The Cognitive Foundations of Attitudes: Some Implications For Multi-Attribute Models


Bobby J. Calder (1975) ,"The Cognitive Foundations of Attitudes: Some Implications For Multi-Attribute Models", in NA - Advances in Consumer Research Volume 02, eds. Mary Jane Schlinger, Ann Abor, MI : Association for Consumer Research, Pages: 241-248.

Advances in Consumer Research Volume 2, 1975      Pages 241-248


Bobby J. Calder, University of Pennsylvania

[Bobby J. Calder is Associate Professor of Marketing and Organizational Psychology at The Wharton School, University of Pennsylvania.]

Multi-attribute models of attitude of the type proposed by Fishbein and Rosenberg have received considerable attention in consumer research. Part of the appeal of these models is probably due to their specification as equations. This level of formalization may be more apparent than real, however. As shown in this paper, these models have not been mathematically derived nor have they been induced from empirical relationships. Rather they were originally based on some rather primitive ideas about the cognitive functioning underlying attitudes. They have subsequently become somewhat autonomous from any theoretical justification except the implicit and ill-defined idea that attitudes are based on beliefs. It is argued that in order to improve multi-attribute models it will be necessary to develop a greater understanding of the cognitive foundations underlying them. It is suggested that multiattribute models are best viewed as "processing rules" and as such may depend on the structural representation of information in the cognitive system. This implies that no single form of a multi-attribute model can be established empirically, and that many forms may be appropriate.

Multi-attribute attitude models have become commonplace in consumer behavior research. They have been applied to attitudes towards products ranging from convenience goods to capital equipment purchases. Although the reason for this popularity is not entirely clear (and might make an interesting study in the sociology of science at that), the usual rationale is that these models provide information about attitudinal structure as well as a basis for behavioral prediction. Knowledge about attitudinal structure, of course, increases the level of our understanding of consumer processes and supplies diagnostic information for decision-making.

Unlike simpler, unidimensional scales, these models are specified in terms of the beliefs about the salient attributes of a product which underlie a person's attitude. Typically the models take a linear compensatory form which may be-expressed as Aj=Ei=1n wi Bij, where Aj is an overall attitude toward brand j, Bij is a belief about some attribute i of the brand, and w is the weight given to belief i.

Many issues have been actively debated in connection with these multiattribute models. For one, the variables involved can be conceptualized quite differently (e.g., Bass and Talarzyk, 1972; Cohen, Fishbein, and Ahtola, 1972; Sheth and Talarzyk, 1972). The weighting factor w may represent the importance of an attribute and the belief factor an evaluation of the brand in terms of that attribute. Or, in a Fishbein-type model, the w may be taken as the affect associated with an attribute and the belief factor as the attribute's cognitive expectancy. Other issues center around (1) whether the weights should even appear in the model, (2) whether a disaggregate model in which a regression is performed on all the terms in the model without summation is more appropriate, and (3) whether cross-sectional or individual level of analyses :;s more suitable (cf. Wilkie and Pessemier, 1973). Finally, even the basic linear compensatory form of the models may be challenged (e.g., Wright, 1972). Instead of a model in which positive and negative attributes cancel each other, conjunctive models in which all attributes must exceed minimum values or disjunctive models in which one or more attributes must exceed some critical value are possible, as well as lexicographic models in which attributes are considered sequentially.

In short, multi-attribute models possess that susceptibility to endless variation and refinement that researchers cannot resist. Plus, they do have predictive ability, But what of their status as compositional models of attitude? In what sense are multi-attribute models related to underlying attitudinal structure? The purpose of this paper is to redirect attention toward such questions and to discuss a few of the issues involved.


