Marketing Underground: an Investigation of Fishbein's Behavioral Intention Model

David E. Weddle, University of California, Los Angeles
James R. Bettman, University of California, Los Angeles
[ to cite ]:
David E. Weddle and James R. Bettman (1974) ,"Marketing Underground: an Investigation of Fishbein's Behavioral Intention Model", in NA - Advances in Consumer Research Volume 01, eds. Scott Ward and Peter Wright, Ann Abor, MI : Association for Consumer Research, Pages: 310-318.

Advances in Consumer Research Volume 1, 1974    Pages 310-318


David E. Weddle, University of California, Los Angeles

James R. Bettman, University of California, Los Angeles

Consumer researchers have studied the construct of attitude and the problem of the relationship between attitude and behavior in great depth. In recent years there have been many studies of multiple attribute attitude models inspired by the models of Rosenberg (1956) and Fishbein (1967). In these models objects are conceptualized as having many attributes, or actions as having many consequences. The attitude toward an object or action is then modeled by an expectancy-value approach. For each attribute or consequence, the likelihood that it is associated with the object or action is assessed, as is its value. The value and likelihood are then multiplied for each attribute or consequence and these products are summed. The most common model used in marketing, for attitude toward an object, can be expressed as EQUATION where Ao is the attitude (affect) toward some object o, ai is the evaluation of attribute i, Bi is the probability that object o possesses attribute i and n is the number of attributes.

The studies performed in marketing using analogous models are well summarized by Wilkie and Pessemier (1972). They note several controversies that have arisen in applying the model, two of which are of particular interest in this study. First, the form of the model which should be used has evoked a great deal of controversy. Several researchers have proposed that the model is better from a predictive standpoint if the evaluative term ai is dropped from the formulation (Sheth and Talarzyk, 1972; Moinpour and MacLachlan, 1971), while others have argued forcefully for its inclusion (Cohen and Ahtola, 1971; Cohen, Fishbein and Ahtola, 1972). Sheth (1970) and Cohen and Ahtola (1971) have also suggested that the model be disaggregated so that the component for each attribute becomes an independent variable. In addition to the controversies on model form, attribute selection problems have received some study. For example, Wilkie and Weinrich (1972) have shown that using more attributes in the model may lead to worse predictive accuracy than if a smaller number were used. These two issues of model form and attribute selection will be examined in this study, using measures closer to those described by Fishbein than used in previous studies, in an attempt to determine whether measurement problems affect the results obtained.

Although several of the studies in marketing have used Ao to predict either reported or actual behavior (Cohen and Ahtola, 1971; Bass, Pessemier and Lehmann, 1972), Fishbein (1972) is careful to point out that Ao is an attitude toward an object, not toward performing a given behavioral act, and hence Ao may not necessarily be closely related to behavior. He proposes a model of behavioral intention to remedy this situation which includes both an attitudinal and a reference group influence component. This model has not been applied in marketing, even though it appears to be a most appropriate model, although researchers such as Sheth (1970) have proposed similar behavioral intention models of their own. Accordingly, a final purpose of this study is to investigate the use of this behavioral intention model in a marketing context.

The particular context used is a relatively unusual buying situation, that of an undergraduate student deciding whether or not to buy a term paper for a nonmajor related course. This situation has received a great deal of attention recently mainly due to debates about the ethics and legality involved, with the result that much of the buying occurs in an 'underground' type of environment. This particular situation was chosen because both internal attitudes and external reference groups should be relevant, and thus it is an appropriate testing situation for the behavioral intention model.

In summary, then, the purpose of this paper is threefold: 1) To present an initial application in marketing of the Fishbein (1967) behavioral intention model for an interesting empirical situation; 2) to reexamine controversies over the appropriate form for multiple attribute models. and 3) to examine problems in selecting appropriate attributes to describe the situation under consideration.


Before examining the particular data collection procedures used, the models investigated are described. First models of attitude are developed and then models of behavioral intention.

