Image Variables in Multi-Attribute ProductEvaluations: Country-of-Origin Effects

GARY M. ERICKSONJOHNY K. JOHANSSONPAUL CHAO*

An empirical investigation is conducted to determine the effects of image variableson beliefs and attitudes in the multi-attribute model framework. Simultaneous equa-tion regression is used to estimate a model linking a particular type of imagevariable, country of origin, to attitudes and beliefs obtained through a survey ofevaluations of automobile alternatives. The results indicate that country of originaffects beliefs but not attitudes.

T he relationship between affect and cognition in theformation of product evaluations continues to be

an area of interest to researchers of consumer behavior.In particular, whether attitude is mediated by cognitionis a subject that is open to debate. One view is that beliefsregarding a product's attributes precede and are respon-sible for the formation of attitude toward the product(Fishbein and Ajzen 1975). Various researchers (cf Cohenand Houston 1972) have questioned this one-way viewand have investigated the possible influence in the otherdirection (affect to cognition) as well. Zajonc (1980) arguesthat cognition may not even be necessary for the for-mation of affective judgments.

A related area concems external influence on the for-mation of beliefs and attitudes. Various studies haveshown the direct effects on perception of communication(Holbrook 1978; Mitcheii and Olson 1981; Toy 1982)and of physical characteristics ofthe product (Holbrook1981; Tybout and Hauser 1981). Other studies have beenconcerned with direct effects on attitudes—in particular,studies by Matlin (1971) and by Moreiand and Zajonc(1979) have shown the influence on affect of subjectiveas well as objective familiarity.

The present study considers the influence of imagevariables on the formation of beliefs and attitudes. Animage variable is defined as some aspect of the productthat is distinct from its physical characteristics but thatis nevertheless identified with the product. Examples ofimage variables include brand name, symbols used in

• Gary M. Erickson is Assistant Professor. Johny K. Johansson isProfessor, and Paul Chao is a Ph.D. candidate, all in the Depanmentof Marketing and International Business, University of Washington,Seattle, WA 98195. Partial funding was provided by the Pacific RimProject at the Graduate School of Business Administration, Universityof Washington.

advertising, endorsement by a well-known figure, andcountry- of origin for markets in which imported brandshave a significant presence. The present paper considersa particular image variable—country of origin—and an-alyzes its effects on the evaiuation of automobile brands.

THE BELIEF-ATTITUDERELATIONSHIP

Probably the most familiar model linking beliefs andattitudes is the Fishbein model, in which attitude is de-termined by beliefs (Fishbein and Ajzen 1975). Zajonc(1980) argues that this may not be the appropriate wayto view the relationship. Zajonc cites a number of ref-erences to support his contentions that affective reactionsare primary, basic, inescapable, irrevocable, and difficultto verbalize, and that affective judgments involve the self,can become separated from content, and need not involvecognition. Moreover, the possibility of a halo effect(Beckwith and Lehmann 1975; Holbrook 1983) suggeststhat causation could proceed in the other direction—i.e.,from attitude to beliefs.

Empirical resolution of these issues requires an esti-mation procedure that allows for possible two-way cau-sation between beliefs and attitude. If latent variables arenot involved, simultaneous equation regression (Johnston1972) is appropriate. (If it is important to assume latentvariables, a technique such as LISREL is needed; seeJoreskog and Sorbom 1982). Simultaneous equationregression has been applied in previous research to helpson out the directions of influence. The empirical resultsof Beckwith and Lehmann (1975) and Holbrook (1983)show that beliefs may indeed influence attitude. Thesestudies also show that attitude can have a halo effect onat least some beliefs. This empirical evidence suggests theneed for simultaneous models in empirical research in-

© JOURNAL OF CONSUMER RESEARCH • Vol. 11 • Seplember 1984

COUNTRY-OF-ORIGIN EFFECTS 695

volving attitudes and beliefs—not models in which in-fluence is directed only one way.

