Predictive Validation and Cross-Validation of the Fishbein, Rosenberg, and Sheth Models of Attitudes


P.S. Raju, Rabi S. Bhagat, and Jagdish N. Sheth (1975) ,"Predictive Validation and Cross-Validation of the Fishbein, Rosenberg, and Sheth Models of Attitudes", in NA - Advances in Consumer Research Volume 02, eds. Mary Jane Schlinger, Ann Abor, MI : Association for Consumer Research, Pages: 405-426.

Advances in Consumer Research Volume 2, 1975      Pages 405-426


P.S. Raju, University of Illinois

Rabi S. Bhagat, University of Illinois

Jagdish N. Sheth, University of Illinois

[P.S. Raju and Rabi S. Bhagat are doctoral students in the Department of Business Administration and Jagdish N. Sheth is Illinois Business Associates Distinguished Professor of Business and Research Professor at the University of Illinois at Urbana-Champaign.]

The present study compares the Fishbein, the Rosenberg and the Sheth models in their ability to predict behavioral intentions of consumers with respect to the 'Pinto' car. Since the Rosenberg model is restricted to the prediction of 'attitudinal affect' it has been extended to the prediction of behavioral intention for the purposes of this study. Three stages have been identified in the comparison process namely predictive validation, cross-validation and validity generalization. Predictive validation deals with the comparison based on one sample, cross-validation extends the comparisons to other samples from the same population, and validity generalization indicates the extent to which the results are applicable to other populations. While most studies have been restricted to the stage of predictive validation, the present study also includes cross-validation. Data were obtained from 243 respondents, comprising of students and housewives, in the Champaign-Urbana area. The results indicate that the Sheth model has high predictive validity and cross-validity, the Fishbein model has a lower predictive validity but high cross-validity, and the Rosenberg model has a low cross-validity so that not much confidence can be placed in its predictive validity over different samples.

Although the concept of attitude-has been researched for almost a century, the development of comprehensive attitudinal models is a relatively recent phenomenon. The recent efforts devoted to the development of these models can be attributed to the following reasons: 1) they provide a systematic framework-for research by specifying variables that relate to the process, 2) they aid in the generation of new hypotheses, and 3) they provide diagnostic information which facilitates the determination of future courses of action.

Based on cognitive consistency theories, Rosenberg (1956) showed that attitudinal affect could be expressed as a simple mathematical function of cognitive elements. After his pioneering work, several researchers in the areas of social psychology and consumer psychology have developed other models of attitudes. However, the emphasis seems to have shifted from predicting affective tendencies toward the object to the prediction of behavioral intention and specific behaviors toward the object. For instance, Fishbein and his associates (Ajzen & Fishbein, 1970(a), 1970(b), 1973; Fishbein, 1967, 1972) have formulated a model in which behavioral intention is predicted from personal beliefs as well as social normative variables. The Rosenberg and the Fishbein models have been very instrumental in fostering the expectancy value tradition in attitudinal research. In the area of consumer behavior, for example, Wilkie and Pessemier (1973) report that, apart from numerous studies conducted in Europe, over forty studies have been published in the United States alone within a period of three years. Unfortunately, there is little, if any, consistency among these studies in terms of the operationalization or measurement of constructs, so that their usefulness in drawing broad generalizations is rather limited. In addition, expectancy-value models have often been criticized for their inherent limitations (Day, 1972; Sheth, 1973; Sheth & Tuncalp, 1974; Wilkie & Pessemier, 1973). Finally, a number of researchers in social psychology have conceded that attitudes toward the object, no matter how they are defined and measured are not the sole determinants of a person's behavioral intention or actual behavior toward the object (Katz & Stotland, 1959; Rokeach, 1968; Triandis, 1971). Similar suggestions have also been made in Consumer Psychology and in Marketing (Howard & Sheth, 1969; Sandell, 1968; Sheth, 1971; Sheth & Raju, 1973). Recently, Sheth (1971, 1974) has proposed a model of brand choice behavior which attempts to integrate diverse thinking on the role of attitudes in the prediction of behavior both in social psychology and in consumer psychology. The model states that behavioral intentions and behavior of an individual are a function of 1) evaluative beliefs about the object's potential to satisfy his needs, wants, or desires, 2) social stereotype of the object as perceived by him, 3) past satisfaction and resultant predisposition (affect) toward the object, and 4) situational events which are either anticipated or which may unexpectedly occur at the time of behavior.


In spite of the availability of a variety of models, little has been done in terms of comparative research on these models. As a result, the researcher is often unclear as to the relative strengths and weaknesses of different models, and their appropriateness for different situations. "Unless we do a comparative study under the same setting, on the same issue, and on the same or essentially the same subjects, it is not possible to claim victory for any model" (Sheth 1972, p. 465). Comparative studies, therefore, need to be conducted over a variety of situations and behaviors before conclusive assertions can be made about the appropriateness of any particular model for a problem area.

