalentacross:H1(a): domestic and foreign hotel brands;H1(b): tourists speaking Mandarin and English as their first lan-guage;H1(c): the regions the tourists reside (Asia, Europe, or NorthAmerica).

H2: The hotel BE model is metrically equivalent across:H2(a): domestic and foreign hotel brands;H2(b): tourists speaking Mandarin and English as their first lan-guage;H2(c): the regions the tourists reside (Asia, Europe, or NorthAmerica).

. Methods

.1. Survey and sampling

Survey data were collected from tourists staying at upscalefour- and five-star or equivalent) hotels operating in 12 majorities in China. These major Chinese cities were hosting numer-us high-end foreign hotel brands and the number was growingapidly. By focusing on upscale properties of both domestic Chi-ese and foreign hotel brands and their guests traveling from aariety of other countries, this study could capture cross-culturalotel business situations. Although collecting data directly in theultures or countries under comparison would have been desirable,he sampled Chinese cities and upscale hotels hosting routinely

wide variety of international travelers could provide alternativeampling situations for a cross-cultural study especially by justi-ying costs and convenience. Such cross-cultural sampling withinhe same country does not necessarily disrupt the enduring culturaldiosyncrasies of the visitors, especially of short haul, and was fre-uently used in previous cross-cultural studies (e.g., Huang et al.,010; Mattila, 2000; McCleary et al., 2006).

This study attempted to balance several sources of data to max-mize market segment representation. The study design included

5 domestic Chinese and 15 foreign hotel brands with no morehan two properties per brand. For a broader coverage of tourists,t least 20 domestic and 20 foreign tourists were sampled at eachroperty. This 2 (brand identity: domestic vs. foreign) × 2 (tourist

itality Management 36 (2014) 156–166

identity: domestic vs. foreign) sampling design provided at least300 respondents for each of the four design cells, for a total of1200 respondents for the study. The survey also included a ques-tion asking the tourist’s first language, which could reflect touristidentity more culturally. The respondents were categorized ad hocinto three global regions of their residence: Asia, Europe, and NorthAmerica.

The sampling efforts produced results somewhat closely asintended. We contacted various hotel corporate offices, propertymanagers, and hotels associations to seek support for the study.As a result, 32 properties agreed to participate and each propertyreceived two sets of questionnaires (30 English and 30 Chineseversions). The hotel operators were directed to use their guest reg-ister to identify 20 domestic and 20 foreign guests and place thequestionnaire in the guest room the night before the guest’s depar-ture. The cover letter instructed the guest to return the completedquestionnaire to the front desk in the supplied envelope sealed.Operators continued the procedure until they secured at least 20domestic and 20 foreign participants, and then returned the ques-tionnaires to the researchers.

4.2. Measures

This study used the same multi-item scales, except for one item,used by Hsu et al. (2012) to measure the model constructs. One ofthe four brand image items, “I have a clear image of the type of peo-ple who would stay at an XYZ hotel,” was dropped from this studyfor two reasons (Kline, 2010). First, the meaning of “the type ofpeople” in the original statement was vague and, second, Hsu et al.(see Table 1) reported a substantial amount of measurement errorfor the item (.53, standard error = .05). Perhaps the vague wordingcaused the sizable measurement error. All measures were opera-tionalized on a 7-point scale (1 = strongly disagree, . . ., 7 = stronglyagree).

5. Findings

5.1. Descriptive data

The respondents (n = 1346) fairly represented both Chinese(48.7%) and foreign (51.3%) nationals as well as Chinese (41.5%) andforeign (55.6%) hotels stayed. Twenty nine hotels returned com-pleted questionnaires covering 12 major Chinese cities, 11 Chinesehotel brands, and 18 foreign brands. The sample included moremales (63.4%) and the majority (85.1%) aged between 26 and 55years. Nearly 51% completed 2- or 4-year college education andabout 42% held a graduate degree. The majority (64%) were married,but about 30% single. Notable was about 52% not reporting theirannual household income. More than one half (50.7%) reportedMandarin and 31% English as their first language. The regions ofresidence included 59.5% in Asia, 11% Europe, and 17.8% NorthAmerica. Business travelers represented 70%, while vacationers20%.

Table 1 summarizes the results of preliminary analyses. Thedemographic variables showed significantly different distributionsacross the three cross-cultural factors. For example, male guestswho stayed at international hotels tended to represent the samplemore than other respondents (p < .01). A similar pattern was evi-dent with English speaking males (p < .01) and Asian males (p < .01)over their counterparts. For age, respondents staying at interna-

tional brands and speaking English as their first language tendedto be older than those staying at Chinese indigenous brands andspeaking Mandarin. Respondents residing in Asia tended to beyounger than their counterparts. These results indicated that the

H. Oh, C.H.C. Hsu / International Journal of Hospitality Management 36 (2014) 156–166 161

Table 1Descriptive sample characteristics (n = 1346).

Brand identity First language Region of residence

Chinese(n = 558)

International(n = 748)

Mandarin(n = 660)

English(n = 407)

Asia (n = 801) Europe(n = 148)

N. America(n = 239)

Gender 2 = 9.83, p = .002 2 = 23.56, p < .001 2 = 24.69, p < .001Male 335 (25.2) 519 (39.1) 392 (36.1) 293 (27.0) 491 (40.7) 109 (9.0) 179 (14.9)Female 228 (17.2) 246 (18.5) 288 (26.5) 112 (10.3) 328 (27.2) 38 (3.2) 60 (5.0)

Age 2 = 19.93, p = .001 2 = 120.75, p < .001 2 = 125.02, p < .00125 or younger 56 (4.3) 46 (3.5) 73 (6.9) 14 (1.3) 81 (6.8) 6 (.5) 8 (.7)26–35 223 (17.2) 226 (20.5) 312 (29.3) 101 (9.5) 357 (30.1) 37 (3.1) 60 (5.1)36–45 164 (12.7) 235 (18.1) 191 (18.0) 132 (12.4) 241 (20.3) 52 (4.4) 71 (6.0)46–55 70 (5.4) 145 (11.2) 75 (7.0) 101 (9.5) 98 (8.3) 38 (3.2) 59 (5.0)56 or older 31 (2.4) 60 (4.6) 15 (1.4) 49 (4.6) 22 (1.9) 16 (1.3) 39 (3.3)