IMAGE EFFECTSAccording to Fishbein and Ajzen (1975, Chapter 5),

there are three kinds of beliefs: descriptive, inferential,and informational. These beliefs are formed in differentways, and all potentially contribute toward what a cus-tomer believes about a product's attributes. Descriptivebeliefs derive from direct experience with the product.The works by Holbrook (1981) and Tybout and Hauser(198l)relate to this kind ofbeiief, in that they link physicalcharacteristics with product perceptions. Informationalbeliefs are those influenced by outside sources of infor-mation such as advertising, friends, relatives, and so on.The research on communication effects (Holbrook 1978;Mitchell and Olson 1981; Toy 1982) is relevant here.

It is the remaining type of belief—inferential—that isof interest in this paper. This type of belief is formed bymaking inferences {correctly or incorrectly) based on pastexperience as this experience relates to the current stim-ulus (Fishbein and Ajzen 1975). For example, beliefsabout a new family-branded product may be influencedby prior experience with other products possessing thesame family brand. Or a person whose experience suggeststhat German cars are durable might infer that since anAudi is a German car, an Audi is a durable car.

As these examples indicate, image variables may haveinferential effects on product beliefs. Little work has beendone in this area. While Huber and McCann (1982) haveshown that inferences can affect how people evaluateproducts, we need empirical work studying inferentialprocesses. In particular, we need to know whether infer-ences are made based on country of origin, brand name,or other image variables.

An image variable could bave direct influence on at-titude, as well. For example, a brand name may provokean emotional reaction which carries over to attitude to-ward the brand- Or a customer may have a bias againsta foreign country that has affective implications for prod-ucts from that country. An empirical study investigatingthe effects of image variables on the product evaluationprocess should allow for direct effects on attitude as wellas on beliefs.

Previous research regarding image variables has gen-erally involved their influence on perceived quality orsome other overall evaluation. Jacoby, Olson, and Had-dock (1971) show that brand name can affect productquality ratings. Szybillo and Jacoby (1974) indicate thatperceived quality can be influenced by store image. Boththese studies control for actual product differences.

Various studies (Bilkey and Nes 1982 provide a review)reveal that a product's country of origin can affect itsevaluation, although generally only single-cue modelshave been used (Bilkey and Nes 1982, p. 93). Many studieshave been concerned with price as an indicator of quality(see Monroe 1973 for a review). However, price appears

to have a positive effect on perceived quality only whenit is the single cue; other image variables, when includedwith price, become more important quality-cue indicators(Zaltman and Wallendorf 1983, Chapter 11). The presentpap)er takes a more detailed look at the product evaluationprocess, investigating the existence of image effects withinthe belief-attitude relationship.

MODELDue to the possibility of two-way influence between

attitudes and beliefs, a system of simultaneous equationsis needed to represent the relationship, including oneequation in which attitude is the dependent variable, andone equation for each attribute in which the belief ratingfor the level of that attribute is the dependent variable.Attitude is an explanatory variable in each belief equation,and the beliefs are explanatory variables in the attitudeequation. Because ofthe possibility that image variablesmight affect attitude as well as beliefs, these variablesbecome explanatory variables in each equation in thesystem.

Two other considerations enter into the developmentof the model. One is that beliefs about the level of aparticular attribute for a product altemative should de-pend upon the true level for that altemative, the truelevel having its effect through direct experience or com-munication. For the product area chosen for study (au-tomobiles), objective values for many of the attributesare available as published information. These objectiveattribute values are entered as explanatory variables inthe belief equations. In this way. image influences onbeliefs can be viewed as biases, since the effects of thetrue values are controlled for in the estimation.

Another consideration is that previous research on theattitude relationship (Matlin 1971; Moreiand and Zajonc1979) indicates that affect is influenced by both subjectivefamiliarity (the subjects think they are familiar with thestimulus) and objective familiarity (actual exposure tothe stimulus). Thus two variables representing these effectsare included as explanatory variables in the attitude equa-tion: self-assessed familiarity with the alternative, andactual ownership as a measure of objective familiarity.