Researchers have often utilized one or more of the following four major criteria in comParing competing models or theories:

1) Description: this indicates the ability of the model to describe the process or phenomenon being studied. The use of flow diagrams can greatly aid the descriptive capacity of a model.

2) Explanation: the extent to which a model can provide the causes from the knowledge of the effects determines its rating on this criterion.

3) Prediction: the capability of a model to predict future occurrences based on the knowledge of certain key concepts is indicated-by the rating on this criterion. While explanatory models are usually predictive, the reverse need not necessarily be true (Kaplan, 1964). For instance Stochastic models of brand choice are predictive but do not explain the process.

4) Prescription: Often referred to as also the diagnostic ability of a model it essentially deals with the capacity of the model to prescribe courses of action to influence the process in a desired manner.

In conformity with the views presented above, the main aim of this study is to compare the Rosenberg, the Fishbein, and the Sheth models in their ability to predict behavioral intentions of consumers with respect to a brand of a durable consumer product. We restrict ourselves to the single criterion of 'prediction' due to two major reasons:

1) It is beyond the scope of this study to attempt a comparison on all the relevant criteria, and

2) The criteria of 'prediction' and 'prescription' are considered most important and have been used frequently in past research to evaluate attitude models. A comparison on the prescriptive criterion would, however, require a longitudinal study since we would be concerned with the implications of the model for future action. As this study is cross-sectional it is infeasible to test the prescriptive efficacy of the models.

An exhaustive comparison of the models on the basis of 'prediction' would require a three stage analysis comprising of predictive validation, cross-validation, and validity generalization. Since all the models use a multiple regression formulation, as will be discussed later, the three stages are described below with respect to that statistical technique.

Predictive validation provides an indication of the extent to which the model is able to predict the criterion variable for a particular sample representative of the population. A model is considered valid if the correlation coefficient and, therefore, the percentage variance of the criterion variable explained is quite large. By predictive validation alone, however, it is impossible to generalize the results even to other samples from the same population unless the sample size is so large as to cover the entire population. Whereas predictive validity is concerned with a single sample, cross-validity requires that the effectiveness of the predictor composite be tested on a separate independent sample from the same population. Mosier (1951) in an excellent paper on cross-validation states:

"Since both combining weights (S weights) and their effectiveness should be determined on samples representative of the group in which the battery (in our case 'model') will be applied, it is obvious that the two samPles should be selected from the same universe." (p.6)

In cross-validation, a single sample is divided randomly into a 'derivation sample' and a 'validation sample' or two independent random samples are drawn from the population. The multiple regression weights and the multiple correlation coefficient are obtained by applying the model to the derivation sample. The obtained regression weights are used in the validation sample to create 'predicted criterion scores' which are then correlated with the actual criterion measures for the validation sample. The extent to which this correlation coefficient is close to that obtained initially in the derivation sample determines the cross-validity of the model. For perfect cross-validity the correlation coefficients should be equal. If a model has good cross-validity it can be presumed that the predictive power will be almost constant over different samples from the same population.

The test for validity generalization is very similar to that for cross-validity except that the multiple regression weights and the multiple correlation coefficient obtained on a sample drawn from one population are tested on a second sample drawn from a different population. The extent of validity therefore signifies the applicability of a model to different populations.

With respect to the typology of validation presented above, it is interesting to note that most validation studies on attitude models have been restricted to the stage of 'predictive validation' thereby providing biased results, with no indication of the generalizability to other samples or other populations. In this study, the validation procedure has been designed to include cross-validation in addition to predictive validation. Validity generalization has, however, not been performed because our interest at this stage is limited to samples from a single population. In summary, this study was designed to accomplish three objectives:

1) A detailed comparison of the Rosenberg, the Fishbein, and the Sheth models both conceptually and in terms of their ability to predict buying intentions of consumers.

2) Predictive validation of the models based on results obtained from the total (derivation + validation) sample.

3) Cross-validation of the models by applying the model coefficients obtained from the derivation sample to the validation sample.


A brief description of each of the three models used in the study is provided. A more elaborate treatment of each model is available in the previous writings of Fishbein (1967, 1972), Rosenberg (1956), and Sheth (1974).

The Rosenberg Model

As a means to understanding the attitude learning and attitude change processes, Rosenberg (1956, 1960) formulated a structural relationship between the attitude toward an object and the beliefs about the object. The relative stable affective orientation [Ao] toward an object or concept is considered to be a function of two cognitive variables: 1) Value Importance [VI]: the degree of satisfaction generated by the attainment of a desired value state; and 2) Perceived Instrumentality [PI]: the extent to which the object or concept leads to or blocks the attainment of the value state. Every individual, therefore, develops an attitude in conformity with his basic system of values. Mathematically, the model is stated as


where n is the number of desired value states.