Annual household income (in US$) 2 = 41.82, p < .001 2 = 11.85, p = .037 2 = 32.01, p < .00150,000 or less 21 (4.8) 30 (4.6) 9 (2.0) 22 (4.8) 35 (6.3) 10 (1.8) 8 (1.4)50,001–75,000 48 (7.4) 52 (8.0) 17 (3.7) 56 (12.3) 30 (5.4) 22 (4.0) 32 (5.8)75,001–100,000 71 (11.0) 80 (12.4) 10 (2.2) 93 (20.5) 48 (8.7) 28 (5.1) 47 (8.5)100,001–125,000 41 (6.3) 67 (10.4) 9 (2.0) 74 (16.3) 26 (4.7) 24 (4.3) 39 (7.1)125,001–150,000 18 (2.8) 51 (7.9) 8 (1.8) 41 (9.0) 16 (2.9) 22 (4.0) 20 (3.6)More than 150,000 30 (4.6) 127 (19.7) 17 (3.7) 98 (21.6) 47 (8.5) 32 (5.8) 66 (12.0)

Education 2 = 34.90, p < .001 2 = 8.58, p = .035 2 = 11.38, p = .077High school or less 40 (3.0) 26 (2.0) 29 (2.7) 17 (1.6) 40 (3.3) 10 (.8) 9 (.7)2 or 4 year college/university 321 (24.1) 361 (27.1) 388 (35.5) 199 (18.2) 441 (36.4) 62 (5.1) 121 (10.0)Graduate degree 192 (14.4) 376 (28.2) 256 (23.4) 191 (17.5) 328 (27.0) 80 (6.6) 109 (9.0)

Marital status 2 = 11.93, p = .018 2 = 24.42, p < .001 2 = 30.05, p < .001Single 186 (14.1) 215 (16.3) 222 (20.5) 100 (9.2) 273 (22.7) 42 (3.5) 57 (4.7)Married without children 121 (9.2) 223 (16.9) 170 (15.7) 123 (11.4) 202 (16.8) 41 (3.4) 74 (6.2)Married with children 231 (17.5) 285 (21.6) 265 (24.5) 157 (14.5) 316 (26.3) 53 (4.4) 99 (8.2)

(1.0)

sc

ur5slTi

5

toiiCnvmwema.mcev1icm

Divorced 19 (1.4) 28 (2.1) 11

ample distributions across the five demographic and three cross-ultural factors were disproportional (p < .01).

Table 2 presents means and standard deviations for individ-al measures across the three cross-cultural factors. Item meansanged from 4.72 (the first brand loyalty item under Europe) to.83 (the first brand quality item under N. America). The meanshowed no noticeable patterns, although they tended to be slightlyower for brand choice intention than for the other constructs.he standard deviations showed substantive spreads for mosttems.

.2. The overall measurement model

To verify the integrity of Hsu et al. (2012) measurement struc-ure, we fit the same measurement model to the overall data, withne item dropped as explained earlier. Table 3 presents the resultsncluding descriptive statistics and indices of construct reliabil-ty and validity. The model fit the data well (2 = 683.4, df = 168;FI = .99; NNFI = .99; RMSEA = .048). The factor loadings were all sig-ificant and substantial (>.82), providing evidence for convergentalidity of the constructs (Bagozzi and Yi, 1988). The measure-ent errors were relatively small (<.33). Composite reliabilitiesere higher than .89 and the amount of variance extracted for

ach construct was higher than .72, both exceeding the suggestedinimum of .7 and .5, respectively (Bagozzi and Yi, 1988; Fornell

nd Larcker, 1981). Cronbach’s alpha for each construct exceeded89, providing evidence for internal consistency of the measure-

ent items within each construct. None of the 21 inter-constructorrelations, squared, was larger than the amount of variancextracted for each construct, which substantiated discriminantalidity of the constructs (Bagozzi and Yi, 1988; Fornell and Larcker,

981). Close examinations of the parameter estimates and valid-

ty/reliability indices confirmed that the overall results were highlyonsistent with those reported by Hsu et al. in both pattern andagnitude.

25 (2.3) 15 (1.2) 13 (1.1) 9 (.7)

5.3. Equivalence across brand IDs

Table 4 reports the key results of equivalence analyses acrossthe cross-cultural factors. For the domestic vs. foreign hotel brandgroups, the simultaneous equivalence of the variance–covariancematrix and the mean vector fit the data well. The separate analyses,however, indicated that the equivalence of the mean vector shouldbe rejected due to significant differences in the mean values. Thegroup-specific fit of the model was sound and produced a slightlyhealthier fit in the foreign brand group. The configural equivalencemodel fit the data well, lending support for H1(a). The fit indices ofthe metric equivalence model also indicated that the two customergroups responded to the measurement items in the same way; infact, all fit indices were more positive than those for the configuralequivalence model (2 = 8.3, df = 14), attesting to the efficiencyof the metric over the configural equivalence model. H2(a) couldnot be rejected.

The two brand groups’ measurement equivalence was accept-able in additional aspects. The scalar equivalence model wasacceptable, although close examinations of the results indicatedthat some means (or intercepts) could differ significantly betweenthe two groups. In the interest of illustrating an alternative model,mean differences in six of the 21 items, suggested by modifica-tion indices, were freely estimated. This partial scalar equivalencemodel improved the model fit significantly (2 = 147.4, df = 6).The fit of the error variance equivalence, albeit satisfactory overall,worsened significantly compared to that of the metric equiva-lence model (2 = 177.2, df = 21), indicating that a number ofmeasurement errors might be unequal between the groups. Sim-ilarly, both the factor variance–covariance (2 = 157.8, df = 28)and the factor mean (2 = 62.1, df = 7) equivalence models

appeared to fit the data acceptably, but their fits deterioratedsignificantly against the factor means model. Thus, two groupsseemed to differ in some factor variances, covariances, and fac-tor means. The partial factor mean equivalence model freeing

162 H. Oh, C.H.C. Hsu / International Journal of Hospitality Management 36 (2014) 156–166

Table 2Means and standard deviations of measurement items (n = 1346).

