A Comparison of Two Behavioral Intention Models

ABSTRACT - The present investigation compared the predictor variables of Fishbein's and Triandis' behavioral intention models. The distinction Triandis makes concerning affect (gut feeling) was supported, that is, affect and attitude were found to measure two dimensions. In addition, the social influence constructs of both models seem to measure two dimensions: (1) what relevant others think and (2) what is appropriate to do.

Citation:

David Brinberg (1981) ,"A Comparison of Two Behavioral Intention Models", in NA - Advances in Consumer Research Volume 08, eds. Kent B. Monroe, Ann Abor, MI : Association for Consumer Research, Pages: 48-52.

Advances in Consumer Research Volume 8, 1981      Pages 48-52

A COMPARISON OF TWO BEHAVIORAL INTENTION MODELS

David Brinberg, University of Maryland

ABSTRACT -

The present investigation compared the predictor variables of Fishbein's and Triandis' behavioral intention models. The distinction Triandis makes concerning affect (gut feeling) was supported, that is, affect and attitude were found to measure two dimensions. In addition, the social influence constructs of both models seem to measure two dimensions: (1) what relevant others think and (2) what is appropriate to do.

INTRODUCTION

In recent years, there has been increased interest in determining the extent to which verbal reports, survey results, and attitude measures can predict behavior. Numerous models have been developed to predict behavior, and recently, some of these models have been compared (Zuckerman and Reis 1978; Jaccard and Davidson 1975; Brinberg 1979). The primary interest of these researchers comparing models was examining the relationship between the predictor variables and the criterion within each model, in other words, the researchers were interested in comparing the outcome of one model's predictions with the outcome of comparable predictions from the other models. For instance, in the studies cited above, the comparison of the different models involved comparing the multiple correlation (i.e., the percent of variance accounted for in the criterion) or the population estimate of the mean square error.

Often ignored are the relationships among two(or more) sets of predictor variables from the two(or more) models. It is possible that the outcome (e.g., multiple correlation) from a set of models are different(same) even though the predictor variables measure the same (different) underlying dimension(s) (Brewer, Campbell, and Crano 1970; Birnbaum and Mellers 1979). Thus, independent of the models outcomes, substantial similarities (or differences) may be found among the models. Since the dependent variables for these models are the same, differences between these models can exist: (1) among the predictor variables or (2) among the relationships of these predictor variables with the criterion. The major focus of this paper will be on the relationships among the predictor variables. The two models selected for comparison in this study are proposed by Fishbein (Fishbein and Ajzen 1975) and Triandis (1977). These models have been examined in a number of cultures (Davidson, Jaccard, Triandis, Morales, and Diaz-Guerrero 1976; Davidson and Jaccard 1975), with a number of behaviors (Songer-Nocks 1976; Zuckerman and Reis 1978; King 1975; Pomazal 1974) as well as compared using different behaviors (Jaccard and Davidson 1975; Seibold and Roper 1979; Brinberg 1979). However, in none of the studies comparing these two models was an attempt made to determine the interrelationships among the various predictor variables.

Fishbein's Model

Fishbein (Fishbein and Ajzen, 1975) states that a person's intention to perform a behavior is determined by: (1) one's attitude towards the performance of the behavior and (2) one's perception of whether significant others think s/he should or should not perform the behavior. This model may be expressed by the following algebraic equation

Intention = Attitude(w₁)+Subjective Norm(w₂)    (1)

The attitude is viewed as a function of the beliefs concerning the act multiplied by the evaluative aspect of the beliefs. This may be represented as follows:

Attitude = Sb_ie_i    (2)

Where b_i = likelihood that performing the behavior will result in some outcome, e_i = evaluation of that outcome. The subjective norm consists of normative beliefs of significant others multiplied by the motivation to comply with these others. This may be represented as follows:

Subjective Norm = SNbMc   (3)

where Nb = belief that a particular referent thinks you should or should not perform the behavior, Mc = motivation to comply with that particular referent.

The w₁ and w₂ constructs are theoretical parameters reflecting the importance of the variables in determining intention. They are generally determined through regression techniques.

