A Comment on the State of Attitude Measurement in Consumer Research: a Polemic

E. H. Bonfield, Temple University
ABSTRACT - The question of attitude-behavior relationships is discussed in light of apparent current thinking that attitude and behavior ought to be more consistent; the position is taken, and substantiated by means of a limited current literature review and tabulation, that consumer researchers too frequently attempt to measure attitude with nonvalidated models; a method for easy validation is suggested; and current treatment of multi-attribute models is attacked.
[ to cite ]:
E. H. Bonfield (1979) ,"A Comment on the State of Attitude Measurement in Consumer Research: a Polemic", in NA - Advances in Consumer Research Volume 06, eds. William L. Wilkie, Ann Abor, MI : Association for Consumer Research, Pages: 238-244.

Advances in Consumer Research Volume 6, 1979      Pages 238-244

A COMMENT ON THE STATE OF ATTITUDE MEASUREMENT IN CONSUMER RESEARCH: A POLEMIC

E. H. Bonfield, Temple University

ABSTRACT -

The question of attitude-behavior relationships is discussed in light of apparent current thinking that attitude and behavior ought to be more consistent; the position is taken, and substantiated by means of a limited current literature review and tabulation, that consumer researchers too frequently attempt to measure attitude with nonvalidated models; a method for easy validation is suggested; and current treatment of multi-attribute models is attacked.

INTRODUCTION

The controversy concerning relationships between attitude and behavior is not unique, of course, to those engaged or interested in consumer research. The battle in the social psychology literature dates back at least to LaPiere (1934) and was explicitly discussed by Bauer at the first American Marketing Association Attitude Research Conference (1966). The problem of why attitudes are not particularly good predictors of behavior has been addressed in terms of definitional nuances, e.g., unidimensional versus multidimensional theoretical and measurement approaches (see Fishbein, 1967b, pp. 477, 478); specificity, e.g., attitude toward an object versus attitude toward an act (see Ryan and Bonfield, 1975, pp. 119, 120); a large variety of scaling techniques; and even questioning whether attitude exists in many cases (Kassarjian, 1978; Kassarjian and Kassarjian, 1978; Lastovicka, 1978). It seems appropriate to argue again, as have Doob (1947) and Fishbein (1967a, b), that in many cases attitude ought not predict behavior.

But problems with consumer researchers' use of attitude go beyond definitional and relational questions. We are in the habit of defining any measure we use as an operationalization of attitude. That is, we will call our models measures of attitude and either fail to validate these models or "validate" them using one-item scales or other measures of unproven validity and reliability. This charge is serious, and the second part of this paper includes documentation of the degree to which the charge is true. A specific, relatively easy validation technique will be suggested. Finally, consumer researchers' use of multi-attribute models will be attacked, not in terms of potential value of this type of model, but in terms of why present models are providing less than satisfactory results.

A DEFINITION OF ATTITUDE

Attitude, like personality, motivation, and learning, is a behavioral construct rather than some physical thing which can be seen, handled, and physically manipulated like copper, lead, or iron. Thus, the definition a theorist accepts for attitude depends on the preferences of the theorist. A definition which may be acceptable to most attitude theorists is that attitude is a learned predisposition to respond positively or negatively with respect to some attitude object, concept, or behavior relative to some object or concept, which is a paraphrase of a definition provided by Fishbein and Ajzen (1975, p. 6). This definition, which is intended to be unidimensional, is confined to what many multidimensionalists would define as affect only. Since "positively or negatively" suggests a dimension with a neutral point or area, it is contended individuals can hold neutral as well as no attitudes toward particular objects, concepts, or behaviors.

Operationally, however, it is fair to say that attitude is whatever it is a Thurstone equal appearing intervals attitude scale measures. [Concern here is only for attitude toward specific objects, concepts, or behaviors. Thus, the term "Thurstone scale" always refers to the equal appearing intervals scale. Thurstone's law of comparative Judgment, utilizing paired comparison data, is viewed as most useful for obtaining relative attitude toward each object within a class of objects without showing degree of positive or negative affect.] That is, any measure of attitude constructed according to the research protocol accepted for Thurstone scales, which includes tests for unidimensionality and reliability, is, by definition, a valid measure of attitude. Moreover, the only acceptable non-Thurstone measures of attitude have been shown to be valid measures of attitude by exhibiting high correlations with Thurstone scales measuring attitude toward the same object, concept, or act. Other techniques considered standard or valid measures of attitude when appropriate research protocols are followed include Guttman scales, Likert scales, and semantic differential attitude scales. No measure of attitude, no matter what its theoretical base, and regardless of whether it has face validity, can be accepted as a valid measure of attitude unless it can be shown to be measuring the same thing as one of the four standard scales.