Construct and measurement itemsa Brand ID Language Region of residence

Chinese Foreign Mandarin English Asia Europe N America

Meanb SD Mean SD Mean SD Mean SD Mean SD Mean SD Mean SD

Brand choice intentionSmarter to choose XYZ 5.16 1.19 5.04 1.27 5.13 1.23 5.11 1.29 5.04 1.25 5.01 1.26 5.20 1.25Superior choice 5.09 1.23 4.91 1.26 5.07 1.27 4.96 1.26 4.97 1.29 4.82 1.22 5.11 1.18Make sense to choose XYZ 5.02 1.27 4.82 1.35 4.93 1.27 4.97 1.40 4.86 1.29 4.75 1.47 5.08 1.34

Brand loyaltyChoose XYZ repeatedly 5.06 1.28 4.78 1.50 4.94 1.36 5.04 1.43 4.87 1.40 4.72 1.56 5.06 1.39Feel good and positive 5.20 1.21 5.20 1.29 5.15 1.25 5.37 1.20 5.09 1.28 5.24 1.29 5.50 1.11Feel pleasant 5.24 1.24 5.12 1.32 5.17 1.27 5.30 1.26 5.09 1.29 5.14 1.31 5.42 1.24

Brand qualityOf high quality 5.45 1.05 5.62 1.09 5.50 1.06 5.69 1.08 5.43 1.08 5.61 1.05 5.83 1.01Sets quality standards 5.32 1.09 5.32 1.22 5.28 1.11 5.49 1.22 5.21 1.15 5.29 1.19 5.66 1.18Of the highest standard 5.20 1.18 5.15 1.29 5.15 1.21 5.36 1.24 5.04 1.26 5.17 1.26 5.50 1.19

Brand awarenessKnow symbol and logo 5.11 1.51 5.33 1.52 5.27 1.46 5.33 1.51 5.16 1.54 5.19 1.58 5.47 1.44Know what it looks like 5.17 1.33 5.18 1.37 5.26 1.27 5.23 1.37 5.14 1.33 5.18 1.40 5.29 1.41Recognize the hotel 5.12 1.40 5.20 1.42 5.28 1.34 5.21 1.39 5.16 1.41 5.07 1.47 5.26 1.39

Brand imagePrestigious 5.26 1.13 5.22 1.25 5.36 1.13 5.22 1.26 5.22 1.21 5.12 1.29 5.44 1.10Sophisticated 5.23 1.16 5.14 1.26 5.32 1.20 5.11 1.23 5.19 1.24 4.90 1.35 5.36 1.09Special 5.13 1.24 4.99 1.31 5.14 1.22 5.14 1.26 5.01 1.28 4.98 1.36 5.21 1.27

Management trustTrust management 5.36 1.17 5.50 1.14 5.41 1.18 5.58 1.11 5.34 1.17 5.58 1.09 5.64 1.10Know what to do 5.36 1.14 5.46 1.16 5.36 1.18 5.60 1.08 5.31 1.17 5.47 1.12 5.66 1.08Good practice 5.25 1.17 5.32 1.21 5.26 1.21 5.44 1.15 5.18 1.21 5.31 1.19 5.54 1.12

Brand reliabilityMeet expectations 5.23 1.14 5.26 1.21 5.20 1.20 5.44 1.15 5.15 1.19 5.35 1.18 5.50 1.18Consistent every time 5.26 1.15 5.27 1.16 5.22 1.17 5.44 1.12 5.18 1.17 5.37 1.16 5.48 1.14Will not disappoint 5.25 1.16 5.22 1.26 5.15 1.26 5.44 1.13 5.12 1.25 5.33 1.17 5.50 1.19

gree)

f(

5

mmpmtgstrb

acwicmlew(at

a Refer to Table 3 for a full description of each measurement item.b All items were measured on a 7-point scale (1 = strongly disagree; 7 = strongly a

our equality constraints could improve the model fit significantly2 = 58.5, df = 4).

.4. Equivalence across languages

The Mandarin vs. English speaking customer groups showedeasurement equivalence. Although the variance–covarianceatrix and mean vector fit the data well, the mean equivalence

ortion could cause much of the misfit implied. The measurementodel fit the data aptly in the English group, but marginally in

he Mandarin group. The configural equivalence model exhibitedood fit, supporting H1(b) that suggested the same measurementtructure for the groups. The metric equivalence model also fithe data well (2 = 36.0, df = 14) indicating that the two groupsesponded to the measurement scales identically. H2(b) could note rejected.

The remaining equivalence analyses resulted in good fit over-ll at the entire model level, although some individual parametersould differ between the groups. The scalar equivalence assumptionas acceptable, even if the partial scalar equivalence model could

mprove the model fit significantly by relaxing six mean equalityonstraints (2 = 107.0, df = 6). The error variance equivalenceodel was tenable in addition to the satisfactory metric equiva-

ence (2 = 50.2, df = 21). While the factor variance–covariancequivalence model fit the data acceptably, some parametersere significantly different between the two language groups

2 = 108.8, df = 28). The factor mean equivalence model wascceptable, albeit worse than the factor means model (2 = 76.6,df = 7); four factor means seemed to have contributed much of

he worsened fit (2 = 72.1, df = 4).

.

5.5. Equivalence across regions of residence

Equivalence comparisons across the three regional groupsrevealed strong possibilities for data pooling. Both the omnibusand separate analyses of equivalence in the variance–covariancematrix and mean vector produced good model fit overall. The mea-surement model fit the data well in both Asian and North Americangroups, but its incremental fit was relatively inferior in the Euro-pean group. The configural equivalence model was acceptable, andso was the metric equivalence model, thereby supporting bothH1(c) and H2(c). Metric equivalence did not significantly worsenthe fit of the configural equivalence model (2 = 46.3, df = 28).