As correctly noted by Wilkie and Pessemier (1973), "the basic linear compensatory model was developed in social psychology as a static approach to describing an existing attitude structure"(p.437). The most immediate antecedents of the model appear to lie in the early work of Fishbein (1967a) and Rosenberg (1956), as is by now widely recognized. Relatively little attention, however, has been paid to the basis for these original formulations. Fishbein (1967a) initially adapted the perspective of Hullian learning theory. He postulated that an individual first forms a concept, associated with which are a number of responses. These responses constitute a habit-family-hierarchy called a belief system. Beliefs contain mediating evaluative responses which summate through classical conditioning and generalization to become an attitude. In dealing with specific behavioral acts, Fishbein (1967b) adapted the language of Dulany's theory of propositional control as well. Beliefs are thought of as analogous to cognitive expectations or hypotheses about reinforcement contingencies, where the reinforcement possesses some affective value. Rosenberg (1956),in contrast, developed his model in connection with a hypothesized tendency for people to maintain a state of psychological balance between the affective and cognitive components of attitude (cf. Calder, 1973). Multiplying the signs of a cognitive element and an affective element reveals whether the pair is consistent or not. Such thinking evidently led Rosenberg to employ the sum of such products as an index of attitude.

It is apparent that the connection between the original formulations of the Fishbein and Rosenberg models and any underlying attitude theory were at best tenuous, and, in any case, bear no resemblance to current discussions of the linear compensatory model. More recently, it has been argued that the model is basically a subjective expected utility (SEU) approach. The functional similarity of the SEU model and the linear compensatory attitude model does not really provide any theoretical rationale for the attitude model, however. Unlike the attitude model, the maximization of expected utility may be derived from more basic choice axioms. Such thinking does provide a clue as to the implicit thinking behind most investigations of the linear compensatory attitude model.

Much of the appeal of these models is probably due to their specifications as equations. This level of formalization may be more apparent than real. These models have not been mathematically derived nor have they been systematically induced from empirical relationships. Rather they were originally based on some rather primitive ideas about the cognitive functioning underlying attitudes. They have subsequently become somewhat autonomous from any theoretical justification except the implicit and ill-defined idea that attitudes are based on beliefs.

Most investigations of the linear compensatory attitude model are perhaps best characterized as "paramorphic representations." Hoffman (1960) first employed this term in describing the use of linear multiple regression equations as models of human judgment. The equations are intended to predict judgments and to explain some aspects of the judgment process (such as cue weighting), but are not intended to serve as models per se of the underlying psychological process. In the case of the multi-attributes models under consideration, a "paramorphic" research strategy is defensible. But it is a far step from the claims of understanding and diagnosing attitudinal structure which are usually advanced for the models.


It is our contention that an "information-processing" approach will be more fruitful than the typical "paramorphic" approach in further investigating multiattribute attitude models, whether of the linear compensatory variety or not. The scope of such an information-processing approach was sketched in the early work of Carl Hovland and has been expounded upon by McGuire (1972). Persuasion is considered a multi-stage process involving information (communication) delivery, attention to the information, comprehension, acceptance of the information, persistence of the acceptance, and behavior consistent with the acceptance. Such a framework serves to emphasize the role of cognitive processes as the underlying determinants of attitude.

Of central importance is the role of comprehension. Traditionally comprehension has been identified mostly with rote learning of communication contents or object (product) attributes. It has been argued, however, by Greenwald (1968) and Calder, Insko, and Yandell (1974) that such a view is much too narrow. Comprehension is an active process. Beliefs do not originate only with external sources of information. Individuals also generate new beliefs in processing information and engage existing beliefs stored in memory.

Although cognitive theories of such comprehension are far from complete, any theory must have two crucial aspects: representation and process. Representation refers to the internal information structure; process refers to the rules for manipulating this structure. Cognitive psychologists have recently begun to make a distinction in terms of representation between memory systems and memory codes. Memory systems involve different mechanisms for storage (e.g., long-term versus short-term stores) and their organizational structure. Memory codes refer to the form in which information is stored. A growing number of studies of imagery have investigated visual (iconic) forms of storage. Linguistic codes have continued of course, to receive the most attention. The processing of the information represented has generally been treated in terms of mathematical models, though there has been a trend toward the use of computer simulations (e.g., Anderson and Bower. 1973).

What such considerations have to do with multi-attribute models of attitudes is this. Multi-attribute models are best thought of as putative processing rules. But by themselves such processing rules do not constitute a cognitive theory. It is necessary to specify what it is they operate on, and this specification must be done in both theoretical and operational terms. Let us assume for our purposes that beliefs are represented in the form of linguistic codes. While we have no really adequate theories of the functional units of such codes, it is not unreasonable to assume a propositional format, not unlike the simple declarative sentence (Mehler, 1963; Fillenbaum, 1966; Crothers, 1972). How might these unitary beliefs be organized in memory? The memory system involved may be called active or operational memory. Items in this system are activated either from new input existing in the short-term store or from previously retained material in the long-term store. As is fairly well known, the size of this memory system appears to be constrained to something like six to eight units, though the units themselves may be more or less complex (Miller, 1956). This does not mean that we are limited to considering only seven or so beliefs at a time. It does mean that the organization of beliefs in active memory is usually more complex than a simPle list.