Historically, it can be said the standard scales, and the theory which underlies them, "own" the attitude conceptualization. While behavioral theorists and researchers can argue the merits of this conceptualization, it appears more reasonable to posit a new concept to encompass what is being measured. It is beyond the scope of this paper to suggest other conceptualizations related to, but not, attitude. A few related notions already in use which are sometimes used either as related to or synonymous with attitude are values, beliefs, preferences, and even intentions.

Under the historically accurate attitude approach, it is unreasonable to expect attitude to be a perfect predictor of behavior. Critics of the attitude concept who base their criticism of attitude measures on weak relationships with behavior are allowing themselves to be blind to the relative place of attitude within the total set of internal and external factors which lead to human behavior. While we can improve our understanding of attitude-behavior relationships, and behavioral predictions, by specifying attitude objects, concepts, and behaviors more precisely, we still should not expect near-perfect attitude-behavior relationships. While it may be possible to develop a measure, calling it a measure of attitude, with high measure-behavior congruence, most likely the measure would be effective for only a short term because it would provide no, or spurious, understanding of the relationship between the measure and the behavior.

The reasons for imperfect attitude-behavior relationships are, for the most part, fairly obvious. First, attitude is only one of a number of factors influencing behavior. Other factors, such as personality or social influence, may act so as to block behavior that attitude would favor. Much social psychological and consumer research has been expended on the Fishbein extended model (see Fishbein and Ajzen, 1975; Ryan and Bonfield, 1975) which is described as a method for studying attitude in relation to other factors as they relate to behavior.

Secondly, attitudes may be weakly held and thus easily changed. In those cases, even if attitude is a good predictor of behavior, only when the attitude and behavior measurements are nearly simultaneous could we expect a strong attitude-behavior relationship to be found.

Third, attitude may, in some cases, not even exist. Since most respondents will provide information on attitude scales unless they have absolutely no knowledge of its subject, it should not be surprising when "attitudes" among these respondents have little or no relationship to their behavior.

Fourth, behavior with respect to an object, concept, or act can be viewed as learned after the attitude is learned (Doob, 1947, as described by Fishbein, 1967b, p. 478). That is, attitude is treated by these theorists as a learned mediating response (Doob, 1947; Rhine, 1958; Fishbein, 1967a, b; Olson and Mitchell, 1975). Once attitude is learned, the individual must also learn what response to make to that attitude.

...that is, there is no innate relationship between the attitude and behavior; one still has to learn a behavioral response. Two people may learn to hold the same attitude toward a given stimulus; clearly, however, they may also learn to make different responses given the same learned attitude.

For example, two students may learn to feel equally favorable toward a given instructor. Furthermore, this feeling may initially elicit the same overt response in both students (e.g., calling the instructor by his first name). From Doob's point of view, the probability that this behavior will persist is a function of the reinforcement the students get for making this response. For example, with respect to one student, the instructor might say, "Well, I'm glad you finally decided to drop that 'professor' nonsense," while he might tell the other student, "I'd prefer it if you wouldn't call me by my first name." If this were the case, Doob's theory would predict that the behavior would continue for the first student, but not for the second (Fishbein, 1967b, p. 478).

Fishbein recognized that, in the given example, the non-reinforced behavior would probably lead to attitude change as well as differential behavior.

Fifth, since attitude is a predisposition to act rather than an act itself, it is possible to view attitude as a wish or impulse. In that sense, attitude may be closely related to the Freudian concept of the id, which, according to Hall and Lindzey (1968), "is able to do only one thing, namely, to wish (p. 249)." If so, and since the wishes of the id are monitored and frequently modified by the ego and superego, not only would we expect much less than perfect attitude-behavior relationships, but we should also expect positive attitudes toward many "socially unacceptable" objects, concepts, and behaviors. This notion suggests indirect measures of attitude, such as the semantic differential, will be most accurate as attitude measures because they are less subject to ego and superego moderating effects which would show up as more socially acceptable attitudes. Alternately, it may be possible individuals have three potentially distinct attitudes with respect to each object, concept, or act; one each associated with the id, ego, and superego.