The three-group comparisons for the remaining equivalencesshowed equality at the model level, but some significant differ-ences at the parameter level. The scalar equivalence model resultedin good fit, although the fit could further improve by allowing 20of the 63 means to differ across the three groups as shown inthe partial scalar equivalence model (2 = 87.4, df = 20). Thethree groups responded to the scales with significantly differentamounts of measurement errors (2 = 199.0, df = 42). The fitof the factor variance–covariance equivalence model was good,even against the basic factor means model (2 = 172.5, df = 56).However, the factor mean equivalence model could fit the data bet-ter (2 = 76.6, df = 14), especially if 10 of the 21 means wereallowed to differ across the groups (2 = 55.5, df = 10).

5.6. Equivalence in structural relationships

As an extension of the factor variance–covariance equivalenceanalysis, this study tested equivalence of the structural relations

H. Oh, C.H.C. Hsu / International Journal of Hospitality Management 36 (2014) 156–166 163

Table 3Measurement model results (n = 1306).

Construct and measurement itema Mean(standard deviation)

Factor loading(standard error)b

Error(standard error)b

Brand choice intention (� = 91; VE = .77; ̨ = .91)Even if other competing brands are not different from XYZ in any way, itseems smarter to choose an XYZ hotel

5.08 (1.24) .87 (–) .24 (.02)

An XYZ hotel is always a superior choice to its rival hotels 4.98 (1.25) .90 (.02) .19 (.02)It makes sense to choose XYZ instead of any other hotel brand, even if theyare the same

4.90 (1.33) .85 (.03) .27 (.02)

Brand loyalty (� = 90; VE = .76; ̨ = .90)I will choose XYZ hotels over and over again without hesitation 4.90 (1.43) .82 (–) .33 (.03)I feel good and positive when I think about staying at an XYZ hotel 5.20 (1.26) .90 (.02) .19 (.02)Thinking about the XYZ hotel makes me feel pleasant 5.17 (1.29) .89 (.02) .20 (.02)

Brand quality (� = 90; VE = .75; ̨ = .90)XYZ is of high quality 5.55 (1.08) .83 (–) .32 (.02)XYZ sets quality standards other hotels should follow 5.32 (1.17) .88 (.03) .23 (.02)I consider XYZ’s quality to be of the highest standard 5.17 (1.25) .89 (.03) .21 (.02)

Brand awareness (� = 92; VE = .79; ̨ = .92)I know what the XYZ symbol or logo looks like 5.24 (1.52) .84 (–) .29 (.03)I know what an XYZ hotel looks like 5.17 (1.35) .91 (.02) .17 (.02)I can easily recognize XYZ hotels among other competing hotels 5.16 (1.42) .91 (.02) .18 (.02)

Brand image (� = 89; VE = .72; ̨ = .89)The XYZ brand is prestigious 5.23 (1.20) .84 (–) .30 (.02)The XYZ hotels tend to attract sophisticated people as guests 5.17 (1.22) .86 (.03) .27 (.02)Staying at an XYZ hotel makes me feel special 5.05 (1.29) .86 (.03) .26 (.02)

Management trust (� = 92; VE = .79; ̨ = .92)I trust the XYZ management 5.43 (1.16) .88 (–) .23 (.02)The XYZ management knows how to do the hotel business 5.42 (1.15) .92 (.02) .16 (.01)The XYZ implements good management practices other hotels can learn 5.29 (1.20) .88 (.02) .23 (.02)

Brand reliability (� = 92; VE = .80; ̨ = .92)The XYZ will meet my expectations every time 5.25 (1.18) .92 (–) .15 (.01)My experience with XYZ will be consistent every time I stay 5.26 (1.16) .88 (.02) .23 (.02)The XYZ will not disappoint me every time 5.24 (1.22) .88 (.02) .23 (.02)

2 = 683.4, df = 168; CFI = .99; NNFI = .99; RMSEA = .048 (90% confidence interval = .045–.052).N y Forn

.

atwestlFw

esetadgTstwaarmr

Mp

ote: � = construct reliability and VE = amount of variance extracted, as suggested ba All items measured on a 7-point scale (1 = strongly disagree; 7 = strongly agree)b Standardized estimates, all statistically significant (p < .01).

cross the groups defined by the three cross-cultural factors. Struc-ural relations equivalence analysis like this is applicable where oneorks with an outright structural model and can extend to mod-

ration analysis for theory testing. The BE model (Fig. 1) had sixtructural relations, with brand loyalty as a complete mediator ofhe five BE constructs toward brand choice intention. This equiva-ence test was nested in the metric equivalence model (Milfont andischer, 2010; Steenkamp and Baumgartner, 1998), which in turnas nested in the configural equivalence model.

Table 5 provides summary results of the structural relationsquivalence model. Both configural and metric equivalence con-traints fit the two brand ID groups acceptably. The metricquivalence constraints, in particular, did not significantly worsenhe model fit (2 = 8.8, df = 14). The structural relations equiv-lence model also fit the data acceptably without significantlyeteriorating the model fit (2 = 10.1, df = 6). For the two lan-uage groups, all three equivalence models fit the data acceptably.he metric equivalence model did not undermine the model fitignificantly (2 = 38.3, df = 14). However, the structural rela-ions equivalence model implied that some structural coefficientsere not equal between the groups (2 = 29.1, df = 6). Finally,

ll three equivalence models also fit the three regional groupscceptably, although RMSEA was slightly inferior for the two moreestrictive models. The model fit did not get worse with either theetric equivalence restriction (2 = 47.7, df = 28) or structural

elations equivalence requirement (2 = 20.6, df = 12).Inequivalence of the structural model between the English- and

andarin-speaking groups suggested a need for disaggregated,arameter-specific equivalence tests. Table 5 provides further

ell and Larcker (1981) and Bagozzi and Yi (1988).

summary results of such tests under the language group results.Four of the six relationships appeared to be significantly differ-ent between the two language groups. Specifically, the effect ofbrand loyalty on brand choice intent was larger in the Mandarin-speaking travelers ( ̌ = .88) than in the English counterpart (.82)(2 = 199.9, df = 1). The Mandarin-speaking group also showeda significantly stronger brand reliability-brand loyalty relationship(.39) than the English-speaking group (.15). However, the rela-tionship between brand image and brand loyalty was stronger inthe English group (.42) than in the Mandarin group (.09) and thebrand trust-brand loyalty relationship followed the same directionof inequivalence (.32 vs. .17). The other two remaining relationshipswere equivalent between the two language groups.