Since Rosenberg was mainly concerned with the cognitive processes underlying attitude formation and change, he has restricted his model to the prediction of attitude. However, our interest in this paper centers on the prediction of behavioral intention. Rosenberg (1960) has indicated, in the context of action predicting capability of attitudes, that although attitudes are not the sole predictors of behavior, they could be effective predictors. He also recognizes the presence of situational factors and restraints that could possibly influence behavior. Since Rosenberg has not extended his model to include these situational factors, we shall restrict ourselves to the prediction of behavioral intention from the attitude score obtained from his model. The Rosenberg model, thus consists of a single determinant construct namely the personal beliefs about the role of the object in attaining differentially valued states or goals.

The Fishbein Model

According to Fishbein (1967, 1972), behavioral intention is a function of two components; one attitudinal and the other normative. The model is represented formally as

B~BI = [Aact] wo + [(NB)(Mc)] w1

where B = behavior, BI = behavioral intention, AaCt = attitude toward the act or behavior, NB = normative belief, i.e., what others expect or say should be done in the situation, Mc = motivation to comply with the normative belief, and w0 and w1 are empirically determined weights. AaCt itself is determined as


where Bi = the individual's belief about the likelihood that the behavior will result in the ith consequence or outcome, ai = the individual's evaluation of outcome i, and n = number of salient beliefs.

There are three aspects of the model which need to be pointed out. First, behavioral intention is considered to be an immediate antecedent of behavior so that other influences that might be active between the two are ignored. In the context of consumer behavior this assumption may not be valid because of the considerable time span that mediates the expression of an intention and its translation into action (Howard and Sheth, 1969; Sheth, 1971). Second, the model explicitly recognizes the importance of variables such as social norms and motivation in influencing behavior. In this sense it is a definite improvement over the Rosenberg model. However, all other variables influencing behavior are presumed to operate indirectly by influencing the two components in the model or their relative weights. Finally the model makes a clear distinction between 'attitude toward the object' and 'attitude toward the act'; only the latter being considered an effective predictor of behavior.

The Sheth Model

In the comprehensive model proposed by Sheth (1971, 1974) behavior [B] is determined by 1) the behavioral intention with respect to the object [BI] , 2) past satisfaction from behavior which results in a predisposition toward the object [S], and 3) Unexpected events [UE] that influence the individual at the time of manifestation of the behavior. Mathematically,

B = w1 [BI] + w2 [S] + w3 [UE].

The unexpected events factor is presumed to be uncorrelated with the predisposition or the behavioral intention and it may either enhance or inhibit the translation of intentions to actions. It is expected that, as the time period between intention and behavior increases, the influence of this factor will correspondingly increase; for example, unexpected events play a major role in the purchase of consumer durable goods.

Although the Sheth model can be used to predict behavior, our interest in this study is-limited to behavioral intention, which is hypothesized to be a function of 1) evaluative beliefs [EB] about the object's potential to satisfy needs, wants, and desires, 2) perceived social stereotype [SS] of the object, 3) predisposition [S] resulting from past satisfaction,and 4) situational influences [AS] that the person anticipates will be effective at the time of behavior. In a functional form.

BI = w1 [EB] + w2 [SS] + w3 [S] + w4 [AS].

Evaluative beliefs represent the individual's perception of the object's capability to satisfy a set of relevant needs, wants and desires. In this respect, evaluative beliefs closely resemble the perceived instrumentality concept in the Rosenberg model, the major difference being in the definition of valued stakes. While Rosenberg model has been limited to more fundamental values of the individual, the valued states in the Sheth model are specific to the class of objects on which research is conducted. Evaluative beliefs are directly related to the instrumental-utilitarian function that attitudes are presumed to perform in Katz's functional formulation (1960). However, evaluative beliefs are presumed to be multidimensional which requires a profile analysis of the object's capability to satisfy diverse functional needs of the individual.

Social stereotype represents the individual's perceptions of the social connotations or social imagery the object possesses. For example, a brand of cigarettes such as Virginia Slims may-be perceived to be more feminine than other brands such as Marlboro although functionally the two cigarettes are highly similar in their characteristics. The social stereotype is presumed to be due to the identification of the object with specific segments of the society based on life cycle factors (age, marital status, children, etc.), on socioeconomic factors (education, occupation, income, housing, mobility, etc.), on life style factors (activities, interests, opinions, and values, etc.), and cultural factors (religion, tradition, language, etc.) and biological factors (sex, race, etc.). The Sheth model presumes that social stereotype is also a multidimensional concept which requires dimensional analysis on a profile of perceptions of the individual about the goal object as a social object.

Anticipated situation pertains to all the relevant situational influences that the individual expects will occur, and all the relevant activities that the individual expects he will engage in at or around the time of manifestation of behavior. Expectation of certain occurrences would therefore enhance the intentions whereas expectation of certain others might inhibit them. Events such as vacations, moving, and marriage are usually anticipated by a person in determining his intentions.

Finally, Satisfaction represents the positive or negative predisposition toward an object based on past experience with the object as instrumental to the attainment of certain desirable goals. Satisfaction is measured by the individual's affective tendency toward the object.