The model can be stated symbolically:

= ao + 2 + +n

+

for /c = 1, 2 , . . . , A", where:

Aij ^ subject /'s {i = 1, 2, . . . , / ) attitude towardaltemative y (_; = 1 , 2 , . . . , / )

B^ji, = Ts belief about the level of attribute k foralternative j

Fn - /'s subjective familiarity toward j

696 THE JOURNAL OF CONSUMER RESEARCH

FIGURE

BELIEF-ATTITUDE MODEL WITH IMAGE EFFECTS

IMAGEVARIABLES

TRUELEVELS

L >

N,j = 1 if / owns an alternative j , 0 if not

Ijh = value of image variable h for alternative j

Tjk = true value for attribute k of altemative j .

The terms u,yo, ",>* are error terms, assumed uncor-related across equations and subjects, and the parametersoi<i.<^k, 5o, 5|, XoA> cnk,&k,yk, Kh are to be estimated bysimultaneous equation regression (Johnston 1972, Chap-ters 12 and 13). In particular, estimates ofthe Xo/, andX̂A will help sort out where the influence of image variablesis felt in the belief-attitude relationship. The model isexhibited graphically in the Figure.

DATATo estimate the model, survey data were gathered from

96 MBA students at the University of Washington con-ceming their attitudes toward, beliefs about, and famil-iarity with certain automobiles. An advantage of usingautomobiles is that the country of origin has become animportant factor in this market, with imports gaining asignificant share. Also, country of origin is relatively easyto identify for this product class.

Two pilot studies were conducted on subsamples ofthe larger sample to elicit automobile makes and modelswhich this group would generally include in a set forfurther consideration. Another purpose ofthe pilot studieswas to identify an initial list of attributes that these peoplesee as salient in their evaluation of cars. The final list ofstimulus objects (automobile models) consisted of fourU.S., two German, and four Japanese models.' This meant

' One "German" car. the VW Rabbit, is actually assembled in Penn-sylvania and was classified as a U.S. model in the data analysis. Altemativeestimation was done, classifying the Rabbit as a German auto. Theresults, in temis of statistical significance of parameter estimates, showedno essential differences.

SUBJECTIVE &OBJECTIVE

FAMILIARITY

that two country-of-origin dummy (0-1) variables wereneeded as image variables—one was created for Germanautos and another for Japanese cars. The 10 models se-lected for the final list were as follows: VW Rabbit, FordMustang, Honda Accord, Chevrolet Citation, Datsun 200SX, Audi 4000, Mazda 626, BMW 320i, Toyota Celica,and Plymouth Reliant.

The final questionnaire began with a brief introductionexplaining the purpose of the study. This was followedby a section asking the respondents for their beliefs aboutattribute levels for all 10 autos. Semantic differential scaleswere used for all rating questions throughout the ques-tionnaire. Respondents were asked to record a rating witha slash through a horizontal line connecting the categories,to provide a continuous nature to the rating.

The questionnaire then elicited importance weights forthe attributes to be used in the selection of a smaller setof attributes considered most important by the sample.The respondents were asked to rate their familiarity witheach auto, after which they provided an overall rating ofeach auto on a five-point semantic scale (with continuousresponses allowed between categories). The final questionsinvolved autos presently owned and a few backgroundvariables. The time taken to complete the questionnaireaveraged about 25 minutes.

As mentioned before, the self-reported importanceweights were used to select a smaller set of importantattributes. Those selected for the empirical analysis were:price, gas mileage, reliability, durability, and workman-ship. True values were obtained when possible from pub-lished sources. Gas mileage and price were coded fromConsumer Reports and Car & Driver. Reliability wasidentified in terms of predicted repair incidence, valuesfor which were obtained from Consumer Reports. Todefine the attributes for which no available ratings couldbe obtained (durability and workmanship, and reliabilityfor the Plymouth Reliant), we relied upon the mean re-

COUNTRY-OF-ORIGIN EFFEOTS 697

sponses of "experts" from the sample. For each auto, fiverespondents were chosen who were closest in their eval-uations of price, gas mileage, and reliability (only the firsttwo attributes for the Plymouth Reliant) to the publishedratings.^

MODEL ESTIMATIONBefore estimation of the model, the responses were

standardized within each individual. Since variations oc-cur among individuals because of idiosyncratic scale use(e.g., differing zero points), standardizing was deemedappropriate. This is of course also the common strategyin work on muUi-attribute models (cf. Bass and Wilkie1973).