The attitude construct developed by Fishbein for use in predicting behavior is attitude toward the act, AaCt, which is intended to be very situation specific, that is, an attitude about a particular action in a given situation (Fishbein, 1972, p. 247). Fishbein develops a model for Aact which parallels that for Ao above:


where bi is the probability that performance of the act will lead to consequence i, ai is the evaluation of consequence i and n is the number of consequences. If the arguments of Sheth and Talarzyk (1972) and others are accepted, then an alternative model of AaCt would be one omitting the evaluative measure ai as in Equation 2:


A third model of Aact, proposed by Cohen and Ahtola (1971), is that the individual consequence components are disaggregated and given weights wi to be empirically estimated: [The wi notation is used in all equations for convenience, and it is not intended that the wi are the same in all equations where they appear.]


Finally, Sheth (1970) proposed a disaggregated version of model 2, where the ai are deleted:


By comparing Models 1 and 2 and Models 3 and 4 some insights can be gained into the value of the ai term. Comparing Models 1 and 3 and Models 2 and 4 shows the effect of disaggregation.

Fishbein's (1967) behavioral intention model is a modified version of Dulany's (1968) theory of propositional control. In this model behavioral intention is a function of attitude toward the act and a normative component dealing with external referents. This component is defined to be the sum of products of two measures for each referent; normative belief (NB) or the respondent's belief about what the particular referent expects him to do, and motivation to comply (MC) or his desire to comply with the referent. The total model is given by:


where BI is behavioral intention, NBj is the normative belief for referent j, MCj is the motivation to com!ply for referent j, m is the number of referents and wo and w1 are empirically estimated weights for the attitudinal and normative components. Also ai, bi and n are as before. [An alternative model, where the NBj MCj terms are disaggregated was investigated, but the results were similar to those of the aggregate model, and are not presented here.]



Alternative versions of this model using the alternative models for AaCt of equations 2, 3 and 4 can also be examined and are given as equations 6, 7 and 8 resPectivelY:

EQUATION (6) - (8)

These eight models are the total set to be examined in this study. Now the data collection procedures must be considered.

An initial pilot study was run to determine relevant consequences of the act of buying a term paper and relevant referents. Based on an analysis of . forty open ended questionnaires, the nine consequences and five referents given in Table 1 were developed, A questionnaire designed to measure ai, bi, NBj, AaCt and BI was then developed. All were measured using 11-point bipolar scales. For example, bi was measured by asking how likely the respondent felt each consequence was if a term paper were purchased, with a scale from -5(highly unlikely) to +5 (highly likely). BI and MCj were measured on the same type of scale. The ai were measured by asking the subjects to evaluate each consequence on a scale from -5 (very bad) to +5 (very good). AaCt and NBj were measured on similar scales.

Subjects were students in several undergraduate university classes. The questionnaire was distributed with return envelopes to 139 students. In the instructions the students were guaranteed anonymity and were asked to complete the questionnaires after leaving the classroom situation. They were also given instructions for returning the completed questionnaire. The total number of returned questionnaires was 70; however, missing answers for one or more components caused rejection of 13 responses, for a total usable sample of 57. This is a rather small sample. However, there are still adequate degrees of freedom for model testing. A more serious problem concerns potential bias in the responses, that is, are students who did not respond more prone to buy term papers? Arguments can be made denying that such a bias exists. First, complete anonymity was assured by the structure of the return process even though no evidence is available as to whether anonymity was perceived. Thus, there is no particular reason for those predisposed to buy term papers to not reply. In any case, the major interest in this study is in the fit of the models rather than the particular coefficients obtained, and there is no theoretical reason to expect the fit of the models to be affected by any possible sample bias. Second, there was a reasonable range of responses for BI. [Approximately 24 of the sample expressed a reasonable likelihood of buying a term paper which does not seem out of the range one would expect.] Hence, it was felt that model testing could proceed.

