Attitude Models in Consumer Research

ABSTRACT - The subject of attitude models has been the topic of many research papers. Much confusion exists about the nature of the Rosenberg, Fishbein, and multi-attribute models and about how they should be operationalized to measure attitudes. The three papers which have been assigned to me have attempted to synthesize different approaches to modeling consumer preferences, to put the Fishbein model within the framework of an expected utility model, and to point out the researchers' failure to validate the attitude scales they have developed. I believe all of the three papers have contributed in their unique ways to the further understanding of attitude models.

Citation:

Sadaomi Oshikawa (1979) ,"Attitude Models in Consumer Research", in NA - Advances in Consumer Research Volume 06, eds. William L. Wilkie, Ann Abor, MI : Association for Consumer Research, Pages: 256-258.

Advances in Consumer Research Volume 6, 1979      Pages 256-258

ATTITUDE MODELS IN CONSUMER RESEARCH

Sadaomi Oshikawa, University of Washington

ABSTRACT -

The subject of attitude models has been the topic of many research papers. Much confusion exists about the nature of the Rosenberg, Fishbein, and multi-attribute models and about how they should be operationalized to measure attitudes. The three papers which have been assigned to me have attempted to synthesize different approaches to modeling consumer preferences, to put the Fishbein model within the framework of an expected utility model, and to point out the researchers' failure to validate the attitude scales they have developed. I believe all of the three papers have contributed in their unique ways to the further understanding of attitude models.

In the following sections, each of the papers will be critically evaluated and their contributions and weaknesses pointed out.

Finally, in the "Directions for Future Research" section, the importance of establishing predictive validity of the models will be emphasized and one hypothesis about why multi-attribute models produce high correlations will be explicated.

THE JAIN, MAHAJAN AND MALHOTRA PAPER

Contributions of the Paper

Jain, Mahajan and Malhotra have done an admirable job of reviewing the literature on multi-attribute models and comparing the differences in the compositional and decompositional approaches. This review article is a worthwhile reading for those who wish to keep themselves up-to-date on the topic of multiattribute models.

Questions with the Paper

While multiattribute models have been accepted by many consumer behavior researchers, their construct validity has never been clearly established. While Fishbein and Rosenberg attitude models have theoretical bases in stimulus-response learning theory and cognitive instrumentality theory respectively, multiattribute models in the form presented in the paper do not seem to have any theoretical base in social psychology. Research efforts need to be directed at validating multiattribute models and explaining why importance weights and respondents' beliefs are related to consumer attitudes.

The authors are interested in the validity of multi-attribute models and suggest obtaining the correlation between the attitude scores and the input data such as independent ratings of preference ordering of the alternatives. Methodological problems of this procedure will be discussed in the "Directions for Future Research" section.

THE HUBER AND LEONE PAPER

Contributions of the Paper

Huber and Leone have addressed themselves to the issue of unipolar and bipolar coding in attitude scales. This issue is important because there is no consensus about which coding scheme to adopt and because the choice of unipolar or bipolar coding makes a difference in the prediction of preference among a set of attitude objects. A simple algebra can show that predicted preference of attitude objects can be reversed by using different coding schemes.

Huber and Leone are to be commended for attempting to resolve the unipolarity-bipolarity coding issue. They hypothesize that Fishbein attitude model is a truncated version of the full expected utility model and that the bipolar coding scheme for association is required by the axioms of probability. If their hypotheses and the underlying assumptions are supported, Huber and Leone will be making a great contribution to the field of consumer research.

The present writer agrees with them when they emphasize the importance of testing the assumptions underlying different attitude models instead of comparing their overall predictive capabilities.

Questions with the Paper

As was indicated, Fishbein developed his original model on the basis of S-R learning theory, and the attitude toward an object is the affect (or liking) toward that object. Since the affect in the original Fishbein model is both conscious and unconscious processes, it is difficult to regard the original Fishbein model as equivalent to highly cognitive expected utility model. Fishbein's attitude-toward-act model, on the other hand, was developed from Dulany's cognitive theory of propositional control, and Huber and Leone are quite right in stating that Fishbein's attitude-toward-act model is analogous to an expected utility model.

However, the authors appear to be in error in stating that the Fishbein model does not include the negative outcome because there is no theoretical reason why the negative outcome cannot be included (Fishbein, 1967, 257-66) and Fishbein and Ajzen (1975, 223) indicate in the footnote that "the model deals only with associative relations between object and attribute, and thus a belief such as '0 is not X' is viewed as an association between the object 0 and the attribute not X."