Since the plan was to pool the respondents for esti-mation, some care was required to justify this empiricalapproach. Fisher (1970) and Bass and Wittink (1975)suggest a single equation 7^-test for pooling based on com-parison of parameter estimates from two different singleequation regressions. This was applied in the present studyby comparing ordinary least squares (OLS) esimates ofthe affect equation based on different subsamples. Indi-viduals were separated from the rest ofthe sample, regres-sions were run with them excluded, and these regressionswere compared with those that included the individuals.

The selection of individuals to test for exclusion fromthe sample was determined on the basis ofthe self-reportedattribute importance ratings. The aim was to identifyrespondents whose importance ratings deviated most fromthe average. The importance ratings of each individualwere correlated with the overall sample mean ratings.Low or negative correlations indicate "outliers" that per-haps do not belong with the rest of the group for theempirical analysis.

Tests were run sequentially, an individual beingdropped for each run (the individual remaining with thelowest correlation). The exercise was stopped when non-significant differences, at the 0.05 level, were noted. Theresult of the sequential /^-tests was that one individualwas dropped from the analysis—the only individuai whoseimportance ratings correlated negatively with the overallsample. This left a total p>ooled sample of 95 respondents.The data are summarized in Table 1, which lists meansand standard deviations, and in Table 2, which showsPearson product-moment correlations.

Strong multicollinearity exists among the belief vari-ables, and this can seriously hamper the estimation. Tocorrect it, a principal components analysis was used to

^ An altemative method of obtaining the missing ratings was to askthose who expressed high levels of familiarity with particular models.This resulted in values that differed in many cases from those obtainedthrough the use of "experts" for the durability and workmanship ratings.These ratings produced less desirable results than those reported in thesubsequent section, in particular, "true" durability based on familiarityshowed a negative effect on beliefs (significant at the 0.10 level). As aconsequence, these results are not rer>orted further. Contact the firstauthor for details, if desired.

TABLE 1

MEANS AND STANDARD DEVIATIONS

Variable

AttitudePrice beliefMileage beliefReliability beliefDurability beliefWorkmanship beliefFamiliarityOwnershipJapanGermanyTrue priceTrue mileageTrue reliabilityTrue durabilityTrue workmanship

Mean

JSW.060.000

-.085.011

mdMS.400.200

8.74024.7003.9003.7003.900

Standarddeviation

1.0001.0001.000

.991

.9691.045

.995

.263

.490

.4002.1053.2281.446

.900

.944

identify uncorrelated factors. This analysis showed a dis-tinct reduction in incremental variance explanation asthe number of factors exceeded two. Eigenvalues were2.49, 1.24, 0.55, 0.39, and 0.32; the first two componentsexplained 74.6 percent ofthe total variance.

The two most important factors show distinct patternsof correlation with the belief variables. One factor isstrongly and positively correlated with four of the beliefvariables—reliability, durability, workmanship, andprice—and consequently could be referred to as a "qual-ity" factor. The other significant factor shows strong pos-itive correlation with gas mileage and negative correlationwith price, so that it appears to be an "economy" factor.The two factors were defined specifically by combiningthe appropriate subsets of belief variables, standardizedand weighted by factor score coefficients:

Quality factor - 0.247 X Price + 0.313 X Reliability+ 0.344 X Durability + 0.350X Workmanship

Economy factor = 0.727 X Mileage - 0.483 X Price

These factors were then used in place ofthe five originalbelief variables in the estimation ofthe system of equa-tions. Of course, to estimate this system by OLS regressionresults in estimates that are inconsistent (Johnston 1972,p. 351). Because of this, the system was estimated bytwo-stage least squares (Johnston 1972, p. 380 ff.). TheTSLS estimates are shown in Table 3, with asymptotic/-statistics.

