Future Directions in Attitude Theory


Michael J. Ryan (1980) ,"Future Directions in Attitude Theory", in NA - Advances in Consumer Research Volume 07, eds. Jerry C. Olson, Ann Abor, MI : Association for Consumer Research, Pages: 356-358.

Advances in Consumer Research Volume 7, 1980     Pages 356-358


Michael J. Ryan, Columbia University


The three papers presented in this session have serious limitations yet each raises important issues. These issues relate to specifying attitudinal antecedents, determining the configuration of beliefs, and relationships between attitudinal components and behavior. Each of the papers will be discussed separately after which an attempt will be made to suggest a unified direction for examining the issues that were raised in each paper.


The title, abstract, and introduction of this paper propose an alternative multiattribute attitude model. Fishbein's equal weighted additive attitude model, cited as the most highly researched of the consumer behavior models, is given as the present paradigm and Anderson's averaging model as the alternative. Information integration theory is, as accurately described by Frey and Kinnear, a method for describing how stimuli are combined, not a theory. Nonetheless, a theory to supplant or modify the additive model, which does not have a strong conceptual basis, would be welcome. Unfortunately, none is forthcoming. Instead, the unsettled adding versus averaging controversy is given against the Fishbein model, the adding model is restated as a regression equation, the averaging model is described, and maximum likelihood parameter estimation procedures are outlined.

The Fishbein model is seen as deficient because it does not deal realistically with objective data. This argument is specious. Whereas there may be objective data concerning the relationship between an object and an attribute, this does not guarantee a congruence with subjective beliefs about the relationship. The concepts of selective and perceptual distortion are commonly accepted, and there is ample evidence that even public policy disclosures may be selectively received or distorted. Thus, Frey and Kinnear's argument that objective data makes subjective expectancies deficient is naive. If there is a strong belief consistent with reality (e.g., the consumer believes that a three bedroom house has three bedrooms) this merely means that the evaluative aspect (ai) is given full weight for the attribute. Ahtola uses this same interpretation in his paper. Rather than show a deficiency, this argument illustrates the generality of the model as it covers situations where beliefs can be or not be congruent with reality.

They also confuse trustworthiness of the source of the information with attributes. Source effects, such as trustworthiness, are discussed in some detail by Fishbein and Ajzen (1975) who view them as antecedents to beliefs.

The maximum likelihood estimation procedures outlined by Frey and Kinnear are not unique to functional measurement. In fact the procedures developed by Joreskog and employed by Bagozzi and Bankrant use this procedure to estimate the path coefficients shown in their Figure 1 and to obtain the overall test of model fit. It would have been interesting to see exactly how these procedures could be applied to obtain an unambiguous test of belief configuration.

In short, this paper covers a good deal of old ground briefly. It has good potential for in depth development in two areas--conceptual development and methodological description--which would probably require two separate papers.


This paper's main value is the setting out of conditions under which the multicomponent model may exist and under which its components may predict behavior. An argument is not given in support of the first hypothesis which essentially says that the single and multiple component model exists simultaneously for the same people in the same situation. It seems these are mutually exclusive alternatives. There is also a problem of casualness in the methodology and/or the reporting as indicated by the following examples.

The second and third hypotheses concern the prediction of behavior yet the behavior occurred before the attitude measurement. Thus, the flow of effects in Figure 1 between the circled variables seems reversed. At best, this is a test of concurrent not predictive validity. The vector G in Equation 3, the relationship among the behaviors and their measures, is not identified, nor are the L or G coefficients in Figure 1 explained or discussed. The ai discussed in the text are not labeled in the Figure. The double headed path between the endogenous variables (B1 and B2) is not identified and the reader must speculate as to whether the values on these paths are coefficients or errors (z) in fitting the model. The Z measurement errors and L coefficients for the Likert and Guttman measures shown in Figure 1 are not mentioned in the text. Although each of the above issues may be considered a minor oversight in itself, when taken together they make it extremely difficult to comprehend the work.

The general analysis and interpretation is also open to question. The single component measurement model and multi-component structural model are supported with type 1 probability errors of .11 and .05, .13 respectively. Since support necessitates accepting the statistical null hypothesis with a low power c2 test, this is a tenuous position. As stated above, the multicomponent model, shown in the left half of their Figure 1, is an alternative to the single component model which would appear as follows:


Not only does it appear illogical to accept both alternatives, but the statistical test also shows the multi-component model fits the data (p = .33) better than the single component model (p = .11). Instead of developing a decision rule, we are asked to accept both a single and four component restricted maximum likelihood factor analysis solution on the same data.

Unfortunately, the four component model is not consistent with the literature review on hypothesis which indicates the following two component measurement model should have been tested and results reported.


Since the four component model is described as the most parsimonious to fit the data, it is possible that other models were tested and fit the data. If so, it would be interesting to know which models were tested, what decision rule was applied to reject others that fit the data, and why such a decision rule was not used to reject the single component measurement model. In addition, the test of the four component measurement model is questionable since only one of the four "underlying dimensions" has a multiple indicator. A four factor solution with five measures where four of the loadings and three of the communalities have been fixed is not compelling.