If the argument that the Fishbein model is a truncated version of expected utility model is based solely on the assumption that the Fishbein model does not include the negative outcome, then Huber and Leone's argument does not seem to be tenable. However, it is still possible that the Fishbein model is a truncated version of the expected utility model for the reasons they have not considered. If that is the case, the authors have contributed considerably to the field of consumer research.

Of the assumptions which must be satisfied, Huber and Leone indicate that the truncated model assumes the independence of utilities across attributes. In consumer research, halo effects are common. Product attributes interact and their utilities are not independent of each other.

Finally, the authors suggest that whether the assumption of the bipolarity of utilities is justified is an empirical question. This can be tested by examining whether probabilities which are coded bipolarly fit the data better than those which are coded unipolarly. However, problem arises when some researchers find the bipolar coding fits the data better (Bettman, Capon and Lutz, 1975; Lutz, 1977) while others find the unipolar coding fits the data better (Ahtola, 1975).

It may be more fruitful to examine whether the assumption of the bipolarity of utilities is justified theoretically and then test it empirically. Huber and Leone have done an excellent job in directing our attention to the right (theoretical) direction.

THE BONFIELD PAPER

Contributions of the Paper

While Jacob Jacoby (1975, 1978) had made the same point, Bonfield made a contribution by pointing out the fact that many researchers have developed but have not validated their own attitude measuring instruments. When the instruments have not been validated, the researchers' empirical results cannot be conclusive.

If the results fail to support the model, the researcher must decide whether the model is not tenable or, while the model is tenable, the measurement procedure is faulty. If the results support the model, on the other hand, he still has to make sure that the results are not due to the artifacts of design or some spurious relationships.

Questions with the Paper

The author states that an attitude measuring instrument can be validated by showing high correlations with Thurstone scales measuring attitude toward the same object, concept, or act. It is true that the researcher who develops a new attitude instrument needs to establish that his instrument measures attitudes as do other validated instruments by means of showing high correlations with those validated instruments. In evaluating the semantic differential as a measure of attitude, for example, Osgood, Suci and Tannenbaum (1957) compared the semantic differential with Thurstone and Guttman scales. This comparison is a necessary condition but is not a sufficient condition for establishing validity because the high correlation may result if the subjects see the connection between those scales and respond consistently to both of them.

In order to establish validity, it is necessary to demonstrate that the researcher can predict correctly future behavior of the respondents on the basis of their scores on that instrument and establish predictive validity (Osgood, Suci and Tannenbaum, 1957, 104-24).

Bonfield is quite correct, on the other hand, in stating that an unvalidated attitude measure with high measure-behavior congruence would provide no understanding of the relationship between the measure and the behavior. It may be incorrect to assert the validity of an instrument solely because of the high congruence since the high congruence may result even though the instrument is not valid.

Finally, Bonfield asks whether researchers should concentrate on multi-attribute models which contain only attitudinal vectors or should develop multi-attribute models which contain not only attitudinal vectors but also other salient vectors. Since an attitude is a predisposition to evaluate an object favorably or unfavorably, only evaluative vectors which attribute good or bad qualities to that object (e.g., delicious looking, healthy, tasty, smells good, economical, hard to digest, in his example) should be included. Definitional vectors like chicken, veal scallops and $8.25 does not contribute to his liking or disliking of this entree. According to Katz and Stotland (1959, 429), "Judgments which are purely cognitive would not fall into the category of attitudes."

When not only evaluative vectors but also other salient vectors such as definitional and social influence vectors are included, the multi-attribute model may perform better in "predicting" behavior not necessarily because the model is better but because there are more predictor variables and consequently R² will be higher. Adjusted R² is a more appropriate statistic to determine which model (the multi-attribute model or the global multi-attribute model) has better "predictive" power.

Almost all researchers evaluate attitude models on the basis of the coefficient of multiple correlation or some other measure of association between the criterion variable (e.g., revealed preference, ranking, etc.) and the predictor variables. This research methodology can produce misleading and untenable results, and the following section will discuss the procedure which will reduce this bias.

DIRECTIONS FOR FUTURE RESEARCH

One of the puzzles in multi-attribute attitude research is the fact that multi-attribute models perform reasonably well in "predicting" criterion attitude even though there is no apparent social psychological theory to support the models. While Fishbein and Rosenberg models are based on S-R learning theory and instrumentality theory respectively, the multi-attribute models do not have a corresponding theoretical base. A possible reason for this good performance is the tendency of the respondent with favorable brand affect to evaluate salient attributes favorably.