For other reasons attitude may not predict behavior, see Cohen (1974, pp. 341-343) and Cohen and Ahtola (1971, p. 346). If behavior is not a validator of attitude measures, a fact tacitly recognized by most consumer researchers, how then are we validating our attitude measures?

VALIDATION OF NONSTANDARD ATTITUDE MEASURES

Consumer researchers have utilized a number of attitude measures, many of which are nonstandard in the sense of not being Thurstone, Guttman, Likert, or semantic differential attitude scales. In particular, consumer researchers appear enamored with multi-attribute attitude models.

A limited review of the literature reveals the reason for the title of this paper. The review includes three sources: (1) all issues of the Journal of Marketing Research from February 1974 to February 1978 which loosely defines a "post-Wilkie-Pessemier (1973) era" for attitude, particularly multi-attribute attitude, measurement; (2) all issues of the Journal of Consumer Research which was first published in 1974; and (3) all Association for Consumer Research Conference Proceedings. An item had to be available April 1, 1978 to be included in the review. [The purpose of this review was to document a trend. Since it is inappropriate to enter into data search with biased expectations, it is hoped some degree of scientific objectivity was maintained throughout the data search.] Some double counting was included since multiple publications of the same data set, except in review articles, were treated as separate studies. The results of the data review are summarized in Table 1.

As can be seen, nearly half the studies utilized scales which were operationalizations of some form of multi-attribute model. There were three measures for which enough information was presented for classification as a standard attitude measure, although in a number of cases the technique could be considered an approximation of a standard scale. Few researchers stated that any attempt was made to validate any attitude scale, but in many cases, terms such as "criterion" and "independent" measure were used in such a way as to suggest high correspondence between the two measures would be acceptable evidence of validity. Of these, only Lutz (1972, 1975) used a true standard measure, semantic differential scales, in such a way as to test the validity of a multi-attribute measure. In the more recent study, Lutz (1975) was treating the multi-attribute model as an indicant of the underlying cognitive structure of attitude rather than a measure of attitude. While Lutz' study is classified as a multi-attribute study with a true validation attempt in Table 1, it is also clear the study could have been classified as one in which a true semantic differential attitude measure was used. At any rate, it is clear that Lutz, for one, knows how to use the semantic differential to measure attitude since in the earlier study, Lutz (1972) had tested a multi-attribute model, called an attitude model, against an appropriately developed semantic differential measure of attitude.

Some researchers (e.g., Arndt and Crane, 1975) specifically addressed themselves to the limitations of their measures while others (e.g., Cox, 1975) did not specify that their operationalizations were attempts to measure attitude. Finally, a number of researchers related their attitude measures to preference ranks. While perhaps related to Thurstone's law of comparative judgment, preference ranks have not been substantiated as an attitude validating construct.

TABLE 1

TYPES OF ATTITUDE SCALES IN CONSUMER RESEARCH STUDIES

The large number of articles, 127, in which attitude measures were utilized precludes study-by-study review. Instead, some examples will be discussed. These examples are not intended to embarrass the researchers, indeed this writer is as guilty as any.

The difficulty of developing a Guttman scale was demonstrated by Pessemier, Bemmaor, and Hanssens (1977). They were forced to reject their hypothesis that their items, concerning willingness to donate body parts, could be scaled in one dimension. Admirably, they not only admitted this problem, but also included ten "attitude scales" as independent predictors of willingness to donate. Each scale, consisting of four items each, was Likert-like, hut the attitude constructs seemed unlikely to be tied, directly or indirectly, to a willingness to donate body parts. Instead, such scale items as "Every woman should be free to have an abortion" and "I try to impress people with my leadership ability" seem more related to psychographics/life style research. There is no quarrel with the researchers' relational findings, but neither is there a validated attitude measure in their article.