Triandis' Model

Triandis (1977) states that the determinants of a person's intention may be expressed by the following set of equations.

Intention = A(w₃)+C(w₄)+S(w₅)    (4)

where A = affect towards the behavior, that is, perceived enjoyment vs. disgust associated with performing the behavior, C = the perceived value of the consequences associated with the behavior, S = social determinants.

As before, w₃, w₄, and w₅ are theoretical parameters reflecting the importance of the constructs in determining intent ion. They, too, are generally determined through regression techniques.

The social determinants component consists of a number of constructs: norms, roles, the actor's self-concept, moral norm, ideals, and contractual agreements. In this study, norms, roles, the self-concept, and the moral norm were used as elements of the social component. The definitions of these constructs will be discussed later.

The second equation of the model specifies the "C" component of the first equation. Following instrumentality theory (Rosenberg 1956), this term is operationalized as follows:

Consequences = Sp_cv_c    (5)

where p_c = likelihood that a behavior will result in some outcome, v_c = evaluation of that outcome.

Comparison of Fishbein's and Triandis' Models

One difference between Fishbein's and Triandis' models concerns a distinction between affect and attitude (evaluation). Triandis (1977) defines affect as the "gut feeling" associated with the behavior and distinguishes this from attitude (evaluation). Evaluation, according to Triandis, is a cognitive judgment based on Sp_cv_c. Triandis (1977) justifies this distinction by arguing that for some types of behaviors (e.g., sin and/or duty), both constructs will make separate contributions in predicting intention. Fishbein, on the other hand, has argued that both affect and attitude measure the same underlying evaluative dimension. Therefore, it is not necessary to postulate separate constructs since each is measuring the same underlying dimension. (It is important to note that Sp_cv_c is equivalent to Sb_ie_i and is used interchangeably throughout this paper.) Based on these hypothesized relationships, one aspect of this study will examine whether a single or multiple dimension can be used to describe affect, attitude, and consequences (Sb_ie_i).

A second difference between these two models involves the normative component in Fishbein's model and the social determinants in Triandis' model. Fishbein's approach implies that the normative component is an indicant of what relevant others think. According to Fishbein (Fishbein and Ajzen 1975), the subjective norm measures "what most people who are important to me think I should do."" This subjective norm is hypothesized to be determined by normative beliefs (what a particular referent thinks I should do) multiplied by the motivation to comply with that referent. Triandis (1977), on the other hand, defines a norm as the appropriateness of performing the behavior. In addition to norms, Triandis includes other constructs in the social determinants: roles, which is the appropriateness of performing the behavior for a person in a particular position in the social system; self-concept, .which consists of beliefs the person perceives concerning the appropriateness of the act for himself(herself); and moral norm, which is the perceived moral obligation to perform the behavior.

All the constructs in both the social determinants of Triandis' model and the normative component in Fishbein's model may be seen to measure a social influence dimension. If this is the case, these constructs should measure the same underlying dimension. Another aspect of this study, then, will examine the relationships among the constructs that are proposed to measure social influence.

Numerous behaviors can be used to test these two models. However, when comparing models, it would be useful to select behaviors that have been examined by each model separately. One behavior that meets this criterion is blood donation (Zuckerman and Reis 1978; Pomazal 1974).

METHOD

Subjects

Subjects were obtained from the participant pool in partial fulfillment of a course requirement. The data were collected at three points in time, during the course of four months, with a total of 96 subjects completing all three phases.

Procedure

In order to obtain the salient beliefs, 100 subjects, who did not participate in the main phase of this study, were asked to indicate the beliefs they associated with donating blood. Twenty-five subjects indicated their beliefs concerning donating blood in each of three contexts (i.e., in the context of the blood bank, the hospital, and the blood mobile) as well as donating blood in general. The use of these different contexts will be made clear shortly. The subjects were asked to indicate the advantages, disadvantages and consequences of donating blood. The modal set of beliefs were similar across all contexts. Thus, only one set of beliefs was used as the .modal salient set.