A number of interesting results are noted. For one, astrong mutual relationship between attitudes and beliefsemerges. There appears to be both a forward eifect ofbeliefs on attitudes as well as a strong halo effect of attitudeflowing back to beliefs. And as expected, the estimationresults show significant effects of exogenous variables.Familiarity—although not ownership (a weak measureof exposure)—affects attitude directly. Also, true attribute

698 THE JOURNAL OF CONSUMER RESEARCH

TABLE 2

CORRELATIONS

M W O TP TM TR TD TW

Attitude (A)Price belief (P)Mileage belief (M)Reliability belief (R)Durability belief (D)Workmanship belief (W)Familiarity (F)Ownership (0)Japan (J)Germany (G)True price (TP)True mileage (TM)True reliability (TR)True durability (TD)True workmanship (TW)

1.000.458.179.548.651.710.455.089.150.354.422.175.390.474.567

1 000-.269

.246

.405

.467

.075

.065-.096

.632

.669-.096

.404

.527

.471

TABLE 3

PARAMETER

1.000.180.076.144.224

-.040.273

-.307-.267

.477-.031-.021

.171

ESTIMATES

1.000.579.567.281.029.243.105.231.124.303.332.337

1.000.648.305.042.059.333.383.100.304.379.425

1.000.288.062.189.402.498.168.467.527.593

1.000.099.062

-.023-.026

.145

.009

.110

.194

1.000-.077

.078

.047-.011-.027-.034-.004

garding annegatively

1.000-.408

.091

.329

.621

.499

.303

auto's

1.000.798

-.263.380.444.583

1.000-.087

.721

.782

.672

. quality, butto perceptions; of a

vooo.122 1.000.072 .899 1.000.319 .799 .789 1.000

a factor that contributescar's economy.

Equation/variableCoefficientestimate f-value

Dependent variable:Attitude

Ouaiity factorEconomy factorFamiliarityOwnershipJapanGermanyConstant

Dependent variable:Ouaiity factor

AttitudeTrue priceTrue reliabilityTnje durabilityTrue workmanshipJapanGermanyConstant

Dependent variable:Economy factor

AttitudeTrue pnceTrue mileageJapanGermanyConstant

.869

.255

.152

.109

.007

.264-.064

.658

.079-.052

.146

.007

.056

.143-1.099

.246-.343

.074

.501

.312

.903

8.66'5.26'4.20'1.38

.081.44- .69

10 74 '2.50'-.922.28'.10.49.75

-3 .62 '

4.71"-11.72"

8.47'5.51'1.67"3.07'

* Slgnincant al the 0 05 level." SignifiCBnt at the 0 10 IBV«.

levels influence beliefs. (Only the appropriate attributes,based on the components of each factor, were includedin each belief equation.) Note in particular the dual func-tion of actual price—a positive determinant of belief re-

Finally, the effect of the image variable, country oforigin, appears to have direct effects on beliefs and noton attitudes. An interesting conclusion is that qualityperceptions are not affected; these are quite well explainedby true price and durability as well as by the affectivehalo bias. Yet beliefs about a car's economy apparentlyare biased by the country-of-origin image, Japanese carshaving somewhat more of an advantage than Germanautos.

It must be emphasized that these conclusions aboutthe relative effects of image variables on beliefs and at-titudes should be viewed in proper perspective. Only oneparticular product class (automobiles) and one imagevariable (country of origin) have been investigated. Theremay be other situations in which image affects attitudedirectly.^

CONCLUSIONThis study of the influence of image variabies on the

product evaluation process indicates that an image vari-able does not appear to be affective in nature: the empiricalresults show that such variables influence belief formationrather than attitude. While previous work has shown thatphysical characteristics and communication—as well asoverall attitude—affect beliefs, the present study showsthat image variables also affect beliefs through inferencesmade by consumers. It also indicates that the effect of

' It may also be that the distinction between an image variable anda belief is not especially clear. The image variable itself may satisfy aneed (e.g., labels on designer jeans), or there may be uncertainty regardingthe true nature ofthe image variable. In the present study, it is assumedthat beliefs relate to physical performance characteristics only, and thatthe countries of origin of the aJtematives are widely known. Automobileswere selected for the study with these conditions in mind.

COUNTRY-OF-ORIGIN EFFECTS 699

image variables on attitude is not direct; any influencethey have appears to be a secondary one acting throughbeliefs.

[Received April 1983. Revised March 1984.]