6. Discussion and implications

The findings generally indicate that Hsu et al. (2012) hotel BEmodel may be generalizable across the three cultural customersegments, as evidenced in the acceptable metric as well as con-figural equivalences. The test results suggest that tourists interpretand respond to the measurement scales of the model in the samemanner regardless of the hotel brands patronized, their first lan-guage spoken, or their continental location of residence. This isparticularly true, despite that the cross-cultural tourist groups hadsignificantly heterogeneous sample distributions on five demo-

graphic variables (see Table 1). These findings stand against thesecular perception that people think and process information dif-ferently depending on such cross-cultural factors as in this study.A clearly defined target of evaluations like BE may elicit some

164 H. Oh, C.H.C. Hsu / International Journal of Hospitality Management 36 (2014) 156–166

Table 4Results of equivalence modeling.

2 df 2/df RMSEA NNFI CAIC CFI

Brand IDEquality of ˙(g)

xx and �(g)x 657.5 252 2.61 0.050 0.99 2717.5 1.00

Equality of ˙(g)xx 575.8 231 2.49 0.048 0.99 2464.1 1.00

Equality of �(g)x 726.7 21 34.60 0.230 0.46 4675.1 0.97

Independent Chinese brands 455.8 168 2.71 0.055 0.99 917.3 0.99Independent Foreign brands 433.3 168 2.58 0.046 0.99 913.2 0.99Configural equivalence 889.2 336 2.65 0.050 0.99 1919.2 0.99Metric equivalence 897.5 350 2.56 0.049 0.99 1813.1 0.99Scalar equivalence 1044.9 371 2.82 0.053 0.99 2132.1 0.99Partial scalar equivalence 897.5 365 2.46 0.047 0.99 2033.8 0.99Error variance equivalence 1074.7 371 2.90 0.054 0.99 1818.6 0.99Factor means model 982.8 364 2.70 0.051 0.99 2127.4 0.99Full factor variance/covariance equivalence 1140.5 392 2.91 0.054 0.99 2056.1 0.99Factor mean equivalence 1044.9 371 2.82 0.053 0.99 2132.1 0.99Partial factor mean equivalence 986.4 367 2.69 0.051 0.99 2106.3 0.99

LanguageEquality of ˙(g)

xx and �(g)x 608.0 252 2.41 0.052 0.99 2617.1 1.00

Equality of ˙(g)xx 507.6 231 2.20 0.047 0.99 2349.3 1.00

Equality of �(g)x 100.3 21 4.78 0.084 0.98 3951.1 1.00

Independent Mandarin version 583.3 168 3.47 0.061 0.99 1055.3 0.99Independent English version 276.2 168 1.64 0.040 1.00 717.7 1.00Configural equivalence 859.5 336 2.56 0.054 0.99 1864.0 0.99Metric equivalence 895.5 350 2.56 0.054 0.99 1788.4 0.99Scalar equivalence 1002.5 371 2.70 0.057 0.99 2062.8 0.99Partial scalar equivalence 895.5 365 2.45 0.052 0.99 2003.7 0.99Error variance equivalence 945.7 371 2.55 0.054 0.99 1671.2 0.99Factor means model 925.9 364 2.54 0.054 0.99 2042.0 0.99Full factor variance/covariance equivalence 1034.7 392 2.64 0.055 0.99 1927.6 0.99Factor mean equivalence 1002.5 371 2.70 0.057 0.99 2062.8 0.99Partial factor mean equivalence 930.3 367 2.53 0.054 0.99 2022.6 0.99

Region of originationEquality of ˙(g)

xx and �(g)x 1073.7 504 2.13 0.053 0.99 3109.9 1.00

Equality of ˙(g)xx 941.5 462 2.04 0.051 0.99 3317.0 1.00

Equality of �(g)x 85.9 42 2.05 0.051 0.99 5855.1 1.00

Independent Asia 611.5 168 3.64 0.057 0.99 1095.7 0.99Independent Europe 318.2 168 1.89 0.078 0.98 696.1 0.98Independent N. America 287.7 168 1.71 0.055 0.99 695.7 0.99Configural equivalence 1217.4 504 2.42 0.060 0.99 2744.6 0.99Metric equivalence 1263.8 532 2.38 0.059 0.99 2564.7 0.99Scalar equivalence 1381.4 574 2.41 0.060 0.99 2581.9 0.99Partial scalar equivalence 1293.9 554 2.34 0.060 0.99 2926.1 0.99Error variance equivalence 1462.7 574 2.55 0.063 0.99 2424.3 0.99Factor means model 1304.8 560 2.33 0.058 0.99 2888.4 0.99Full factor variance/covariance equivalence 1477.2 616 2.40 0.059 0.99 2608.4 0.99Factor mean equivalence 1381.4 574 2.41 0.060 0.99 2581.9 0.99Partial factor mean equivalence 1325.9 564 2.35 0.058 0.99 2877.2 0.99

Note: RMSEA, root mean square error of approximation; NNFI, non-normed fit index; CAIC, consistent Akaike Information Criteria; CFI, comparative fit index.

Table 5Results of equivalence tests of structural relationships.

Groups Models 2 df 2/df 2 df 2/df RMSEA CAIC

Chinese vs. foreign Configural equivalence 1099.9 346 3.2 .06 2048.2Metric equivalence 1108.7 360 3.1 8.8 14 0.6 .06 1942.5Structural relations equivalence 1118.8 366 3.1 10.1 6 1.7 .06 1903.6

Mandarin vs. English Configural equivalence 1007.1 346 2.9 .06 1931.9Metric equivalence 1045.4 360 2.9 38.3 14 2.7 .06 1858.6Structural relations equivalence 1074.5 366 2.9 29.1 6 4.9 .06 1839.9Loyalty → choice int. 1245.3 361 3.5 199.9 1 199.9 .07 2050.5Quality → loyalty 1045.5 361 2.9 0.1 1 0.1 .06 1849.7Awareness → loyalty 1045.7 361 2.9 0.3 1 0.3 .06 1850.9Image → loyalty 1077.2 361 3.0 31.8 1 31.8 .06 1882.4Trust → loyalty 1077.9 361 3.0 32.5 1 32.5 .06 1883.2Reliability → loyalty 1049.6 361 2.9 4.2 1 4.2 .06 1854.9

