Comments on Investigations of the Theory of Reasoned Action

Thomas J. Page, Jr., University of Wisconsin-Madison
ABSTRACT - Three studies dealing with some aspect of the theory of reasoned action are discussed. Possible explanations for the results of all three studies are offered as are reasons for the contradictory results of the tests of the sufficiency assumption. Directions for future research that may help to resolve the issues are provided.
[ to cite ]:
Thomas J. Page, Jr. (1983) ,"Comments on Investigations of the Theory of Reasoned Action", in NA - Advances in Consumer Research Volume 10, eds. Richard P. Bagozzi and Alice M. Tybout, Ann Abor, MI : Association for Consumer Research, Pages: 106-109.

Advances in Consumer Research Volume 10, 1983      Pages 106-109


Thomas J. Page, Jr., University of Wisconsin-Madison

[The author wishes to thank Robert E. Burnkrant and Paul W. Miniard of The Ohio State University for their helpful comments during the preparation of this paper.]


Three studies dealing with some aspect of the theory of reasoned action are discussed. Possible explanations for the results of all three studies are offered as are reasons for the contradictory results of the tests of the sufficiency assumption. Directions for future research that may help to resolve the issues are provided.


Perhaps the most widely researched model in consumer behavior is Ajzen and Fishbein's (1980) theory of reasoned action. One needs only to look at the marketing consumer behavior and social psychology journals of the last several years for ample evidence of this research stream. Furthermore this line of research shows no signs of letting up as evidenced by the studies presented here and numerous others currently in progress (e.g. Burnkrant and Page 1982; Miniard and Page 1982). One reason the theory is so-widely investigated is the fact that the relationships are stated very explicitly. It is interesting to note however that while this explicitness stimulates an abundance of research on the theory it does not necessarily increase support for the stated relationships. Evidence of this is provided by two of the papers discussed here. Both test the sufficiency assumption of the theory but they obtain diametrically opposed results. Thus while the theory appears to be clearly formulated there are issues that still must be resolved.

The discussion of the three papers will not deal with methodological problems of the studies since the authors readily acknowledge shortcomings and problems of their studies and none of them appears to be extremely critical. What is important at this point are the conceptual issues that may account for the results obtained in the studies. Loken's paper on the effects of purchase information or attitude toward the behavior is discussed first, then Loken's and Crosby and Muehling's tests of the sufficiency assumption are examined.


Given that Loken sets out to manipulate beliefs about the store (Ebses) and look at its effects on attitude toward the product (Ap) and toward purchase (AB) she accomplishes what she sets out to do. She contends that positive and negative store information were expected to affect AB but not Ap. Her results showed a significant effect of her manipulation on Ebses (p < .001), a marginally significant effect on AB (p = .056) and no effect on BI (p = .28) or Ap (F < I). What follows are possible explanations for these results.

Ajzen and Fishbein (1980) state that "...consumer behavior involves at least a target and action elements (e.g. buying or using a brand or product) and often also content and time elements (p. 160). In Loken's manipulations, the target (i.e. Zipp) is held constant, and the context is manipulated (i.e., favorable or unfavorable information about Larsen's), but the action (i.e. buying Zipp) is not mentioned nor is the time element. What really is being manipulated then, is attitude toward an object which in this case is the store where the product is purchased. This brings up two points concerning correspondence between her manipulations and measures. First since the act of buying Zipp is not contained in the manipulations of context it is possible that this omission weakened any possible effects the manipulations might have had on the attitude toward the act of buying Zipp and on BI. Attitudes toward objects specify a given target but generalize across action and the remaining elements and with this lack of specificity in the remaining elements strong effects of store information on AB and BI cannot be expected (Ajzen and Fishbein 1980 p. 84). Perhaps stronger effects may have been obtained if for example the manipulations were of the form buying Zipp a new beverage sold at Larsen's would be nutritious rather than Larsen's supermarket carries Zipp a new beverage described by consumers as nutritious. In other words the more specific the manipulations are made with regard to action as well as context and target the greater the effects on AB and BI are likely to be.

The second point regarding correspondence concerns the measures of AB and BI. Loken states that her measures of AB and BI did not include mention of Larsen's. This means that context was not included in her measures or either construct. In other words. her measures did not correspond to her manipulations of context. Presumably measures of the form "Buying Zipp at Larsen's would be: good-bad" rather than just saying "Buying Zipp would be: good-bad" may have been more sensitive to her manipulations. In summary the nonsignificant effects obtained in her experiment may be due to lack of correspondence in that her manipulation mentions context and not action but her measures mention action and not context.