Hawkins, Albaum, and Best (1974) provided a test of the Stapel scale using the semantic differential as a validating construct. Unfortunately, their semantic differential scales had six rather than seven scale positions and included such nonstandard items as "helpful employees-unhelpful employees" and "limited selection-wide selection." Also, no internal validation was provided to develop a semantic differential attitude measure. Hawkins, et al., did not claim to be testing an alternate attitude measure, however, but an alternate to the semantic differential because they felt the Stapel scale was easier to administer in telephone interviews. Appropriate replication of this study testing for Stapel-semantic differential attitude scale equivalence would be useful. Such studies should include a semantic differential telephone administration treatment which was not included among the original treatments.

In an early consumer research test of the Fishbein extended model, Bonfield (1974) utilized a multi-attribute model described by Sheth (Howard and Sheth, 1969; Sheth, 1969). Since the Sheth model is nonstandard, it should have been validated. No report of a validation attempt was shown. More recently, Holbrook (1977) has used a "nine-position general liking scale (p. 168)" with "do not like at all" and "like very much" as endpoints as a criterion variable for a multi-attribute attitude model. While the criterion scale has face validity as a measure of affect, it suffers, nonetheless, from a lack of validation against a standard scale. Moreover, as a one-item criterion measure, it is impossible to test for internal reliability. More seriously, perhaps, the criterion scale may suffer from ceiling effect on the positive side at least. That is, among well-liked singers, all would likely have top scores among individuals, even though their actual attitudes would not be equally positive. Bonfield and Holbrook are only two examples of researchers who have used sophisticated theoretical and mathematical concepts, such as Holbrook's optimally scaled model, yet failed in the critical validation aspects of the models they used.

If, as is contended with the examples discussed, most consumer research attitude measures have not been shown to measure attitude, how should the researcher go about checking the validity of nonstandard attitude measures? [A number of sources can be found which discuss validity more thoroughly than is possible here. One such source, Zaltman, Pinson, and Angelmar (1975, pp. 39-47) is directly tied to consumer research.] The development of a standard scale measuring attitude toward the same object, concept, or behavior as the nonstandard scale is one viable strategy. The standard measure acts as an independent measure of concurrent and/or convergent validity (Zaltman, et al., 1975, pp. 39-47). Because Thurstone, Guttman, and Likert scales are time consuming, and sometimes difficult, to construct, semantic differential attitude scales are recommended as a validating construct. Since most multi-attribute models have been operationalized utilizing semantic differential type scales, convergent validity is sacrificed when semantic differential methodological protocols are utilized. Following an appropriate research protocol, semantic differential scales are easily selected (i.e., no development is necessary), easily administered, and, with computers and factor analysis programs, easily analyzed.

It should be recognized, however, true semantic differential scales utilize standard instructions which can be paraphrased, have seven places for each scale-item, and only bipolar scale item names selected from a specific set can be used. This information can easily be found in Osgood, Suci, and Tannenbaum (1957). Some 10 to 20 scales should be selected, most associated, a priori, with the evaluative dimension, but with some associated with the activity and potency dimensions of meaning. Once data is collected, factor analysis--actually principal components analysis with varimax rotation--will reveal those scales with the highest loadings on the evaluative dimension, the only semantic dimension representing attitude. If high enough loadings are found for three to five evaluative scales, scores on these scales can be summed or averaged for each individual as an attitude criterion variable. High factor loadings can be taken as indicants of scale reliability rendering additional tests for reliability in the criterion measure superfluous. Since attitude structure may be different across objects, concepts, and acts, as well as across respondent groups, separate factor analyses should be performed for each regardless of the original set of scales selected.

While the accurate adherence to a standard measure protocol can be considered a validation procedure in and of itself, even standard scales should be checked for validity. Lundstrom and Lamont (1976), for example, checked both internal and external validity of their Likert-type scale designed to measure consumer discontent. Validity testing included sampling from two groups expected to hold favorable, i.e., contented, and unfavorable, i.e., discontented, attitudes. The former group consisted of members of Rotary and Kiwanis, business and professional organizations, while the latter group consisted of members of a consumer activist organization and people who had lodged complaints with a Better Business Bureau. Internal validity testing consisted of correlating item scores with total scores while external validity was tested by comparing attitude scores of the two groups. Testing external validity using known groups is also a viable strategy for nonstandard attitude measures.