For a more elaborate treatment of the constructs the reader is referred to the earlier writings of Sheth (1971, 1974).


In this section, we will briefly point out similarities and differences among the models with respect to three broad areas; number of constructs, definition and measurement of constructs and operationalization of the models.

Number of Constructs

The extended Rosenberg model used in this study is the simplest of the three models since it uses only one construct namely 'attitudinal affect' in the prediction of behavioral intention. While recognizing the importance of the attitudinal variable, Fishbein also incorporates in his model social normative influences and hence takes into account the fact that behavior is not a function of attitude alone. While definitely an improvement over the Rosenberg model, the Fishbein model still does not consider the effect of other important variables such as situational influences. The 'evaluative belief' and the 'social stereotype' constructs in the Sheth model (restricting to the behavioral intention level) are roughly the counterparts of the 'attitudinal' and 'social normative' constructs respectively of the Fishbein model. Sheth, however, has two additional constructs that make the model more comprehensive. One of these constructs 'predisposition resulting from past satisfaction' takes into account the fact that affect need not always be a result of the beliefs held by the individual. The other construct 'Anticipated situations' recognizes the importance of situational influences that a person is able to foresee in determining his intentions.

Finally, in contrast to the assumption made by Fishbein that behavioral intention is an immediate antecedent of behavior Sheth explicitly considers unexpected events to intervene between the two. However, research still needs to be done to effectively operationalize this construct. Recent research efforts concerning situational influences on behavior (Belk, 1974; Sandell, 1968) have provided valuable insight into the problem.

Definition and Measurement of Constructs

The models can be compared with respect to three distinct types of constructs that are utilized in the prediction of behavioral intention: 1) Attitudinal 2) Social and 3) Situational. In addition, there are important differences between the Fishbein and the Sheth models in the measurement of behavioral intention. All three models use the attitude type construct, only the Fishbein and the Sheth models use the social type construct, and the situational type construct is used only by the Sheth model. As a result, no comparison wan l)e made with respect to the measurement of situational constructs, the comparison is restricted to the Fishbein and the Sheth models for the social type constructs and all three models are compared with respect to the attitudinal construct.

All the three models consider 'attitude' to be a univariate unidimensional construct; but measure it in different ways. In the Rosenberg model it is a summated product of perceived instrumentality and value importance. Rosenberg recommends a 21 point scale ranging from -11 to +11 to measure value importance and a 11 point scale ranging from -5 to +5 to measure perceived instrumentality. It has been suggested, especially after the work of Miller (1956), that such scales might be inappropriate because a person usually cannot distinguish more than five to nine different levels of a concept (Sheth & Park, 1973; Tuncalp, 1973). In this study, therefore, these components have only been measured using scales ranging from -3 to +3, although the wording of the scales is consistent with the recommendations of Rosenberg. Fishbein also uses the expectancy-value approach and the two components of 'belief' and 'evaluation' are measured by scales ranging from -3 to +3. However, Fishbein measures 'attitude toward the act' as opposed to 'attitude toward the object'. In the Sheth model, the evaluative beliefs are not combined into a single construct called 'attitude', but there are essentially the counterpart of such a construct as used by Rosenberg and Fishbein. The evaluative beliefs are measured by scales ranging from l to 7 and a further distinction is that all the scales need not necessarily be bipolar. They are anchored at the ends by the extremities of the normal range of the belief that a person uses. For example, the scale for the evaluative belief regarding durability of Pinto ranges from 'very durable' to 'only average in durability' rather than from 'very durable' to 'not at all durable'. Further differences in the wording of the scales for the three models will become evident from the sample of scales that will be presented in the 'Method' section.

The 'social stereotype' construct considered by Sheth is quite different from the social normative construct as defined by Fishbein. While Fishbein is mainly concerned with what others want the person to do and whether the person wants to comply with their wishes, Sheth is concerned with the social connotations and social imagery of the object. Again Fishbein uses scales ranging from -3 to +3 to measure each component of the social normative construct whereas Sheth uses scales ranging from 1 to 7 to measure social stereotypes.

Finally, a major distinction between the Fishbein an(l the Sheth models is in the measurement of behavioral intention, Fishbein has convincingly argued that for better prediction, behavioral intention should be measured with respect to a specific object rather than a generalized group of objects (Fishbein, 1967). Thus, buying intention should be measured with respect to 'Pinto' rather than toward the general product category of 'automobile'. While Sheth agrees with this point, he measures buying intention not only with respect to a specific brand such as 'Pinto' but also qualifies it with respect to the need or motivational level of the buyer. His scale, therefore, measures how seriously a person would consider buying a 'Pinto' if he were to buy an automobile. Since Rosenberg does not provide a scale for measurement of behavioral intention both the Fishbein and the Sheth scales have been used as criterion measures in the Rosenberg model. The Sheth scale, however, has been recoded from -3 to +3 (instead of l to 7) when used on the Rosenberg model to make it compatible with the way 'perceived instrumentality' and 'value importance' are measured.

