Advances in Consumer Research Volume 7, 1980 Pages 339-344
SINGLE COMPONENT VERSUS MULTICOMPONENT MODELS OF ATTITUDE: SOME CAUTIONS AND CONTINGENCIES FOR THEIR USE
Richard P. Bagozzi, Massachusetts Institute of Technology
Robert E. Burnkrant, Ohio State University
The convergent and predictive validity of both the single component and affective-cognitive models of attitude were examined in the context of attitudes toward political participation. Using a multivariate structural equation methodology, it was found that it is not possible to discriminate between the single component and affective-cognitive models based on convergent validity alone. Second, when affective and cognitive dimensions were treated as indicators of a single attitude construct, attitude was not found to significantly predict actual and self-reported behavior. However, as a third hypothesis, when attitude was examined as a multi-dimensional affective/cognitive construct, cognitions, but not affect, were found to be the primary predictors of behavior. Overall the results indicate that cognitions may be the principle determinant of behavior for some attitude objects in certain circumstances and that affect, while highly correlated with cognitions, are nevertheless spuriously related to behavior. To explain the findings, arguments as to the antecedents of attitude and the conditions surrounding attitude-behavior correspondence are used.
Researchers in both psychology and consumer behavior hold widely differing views regarding the content of attitude. A number of researchers maintain that attitude is a complex construct consisting of separate affective, cognitive, and conative dimensions (e.g., Rosenberg and Hovland, 1960; Ostrom, 1969; Bagozzi, Tybout, Craig, and Sternthal, 1979). Others consider attitude to represent a single dimension of affect for or against a psychological object (i.e., Thurstone, 1931). Those who maintain this latter position often consider attitudes to be comprised of beliefs and the evaluations associated with these beliefs (e.g., Fishbein and Ajzen, 1975).
The distinction between a multicomponent and a single component model of attitude is an important one for both consumer behavior and psychology because these positions have unique implications for both attitude change strategies and the attitude-behavior relationship. If a single component model of attitude is a valid representation, then any single valid measure of attitude would be an equally appropriate predictor of behavioral criteria; but if a multicomponent model is valid, both cognitive and affective measures should be employed as simultaneous predictors of behavior. In consumer behavior, attitude change researchers have almost invariably assumed that it is necessary to change beliefs in order to change attitude. However, according to a multi-component conceptualization, a belief change, or cognitive learning, strategy would be only one of several options available to the advertiser. These and other related distinctions between the two models will be treated in more detail subsequently.
The present research is intended to demonstrate that this issue of the single component versus multicomponent model of attitude is not a simple question that can be answered in a yes/no dichotomy. Rather, it is proposed that the validity of the attitude models depends on the conditions surrounding attitude formation, the characteristic of the attitude object, and the degree of definitional specificity between attitude and behavior. After briefly reviewing the literature supporting each model and suggesting conditions under which the multi-component model may or may not be applicable, hypotheses are formulated, and a test of the single component versus multicomponent model of attitude is made.
EVIDENCE FOR THE MULTICOMPONENT MODEL
Although a vast literature exists relevant to the single component model of attitude (cf., Eagley and Himmelfarb, 1974, 1978; Schuman and Johnson, 1976), very little is known about the validity of the multicomponent model. In an early study, Woodmansee and Cook (1967) found no evidence for the multicomponent model in their investigation of racial attitudes. However, they made no effort to assess either discriminant or predictive validity, and the procedure that they used to determine convergent validity (i.e., factor analysis with varimax rotation) can be criticized because it may not yield a unique solution (cf., J÷reskog, 1969). Ostrom (1969) and Kothandapani (1971), in contrast, report the attainment of convergent and discriminant validity when the attitude object was the church and birth control, respectively. However, in a reanalysis of both data sets using the analysis of covariance structures methodology, Bagozzi (1978) found that convergent and discriminant validity could only be sustained for the data of Ostrom but not Kothandapani.
To explain the discrepancy, Bagozzi maintained that separate learning experiences can be identified as antecedents to each attitudinal component (cf., Greenwald, 1968). The affective component is thought to arise largely through classical conditioning where unconditioned and conditioned stimuli become paired to elicit emotional responses. Similarly, cognitions are formed primarily through observational or cognitive learning where the perceived temporal and/or spatial connections between certain persons, events, or things are linked in thought. Finally, either affective or cognitive reactions can arise through instrumental learning or operant conditioning wherein particular actions are found to lead to rewarding or punishing outcomes. For separate components of attitude to exist, then, it was argued that it must be possible to identify unique determinants of each component.