Baker and Churchill (1977) utilized semantic differential-like methodology in an apparently successful attempt to measure affective, cognitive (belief), and conative (intention) dimensions of reactions to advertisements. Rather than factor analysis, however, these researchers utilized item-to-total correlations, thus creating a need for a validating measure which was not included. Other researchers (e.g., Barnes, 1978) have also attempted to tap these three dimensions.

It would be impossible to exhaust validation possibilities in this paper. Hopefully, the argument that validation is both needed and possible can be accepted by consumer researchers interested in measuring attitudes.

WHY MULTI-ATTRIBUTE MODELS DON'T WORK

Actually, appropriate multi-attribute models may "work." First, one or more forms of the model may be a valid measure of attitude, but appropriate evidence supporting such a model has not been presented. Even if they cannot measure attitude, appropriate multi-attribute models may be valuable as indicants of viable strategies for changing attitudes and/or behavior through their presentation of a picture of the cognitive structure underlying attitude. At the present time, we have scant knowledge of which operational form of such a model, or which statistical manipulation of the data collected, if any, is a valid measure of attitude. There are, however, reasons to expect weak relationships between multi-attribute models and criterion, standard attitude measures.

Specifically, two homogeneity assumptions are usually, tacitly, made relative to the operationalization of multi-attribute models. These assumptions are (1) all individuals will have the same set of attributes underlying their attitudes and (2) all objects within a class of objects (e.g., brands within a product class) will have the same attributes. Conversely, the problems of heterogeneity of salient attributes across groups of people and heterogeneity of attributes across objects, concepts, or behaviors within classes are simply not addressed. The problem of determining salient attributes is not addressed here. For discussions of that problem see Ryan and Bonfield (1975, pp. 121, 122) and Ryan and Etzel (1976).

Homogeneity of salient attributes within groups should be a reasonable assumption, otherwise it would probably be useless to deal in multi-attribute models. Nonetheless, since, as Wilkie and Weinreich (1972) have shown, perfect homogeneity of salient attributes across members within a group is unlikely, multi-attribute models will be relatively poor predictors of attitude as measured using a standard procedure. In addition, no case was found in which the homogeneity assumption within groups was tested. Any researcher desiring to test this assumption should be most careful in selecting variables which differentiate groups. Inappropriate differentiating variables will tend to magnify within group heterogeneity.

A more serious error is to treat all members of all groups as having the same set of salient attributes, i.e., a problem of heterogeneity of salient attributes across groups. Upper income group members are less likely, for example, to regard price as a salient attribute of sirloin strip steak than low income group members. Assuming product knowledge, it is likely both groups would rate the object the same on multi-attribute scales even though the price scale(s) would have no bearing on the upper income groups' attitudes. Only Ryan and Etzel (1976) appear to have tested for heterogeneity of salient attributes across groups and across brands. Cox (1975) assumed heterogeneity of salient attributes, but did not test this assumption.

The likelihood of heterogeneity of salient attributes across objects in a class may not be so obvious. For demonstration purposes, it is useful to turn to Lewinian conflict theory as discussed in almost any basic psychology text. Lewin's field theory is concerned with choice. In any choice situation, the individual is motivated to move from one state which is negatively valued, say "hungry," to some other state which is positively valued, "not hungry," for example. Other motives will be operative at the same time, for example, to move from "thirsty" to "not thirsty" and from "need to write manuscript" to "finished manuscript." In moving from a state of "hungry" to "not hungry." the individual may choose from a vast array of means, called pathways in field theory. Any one pathway may not be mutually exclusive compared to other pathways, just as several highway routes from Chicago to Dallas may require driving part of the way on the same stretch of highway as other routes.

Suppose the individual selected a pathway which included eating in a restaurant. The selection of which restaurant follows a decision to eat at a restaurant rather than a fast food franchise, a delicatessen, or at home, among others, of course, but for our example, it will be useful to have our individual at the point of selecting an entree at the chosen restaurant. Each item on the menu has associated with it a number of vectors, which in terms of the present example, are defined as field forces leading toward or away from a specific choice. In Figure 1, vectors are represented by arrows. Each vector has a property called valence which by sign, plus or minus, indicates the direction of the vector; and by degree of positiveness or negativeness, indicates the strength of the vector. Strength of each vector is indicated by the numerical value placed next to the signs in Figure 1. Obviously, a vector is analogous to an attribute. Figure 1 is an example of approach-approach conflict, that is, all the relevant choices have equal and positive net valence. [Other field theory conflict types are not treated here. In avoidance-avoidance conflict, all relevant choices have equal and negative net valences. With ambivalence, the choice is whether to perform a particular, general behavior rather than which of two or more behaviors to perform. Thus, an individual may be ambivalent as to whether to buy a color television set. Ambivalence occurs when the positive and negative valence associated with the vectors cancel each other out leaving a net valence of zero.] If any entree choice had a positive net valence greater than all others, that entree would be easily chosen, although some conflict would still be felt since the positive valence associated with unchosen alternatives would be lost and, at the same time, the negative valence associated with the chosen alternative would be obtained along with the positive valence.