Measuring Instrument

All the constructs in both models were measured on 7 point scales. The belief statements were measured on a subjective probability scale (e.g., Donating blood at the blood bank this semester would be painful: likely-unlikely). The evaluation of those beliefs were measured on an evaluative scale (e.g., Doing something painful is: good-bad). Motivation to comply with a referent was measured by asking the subject, "Generally speaking, I want to do what (Referent X) thinks I should do: want to do-want not to do." Normative beliefs were measured by asking the subject, "(Referent X)thinks I should donate blood at the blood bank this semester: likely-unlikely." A person's attitude toward the act (e.g., Donating blood at the blood bank this semester is:) was measured on three evaluative scales (i.e., good-bad, nice-awful, favorable-unfavorable). The subjective norm was measured by asking the subject what significant others think s/he should or should not do (e.g., Most people who are important to me think I should donate blood at the blood bank this semester: likely-unlikely). The subject's intention was measured on a subjective probability scale (e.g., I intend to donate blood at the blood bank this semester: likely-unlikely).

The affect construct in Triandis' model was measured using scales intended to measure the person's "gut" feeling (i.e., enjoyable-disgusting; pleasant-unpleasant; exciting-nauseating). Roles, as conceptualized by Triandis measure the appropriateness of the behavior for a person in a particular position in the social system (e.g., It is appropriate for students to donate blood at the blood bank this semester: appropriate-inappropriate). Moral norm was measured by asking the subject, "I have a moral obligation to donate blood at the blood bank this semester: have a moral obligation-have no moral obligation.'' The attributes used to measure the self-concept were obtained in the aforementioned elicitation phase (e.g., I consider myself the type of person who is concerned: likely-unlikely).

RESULTS

In order to determine the generalizability of these models across situations, blood donation was examined in the context of the blood bank, the hospital, and the blood mobile. The predictive accuracy of both models was assessed and found not to be systematically different (R² for both models was approximately 30%). Since the main intent of this paper is the examination of the interrelationships among the different predictor variables of both models, time predictive accuracy of these models will not be discussed. The reader interested in a detailed presentation of the predictive accuracy of these models should see Brinberg (1979).

As mentioned earlier, Triandis draws a distinction between affect and attitude whereas Fishbein sees both constructs as measures of the same underlying evaluative dimension. In order to examine this theoretical distinction, two approaches may be used. The first would be to place the attitude and affect measures in a regression analysis and determine whether both make independent contributions in predicting intention. However, as pointed out by Brewer et al. (1970), Kenny (1976) and Birnbaum and Mellers (1979), interpreting the pattern (significance) of regression weights may lead the researcher to mistakenly interpret that more than one factor underlies a set of constructs. Since the regression technique does not allow for the effects of measurement error, a researcher may conclude that two (or more) constructs exist, when, in fact, the constructs are imperfect measures of the same underlying dimension.

To circumvent some of the limitations of the regression technique, Brewer et al. (1970) suggest using a factor analytic approach to determine the number of factors underlying a set of constructs. Since a unique algebraic solution is always available for three variables (Brewer et al. 1970), it was decided to factor analyze the scales used to measure affect and attitude as well as Sb_ie_i in order to determine whether a single factor can be used to account for the relationships among these constructs. Table 1 contains a varimax rotation of a principal components analysis of attitude, affect, and consequences.

TABLE 1

VARIMAX ROTATION OF PRINCIPAL COMPONENTS ANALYSIS OF ATTITUDE, AFFECT AND CONSEQUENCES FOR THE INTENTION TO DONATE BLOOD

A maximum likelihood factor analysis (Lawley and Maxwell 1971; Joreskog 1967; Humphreys and Montanelli 1975) was used to determine the number of factors underlying these concepts since it provides a c² goodness of fit test for any K factor model. If the c² value obtained from this analysis is significant for K factors, this indicates that at least K+1 factors are needed to adequately fit the data, that is, a significant c² indicates that a significant portion of the variance can still be systematically accounted for by extracting another factor. Thus, the number of factors used as an adequate fit of the data occur for the first non-significant c² value.