Asia vs. Europe vs. N. America Configural equivalence 1419.1 519 2.7 .06 1931.9Metric equivalence 1466.8 547 2.7 47.7 28 1.7 .07 2646.4Structural relations equivalence 1487.3 559 2.7 20.6 12 1.7 .07 2570.0

Note: For all models, NNFI = .99 and CFI = .99.

f Hosp

gbtmiftb

tgmsmllcmiaioma

arit(barptbegclaopeo

vTtBodosermcitw

clwE

H. Oh, C.H.C. Hsu / International Journal o

eneralizable fashions of information processing across culturalorders. A theoretical implication is that theories developedhrough Hsu et al.’s model may be applicable to broad cultural seg-

ents examined in this study. Practically, brand managers of bothnternational and Chinese domestic hotels may rely on the modelor gauging the equity value of their brands with such diverse cus-omer groups. That is, culture-specific models of hotel BE may note necessary.

The hotel BE model may face challenges in generalizing beyondhe configural and metric equivalences across the cultural touristroups. The results generally supported both the configural andetric generalizability of the model, but a number of parameter-

pecific differences were evident in the observed item means,easurement errors, latent factor variances and covariances, and

atent factor means. Such partial differences were comparativelyess salient in both the configural and metric equivalence matri-es. For theoretical reasons, complete metric generalizability of theodel is highly desirable because it becomes a baseline model

n which other equivalence models are nested or built, therebyllowing legitimate comparisons across groups. Practically speak-ng, metric equivalence provides efficiency, that is, more degreesf freedom or fewer parameters to calculate, in cross-cultural BEodeling. It also requires less arduous endeavor in interpreting

nd comparing BE rated by customers from different cultures.Theoretical relations among the BE constructs seem generaliz-

ble in selected cross-cultural contexts. In many substantive areas,esearchers work with well-defined causal models, and often ofnterest to them is generalizability of the posited theoretical rela-ionships of the models. The structural relations in Hsu et al.’s model2012) seemed equal in both pattern and strength across the tworand ID and three regional groups. However, such omnibus equiv-lence was not tenable in the two language groups, although someelations could still be equivalent. We did not pursue extensiveairwise comparisons of the parameters across the brand ID andhree regional groups because the entire structural model could note said unequal among the groups. We instead attempted to testquivalence of each relationship parameter between the two lan-uage groups for illustrative purposes to see which relationship(s)ontributed to the structural inequivalence (see Table 5). Neverthe-ess, such an ad hoc procedure could rely on chance outcomes in thebsence of a priori theoretical exposition. As long as theoreticallyr practically justifiable, researchers should conduct pairwise com-arisons and test parameter-specific equivalences. Doing so willnrich theoretical development as well as practical understandingf the cross-cultural applicability of the structural relations.

In sum, the hotel BE model proposed by Hsu et al. (2012) pro-ides a fertile ground for future BE research and industry practice.he model demonstrated a relatively robust measurement struc-ure across the three sets of cultural factors. That is, the sevenE constructs could be defined and measured with the same setsf scales in the cross-cultural situations. Moreover, visitors withifferent cultural backgrounds seem to share ways to expressr perceive their underlying BE-related notions as shown in theatisfactory metric equivalences. While additional cross-culturalxaminations are desirable, this study implies that highly focusedesearch models like the hotel BE model analyzed in this studyay induce the same evaluative schema universally shared among

onsumers regardless of cultural heterogeneities. Practically speak-ng, multinational hotel operators may use the hotel BE model forracking and benchmarking hotel BE in their international venturesithout being overly discouraged by potential cultural differences.

Cross-cultural research, and research involving multigroup

omparisons in general, needs to examine measurement equiva-ence prior to comparing the groups. The conclusions of a study

ithout evidence of metric equivalence may be weak (Horn, 1991).stablishing cross-cultural instrument equivalence is not “a matter

itality Management 36 (2014) 156–166 165

of translating the instrument from one language into another. . .theprocess extends far beyond the issue of translation [and backtranslation] and involves a comprehensive and rigorous series ofprocedures that test for the validity of the measure’s scores withinthe new cultural context, as well as for its structural and measure-ment equivalence with the original instrument and culture” (Byrne,2008, p. 874). This study attempted to synthesize the equivalenceanalysis literature in a non-technical manner and illustrate howone could implement a series of equivalence tests using a hotelBE model. Application examples are yet rare in the hospitality andtourism literature, and this study’s illustration may be helpful tomany upcoming studies of not only cross-cultural comparisonsbut also cross-cultural theory and scale developments. Additionalefforts are necessary to broaden hospitality and tourism theoriesacross cultural and other borders through rigorous assessment pro-cedures as illustrated in this study.

7. Conclusion

This study illustrated an equivalence analysis procedure. Theequivalence analysis literature is at times highly technical and con-ceptual. This study focused on illustrating a streamlined procedurein a less technical manner by using a recently proposed hotel BEmodel. The findings provide empirical evidence for the model’svalidity across the three cultural factors, although cautions arenecessary when trying to generalize the hotel BE model beyondmetric equivalence. While future research needs to provide strongtheoretical justifications for such metric equivalence, this studyadds a useful piece of information to the BE literature, especiallybecause cross-cultural validation efforts for previously proposed BEmodels are practically non-existent in the hospitality and tourismdiscipline. The methodological illustration also encourages moresystematic, routine applications in many topical areas of hospi-tality and tourism research so as to advance theoretical progressand practical utility of models and theories. The procedure isequally applicable to theory building research in that it is generallyinstrumental to both testing moderating effects and determiningboundary conditions of theories.

Several issues draw attention in interpreting the findings of thisstudy. First, as of today the equivalence analyses into hotel BE arerather methodologically motivated than theoretically guided duelargely to the lack of relevant theoretical developments. Rigorousconceptualizations of various equivalences are desirable, althoughempirical evidence associated with newly developed instrumentslike the one examined in this study often accumulates into strongtheoretical knowledge over time (Chin et al., 2003). Researchersneed to give more considerations to the international nature of thehospitality business and attempt to provide additional conceptualunderpinnings to address potential demands for generalizability oftheir propositions across various “borders.”

