Some Issues in Attitude Theory and Measurement

Olli T. Ahtola, University of Florida
ABSTRACT - The three papers in this session attempt to clarify some important issues relevant to the study of attitudes. With minor exceptions I would call the points of view taken to be mainly methodological. Very little is said about underlying theoretical notions.
[ to cite ]:
Olli T. Ahtola (1979) ,"Some Issues in Attitude Theory and Measurement", in NA - Advances in Consumer Research Volume 06, eds. William L. Wilkie, Ann Abor, MI : Association for Consumer Research, Pages: 253-255.

Advances in Consumer Research Volume 6, 1979      Pages 253-255

SOME ISSUES IN ATTITUDE THEORY AND MEASUREMENT

Olli T. Ahtola, University of Florida

ABSTRACT -

The three papers in this session attempt to clarify some important issues relevant to the study of attitudes. With minor exceptions I would call the points of view taken to be mainly methodological. Very little is said about underlying theoretical notions.

Bringing in the basic theories would have been very useful, because the papers attempt to relate different approaches to attitude measurement to each other, which cannot be accomplished adequately by only pointing out similarities or differences in resultant mathematical forms or operational measures and methods chosen. Everybody agrees that some differences exist, but it remains still unclear whether the differences are merely operational issues or based on differences in underlying theories. Bonfield seems to take one extreme position, that is, if you do not operationalize and measure attitudes identically to "traditional" methods or do not demonstrate high correlation with a "traditional'' measure, you should not call the concept you are measuring "attitude". At the other extreme, the authors of the other two papers seem to feel that many different models are not really different but only incomplete or truncated versions of more complete models.

THE BONFIELD PAPER

Contributions of the Paper

Even though we all know it, it is useful to be reminded how important it is to validate our models and measuring instruments carefully and thoroughly, and to receive suggestions how to do it for attitude models. It is also good to be reminded that an attitude is a construct which need not to be directly related to overt behaviors. Furthermore, the often made assumption for methodological convenience, i.e., that the same attributes are salient for all subjects in our population and for all alternatives in a choice set, may not be a good assumption to make in many cases. It is also a good suggestion that for behavior prediction perhaps other models than multiattribute attitude models, might be more useful.

Questions with the Paper

I feel the author is taking a very extreme approach when he argues that the Thurstone equal appearing intervals attitude scale is a measure of attitudes by definition and that all other methods must correlate with it in order to be valid measures of attitudes. Personally, I strongly disagree that any method to measure a construct, especially a hypothetical construct is valid by definition. I think, everybody agrees that at least one of the dimensions of attitude is affect, but the fact that Thurstone was perhaps the first who proposed and tried to justify a method to measure affect toward objects on an interval scale, does not give this method any seniority-based right to claim ownership of the concept it was intended to measure.

Of course, it is sensible to find out if a new method converges with the already established methods as a part of the validation process, especially if the new method is based on the same conceptualization of the construct as the established methods. However, I do not feel this comparison is absolutely necessary, nor do I feel it is adequate. To the best of my knowledge, the Guttman scales, for example, became quite established before anybody correlated them with the Thurstone scales. What makes a method valid, including the Thurstone scales, is that it follows logically from the underlying theory, predicts what it is theoretically supposed to predict and does not covary with theoretically unrelated variables. (Campbell and Fiske, 1959; Cronbach and Meehl, 1955)

It is better to consider some methods to measure attitudes to be less (or poorly, or perhaps incorrectly) validated than some other methods, than to categorize them into the dichotomy "validated/unvalidated" based on some weak criterion such as the one suggested by the author. Also, if the developers of these "unvalidated" methods would be required to label the resultant measures as measures of something that has not been measured before, as the author suggests, the result would be an utter chaos. Such labels as "values", "beliefs", "preferences" or "intentions" would not do because methods to measure them have already been proposed.

The rest of the discussion under the subtitle "A Definition of Attitude" is devoted to the reasons why attitudes do not necessarily predict overt behaviors. During this lengthy discussion, the reader is kept wondering what this issue has to do with attitude measurement. Finally, in the last sentence the reader is given a clue. The author is probably trying to tell that even if the "attitude" measure does not predict a certain overt behavior it is not necessarily invalid nor is the "attitude" measure necessarily valid if it happens to predict the behavior quite well. I think the author is quite right here if I interpret him correctly, but this part of the paper is unclearly written. During this discussion the author also, for some reason, tries to draw parallels between attitude and id. I think this is quite unnecessary and probably wrong. Quoting the same authors (Hall and Lindzey, 1968) id is conceptualized as the first, obscure, inaccessible, and unorganized part of personality, with no direct relations with external world, and with immortal contents. These characteristics certainly do not seem consistent with the typical conceptualization of attitude. Those who are interested in the psychoanalytic approach to attitudes should find Sarnoff (1960) interesting.