FIGURE 1

A HYPOTHETICAL RESTAURANT ENTREE EXAMPLE OF APPROACH-APPROACH CONFLICT

This example has two major implications for understanding why multi-attribute models are not good predictors of behavior. First, in Figure 1, each choice has a unique set of associated vectors and no one vector name is associated with all choices--clearly a case of heterogeneity of salient attributes. Entrees on a menu, it can be argued, are different from brands in a product class. However, it is contended the difference is one of number of heterogeneous vectors rather than whether heterogeneity exists. For example, the vector "Superstar in Rent-A-Car" cannot, or more accurately, should not, be associated with Avis, although "Not Superstar in Rent-A-Car" can. "Marlboro Country," or even "Manly Cigarettes," should be a frequently found vector for Marlboro cigarettes unlikely to be found for Winston cigarettes. The fact is much marketing effort is expended to create these unique vectors which are called differential advantages, yet nearly every study which has utilized multi-attribute models across brands in a product class has assumed each brand has exactly the same set of attributes/vectors as every other brand in the product class.

The second implication shown by the example depicted in Figure 1 is that not all vectors are attitudinal or evaluative in content. Some vectors represent values, e.g., healthy, economical; some represent social influence, e.g., spouse will approve, doctor expects me to order this; some represent definitional aspects, e.g., chicken, veal scallops, $8.25; and some are clearly evaluative, e.g., delicious looking, spicy, tasty. The second implication leads to a dilemma: (1) Should researchers concentrate on multi-attribute models in which only attitudinal vectors are considered in attempts to measure attitude as one of many influences on behavior? (2) Or, should researchers attempt to develop more global multi-attribute models which include all salient vectors in order to predict and understand behavior, but not attitude? The bulk of consumer research thus far appears to have been attempts to answer the first question.

It is hypothesized research aimed at answering the second question will lead to two results. First, more global multi-attribute models will lead to better prediction of attitude. This expectation follows from findings that attitude is not independent of other influences on behavior. For example, in their review of research undertaken following the Fishbein extended model, as well as in their further joint research, Ryan and Bonfield (1975, 1977) have shown evidence that while attitude and social influence provide independent explanations of variance in intentions, a joint, or interaction, effect is also present.

Secondly, more global multi-attribute models will lead to better predictions of intentions and possibly behavior than, for example, Fishbein's extended model. Bither and Wright (1977) have utilized a global multi-attribute model to explore preferences for information sources. An interesting re-analysis of their data might test the homogeneity assumptions. A model and procedure suggested by Bernardo and Blin (1977) also appears to have potential value as a global model, but seems to rest on a homogeneity of attributes assumption. Stanton and Lowenhar (1974) have defined a need-press model--in which press is treated as an alternative's ability to satisfy needs--which was operationalized as a multi-attribute model. While the Stanton and Lowenhar model was not a predictor of rank order preference, relative preference, or attitude as measured using a Thurstone scale, it was not tested as part of a global multi-at-tribute model for the predictions of intentions or behavior. It is assumed researchers would continue to attempt to find optimum combination rules for data collected following a more global model.

Such a model may extract a cost, however, in terms of understanding. That is, a global multi-attribute model may obscure relationships among attitudes, social influences, needs, and other influences on behavior. At the same time, more global, multi-attribute models with appropriate identification of vector-attributes can have strategy directing, strong relationships with behavior which are not devoid of potential for understanding behavior. In that sense, such a model--which is not an attitude model--would be in line with Kassarjian's (1978) call for simpler models.