According to one hypothesis (Mandler, 1967), a given unit can itself be expanded to seven or so units. The resulting organization is a hierarchical structure of units which are successively redefined at higher levels of organization. Such a pattern is not a list but a list structure (cf. Newell and Simon, 1963; Reitman, 1965). Thus, beliefs relevant to a given topic might be organized so that concepts of increasingly greater specificity are encountered as one moves down through the structure. At the top are very general topics while beliefs at subsequent levels become much less general. We can complete this picture by reconciling these constraints with the seeming ability of people to integrate an endless amount of information in the long run by following Bower (1970) in envisaging metaorganizations of linked hierarchical list structures.

Now consider how the typical multi-attribute attitude model is operationalized. Respondents are given a list of beliefs on a piece of paper which is thought to correspond with those beliefs which are in fact activated by them in considering a given product. There are two things which could happen, depending on whether one takes a pessimistic or an optimistic point of view. Respondents may faithfully search their own belief structure and report on the belief which most closely matches the statement they are presented with. The researcher then applies his own multi-attribute attitude model to this data and derives a prediction of the respondent's attitude. The problem is that this attitude model may or may not make any sense in terms of the belief structure which the person must process in expressing an attitude, It may or may not be a reasonable model of the cognitive processing strategy. Take the case of the typical linear compensatory model. If the person has not had much exposure to a product or time to think about it, so that his belief structure is very simple, perhaps even as simple as the list presented to him, the linear compensatory model may do quite well, This model is an effective way of integrating small, unorganized information structures. If, however, the person has a more complex belief system along the lines of the possibilities we have discussed, only a disjunctive or lexicographic model would be efficient as a cognitive processing strategy. In short, considerable attention must be given to the cognitive representation of beliefs before any processing model can be postulated.

This was the optimistic viewpoint, now for the pessimistic, In giving respondents a list of beliefs to process, we may be presenting them with a very simple problem-solving task. Shortly after working their way through the beliefs, they are asked to express their attitude. In reflecting on their attitude, it is just possible that instead of processing their own belief structures they simply process the information they have been presented with. The linear compensatory model may be expected to do quite well in this case. Unfortunately, respondents are simply working on a task quite different from the one intended.


It is our feeling that most applications of multi:attribute attitude models have been arbitrary and indiscriminate. Models have been applied with little consideration for the internal structure being processed or, even worse, he task people are actually performing. The answer to these difficulties lies in two directions. For one, at least as much attention needs to be paid to theoretical accounts of the organization of beliefs as to putative processing models. Further work along the lines initiated by Bettman (1974a,b) in the area of decision net models should prove useful, but there is a great need for innovative work. Second, it is doubtful that simply presenting respondents with lists of beliefs should be further propagated as a research strategy. Considerable effort must be directed toward eliciting beliefs from respondents along with information about how these beliefs are structured in memory.

In closing, it should be noted that we are not arguing that presenting respondents with a list of beliefs is never a defensible research strategy. When the objective, for instance, is to measure the attitude of respondents, endorsement of beliefs as in a Thurstone scaling procedure has considerable merit. The beliefs endorsed are being used as indications of an underlying attitude, just as any other set of behaviors might, with sufficient theoretical justification, be used as an indication of attitude. This is a quite different objective, however, than that of testing a multi-attribute model of attitude.


Anderson, J. and Bower, G. Human associative memory. Washington, D.C.: Winston, 1973.

Bass, F. and Talarzyk, W. An attitude model for the study of brand preference. Journal of Marketing Research, 1972, 9, 93-96.

Bettman, J. Decision net models of buyer information processing and choice Findings, problems, and prospects. In G. Hughes and M. Ray (Eds.), Buyer/ consumer information processing. Chapel Hill: University of North Carolina Press, 1974. (a)

Bettman, J. Toward a statistics for consumer decision net models. Journal of Consumer Research, 1974, 1, 71-80. (b)

Bower, G. Organizational factors in memory. Journal of Cognitive Psychology, 1970, 1, 18-46.