Operationalization of Models

The major difference in the operationalization of the models is the expectancy-value approach adopted by Rosenberg and Fishbein versus the factor analytic approach adopted by Sheth. In the expectancy-value approach, the attitude is determined by the summation of the two components of expectancy and value over the whole range of values (or expectancies). In the process, the positive and negative components cancel each other out leaving a simplified index of the cognitive structure. Such an assumption may not be true. Researchers have suggested that a disaggregated version of the model might perform better than the summated version (Cohen & Ahtola, 1971; Lutz & Howard, 1972; Sheth, 1970). In contrast to this, Sheth performs a dimensional analysis on the profile of evaluative beliefs by the technique of principal components analysis. The resultant orthogonal dimensions of the evaluative beliefs are then utilized as independent predictors of behavioral intention. Such a factor analytic approach has several advantages:

First, it is a compromise between the aggregated version and the completely disaggregate version. The disadvantage of the former has already been pointed out. In the latter, all the beliefs are used as independent predictors of behavioral intention. This could result in too many independent variables especially in case-s where there are a number of salient beliefs e.g., automobile purchase. By utilizing only the orthogonal dimensions of the beliefs, the number of predictor variables can be reduced considerably without sacrificing too much predictive power. Second, the factor analytic procedure takes into account the multicollinearity among belief items. Thus, in the aggregated version the correlation between belief items leads to double counting of certain dimensions of the beliefs and in the completely disaggregated version it leads to unstable regression weights. These problems are not encountered in Sheth's approach. Finally, the factor analytic approach is very useful from a policy viewpoint. By using attitude models, the marketer is able to gain some insight as to which beliefs will have to be changed to create a desired change in attitude. However, it is of ten impossible to influence one belief alone without any impact on other cognitive elements. To avoid undesirable consequences on the beliefs that do not need to be influenced, it is necessary for the marketer to understand the dimensions of the cognitive structure and the beliefs that group together under any given dimension. Different promotional and marketing strategies can then be assessed in terms of their effects on the total cognitive structure opposed to only certain specific beliefs.


The models were empirically tested by collecting data on 243 respondents in the Champaign-Urbana area. The respondents were mainly students, student wives, and housewives . An indepth interview with about twenty-five housewives and twenty-five students indicated that the Pinto car (a product of Ford Motor Company) was a salient attitude object for this group of respondents. A separate pilot study involving 40 respondents was performed to obtain salient beliefs pertaining to the purchase of an automobile. Twelve belief items mentioned most frequently were finally selected for the main study. Similar belief items have been utilized as product attributes in past studies (Alpert, 1971). A sample of the scales used for measurement of the various constructs is provided:

Rosenberg Model

1) Value importance (VI)

Please check each scale below so as to indicate the extent to which each consequence that is associated with buying an automobile gives you satisfaction or dissatisfaction.

Economy of operation:

Gives me maximum satisfaction  :_:_:_:_:_:_:_: Gives me maximum dissatisfaction

(11 other value importances measured similarly).

2) Perceived instrumentality (PI)

Please check each scale so as to indicate to is attained or blocked by buying a Pinto.

Economy of operation:

Completely blocked  :_:_:_:_:_:_:_: Completely attained.

Fishbein Model [We thank Professor Martin Fishbein of the Department of Psychology, University of Illinois for his help in the wording and scale construction aspects of his model.]

1) Belief (Bi)

My buying a Pinto would mean buying an automobile that is economical to operate.

Probable   :_:_:_:_:_:_:_:  Improbable

(eleven other beliefs were measured similarly).

2) Evaluation of belief (ai)

Buying a car that is economical to operate is

Good  :_:_:_:_:_:_:_: Bad

(eleven other evaluations were measured similarly).

3) Normative belief (NB)

Others who are important to me think

I should  :_:_:_:_:_:_:_:  I should not

                 buy a Pinto.

4) Motivation to Comply (Mc)

Ingeneral, I want to do   :_:_:_:_:_:_:_:  In general, I don't want to do

         what others who are important to me think I should do.

5) Behavioral Intention (BI)

I would  :_:_:_:_:_:_:_: I would not

              buy a Pinto.

Sheth Model

1) Evaluative Beliefs (EB)

Pinto is economical  :_:_:_:_:_:_:_: Pinto is expensive to operate.

(eleven other evaluative beliefs were measured similarly).

2) Social Stereotype (SS)

Pinto is meant for young people only.

Strongly agree  :_:_:_:_:_:_:_: Strongly disagree

(eight other social stereotypes were measured similarly).