Second, interpretation of the resulting fit indices in equivalenceanalyses necessitates both global and local judgments. While manymethodologists commonly prescribe the use of multiple indicesand often some specific cutoff points for each index, global modelfit indices do not point to which equality constraints are eitheracceptable or unacceptable. Large model size (i.e., a large numberof parameters), coupled with large sample size, often makes theincremental fit indices insensitive to differences at the individualparameter level. In contrast, the chi-square difference test is sen-sitive to sample sizes in the same way as the likelihood ratio andchi-square tests are. This is a gray area. Judicious judgments and

conclusions are necessary on fit indices with careful considerationsgiven to study-specific goals.

Finally, the range of generalizability is subject to the group-ing variables under investigation. There is an infinite number of

1 f Hosp

giosphcrir

R

A

AA

AA

B

B

B

B

B

B

B

B

C

C

C

C

D

D

F

F

G

H

H

H

H

H

66 H. Oh, C.H.C. Hsu / International Journal o

rouping variables in cross-cultural research settings and exam-ning them all is unrealistic. Nonetheless, a relatively small setf variables is likely of substantive interest in most cases. Thistudy employed three far-reaching factors that could broadly definelausible cross-cultural differences. Unlike popular perceptions,owever, these three variables appeared not to interfere with bothonfigural and metric generalizability of the BE model. Futureesearch may build upon this study to include other potentiallynfluential cross-cultural factors, desirably based on a priori theo-etical justifications.

eferences

aker, D.A., 1991. Managing Brand Equity: Capitalizing on the Value of a Brand Name.Free Press, New York.

aker, D.A., 1996. Building Strong Brands. Free Press, New York.hmed, Z.U., Johnson, J.P., Chew, P.L., Tan, W.F., Ang, K.H., 2002. Country-of-origin

and brand effects on consumers’ evaluations of cruise lines. International Mar-keting Review 19 (2/3), 279–302.

kaike, H., 1987. Factor analysis and AIC. Pschychometrika 52 (3), 317–332.nderson, J.C., Gerbing, D.W., 1988. Structural equation modeling in practice:

a review and recommended two-step approach. Psychological Bulletin 103,411–423.

agozzi, R.P., Yi, Y., 1988. On the evaluation of structural models. Journal of theAcademy of Marketing Science 16 (1), 74–94.

ailey, R., Ball, S., 2006. An exploration of the meanings of hotel brand equity. TheService Industries Journal 26 (1), 15–38.

entler, P.M., 2005. EQS 6 Structural Equations Program Manual, Retrieved fromhttp://www.mvsoft.com (20.08.11).

oo, S., Busser, J., Baloglu, S., 2009. A model of customer-based brand equity and itsapplication to multiple destinations. Tourism Management 30 (2), 219–231.

yrne, B.M., 2001. Structural Equation Modeling with AMOS: Basic Concepts, Appli-cations, and Programming. Erlbaum, Mahwah, NJ.

yrne, B.M., 2006. Structural Equation Modeling with EQS: Basic Concepts, Applica-tions, and Programming, 2nd ed. Erlbaum, Mahwah, NJ.

yrne, B.M., 2008. Testing for multigroup equivalence of a measuring instrument: awalk through the process. Psicothema 20 (4), 872–882.

yrne, B.M., Watkins, D., 2003. The issue of measurement equivalence revisited.Journal of Cross-cultural Psychology 34, 287–321.

armines, E.G., McIver, J.D., 1981. Analyzing models with unobserved variables:analysis of covariance structures. In: Bohinstedt, G.W., Borgatta, E.F. (Eds.), SocialMeasurement: Current Issues. Beverly Hills, CA, Sage, pp. 65–115.

han, H., Wan, L.C., Sin, L.Y., 2009. The contrasting effects of culture on consumertolerance: interpersonal face and impersonal fate. Journal of Consumer Research36 (2), 292–304.

hin, W.W., Salisbury, W.D., Gopal, A., Newsted, P., 2003. Authors’ reply: whenshould data overrule theory in scale development? Information SystemsResearch 14 (4), 360–363.

hung, J.Y., Kyle, G.T., Petrick, J.F., Absher, J.D., 2011. Fairness of prices, user fee policyand willingness to pay among visitors to a national forest. Tourism Management32 (5), 1038–1046.

eng, X., Doll, W., Hendrickson, A.R., Scazzero, J.A., 2005. A multi-group analysis ofstructural invariance: an illustration using the technology acceptance model.Information & Management 42, 745–759.

oll, W.J., Hendrickson, A., Deng, X., 1998. Using Davis’s perceived usefulness andease-of-use instruments for decision making: a confirmatory and multigroupinvariance analysis. Decision Sciences 29 (4), 839–869.

orgas-Coll, S., Palau-Saumell, R., Sánchez-García, J., Callarisa-Fiol, L.J., 2012. Urbandestination loyalty drivers and cross-national moderator effects: the case ofBarcelona. Tourism Management 33 (6), 1309–1320.

ornell, C., Larcker, D.F., 1981. Evaluating structural equation models with unob-servable variables and measurement error. Journal of Marketing Research 18(1), 39–50.

ao, F., 2006. Language is culture – on intercultural communication. Journal ofLanguage and Linguistics 5 (1), 58–67.

allak, R., Brown, G., Lindsay, N.J., 2012. The place identity – performance relation-ship among tourism entrepreneurs: a structural equation modeling analysis.Tourism Management 33 (1), 143–154.

an, C., Terpstra, V., 1988. Country-of-origin effects for uni-national andbi-national products. Journal of International Business Studies 19 (2),235–255.

an, H., Hsu, L., Sheu, C., 2010. Application of the theory of planned behavior to greenhotel choice: testing the effect of environmental friendly activities. TourismManagement 31 (3), 325–334.

orn, J.L., 1991. Comments on issues in factorial invariance. In: Collins, L.M., Horn,J.L. (Eds.), Best Methods for the Analysis of Change. American PsychologicalAssociation, Washington, DC, pp. 114–125.

orn, J.L., McArdle, J.J., 1992. A practical and theoretical guide to measurementinvariance in aging research. Environmental Aging Research 18 (3/4), 117–144.

itality Management 36 (2014) 156–166

Hsu, C.H., Oh, H., Assaf, G., 2012. A customer-based brand equity model for upscalehotels. Journal of Travel Research 51 (1), 81–93.