Calder, B. Cognitive consistency and consumer behavior. In H. Kassarjian and T. Robertson (Eds.), Perspectives in consumer behavior. Glenview, Ill.: Scott, Foresman, 1973.

Calder, B., Insko, C. and Yandell, B. The relation of cognitive and memorial processes to persuasion in a simulated jury trial. Journal of Applied Social Psychology, 1974, 4, 62-93.

Cohen, J., Fishbein, M. and Ahtola, O. The nature and uses of expectancy-value models in consumer attitude research. Journal of Marketing Research, 1972, 9, 456-460.

Crothers, E. Memory structure and the recall of discourse. In R. Freedle and J. Caroll (Eds.), Language comprehension and the acquisition of knowledge. Washington, D.C.: Winston, 1972.

Fillenbaum, S. Memory for gist: Some relevant variables. Language and Speech, 1966. 9. 217-227,

Fishbein, M. A behavior theory approach to the relations between beliefs about an object and the attitude toward the object. In M. Fishbein (Ed.), Readings in attitude theory and measurement. New York: Wiley, 1967. (a)

Fishbein, M. Attitudes and the prediction of behavior. In M. Fishbein (Ed.), Readings in Attitude theory and measurement. New York: Wiley, 1967. (b)

Greenwald, A. Cognitive learning, cognitive response to persuasion, and attitude change. In A. Greenwald, T. Ostrom and T. Brock (Eds.), Psychological foundations of attitudes. New York: Academic Press, 1968.

Hoffman, P. The paramorphic representation of clinical judgment. Psychological Bulletin, 1960, 57, 116-131.

Mandler, G. Organization and memory. In K. Spence and J. Spence (Eds.), The psychology of learning and motivation, I. New York: Academic Press, 1967.

McGuire, W. Attitude change: The information-processing paradigm. In C. McClintock (Ed.), Experimental social psychology. New York: Holt, 1972.

Mehler, J. Some effects of grammatical transformations on the recall of English sentences. Journal of Verbal Learning and Verbal Behavior, 1963, 2, 346-351.

Miller G. The magical number seven, plus or minus two: Some limits on our capacity for processing information. Psychological Review, 1956, 63, 81-97.

Newell, A. and Simon, H. Computers in psychology. In R. Luce, R. Bush, and E. Galanter (Eds.), Handbook of mathematical psychology, I. New York: Wiley, 1963.

Reitman, W. Cognition and thought: An information-processing analysis, New York: Wiley, 1965.

Rosenberg, M. Cognitive Structure and attitudinal affect. Journal of abnormal and social psychology, 1956, 53, 368-372.

Sheth, J. and Talarzyk, W. Perceived instrumentality and value importance as determinants of attitudes.Journal of Marketing Research, 1972, 9, 465-467.

Wilkie, W. and Pessemier, E. Issues in marketing's use of multi-attribute attitude models. Journal of Marketing Research, 1973, 10, 428-441.

Wright, P. Consumer judgment strategies: Beyond the compensatory assumption. In M. Venkatesan (Ed.), Proceedings of the third annual conference of the Association for Consumer Research, 1972.



Bobby J. Calder, University of Pennsylvania


NA - Advances in Consumer Research Volume 02 | 1975

Share Proceeding

Featured papers

See More


G10. The Effects of self-construal on evaluations of brand logo colors

Eunmi Jeon, Sungkyunkwan University
Myungwoo Nam, Georgia Tech, USA

Read More


Ecce Machina Humana: Examining Competence and Warmth in Consumer Robots The two fundamental social judgment dimensions-competence and warmth-are as relevant for judging consumer robots as for humans. We find that competence has an increasing positive eff

Read More


Social Class and Prosocial Behaviors

Yan Vieites, Brazilian School of Public and Business Administration, Brazil
Eduardo B. Andrade, FGV / EBAPE
Rafael Burstein Goldszmidt, Brazilian School of Public and Business Administration, Brazil

Read More

Engage with Us

Becoming an Association for Consumer Research member is simple. Membership in ACR is relatively inexpensive, but brings significant benefits to its members.