Multiple regression was used to analyze Models (1) through (8), given the measures described above. [The model numbers refer to the equations in the text.] In addition to using all nine consequences and all five referents in the models, a reduced set of consequences and referents was used. The literature on multiple attribute models suggests that only salient consequences and referents should be used since summing up nonsalient items may only add error to the aggregative models. Wilkie and Weinrich (1972) confirm this assertion that more attributes are not necessarily better for prediction. [Of course this is true only for the aggregated models. For the disaggregated models, eliminating variables can only reduce R2. However, adjusted values of R2 may increase, due to the different numbers of variables being used.] Accordingly, the individual correlation coefficients between the ai bi, NBj and MCj and the dependent variables BI and AaCt were examined. The ad hoc rule used for selecting consequences and referents was to select those with significant correlations. Although this method is ad hoc, it was felt that correlation magnitude should be related to salience. Ideally this method for attribute selection should have been tested using a split sample technique, since the method does capitalize on the data. However, the sample size was too small to permit this. Based on this analysis three consequences and two referents were selected for the reduced set; consequences referring to avoiding busywork (C3), cheating (C6) and leaving time for more relevant studies (C8) and referents of friends (R2) and parents (R5). The regressions for Models (L) through (8) were also run for this reduced set of variables.




A summary of the results of the two sets of runs is given in Table 2 where adjusted R2 values (Kmenta, 1971, p. 365) and significance levels are presented. All models were run for 57 subjects. The results for the reduced set case are in general much more encouraging than for the entire set of variables. Significance levels are much higher, and the R2 values for the aggregated models are much better. This supports the results of Wilkie and Weinrich (1972) who also found smaller numbers of attributes performed better. Since the models for the reduced set perform better, they are the models analyzed with respect to model form and any substantive insights.

The best model for AaCt is the disaggregated form suggested by Cohen and Ahtola, Model (3). The original Fishbein model, Model (1), also does well. The models eliminating the ai term, Models (2) and (4), do slightly worse. These differences in R2 are quite small, and given that only 57 subjects were used, must be considered only suggestive. For the behavioral intention models the results are more conclusive. The models using beliefs only, Models (6) and (8), do substantially worse. Using Williams test for equality of dependent correlation coefficients as suggested by Dunn and Clark (1971), the correlation for Model (5) is significantly different from the correlation for Model (6) at p<.07. None of the other pairs of models had correlations significantly different at p<.10. The best model for behavioral intention is the original Fishbein model, Model (5), with the disaggregated version, Model (7), quite close in performance. The prediction of behavioral intention is also reasonably good as shown by the adjusted R2 values.

One rationale for using multiple attribute attitude models is the insights that can be obtained by examining the estimated coefficients of the models. Standardized regression weights and their significance levels are presented in Table 3 for the reduced variable set models. The most interesting findings here are that in the disaggregated models using both ai and bi, Model (3) for AaCt and Model (7) for BI, the consequences of cheating and leaving time for more relevant studies are the most important for prediction. The coefficients for avoiding busywork are not significant. Secondly, in the Fishbein model for behavioral intentions, Model (5), the attitudinal component is more important for prediction than the normative component (with standardized weights of .49 and .24 respectively). This is somewhat surprising, since normative influences might be thought to be quite strong in this situation.


The performance of the Fishbein behavioral intention model,Model (5), is quite encouraging in this study. The fit of the model is relatively good and on theoretical grounds it has been argued that the components of the model, AaCt and the normative component, are more suited for examining behavioral phenomena than Ao which has been the main model used in marketing. The behavioral intention model provides a particularly appropriate framework for studying consumer choice since it models both internal feelings (AaCt) and external reference group effects (NBj MCj). These external effects in particular have not been formally included in most consumer choice models. One possible problem with the approach is that the consequences used in AaCt may not be independent of the normative component. For example, the ai values for cheating, hurting fellow students, leaving time for more relevant studies and so on might very well be influenced by what referents think. Perhaps Kelman's (1961) partition of processes of social influence into compliance, identification and internalization is relevant here. Compliance refers to accepting influence from a referent because a favorable reaction is desired. Identification occurs when influence is accepted from a referent because it maintains a desired role relationship with the referent. Finally, internalization influence is accepted by a person because it agrees with his value system. In this terminology, internalization and perhaps identification may be subsumed under Aact, whereas compliance is modeled by the NB MCj. This might explain why the attitudinal component was more important than the normative component for prediction.