3) Predisposition or Past Satisfaction (S)

Please indicate the extent to which you are favorably or unfavorably disposed toward the Pinto

Most favorable toward Pinto   :_:_:_:_:_:_:_:  Most unfavorable toward Pinto

4) Anticipated Situation (AS)

Three different scales were used to measure personal (AS1), buying (AS2), and financial (AS3) situations relating to the purchase of a car. Personal (AS1): Is it conceivable that you might change your intention to buy or not to buy an automobile because of some unforseen events (for example, moving, getting married, birth in the family, unanticipated change in your financial status or deciding to take vacations) you did not anticipate may occur in the next six months?

Not at all conceivable   :_:_:_:_:_:_:_:  Very much conceivable

Buying (AS2): Is it conceivable that you might change your intentions to consider or not to consider buying a Pinto due to some unforseen events including a good buy on some other car or friend's advice to reconsider your plans and intentions?

Not at all conceivable   :_:_:_:_:_:_:_:  Very much conceivable

Financial (AS3): If you were to buy an automobile in the next six months, would you have any financial problems?

I simply cannot afford it   :_:_:_:_:_:_:_:  I can easily raise money to pay for it

5) Behavioral intention (BI)

If you were to buy an automobile, how seriously would you consider buying a Pinto?

Definitely would consider buying a Pinto   :_:_:_:_:_:_:_:   Definitely would not consider buying a Pinto

Since the Sheth model utilizes factors scores for both evaluative beliefs and social stereotypes, principal components analyses were first performed independently on the respective correlation matrices using the total sample. The analyses yielded three major factors in each case and rotated factor loadings were obtained by the technique of orthogonal varimax rotation. Factor scores for each individual were obtained from the rotated factor loadings matrix and these scores were kept invariant for all further analyses. Thus, even when the total sample was split into two for purposes of cross-validation, each individual retained the same factor scores. It is essential that the factor scores be invariant because, in cross-validation, we are concerned with the stability of regression weights which will be confounded by the change in factor scores themselves if these are calculated separately for the derivation and the validation samples.

The predictive validity of each model was determined by a multiple regression technique using behavioral intention as the criterion and appropriate variables as predictors. In the case of the Rosenberg model alone this reduces to simple regression analysis since there is only one predictor variable. For purposes of clarity, the regression equations are reproduced below:

1) Rosenberg's model using Fishbein's behavioral intention scale.


2) Rosenberg's model using Sheth's behavioral intention scale.


3) Fishbein model:


4) Sheth model:

BI = b1 [EB1] + b2 [EB2] + b3 [EB3] + b4 [SS1] + b5 [SS2] + b6 [SS3] + b7 [S] + b8 [AS1] + b9 [AS2] + b10 [AS3] + K4

where P1, P2; w0, w1; and b1, b2 --- b10 are regression weights and K1, K2 K3 and K4 are constant intercept values, all determined empirically. In the Fishbein model Aact was determined by a summated product of Bi and ai. In the Sheth model, it can be noticed that there are three evaluative belief factors (EB1, EB2, and EB3) and three social stereotype factors (SS1, SS2, and SS3).

The cross-validity of the models was finally determined by utilizing the regression weights obtained from the derivation sample to predict the criterion values in the validation sample. The predicted criterion values were then correlated with the actual criterion values to determine a cross-validation correlation coefficient.


The results of predictive validation for the Rosenberg model are summarized in Table 1. It can be seen that attitude is a significant predictor of behavioral intention explaining a minimum of about 15 percent of variance in the total sample using Fishbein's behavioral intention scale and a maximum of about 29 percent of the variance in the derivation sample using Sheth's behavioral intention scale. The cross-validation results of the Rosenberg model are not very impressive. The cross-validity correlation coefficient with Fishbein's behavioral intention scale was about 0.23 as opposed to 0.51 obtained in the derivation sample and with Sheth's behavioral intention scale it was about 0.32 as opposed to 0.54 obtained in the derivation sample. Hence we cannot presume with confidence that the performance of the Rosenberg model will be consistent over different samples from the same population.



The predictive validation results for the Fishbein model are presented in Table 2. One interesting finding is that of the two components in the model, 'attitude toward the act' is a significant predictor of behavioral intentions whereas the social normative component is not. Both the predictors, however, account for about 22 percent variance in behavioral intention in the total sample and about 24 percent in the derivation sample. The predictive ability is, thus, quite stable. Further confirmation of this fact is obtained when the cross-validation correlation coefficient is found to be about 0.432, which is not a substantial drop from the value of 0.487 obtained in the derivation sample. We could, therefore, expect the Fishbein model to consistently predict about 23 percent variance in behavioral intention in other samples drawn from the same population.



Table 3 shows the rotated factor structure of evaluative beliefs obtained for the Sheth model on the total sample. Three major factors were extracted explaining about sixty percent of the total variance. Based on the loadings the factors can be interpreted as follows:

The first factor with high loadings on items such as durability, handling, safety, ride, acceleration, and resale value pertains to an overall sense of the 'quality' dimension of Pinto. Beliefs related to luxury, size of engine, pollution, buying cost, and economy of operation load on the second factor which can be interpreted as a 'luxury' dimension of Pinto. The last factor comprising of beliefs pertaining to sportyness, economy of operation, and handling represents the 'sportyness' dimension of Pinto. The rotated factor structure of social stereotypes presented in Table 4 can also be interpreted in a similar manner. The first factor stereotypes Pinto as a car for 'young unmarried people', the second as a car for 'young people with moderate income', and the third as a car for 'older people with low income'. The three factors together account for 56 percent of the total variance in the social stereotypes.