REFERENCESBass, Frank M. and William L. Wilkie (1973), "A Comparative

Analysis of Attitudinal Predictions of Brand Preference,"Journal of Marketing Research. 10 (August), 262-269.

and Dick R. Wittink (1975), "Pooling Issues and Meth-ods in Regression Analysis with Examples in MarketingResearch," Journal of Marketing Research. 12 (November),414-425.

Beckwith, Neil E. and Donald R. Lehmann (1975), "The Im-portance of Halo Effects In MuIti-Attribute Attitude Mod-els," Journal of Marketing Research, 12 (August), 265-275.

Bilkey, WarTen J. and Erik Nes (1982). "Country-of-Origin Ef-fects on Product Evaluations," Journal of InternationalBusiness Studies, 1 (Spring/Summer), 89-99.

Cohen. Joel B. and Michael J. Houston (1972), "Cognitive Con-sequences of Brand Loyalty," Journal of Marketing Re-search. 9 (Eebruary), 97-99.

Eishbein, Martin and Icek Ajzen (1975), Belief, Attitude, In-tention and Behavior: An Introduction to Theory and Re-search, Reading, MA: Addison-Wesley.

Eisher, Eranklin M. (1970), "Tests of Equality Between Sets ofCoefficients in Two Linear Regressions: An ExpositoryNote," Econometrica. 38 (March), 361-366.

Holhrook, Morris B. (1978), "Beyond Attitude Structure: To-ward the Informational Determinants of Attitude," Journalof Marketing Research. 15 (November), 545-556.— (1981), "Integrating Compositional and Decomposi-tional Analyses to Represent the Intervening Role of Per-ceptions in Evaluative Judgments," Journal of MarketingResearch, 18 (Eebruary), 13-28.

(1983), "Using a Structural Model of Halo Effect toAssess Perceptual Distortion Due to Affective Overtones,"Joumai of Consumer Research, 10 (September), 247-252.

Huber, Joel and John McCann (1982), "The Impact of Infer-ential Beliefs on Product Evaluations," Journal of Mar-keting Research. 19 (August), 324-333.

Jacoby, Jacob, Jerry C. Olson, and Rafael A. Haddock (1971),"Price, Brand Name, and Product Composition Charac-teristics as Determinants of Perceived Quality," Journal ofApplied Psychology, 55 (December) 570-579.

Johnston, J. (1972). Econometric Methods (2nd edition). NewYork: McGraw-Hill.

Joreskog, Karl G. and Dag Sorbom (1982), "Recent Devel-opments in Structural Equation Modeling," Journal ofMarketing Research, 19 (November), 404-416.

Matlin, Margaret W. (1971), "Response Competition, Recog-nition, and Affect," Journal of Personality and Social Psy-chology, 19 (September), 295-300.

Mitchell, Andrew A. and Jerry C, Olson (1981), "Are ProductAttribute Beliefs the Only Mediator of Advertising Effectson Brand Attitude?" Journal of Marketing Research. 18(August), 318-332.

Monroe, Kent B. (1973), "Buyers' Subjective Perceptions ofPrice," Journal of Marketing Research. 10 (February), 70-80.

Moreiand, Richard L. and Robert B, Zajonc (1979), "ExposureEffects May Not Depend on Stimulus Recognition," Jour-nal of Personality and Social Psychology. 37 (June), 1085-1089.

Szybillo, George J. and Jacob Jacoby (1974), "Intrinsic versusExtrinsic Cues as Determinants of Perceived ProductQuality," Journal of Applied Psychology. 59 (February),74-78.

Toy, Daniel R. (1982), "Monitoring Communication Effects:A Cognitive Structure/Cognitive Response Approach,"Journal of Consumer Research, 9 (June), 66-76.

Tybout, Alice M. and John R. Hauser (1981), "A MarketingAudit Using a Conceptual Model of Consumer BehaviorApplication and Evaluation," Journal of Marketing, 45(Summer), 82-101.

Zajonc, Robert B, (1980), "Eeeling and Thinking: ReferencesNeed No Inferences," American Psychologist, 35 (Febru-ary), 151-175.

Zaltman, Gerald and Meianie Wallendorf (1983), ConsumerBehavior: Basic Eindings and Management Implications.(2nd edition). New York: John Wiley.