The author suggests the use of a properly constructed evaluative dimension of the Semantic Differential as the validation measure for new attitude scales. I think, because of its relative simplicity it is a sensible part of the validation process. However, as I already stated much more is needed to establish strong validity. Also, it is so well established that scales like "good-bad", "pleasant-unpleasant", "satisfactory-dissatisfactory" always load highly on the evaluative dimension, that if they are used without factor analytic validation, no big harm is committed. Actually, if those scales would not all have high loadings on the same factor, I would not accept that factor as the evaluative dimension of the Semantic Differential.

As I questioned the need to bring psychoanalytic notions to the discussion, I am questioning the usefulness of bringing in Field Theory to complicate the issue. Field Theory is a comprehensive theory of motivation and it is not very clear how to relate its components to attitudes. None of the multiattribute attitude theories (e.g., Fishbein, Rosenberg) state that same attributes are used by all people for all alternatives in a choice set.

Also Field Theory is not needed to justify the inclusion of "nonevaluative" attributes. Fishbein (1967, p. 396) states that any attribute (belief) associated with the attitude object can be a determinant of the attitude. However, if we get out of the realm of attitudes per se, and move toward behavior prediction then perhaps Field Theory provides more direct guidance.

In summary, even though I feel this paper correctly points out deficiencies in attitude measurement in consumer behavior, I am not equally enthusiastic about most of its suggestions. Also, I feel, unnecessary theoretical confusion was added by bringing into discussion in an ad hoc fashion some basic notions from psychoanalytic and Field Theory.

THE HUBER AND LEONE PAPER

Contributions of the Paper

The authors provide reasons why the Fishbein model often provides fairly accurate predictions even though it is not conceptually correct. They also provide a model which is more sound, even though, in my opinion, it is still not quite conceptually correct either.

Questions with the Paper

The authors are incorrect in stating that the Fishbein model does not include negative outcomes. Any belief statement about the attitude object is acceptable to the model whether it deals with positive or negative outcomes. Perhaps the authors actually mean that after an outcome has been determined, its opposite (complement) is not incorporated in the model even though it in fact belongs there. However, Fishbein manages to salvage the model by measuring probabilities in bipolar fashion, because the "utility/disutility" of this opposite (complement) consequence is (-1) times the "utility/disutility" of the included consequence. Even though, personally I partly agree with the authors' explanation why the model works, I am fairly confident Fishbein would not. I am confident he does not see the bipolarity of belief strengths as a method-logical shortcut to truncate the true underlying model. I think he would argue that a salient disassociation from something good is bad irrespective of the value of its opposite and that a disassociation from something bad is good irrespective of the value of its opposite.

My personal problem with the paper is that I do not believe the assumption of "bipolarity of utilities" is in many cases realistic, nor do I feel it is adequate to include only a salient outcome and its opposite in the model. Most outcomes are not perceived to be simply dichotomous, but in fact have several levels, of which all salient ones must be incorporated into the model.

Why make the unreasonable assumptions of dichotomy of attribute levels and bipolarity of utilities if it is not necessary. Ahtola (1975) has proposed the model which does not rely on these assumptions. A good example where a dichotomous salient attribute does not meet the required "bipolarity of utilities" is sex. Sex of an insurance salesperson may be a salient attribute to a client. He may feel that female is OK but male would be even better or vice versa. Thus, both attribute categories have positive or at least non-negative "utility". Such attribute as "temperature of tea" is perceived to have several levels and it is certainly unlikely that most people would evaluate the extreme categories (hot/cold) to have opposite evaluations. In summary, if the purpose of this paper is to suggest a reason why the Fishbein model sometimes works in spite of being conceptually incorrect, I agree with the authors, but if the purpose is to justify the conceptualization of the Fishbein model, I disagree with the authors. It is not clear which stand the authors are taking.

THE JAIN, MAHAJAN, AND MALHOTRA PAPER

Contributions of the Paper

This paper is especially useful for someone interested in multiattribute preference modeling, but not willing to go through the trouble of reading the original literature. An amazing amount of literature is described and compared in only four pages.