Loken found that the effects of positive and neutral store information on AB were not significantly different but negative store information produced a significantly lower AB. She states that negative information may be weighted more heavily in the purchase decision and this may account or the effect of negative purchase information on AB . However she was surprised that positive store information did not increase AB above the neutral condition. It is possible that when people lack useful information as was the case in the neutral condition they will assume it is favorable which would account for the lack of a difference between the two conditions. Furthermore without a manipulation check or measure showing that the two conditions produce different effects on Ebses it is quite possible that they did not and therefore would not produce differences in AB.

The findings reported here would have been more informative if the correspondence problems discussed above were cleared up. Also it would have bee? informative to manipulate Ap to see if an interaction occurred that may have produced a significant effect on AB and BI. Nevertheless her results are interesting in that they show that the addition or information unique to the purchase situation caused the correlations between Ap and AB and between Ap and BI to decrease. This does not necessarily say, however that the effects of store information are fully mediated by AB or BI as the sufficiency assumption discussed below states.


In this section the two papers testing the effects of external variables on intention or behavior are discussed. Loken's paper is dealt with first because while her method of testing the assumption (i.e. causal modeling using LISREL) is preferred by this writer there appears to be a crucial step in the analysis omitted. Then the Crosby and Muehling paper is discussed and the possible reasons their results differ from Loken's are explored.

Loken's Test

Token s results show that external variables as she has operationalized them do not have a significant effect on behavior above that mediated by AB and SN. However there are two possible reasons for these results that must be dealt with before any inferences about the validity of the sufficiency assumption can be made. First it is necessary to show that a relationship between externaL variables and behavior exists. If no relationship exists then the claim that intervening variables such as AB or SN act as mediators is not tested since there is no relationship to mediate. A demonstration of such a relationship is crucial for such external variables as the number of people who presently live in the respondent s household. In Loken's case such a relationship could be shown by a high correlation between her measures of external variables and behavior.

A second issue that must be dealt with is that several of the external variables (EV) may not be different from AB. This is especially critical for the three external variables that are measures of attitude (AEV) and use the same evaluative semantic differentials as were used to measure AB. In other words her measures of AEV and AB may not exhibit discriminant validity. This would be B very easy to test using confirmatory factory analysis as performed by LISREL (Joreskog and Sorbom 1981). The hypotheses to be tested would be that the measures of AEV load on the same factor as the measures of AB. If discriminant validity is to be demonstrated this one factor model should be rejected in favor of a two correlated factor model with the indicators or A loading on one factor and the indicators of AEV on the other factor. The basis for demonstrating discriminant validity would be that the one factor model does not produce an acceptable chi-square and the two factor model produces an acceptable chi-square that is a significant improvement over the one factor model. (See Burnkrant and Page 1982 for a demonstration or this test.)

If the one factor model is not rejected it would indicate that the measures or AB and AEV are really measuring the same underlying construct. Therefore it would be unlikely that AEV would have a direct effect on B over and above that of AB especially if the effects of AB on B are fully mediated by BI as Ajzen and Fishbein (1980) contend (although this has been questioned by Bentler and Speckart (1979 1981)). If the two factor model is supported it would indicate that AEV and AB are measuring distinct constructs. In this case a path from AEV to B may or may not improve the fit of the model to the data, but in Loken's case it did not. Therefore she contends that AEV has no direct effect on B. A lack or discriminate validity may or may not be the case in this instance. However her results would have been greatly strengthened if discriminant validity between the constructs had been demonstrated as described above at the outset.

The sufficiency assumption states that external variables affect intentions and behavior only to the extent that they affect AB or SN. Loken has shown that the effects of her external variables on behavior are mediated by A and SN. However it would have been interesting to have examined at least three other path analysis models after assessing discriminate validity as suggested above. First a model with a path from the external variable to BI rather than B should be investigated. If as Ajzen and Fishbein (1980) theorize behavior is fully mediated by intention then it is unlikely that external variables would have any direct effect on B. It may be however that external variables nave an effect on BI that is not mediated by AB or SN. This would be shown by a significant Path estimate from EV to BI in a model that produces an acceptable chi-square and is a significant improvement in fit over a model without the path.

A second pair of models that would be of interest would be to compare a model in which EV had a direct effect only (i.e. not mediated by AB or SN) on B or BI with a model in which EV is mediated by AB and/or SN. If the sufficiency assumption is valid in stating that external variables affect B or BI only through AB or SN, then the model that hypothesizes a direct unmediated affect of external variables on B or BI should be rejected. Furthermore the mediated model should be a significant improvement in fit.