In addition to showing the potentialities of the behavioral intention model, several methodological issues were studied. The results showed rather dramatically that a smaller number of more salient attributes performs much better than a larger number of attributes, replicating results obtained by Wilkie and Weinrich (1972). How attributes should be selected a priori is still an open question, as the method used in this study was quite ad hoc. Wilkie and Weinrich (1972) suggest using the concept of determinism proposed by Myers and Alpert (1968).

Finally, the use of both evaluative and belief components was supported rather than the beliefs-only models proposed by Sheth and Talarzyk (1972). This finding was particularly pronounced for the behavioral intention models. The disaggregation of the AaCt component as suggested by Cohen and Ahtola (1971) looks promising although the aggregated models perform about as well. Given this comparability in performance, the added diagnostic power of the disaggregated model may be desirable.


Ajzen, Icek, and Martin Fishbein. Attitudinal and Normative Variables as Predictors of Specific Behaviors. Journal of Personality and Social Psychology, 27 (July 1973), 41-57.

Bass, Frank, M. Edgar A. Pessemier, and Donald R. Lehmann. An Experimental Study of The Relationships between Attitudes Brand Preference, and Choice. Behavioral Science, 17 (November 1972), 532-41.

Cohen, Joel B., and Olli T. Ahtola. An Expectancy X Value Analysis of the Relationship between Consumer-Attitudes and Behavior. In D. Gardner, Ed., Proceedings of the Second Annual Conference of the Association for Consumer Research, 1971, 344-364.

Cohen, Joel B., 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), 456-460.

Dulany, Don E. Awareness, Rules, and Propositional Control: A Confrontation with S-R Behavior Theory. In T. Dixon and D. Horton, Eds., Verbal Behavioral and General Behavior Theory. New York: Prentice Hall, 1968.

Dunn, Olive Jean, and Virginia Clark. Comparison of Tests of the Equality of Dependent Correlation Coefficients. Journal of the American Statistical Association, 66 (December 1971), 904-908.

Fishbein, Martin. Attitude and the Prediction of Behavior. In Martin Fishbein, Ed., Readings in Attitude Theory and Measurement. New York: John Wiley & Sons, 1967, 477-492.

Fishbein, Martin. The Search for Attitudinal-Behavioral Consistency. In Joel B. Cohen, Ed., Behavioral Science Foundations of Consumer Behavior. New York: The Free Press, 1972, 245-252.

Kelman, Herbert C. Processes of Opinion Change. Public Opinion Quarterly, 25 (1961), 57-78.

Kmenta, Jan. Elements of Econometrics. New York: MacMillan 1971.

Moinpour, Reza, and Douglas L. MacLachlan. De Relations Among Attribute and Importance Components of Rosenberg-Fishbein Type Attitude Model: An Empirical Investigation. In D. Gardner, Ed., Proceedings of the Second Annual Conference of the Association for Consumer Research, 1971, 365-75.

Myers, James H., and Mark I. Alpert. Determinant Buying Attitudes: Meaning and Measurement. Journal of Marketing, 32 (October 1968), 13-20.

Rosenberg, Milton J. Cognitive Structure and Attitudinal Affect. Journal of Abnormal and Social Psychology, 53 (1956), 367-372.

Sheth, Jagdish N. An Investigation of Relationships Among Evaluative Belief, Affect, Behavioral Intention and Behavior. Unpublished paper, University of Illinois, 1970.

Sheth, Jagdish N., and W. Wayne Talarzyk. Perceived Instrumentality and Value Importance as Determinants of Attitudes. Journal of Marketing Research, 9 (February 1972), 6-9.

Wilkie, William L., and Edgar A. Pessemier. Issues in Marketing's Use of Multi-Attribute Attitude Models. Institute Paper 365, Krannert Graduate School of Industrial Administration, Purdue University, 1972.

Wilkie, William L., and Rolf P. Weinreich. Effects of the Number and Type of Attributes Included in an Attitude Model: More is Not Better. In M. Venkatesan, Ed., Proceedings of the Third Annual Conference of the Association for Consumer Research, 1972, 325-340.