The six variable structural model in Figure 1 is shown to fit the data with no relationship among the four exogenous variables yet a relationship is alluded to in explaining the negative sign on a4. It is not apparent how these components can be both independent and multi-collinear. If the four component model shown in Figure 1 provided the most parsimonious fit, there is no multicollinearity worth mentioning. If there is multi-collinearity there may be less than four dimensions, but this seems unlikely if four components provided the most parsimonious fit. This seems illogical.

Bagozzi and Burnkrant have provided an interesting argument as to what conditions lead to a single or multiple component attitude model. Rather than test this proposition under varying conditions, they claim support for a single and four component measurement model and a four component predictive model under one condition conducive to a two component model. In summary, the introduction and conclusion are most interesting, the middle part most confusing.


Whereas a good deal of work involves simplifying multiattribute models, Ahtola counters this trend with the suggestion that a third component, uncertainty, should be considered a correction factor for his complex vector model. Despite the supporting evidence, which is mitigated by the way in which uncertainty was operationalized, there is a plausible alternative to his "correction factor" argument.

The evaluative aspect may enter an attitude model in three ways. First, as the overall affect, in the Bagozzi and Burnkrant use of the term, toward the attitude object. Second, as the evaluative aspect of a concept consistent with Ahtola's interpretation (aij). Third, as the evaluation of the perceived relationship between the attitude object and a concept. The third interpretation merely defines a level of specificity illustrated by the following example. A housewife may negatively evaluate pollution and strongly believe dishwashing detergents pollute yet she may not negatively evaluate the relationship between dishwasher detergent usage and pollution because of a belief that it contributes infinitesimally to overall pollution. Thus, the more specific evaluation seems appropriate. If the consumer is operating at this specific level, the evaluation is applied to the existing belief and if the belief is uncertain this would be reflected in the evaluation. This argument is not inconsistent with Ahtola's findings that the higher the uncertainty the stronger the negative evaluation. However, the argument together with his findings suggest that uncertainty is accounted for in the evaluative component. Thus, including uncertainty as a third component in order to effect a correction is a form of double counting. It seems uncertainty is more properly viewed as an antecedent variable.

The information theory notion used to operationalize uncertainty is inappropriate since its intended use is to transform probabilities to continuous functions not to tap a mental construct. Consequently, the author has a form of tautology arising from the use of two labels (uncertainty, probability) for the same construct. Hence, uncertainty (probability) operationalized as a continuous transformation of probability (uncertainty) does not produce a measure (indication of a construct) independent of the probability measure. Nor was this the intention of the information theorists who interchange the terms "probability" and "uncertainty" and use the log2 transformation merely to produce a continuous function appropriate for parametric statistical procedures.

In summary, an interesting idea with questionable empirical support more properly viewed as a search for an antecedent rather than third variable component.


A recent monograph (Fishbein and Ajzen 1975) provides an explicit conceptualization of the relationships among the variables treated in these three papers. Contrary to Bagozzi and Burnkrant's parallel relationships among four components, Fishbein and Ajzen traditionally view attitude as a tripartite concept where the components are ordered sequentially. They also modify the traditional cognitive component to be beliefs times evaluation--ai or specific ai as opposed to overall affect. Utilizing Ahtola and Fry and Kinnear's antecedents and Bagozzi and Burnkrant's nomenclature a structural model would appear as follows.


The observed indicators are omitted although it is relatively easy to specify them for all variables except Acog which is a combination of other constructs--beliefs and evaluations. Hopefully, the methodology referred to by Fry and Kinnear, since it appears consistent with Joreskog's framework, will provide an extension allowing an isomorphic modeling of the structural configuration of beliefs and evaluations and their relationships to empirical indicants. The Fishbein and Ajzen hierarchy should only be considered as a starting point since, as Bagozzi and Burnkrant have argued, there may be paths to B in addition to a5 under certain conditions. Other attitude theorists, most notably conditioning theorists, would also argue that belief does not necessarily precede affect.

Each of the three papers addresses an important issue. Despite their seemingly different orientations, they fit well into a unifying conceptual framework the recognition of which should lead to orderly future inquires and hopefully to a unified theory.


Fishbein, M. and Ajzen, I. (1975), Belief, Attitude, Intention and Behavior, Reading: Addison-Wesley.



Michael J. Ryan, Columbia University


NA - Advances in Consumer Research Volume 07 | 1980

Share Proceeding

Featured papers

See More


Is CSR for Sale? Investigating the Effects of Acquisition of Socially Responsible Brands on CSR Perceptions

Bianca Grohmann, Concordia University, Canada
Argiro Kliamenakis, Concordia University, Canada
H. Onur Bodur, Concordia University, Canada

Read More


A7. Credible Critters: Source and Message Expectancy Violation and Influence on Perceived Trustworthiness and Credibility

Justin Graeber, University of Texas at Austin, USA

Read More


Signaling Fun: Anticipated Sharing Leads to Hedonic Choice

Nicole Kim, University of Maryland, USA
Rebecca Ratner, University of Maryland, USA

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.