Crosby and Muehling's Test

Crosby and Muehling also test the sufficiency assumption of the theory of reasoned action and they obtain results that are contradictory to those obtained by Loken. They find that external variables do have a significant direct effect on intention (Note: Their results are not directly comparable to Loken's because she looked at effects of external variables on behavior and not intention.) In fact they find support for the direct effects only model described above in that self-concept had a direct effect on intention and no mediating or moderating effect.

What is desperately needed here is an overall measure of the goodness of fit of the model to the data such as the chi-square statistic provided b LISREL. Even though there may be a significant path from an external variable to intention it does not necessarily mean that such a model provides an adequate representation of the underlying structure of the data and therefore inferences based on regression coefficients alone could be misleading. Crosby and Muehling should compare the did of their model with the fit obtained from testing the same two models described in Loken. In fact the same set of nested model comparisons advocated in the discussion of Loken's research above could be performed except for those involving behavior. Such comparisons or nested models with their adequate sample size would provide a good test of the sufficiency assumption. If adding the path from UV to BI (i.e. going from model to model 1 in Figure 2 of Loken) produces a significant path estimate and a significant improvement in fit over the model without the path then the sufficiency assumption has been shown to be invalid.


One thing any discussion of the two papers testing the effects of external variables must do is attempt to explain why such inconsistent results were obtained. Loken found that her external variables had no direct effects whereas Crosby and Muehling found a direct effect for several external variables.

One or the major differences between the studies is that Loken looks at the effect of external variables on behavior and Crosby and Muehling look at the effect on intention. As pointed out earlier in the discussion of Loken's research this may account for some of the differences. Second Loken uses several external variables that appear to be very similar to her AB measures which may bias her results against finding a direct effect. As mentioned above a demonstration of discriminant validity is needed here. Discriminant validity does not appear to be as critical a problem for Crosby and Muehling since their external variables do not appear to be highly similar to their model variables. Third the two papers test different operationalizations of the model. Loken measures Ebiei, AB and SN and Crosby and Muehling measure ENBjMCj and AB. It is not clear that these differences wound account for the discrepant findings but it is certainly worth testing. Ideally any future research would Rather measures of Ebiei. ENBjMCj, AB, SN both BI and B. The next step in testing this aspect of the theory would then be to gather multiple indicators of each construct and perform a structural equation analysis with latent variables. This approach would allow the assessment of the convergent and discriminant validity of the constructs of the model (Burnkrant and Page 1982), as well as take into account the effects of measurement error.


The theory of reasoned action has much to contribute to our understanding of consumer behavior. The studies resorted here have helped add to this contribution. Loken's study of the effects of purchase information on AB and BI has shown that while the effects are not significant negative information can reduce purchase intentions without affecting attitude toward the product. Perhaps the greatest contribution however is made by the fact that the two tests of the sufficient assumption arrive at different conclusions. In a sense this is ideal because it should stimulate further research into the issue perhaps along the lines outlined above and eventually the question may be resolved.


Ajzen Icek and Martin Fishbein (1980) Understanding Attitudes and Predicting Behavior Englewood Cliffs New Jersey: Prentice-Hall.

Bentler Peter M. and George Speckart (1979) Models of Attitude Behavior Relations Psychological Review 86 5 p. 459-464.

Bentler, Peter M. and George Speckart (1981) Attitudes Cause Behaviors: A Structural Equation Analysis Journal of Personality and Social Psychology 40 2 p. 226-238.

Burnkrant Robert E. and Thomas J. Page Jr. (1982) An Examination of the Convergent Discriminant and Predictive Validity of Fishbein's Behavioral Intention Model Journal of Marketing Research November.

Crosby lawrence A. and David D. Muehling (1983) "External Variables and the Fishbein Model: Mediation Moderation or Direct Effects?" in Richard D, Bagozzi and Alice M. Tybout (eds.) Advances in Consumer Research Vol. X.

Joreskog Karl C. and Dag Sorbom (1981) LISREL: Analysis of Linear Structure Relationships by the Method of Maximum Likelihood Chicago Illinois: National Educational Resources Inc.

Loken, Barbara (1983a) The Theory of Reasoned Action: Examination of the Sufficiency Assumption for a Television Viewing Behavior in Richard P. Bagozzi and Alice M.. Tybout (eds.) Advances in Consumer Research Vol. X.

Loken, Barbara (1983b) Effects of Uniquely Purchase Information on Attitudes Toward Objects and Attitudes Toward Behaviors in Richard P. Bagozzi and Alice M. Tybout (eds.) Advances in Consumer Research Vol. X

Miniard Paul W. and Thomas J. Page Jr. (19&9) Causal Relationships in the Fishbein-Ajzen Behavioral Intention Model Working Paper Series College or Administrative Science. The Ohio State University, WPS 87-39.