For the factor analyses reported in Table 1, all yielded a significant c² for the first factor (p < .01) and all but one yielded a non significant c² for the second factor. The significant c² for the second factor was in the context of the blood mobile at time 3 (c² = 17.63, p < .05). The X² for the third factor was non-significant.

The findings from the maximum likelihood factor analysis indicate that two factors consistently provide a sufficient fit of the data. Based on the pattern of factor loadings found in Table 1, the first factor appears to measure an affect dimension. That is, the bi-polar adjectives specified a priori to measure affect (Triandis 1977), all loaded on the same dimension. The second factor appears to measure an evaluative dimension since the good-bad scale (which Fishbein uses as a measure of attitude) consistently has a high loading.

An alternative explanation to account for the pattern of factor loadings (i.e., the emergence of two factors) is that it is an artifact of the response language used by the subject. In other words, two dimensions may emerge since one set of adjectives (e.g., affect) have a greater subjective range than another set (e.g., attitude). However, the plausibility of this alternative explanation is reduced since an independent assessment of attitude (i.e., Sb_ie_i) is consistently loaded with the attitude scales. This provides additional evidence for describing the second factor as evaluative.

Thus, based on the pattern of factor loadings as well as the maximum likelihood factor analysis, Triandis' distinction between affect and attitude is supported since two factors, not one, are needed to adequately describe the data.

The relationships among the constructs in Triandis' social determinants component as well as Fishbein's normative component also need to be examined in order to determine the number of dimensions underlying these concepts. Triandis has postulated that moral norm, roles, norms, and the self-concept can be summed to form an index of social determinants. This implies that these constructs measure the same underlying dimension. In order to determine the relationships among these constructs (along with Fishbein's normative measures) the same factor analytic techniques described earlier were used. Tables 2 contains the varimax rotation of a principal components analysis of the constructs that measure social influence.

A maximum likelihood factor analysis was performed in order to determine the number of factors that would adequately describe the data. In all cases, the c² value for the first factor was highly significant(p < .01). and non-significant for the second factor. These patterns of results indicate that a two factor solution can be used to describe the constructs that measure social influence. This is contrary to Triandis' assumption that the moral norm, roles, norms, and the self-concept are measuring a single dimension. Based on the pattern of factor loadings found in Tables 2, the first factor appears to measure a dimension that deals with the appropriateness of the behavior, since norms and roles have a high loading on it. The second factor appears to measure what "relevant others" think since the subjective norm and SNbMc have a high loading on this dimension. Interestingly the moral norm generally has a high loading on this dimension. This suggests that people view a moral obligation as something society thinks they should do, in other words, measuring what relevant others think.

TABLE 2

VARIMAX ROTATION OF PRINCIPAL COMPONENTS ANALYSIS OF SOCIAL INFLUENCE CONSTRUCTS FOR THE INTENTION TO DONATE BLOOD

In summary, a single factor solution was not adequate to describe the relationships among the social influence constructs. Based on the maximum likelihood factor analysis, two factors were adequate to describe the data. These factors appear to measure: (1) what relevant others think and (2) what is appropriate to do.

DISCUSSION

Based on the factor analyses of attitude, affect, and consequences two dimensions, not one, appear to be sufficient in describing the relationships among these constructs. This result is contrary to Fishbein's assumption that all three are measuring the same underlying dimension and supportive of Triandis' assumption that these constructs may be separated into two dimensions; affect and attitude.

A similar analysis of the social determinants in Triandis model and the normative components in Fishbein's model shows that two dimensions, not one, are needed to account for the relationships among these constructs. This is inconsistent with Triandis' assumption that the constructs of the social determinants component measure a single underlying dimension. The two dimensions appear to measure: (1) what relevant others think and (2) what is appropriate to do.

An implication of these findings is that neither model, alone, can account for the relationships among these constructs. This suggests that a composite model, incorporating features of both models may be used to predict intention. This composite model would contain: (1) an affect construct, (2) an attitude construct, (3) a "relevant others" construct, and (4) an "appropriateness" construct. A preliminary analysis of this composite model indicated that it was an accurate predictor of intention and consistently was more effective in predicting intention than either Fishbein's or Triandis' model. However, the magnitude of this difference was small and it is possible that this composite model was "better" since it was tested in the same sample in which it was developed. Future research is necessary comparing this composite model with Fishbein's and Triandis' model in order to determine whether it is useful to integrate these two approaches in predicting intention.