CONCLUSION

While starting on a highly critical note, this paper ends more hopefully. It should be clear that multi-attribute models of attitude can be developed which are valid, which provide an understanding of attitude, and which lead to strategies for changing attitude. At the same time, it should be clear that attitudes alone do not control behavior.

Naturally, all problems associated with multi-attribute models cannot be discussed within the limited scope of this paper. Such important topics as consistency of attribute meaning (Gensch and Golob, 1975), attribute content (Ahtola, 1975; Etter, 1975), and specific combination rules, which include the question of weighting, have not been addressed.

Finally, a more global multi-attribute model which is not a model of attitude per se may lead to a parsimonious model for prediction, influence, and understanding of consumer behavior.

REFERENCES

Ahtola, Olli T. (1975), "The Vector Model of Preferences: An Alternative to the Fishbein Model," Journal of Marketing Research, 12, February, 52-9.

Arndt, Johan and Edgar Crane (1975), "Response Bias, Yea-Saying, and the Double Negative," Journal of Market-lng Research, 12, May, 218-20.

Baker, Michael J. and Gilbert A. Churchill, Jr. (1977), "The Impact of Physically Attractive Models on Advertising Evaluations," Journal of Marketing Research, 14, November, 538-55.

Barnes, James G. (1978), "A Hierarchical Model of Source Effect in Retail Newspaper Advertising," in Advances in Consumer Research, Volume V, ed. H. Keith Hunt, Association for Consumer Research, 235-42.

Bauer, Raymond A. (1966), "Attitudes, Verbal Behavior, and Other Behavior," in Attitude Research at Sea, ed. Lee Adler and Irving Crespi, Chicago: American Marketing Association, 3-14.

Bernardo, John J. and J. M. Blin (1977), "A Programming Model of Consumer Choice Among Multi-Attributed Brands," Journal of Consumer Research, 4, September, 111-9.

Bither, Stewart and Peter Wright (1977), "Preference Between Product Consultants: Choices vs. Preference Functions," Journal of Consumer Research, 4, June, 39-47.

Bonfield, E. H. (1974), "Attitude, Social Influence, Personal Norm, and Intention Interactions as Related to Brand Purchase Behavior," Journal of Marketing Research, 11, November, 379-89.

Cohen, Joel B. (1974), "Toward an Integrated Use of Expectancy-Value Attitude Models," in Buyer/Consumer Information Processing, ed. G. David Hughes and Michael L. Ray, Chapel Hill: University of North Carolina Press, 331-46.

Cohen, Joel B. and Olli T. Ahtola (1971), "An Expectancy X Value Analysis of the Relationship Between Consumer Attitudes and Behavior," Proceedings of the 2nd Annual Conference of the Association for Consumer Research, ed. David M. Gardner, 344-64.

Cox, Eli P., III (1975), "Family Purchase Decision Making and the Process of Adjustment," Journal of Marketing Research, 12, May, 189-95.

Doob, Leonard W. (1947), "The Behavior of Attitudes," Psychological Review, 54, 135-56.

Etter, William L. (1975), "Attitude Theory and Decision Theory: Where is the Common Ground?" Journal of Marketing Research, 12, 481-3.

Fishbein, Martin (1967a), "A Behavior Theory Approach to the Relations Between Beliefs about an Object and the Attitude Toward the Object," in Readings in Attitude Theory and Measurement, ed. Martin Fishbein, New York: Wiley, 389-400.

Fishbein, Martin (1967b), "Attitude and the Prediction of Behavior," in Readings in Attitude Theory and Measurement, ed. Martin Fishbein, New York: Wiley, 477-92.

Fishbein, Martin and Icek Ajzen (1975), Belief, Attitude, Intention and Behavior: An Introduction to Theory and Research, Reading, Mass.: Addison-Wesley.

Gensch, Dennis H. and Thomas F. Golob (1975), "Testing the Consistency of Attribute Meaning in Empirical Concept Testing," Journal of Marketing Research, 12, August, 348-54.

Hall, Calvin S. and Gardner Lindzey (1968), "The Relevance of Freudian Psychology and Related Viewpoints for the Social Sciences," in The Handbook of Social Psychology, Second Edition, ed. Gardner Lindzey and Elliot Aronson, Reading, Mass.: Addison-Wesley, Volume One, 245-319.