This cognitive consistency, also called halo effect by Beckwith and Lehmann (1975), tends to increase the correlation between the criterion variable and the predictor variables and tends to overestimate the "predictive'' capability of multi-attribute models. Consistency is a valued trait, and respondents may consider the attitude measuring instrument as a test of their ability to be consistent and respond in such a way to be evaluated as intelligent (i.e., evaluation apprehension).

Since the number of attitude objects is ordinarily small (typically less than 20) and attributes are concrete rather than abstract, it is not difficult for the respondents to get the connection between attitude objects and salient attributes and respond consistently. When the correlation is reasonably high, the researcher concludes that the model's predictive performance is satisfactory or the model has predictive capability (Bruno and Wildt, 1975; Churchill, 1972; Holbrook, 1977). This usage of the word "predictive" appears to be based upon the fact that the overall affect as the criterion variable is "predicted" by predictor variables. Some writers conclude that the model has "predictive validity'' when the correlation is high (Wildt and Bruno, 1974).

However, strictly speaking, predictive validity is the ability to foretell the future behavior of respondents on the basis of their scores on the scale. In almost all studies of multi-attribute models, the criterion variable has been measured immediately before or soon after the predictor variables have been measured. This procedure not only fails to test the predictive validity but also causes the respondents to become aware of the relationships between the criterion and the predictor variables and produces artifactually high correlation coefficients.

The criterion variable and the predictor variables need to be measured at two different time periods sufficiently far apart so that the respondents do not get the connection between the two variables, and irrelevant attitude objects and attributes should be included in order to disguise the true purpose of the study.

One procedure which would enable the researchers to test predictive validity is the one Osgood, Suci and Tannenbaum (1975) used to evaluate the semantic differential. This method has the advantages of preventing the respondents' attempts for consistency from distorting the correlation and of determining the directionality of causality between the overall affect and the cognitive structure regarding the salient attributes.

Finally, if the criterion and the predictor variables are measured at about the same time, the respondents are more likely to become aware of the connection between the two variables when responding to the multi-attribute scales than when responding to the Fishbein or the Rosenberg scale. As was indicated already, the attribute beliefs and importances are usually more concrete concepts while the counterparts in the Fishbein and the Rosenberg scales tend to be more abstract. The relationship between the criterion variable and the predictor variables is more apparent in the multi-attribute scales. Having rated X brand more satisfactory than Y brand on salient attributes, the respondent may express greater affect toward X brand than toward Y brand in order to appear consistent and intelligent. The relationship in the other two scales is less obvious. This artifact may explain the relatively satisfactory performance of multi-attribute models.

REFERENCES

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

N.E. Beckwith and D. R. Lehmann, "The Importance of Halo Effects in Multi-Attribute Attitude Models," Journal of Marketing Research, 12 (August 1975) 265-75.

J. Bettman, N. Capon and R. J. Lutz, "Cognitive Algebra in Multi-Attribute Attitude Models," Journal of Marketing Research, 12 (May 1975), 151-64.

A. V. Bruno and A. R. Wildt, "Toward Understanding Attitude Structure: A Study of the Complimentarity of Multi-Attribute Attitude Models," Journal of Consumer Research, 2 (September 1975), 137-45.

G. A. Churchill, Jr., "Linear Attitude Models: A Study of Predictive Ability," Journal of Marketing Research, 9 (November 1972), 423-6.

Martin Fishbein, Readings in Attitude Theory and Measurement, ed., (New York: Wiley, 1967)

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

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

Jacob Jacoby, "Consumer Research: Telling It Like It Is," in Advances in Consumer Research, Vol. 3, B. A. Anderson, ed. (Atlanta: Georgia State University, 1976), 1-11.

Jacob Jacoby, "Consumer Research: A State of the Art Review," Journal of Marketing, 42 (April 1978), 87-96.

D. Katz and E. Stotland, "A Preliminary Statement to a Theory of Attitude Structure and Change," in Psychology: A Study of Science, Vol. 3, S. Koch (ed.), (New York: McGraw-Hill, 1959), 423-75.

R. J. Lutz, "An Experimental Investigation of Causal Relations Among Cognitions, Affect, and Behavioral Intention," Journal of Consumer Research, 3 (March, 1977), 197-208.

C. E. Osgood, G. J. Suci and P. H. Tannenbaum, The Measurement of Meaning (Urbana: University of Illinois Press, 1957).

A. R. Wildt and A. V. Bruno, "The Prediction of Preference for Capital Equipment Using Linear Attitude Models," Journal of Marketing Research, 11 (May 1974), 203-5.

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