Finally, past research (e.g., Jaccard and Davidson 1975, Seibold and Roper 1979) has focused on the comparison of the predictive accuracy of these two behavioral intention models and has found both models to be superior to the other in predicting intention. What the present study suggests is that substantial similarities and differences may be found among these models by examining the relationships among the predictor variables. Therefore, it is strongly recommended that researchers interested in comparing different models (which have the same criterion), examine the relationships among the independent variables in addition to examining the relative predictive efficiency of these models.

REFERENCES

Birnbaum, M. H., and Mellers, B. A. (1979), "Stimulus Recognition May Mediate Exposure Effects," Journal of Personality and Social Psychology, 37, 391-394.

Brewer, M. B., Campbell, D. T., and Crano, W. D. (1970), "Testing a Single Factor Model as an Alternative to the Misuse of Partial Correlations in Hypothesis Testing,", Sociometry, 33, 1-11.

Brinberg, D. (1979). "The Comparison of Three Attitude Models for the Prediction of Blood Donation Behavior," Unpublished Doctoral Dissertation.

Davidson, A. R. and Jaccard, J. J. (1975), "Population Psychology: A New Look at an Old Problem," Journal of Personality, and Social Psychology, 31(6), 1073-1082

Davidson, A .R., Jaccard, J. J., Triandis, H. C., Morales, M. L. and Diaz-Guerrero, R. (1976), "Cross-Cultural Model Testing: Toward a Solution of thee Etic-Emic Dilemma," International Journal of Psychology, 11, 1-13.

Fishbein, M. and Ajzen, I. (1975), Beliefs, Attitudes, Intention and Behavior: An Introduction to Theory and Research, Reading, Mass: Addison-Wesley.

Humphreys, L. G. and Montanelli, R. G. (1975), "An Investigation of the Parallel Analysis Criterion for Determining the Number of Common 'Factors", .Multivariate Behavioral Research, 10, 193-206. m ,

Jaccard, J. J. and Davidson, A. R. (1975), "A Comparison of Two Models of Social Behavior: Results of a Survey Sample", Sociometry, 38, 491-517.

Joreskog, K. G. (1967), "Some Contribution to Maximum Likelihood Factor Analysis" Psychometrics, 32, 443-482

Kenny, D. A. (1976), "Data Analysis in the Social Psychological Experiment", Representative Research in Social Psychology, 7(2), 120-133.

King, G. W. (1975), "An Analysis of Attitudinal and Normative Variables as Predictors of Intentions and Behaviors" Speech Monographs, 42, 237-224

Lawley, D. N. and Maxwell, A. G,. (1971), "Factor Analysis as a Statistical Method," American Elsevier, New York.

Pomazal, R. S. (1974), "Attitudes, Normative Beliefs, and Altruism: Help for Helping Behavior ," Unpublished Doctoral Dissertation, University of Illinois.

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

Schwartz, S. H. and Tessler, R. C. (1972), "A Test of a Model for Reducing Attitude Behavior Discrepancies," Journal of Personality and Social Psychology, 24, 349-364.

Seibold, D. R., and Roper, L. C. (1979), "Psychosocial Determinants of Health Care Intentions: Test of the Triandis and Fishbein Models," In D. Nimmo (ed.), Communication Yearbook 3, New Brunswick, NJ Transaction Books.

Songer-Nocks, E. (1976), "Situational Factors Affecting the Weighting of Predictor Components in the Fishbein Model," Journal of Experimental Social Psychology, 12(1), 56-70.

Triandis, H. C. (1977), Interpersonal Behavior, Monterey, CA. Brooks/Cole.

Zuckerman, M. and Reis, H. T. (1978), "Comparison of Three Models for Predicting Altruistic Behavior," Journal of Personality and Social Psychology, 36(5), 498-510.

----------------------------------------