Table 5 summarizes the predictive validation results for the Sheth model. There are ten predictor variables in the regression equation; three evaluative belief factors, three social stereotype factors, prior predisposition, and three anticipated situational variables. The three anticipated situational variables pertain to anticipated personal situations (such as moving, marriage, etc.), anticipated buying situation (such as expecting a better buy), and anticipated financial situation. The multiple correlation coefficient for the total sample is about 0.73 and for the derivation sample is about 0.75, explaining about fifty-three percent and about fifty-six percent respectively of the variance in behavioral intention. The four significant predictors of behavioral intention are prior predisposition, anticipated buying situation, the 'quality' dimension of evaluative beliefs, and the 'sportyness' dimension of evaluative beliefs.



The cross-validation correlation coefficient obtained was about 0.665 pointing to the fact that the high correlation coefficients are not due to idiosyncracies of the sample and that the model can be expected to perform equally well in other samples from the same population.

Since the three models use different numbers of predictor variables, the coefficient of determination (R2) cannot be compared directly. Adjusted R2 values were, therefore, calculated in each case to account for the number of predictor variables and degrees of freedom. The formula used was

Adjusted R2 = 1 - (1-R2) N-1/N-n

where N=sample size and n=number of predictors. Since the Rosenberg model uses only one predictor, the adjusted R2 is not different from the original R2. It can be seen that even for the Fishbein and the Sheth models the decrease in R2 due to adjustment is very minimal indicating that, especially in the case of the Sheth model, the correlation coefficients are not spuriously high due to the number of predictors.


The results clearly indicate that the Sheth model has a higher predictive validity than the Fishbein model. However, both models perform equally well in terms of cross-validation. The Rosenberg model was found to vary considerably in its predictive power over different samples. This points to the fact that although the Rosenberg and the Fishbein model are mathematically similar in the measurement of attitude, the nature of the constructs produces vastly different results. The better performance of the Fishbein model cannot be attributed to the inclusion of the normative beliefs component because it did not contribute sufficiently to the prediction of behavioral intention. The improved performance of the Sheth model over the Fishbein model, on the other hand, can be attributed to a more complete treatment of variables involved in the process, the factor analytic approach adopted and the measurement of behavioral intention incorporating needs and motives.

Apart from the relative predictive powers of the models, the study also provides other interesting conclusions.

1) Attitudes are effective predictors of behavioral intentions. This is confirmed by all the three models in this study. Further, both attitude toward the act (Fishbein model) and attitude toward the object (called predisposition in Sheth's model) prove to be significant indicating that either they are both important or that the distinction is not useful.

2) Anticipated situational influences are important for prediction of behavioral intention and behavior at least in the consumer behavior context. The fact that anticipated buying situation was a significant predictor of behavioral intention in the Sheth model suggests that buying an automobile might be largely influenced by the kind of deal that a person gets on a car or his expectation that some other alternative will become available before the time of purchase.

3) The poor predictive ability of social beliefs in both Fishbein and Sheth models is somewhat surprising in the situation of buying automobiles. Our conclusion is that Pinto is probably a universal car devoid of any stereotype and that the sample used was homogeneous with respect to life cycle, socioeconomic status and life style which might have minimized stereotype differences toward Pinto.


This study has attempted to demonstrate the importance of comparing and cross-validating attitude models. The authors agree that one study alone, such as this, is not sufficient to prove conclusively the superiority of any model. The need for replication of such studies over a variety of behaviors and situations is, therefore, imminent. It is possible that one model might be appropriate in a buying situation whereas another might be appropriate in a social situation. Once the models are cross-validated, the next step would be to determine the extent to which they are applicable to different populations, which is labeled as 'validity generalization' in this study. Finally, it is-hoped that the extensive validation and comparison of existing models will lead to a better understanding of consumer choice processes.


Ajzen, I., & Fishbein, M. Attitudinal and normative variables as predictors of specific behaviors: A review of research generated by a theoretical model. Paper presented at the Attitude Research and Consumer Behavior Workshop, University of Illinois, Urbana, 1970.(a)

Ajzen, I., & Fishbein, M. The prediction of behavior from attitudinal and normative variables. Journal of Experimental Social Psychology, 1970, 6, 466-87.

Ajzen, I., & Fishbein, M. of specific behavior. Attitudinal and normative variables as predictors Journal of Personality and Social Psychology, 1973, 27, 41-52.

Alpert, M.I. A canonical analysis of personality and the determinants of automobile choice. Combined Proceedings of the American Marketing Association, Spring and Fall Conferences, 1971, 312-16.