Hawkins, Del I., Gerald Albaum, and Roger Best (1974), "Stapel Scale or Semantic Differential in Marketing Research," Journal of Marketing Research, 11, August, 318-22.

Holbrook, Morris B. (1977), "Comparing Multiattribute Attitude Models by Optimal Scaling," Journal of Consumer Research, 4, December, 165-71.

Howard, John A. and Jagdish N. Sheth (1969), The Theory of Buyer Behavior, New York: Wiley.

Kassarjian, Harold H. (1978), "Presidential Address, 1977: Anthromorphism and Parsimony," in Advances in Consumer Research, Volume V, ed. H. Keith Hunt, Association for Consumer Research, xiii-iv.

Kassarjian, Harold H. and Waltrud M. Kassarjian (1978), "Attitude Under Low Involvement Conditions,'' In Attitude Research Plays for High Stakes, John C. Maloney and Bernard Silverman, Chicago: American Marketing Association, forthcoming.

Lastovicka, John L. (1978), "Are Attitude Models Appropriate for Mass TV Advertising?" Paper presented at the American Marketing Association Ninth Annual Attitude Research Conference, Tarpon Springs, Fla.

LaPiere, Richard T. (1934), "Attitudes versus Actions," Social Forces, 13, 230-7.

Lundstrom, William J. and Lawrence M. Lamont (1976), "The Development of a Scale to Measure Consumer Discontent,'' Journal of Marketing Research, 13, November, 373-81.

Lutz, Richard J. (1972), "Investigating the Feasibility of Personalized Rapid Transit: An Experimental Approach,'' in Proceedings of the Third Annual Conference of the Association for Consumer Research, ed. M. Venkatesan, 800-6.

Lutz, Richard L. (1975), "Changing Brand Attitudes Through Modification of Cognitive Structure," Journal of Consumer Research, 1, March, 49-59.

Olson, Jerry C. and Andrew A. Mitchel (1975), "The Process of Attitude Acquisition: The Value of a Developmental Approach to Consumer Attitude Research," in Advances in Consumer Research, Volume 2, ed. Mary Jane Schlinger, Association for Consumer Research, 249-64.

Osgood, Charles E., George J. Suci, and Percy H. Tannenbaum (1957), The Measurement of Meaning, Urbana: The University of Illinois Press.

Pessemier, Edgar A., Albert C. Bemmaor, and Dominique M. Hanssens (1977), "Willingness to Supply Human Body Parts: Some Empirical Results," Journal of Consumer Research, 4, December, 131-40.

Rhine, Ramon J. (1958), "A Concept-Formation Approach to Attitude Acquisition," Psychological Review, 65, 362-70.

Ryan, Michael J. and E. H. Bonfield (1975), "The Fishbein Extended Model and Consumer Behavior," Journal of Consumer Research, 2, September, 118-36.

Ryan, Michael J. and E. H. Bonfield (1977), "Fishbein's Extended Model: A Test of External and Pragmatic Validity,'' New York: Columbia University Graduate School of Business, Research Paper Number 168.

Ryan, Michael J. and Michael J. Etzel (1976), "The Nature of Salient Outcomes and Referents in the Extended Model," in Advances in Consumer Research, Volume III, ed. Beverlee B. Anderson, Association for Consumer Research, 485-90.

Sheth, Jagdish N. (1969), "Using Factor Analysis to Estimate Parameters," Journal of the American Statistical Association, 64, September, 808-22.

Stanton, John L. and Jeffrey A. Lowenhar (1974), "A Congruence Model of Brand Preference: A Theoretical and Empirical Study," Journal of Marketing Research, 11, November, 427-33.

Wilkie, William L. and Roll P. Weinreich (1972), "Effects of the Number and Type of Attributes Included in an Attitude Model: More Is not Better," in Proceedings of the Third Annual Conference of the Association for Consumer Research, ed. M. Venkatesan, 325-40.

Wilkie, William L. and Edgar A. Pessemier (1973), "Issues in Marketing's Use of Multi-Attribute Attitude Models," Journal of Marketing Research, 10, November, 428-41.

Zaltman, Gerald, Christian R. A. Pinson, and Reinhard Angelmar (1973), Metatheory and Consumer Research, New York: Holt, Rinehart and Winston.

----------------------------------------