Questions with the Paper

By attempting to do so much in only four pages it is natural that several important issues are touched very lightly or totally left out. Also, perhaps for the same reason, the paper is very descriptive with little critical analyzing.

Somewhat surprisingly, one of the issues totally left out is the theoretical justification for the model(s) and the implications of the theory/theories to the issues discussed. Also, the well known and closely related "Expectancy-Value" type of multiattribute preference models (see, for example, Cohen, Fishbein and Ahtola, 1972) have not at all been discussed in the paper. It is unfortunate because these models have stronger theoretical bases than the models described in the paper.

At the end of the discussion of attribute identification the authors recommend the use of Kelley's Repertory grid type of method. At the same time, they also emphasize the need to keep the number of attributes to the minimum. One problem with the grid method is that it tends to make previously nonsalient attributes salient when respondents struggle to find differences where they otherwise would not perceive them. This, of course, is a threat to the external validity and adds to the number of attributes. I personally prefer a free elicitation method (Ahtola, 1973).

It might be of interest to note that the "fundamental additive model" proposed by the authors is identical to Ahtola's Vector model (Ahtola, 1975) with only one difference. This crucial difference is that I defined Xijlk as the lth consumer's subjective probability that kth alternative possesses attribute i at the level j. My model was developed to solve some conceptual problems I felt were present in the Fishbein model.

Going back to the models discussed in the paper, I find it difficult to see the close connection between the "general model" and the "traditional compositional model" and for that reason the connection between the commonly used compositional and decompositional models. In the fundamental model anjl is a measure of value while in the traditional compositional model it is a measure of attribute importance. These two concepts are not identical (Cohen, Fishbein and Ahtola, 1972).

Also, the "traditional compositional models" vary widely in how Xilk is conceptualized. Very commonly, it is defined in terms of how satisfactory the lth consumer feels alternative k is on attribute i. That is, Xilk is often a measure of evaluation in the traditional model while in the fundamental (and compositional) model it is a measure of perception.

Perhaps in their eagerness to show that the commonly used compositional and decompositional multiattribute preference models are based on the same underlying "fundamental model" the authors have not recognized some fundamental differences among the compositional models and between these models and the proposed "fundamental model".

The only question I have about decompositional models is why so much energy is spent to derive interval scaled measures from ordinal responses. Why not measure the responses on interval scales in the first place?

DIRECTIONS FOR FURTHER RESEARCH

Research using or developing multiattribute attitude models has suffered from lack of theoretical justification, from lack of proper validation, as Bonfield so forcefully demonstrated, and from the bewildering number of slightly different conceptualizations and operationalizations. Very little conceptual and/or empirical work has been done to help the potential users of these models to select the most valid and useful one for their purposes. To some extent Jain, Mahajan, and Malhotra attempted to provide this sort of assistance. Much more work needs to be done, however. The comparative studies need to be sensitive to real differences among the models, describe the theoretical bases of the models, evaluate and empirically test the reliability and validity of alternative operationalizations of the same model, and demonstrate conceptually and empirically why a given approach is superior in a given situation, taking into consideration the purpose of the research.

REFERENCES

Olli T. Ahtola, An Investigation of Cognitive Structure within Expectancy-Value Response Models (Doctoral Dissertation, University of Illinois) Ann Arbor, Michigan, University Microfilms, 1973, No. 74-5517.

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

Donald T. Campbell and Donald W. Fiske, "Convergent and Discriminant Validation by the Multitrait-Multimethod Matrix," Psychological Bulletin, 56 (1959), 81-105.

Joel B. Cohen, Martin Fishbein and Olli T. Ahtola, "The Nature and Uses of Expectancy-Value Models in Consumer Attitude Research," Journal of Marketing Research, 9 (November 1972), 456-60.

Lee J. Cronbach and Paul E. Meehl, "Construct Validity in Psychological Tests," Psychological Bulletin, 52 (1955), 281-302.

Martin Fishbein, "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, M. Fishbein (ed.), (New York: John Wiley & Sons, 1967), 389-400.

Calvin S. Hall and Gardner Lindzey, "The Relevance of Freudian Psychology and Related Viewpoints for the Social Sciences," in The Handbook of Social Psychology, 2nd Edition, G. Lindzey and E. Aronson (eds.), (Reading, Mass.: Addison - Wesley, 1968), Volume One, 245-319.

Irving Sarnoff, "Psychoanalytic Theory and Social Attitudes,'' Public Opinion Quarterly, 24 (1960), 251-79.

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