Hu, L., Bentler, P.M., 1999. Cutoff criteria for fit indexes in covariance structure analy-sis: conventional criteria versus new alternatives. Structural Equation Modeling6 (1), 1–55.

Huang, S., Hsu, C.H., Chan, A., 2010. Tour guide performance and tourist satisfac-tion; a study of the package tours in Shanghai. Journal of Hospitality & TourismResearch 34 (1), 3–33.

Jöreskog, K.G., Sörbom, D., 2006. LISREL 8.8: Structural Equation Modeling with theSIMPLIS Command Language. Scientific Software International, Chicago, IL.

Kayaman, R., Arasli, H., 2007. Customer based brand equity: evidence from the hotelindustry. Managing Service Quality 17 (1), 92–109.

Keller, K.L., 1993. Conceptualizing, measuring, and managing customer-based brandequity. Journal of Marketing 57 (1), 1–22.

Keller, K.L., 2003a. Strategic Brand Management: Building, Measuring, and ManagingBrand Equity, 2nd ed. Upper Saddle River, NJ, Prentice Hall.

Keller, K.L., 2003b. Brand synthesis: the multidimensionality of brand knowledge.Journal of Consumer Research 29 (4), 595–600.

Kish, P., Riskey, D.R., Kerin, R.A., 2001. Measurement and tracking of brand equity inthe global marketplace: the Pepsico experience. International Marketing Review18 (1), 91–96.

Kline, R.B., 2010. Principles and Practice of Structural Equation Modeling, 3rd ed.Guilford, New York.

McCleary, K.W., Weaver, P.A., Hsu, C.H., 2006. The relationship between interna-tional leisure travelers’ origin country and product satisfaction, value, servicequality, and intent to return. Journal of Travel & Tourism Marketing 21 (2/3),117–130.

Milfont, T.L., Fischer, R., 2010. Testing measurement invariance across groups:applications in cross-cultural research. International Journal of PsychologicalResearch 3 (1), 111–121.

Kim, D., Wen, L., Doh, K., 2010. Does cultural difference affect customer’s response ina crowded restaurant environment? A comparison of American versus Chinesecustomers. Journal of Hospitality & Tourism Research 34 (1), 103–123.

Kim, H., Kim, W.G., 2005. The relationship between brand equity and firms’ per-formance in luxury hotels and chain restaurants. Tourism Management 26 (4),549–560.

Kim, W.G., Jin-Sun, B., Kim, H.J., 2008. Multidimensional customer-based brandequity and its consequences in midpriced hotels. Journal of Hospitality &Tourism Research 32 (2), 235–254.

Line, N.D., Runyan, R.C., 2012. Hospitality marketing research: recent trends andfuture direction. International Journal of Hospitality Management 31 (3),477–488.

Mattila, A.S., 2000. The impact of culture and gender on customer evaluations ofservice encounters. Journal of Hospitality & Tourism Research 24 (2), 263–273.

Motameni, R., Shahrokhi, M., 1998. Brand equity valuation: a global perspective.Journal of Product & Brand Management 7 (4), 275–282.

Ployhart, R.E., Oswald, F.L., 2004. Applications of mean and covariance structureanalysis: integrating correlational and experimental approaches. OrganizationalResearch Methods 7 (1), 27–65.

Prasad, K., Dev, C.S., 2000. Managing hotel brand equity: a customer-centric frame-work for assessing performance. Cornell Hotel and Restaurant AdministrationQuarterly 41 (3), 22–31.

Rock, D.A., Werts, C.E., Flaugher, R.L., 1978. The use of analysis of covariance struc-tures for comparing the psychometric properties of multiple variables acrosspopulations. Multivariate Behavioral Research 13 (October), 403–418.

Shanahan, K.J., Hyman, M.R., 2007. An exploratory study of desired hotel attributesfor American tourists vacationing in China and Ireland. Journal of Vacation Mar-keting 13 (2), 107–118.

Shimp, T.A., Sharma, S., 1987. Consumer ethnocentrism: construction and validationof the CETSCALE. Journal of Marketing Research 24 (3), 280–289.

So, K.K.F., King, C., 2010. When experience matters: building and measuring hotelbrand equity-the customers’ perspective. International Journal of ContemporaryHospitality Management 22 (5), 589–608.

Steenkamp, J.-B.E.M., Baumgartner, H., 1998. Assessing measurement invariance incross-cultural consumer research. Journal of Consumer Research 25, 78–90.

Tsai, H., Cheung, C., Lo, A., 2010. An exploratory study of the relationship betweencustomer-based casino brand equity and firm performance. International Jour-nal of Hospitality Management 29 (4), 754–757.

Vandenberg, R.J., Lance, C.E., 2000. A review and synthesis of the measurementequivalence literature: suggestions, practices and recommendations for orga-nizational research. Organizational Research Methods 3, 4–70.

van de Vijver, F., Leung, K., 1997. Methods and Data Analysis for Cross-culturalResearch. Sage Publications, Thousand Oaks, CA.

Wall, M., Heslop, L.A., 1986. Consumer attitudes toward Canadian-made versusimported products. Journal of the Academy of Marketing Science 14 (2), 27–36.

Wheaton, B., Muthen, B., Alwin, D.F., Summers, G.F., 1977. Assessing the reliabilityand stability in panel models. In: Heise, D.R. (Ed.), Sociological Methodology.Jossey-Bass, San Francisco, pp. 84–136.

Xu, J.B., Chan, A., 2010. A conceptual framework of hotel experience and customer-based brand equity: some research questions and implications. InternationalJournal of Contemporary Hospitality Management 22 (2), 174–193.

Yoo, B., Donthu, N., 2002. Testing cross-cultural invariance of the brand equity cre-ation process. Journal of Product & Brand Management 11 (6), 380–398.