Belk, R.W. An exploratory assessment of situational effects in buyer behavior. Journal of Marketing Research, 1974, 11, 156-63.

Cohen, J.B., & Ahtola, O.T. An expectancy x value analysis of the relationship between consumer attitudes and behavior. Proceedings of the Second Annual Conference of the Association for Consumer Research, 1971, 344-64.

Day, G.S. Evaluating models of attitude structure. Journal of Marketing Research, 1972, 9, 279-86.

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

Fishbein, M. The search for attitudinal behavioral consistency. In J. Cohen (Ed.), Behavioral Science foundations of consumer behavior. Glencoe, Ill.: Free Press, 1972.

Howard, J.A., & Sheth, J.N. The theory of buyer behavior. New York: John Wiley, 1969.

Kaplan, A. The conduct of inquiry: Methodology for behavioral science. San Francisco: Chandler Publishing Co., 1964.

Katz, D. The functional approach to the study of attitudes. Public Opinion Quarterly, 1960, 24, 163-204.

Katz, D., & Stotland, E. A preliminary statement to the theory of attitude structure and change. In S. Koch (Ed.), Psychology: A study of a science. Vol. 3. New York: McGraw-Hill, 1959.

Lutz, R.J., & Howard, J.A. Toward a comprehensive view of the attitude-behavior relationship: The use of multiple set canonical analysis. Proceedings of the Social Statistics Section, American Statistical Association, 1971, 215-24.

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

Mosier, C.I. Problems and designs of cross-validation. Educational and Psychological Measurement, 1951, 11, 5-11.

Rokeach, M. Beliefs, attitudes, and values: A theory of organization and change. San Francisco: Jossey-Bass, 1968.

Rosenberg, M.J. Cognitive structure and attitudinal affect. Journal of Abnormal and Social Psychology, 1956, 53, 367-72.

Rosenberg, M.J. A structural theory of attitude dynamics. Public Opinion Quarterly, 1960, 24, 319-40.

Sandell, R.G. Effects of attitudinal and situational factors on reported choice behavior. Journal of Marketing Research, 1968, 5, 405-8.

Sheth, J.N. An investigation of relationships among evaluative beliefs, affect, behavioral intention, and behavior. University of Illinois, College of Commerce and Business Administration. Working paper, 1970.

Sheth, J.N. Affect, behavioral intention, and buying behavior as a function of evaluative beliefs. In P. Pellemans (Ed.), Insights in consumer and market behavior. Belgium: Numur University Publications Universitaires, 1971.

Sheth, J.N. Reply to comments on the nature and uses of expectancy-value models in consumer attitude research. Journal of Marketing Research, 1972, 9, 462-65.

Sheth, J.N. Brand profiles from beliefs and importances. Journal of Advertising Research, 1973, 13(1), 37-42.

Sheth, J.N. A field study of attitude structure and attitude-behavior relationship. In J.N. Sheth (Ed.), Models of buyer behavior: Conceptual, quantitative, and empirical. New York: Harper and Row, 1974.

Sheth, J.N., & Park, C.W. Equivalence of Fishbein and Rosenberg theories of attitudes. University of Illinois, College of Commerce and Business Administration. Working paper #108, 1973.

Sheth, J.N., & Raju, P.S. Sequential and cyclical nature of information processing models in repetitive choice behaviors. Proceedings of the Fourth Annual Conference of the Association for Consumer Research, 1973, 348-58.

Sheth, J.N., & Tuncalp, S. Structural assumptions underlying Fishbein's expectancy-value model of attitudes. University of Illinois, College of Commerce and Business Administration. Working Paper #207, 1974.

Triandis, H.C. Attitude and attitude change. New York: John Wiley, 1971.

Tuncalp, S. A study of the relationship between consumer's perceived quality judgements about a product based on the extrinsic and intrinsic product attributes and the underlying cognitive structure: An empirical investigation. Unpublished doctoral dissertation, University of Illinois, 1973.

Wilkie, W.L., & Pessemier, E.A. Issues in marketing's use of multi-attribute attitudinal models. Journal of Marketing Research, 1973, 10, 428-41.



P.S. Raju, University of Illinois
Rabi S. Bhagat, University of Illinois
Jagdish N. Sheth, University of Illinois


NA - Advances in Consumer Research Volume 02 | 1975

Share Proceeding

Featured papers

See More



Evan Polman, University of Wisconsin - Madison, USA
Sam J. Maglio, University of Toronto Scarborough

Read More


Liminality, Portals, and Narratives of Transformation

Laetitia Mimoun, HEC Paris, France
Fleura Bardhi, City University of London, UK

Read More


I10. Can Body Size Influence the Judgments of Warmth and Competence?

Trang Thanh Mai, University of Manitoba, Canada
Olya Bullard, University of Winnipeg
Luming Wang, University of Manitoba, Canada

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.