Applying this argument, Bagozzi suggested that separate antecedents could be identified in the formation of attitudes toward the church, but not necessarily toward birth control. For attitudes toward the church, it was argued that a common set of antecedents most likely existed for those holding more-or-less favorable attitudes (e.g., through cultural norms, family reinforcement, and socialization); these determinants probably operated from an early age; and considerable thought, reflection, discussion, and conduct consistent with one's attitude occurred throughout life. As a result, particularly for those expressing favorable attitudes toward the church, one expects experiences to have occurred involving classical and operant conditioning, cognitive learning, and moral reinforcement. For attitudes toward birth control, one does not necessarily expect consistent, effective antecedents to exist; any antecedents that might have been present probably began later in life; and much less thought, discussion and behavioral experiences probably occurred. Given that birth control practices are private topics and that the sample in Kothandapani's study consisted of low income black women (half of whom were nonusers of contraceptives), it is reasonable to expect that strong antecedents were not operative to the degree required for the formation of separate, crystallized attitudinal components. The data supported these hypotheses (Bagozzi, 1978).
If separate antecedents shape affective and cognitive reactions toward an attitude object, then the correlations among affective and cognitive responses must not only be high and statistically significant (suggesting convergent validity), but the pattern of correlations must be consistent among and across all pairs of responses. The latter requirement is similar to Campbell and Fiske's (1959, p. 84) fourth desideratum with regard to discriminant validity. The implication of this for attitude structure is that, for those attitudes achieving clear crystallization of affective and cognitive dimensions, the two-dimensional confirmatory factor analysis model must be capable of representing attitude, while the one-dimensional confirmatory factor analysis model should not be sufficient to represent attitude. Bagozzi and Burnkrant (1979) tested this hypothesis on data originally collected by Fishbein and Ajzen (1974), where the attitude object was "being religious." As predicted, even though the intercorrelations among the two affective and three cognitive scales were high (r = .519 to r = .878) and statistically significant (all were at the .01 level or better), the single component attitude model had to be rejected in two separate samples, while the bicomponent affective/cognitive model could not be rejected in either sample. This is an important finding because it demonstrates that not only is there evidence for a bicomponent model (using Campbell and Fiske's first and fourth desiderata), but the bicomponent model can exist even when the single component model does not, in certain instances.
The above discussion has dealt primarily with the internal consistency of the multicomponent attitude construct, especially with respect to convergent and discriminant validity. The validity of the construct must also be assessed in a predictive context, however. Following Campbell (1960), nomological validity refers to the degree to which predictions from a construct are confirmed when it is part of a larger theoretical network. In this sense, a construct achieves nomological validity when it functions as both an independent and dependent variable, and hypotheses relevant to its role in either case are confirmed. Further, there must be theoretical reasons underlying the predictions as to the mechanisms or laws governing these relationships.
The research with regard to the predictive validity of the multicomponent model is sparse, but promising. Norman (1975), for instance, found that affective-cognitive consistency moderates the relationship between attitude and behavior. His study showed that the affective, and to a lesser extent the cognitive, component of attitude toward "volunteering as a subject for psychological research" predicted subsequent behavior at significantly stronger levels for those high as opposed to low in affective-cognitive consistency. However, these findings should be viewed with caution.
While it is true that affect functioned as a significant predictor of behavior in the high, but not low, affective-cognitive consistency group, it was also the case that cognitions were significant predictors of behavior for both groups overall. Because Norman only performed univariate tests, it is impossible to determine whether the attitude-behavior relationship was spurious. Given the methodology and findings, it is impossible to say whether the observed differences between groups was due to the confounding influence of the strong cognitive-behavior relationship and the collinearity of cognitions with affect. Further, given the attitude object, it is likely that no separate, strong, or well-developed antecedents to the attitude dimensions existed, that little experience with the attitude object would be expected, and that the attitude object would carry a low salience for subjects. Ail of these factors argue against the development of distinct attitudinal components and lend support to the claim that the observed differences in prediction between groups could have been spurious.
Perhaps the strongest evidence for the predictive validity of the multicomponent model can be found in Bagozzi and Burnkrant (1979). These authors argued that, even though the individual univariate correlations between each affective and cognitive measure of attitude on the one hand and the self-reported behaviors and behavioral intentions on the other hand were high (r = .444 to r = .727) and statistically significant (all were significant at the .01 level or better), because the pattern of correlations should not sustain the single component model, the predictive validity of this model should also be rejected. This was, in fact, found to be the case in the two samples tested. In contrast, when the multicomponent attitude model was examined for its predictive validity using the multivariate statistical causal modeling methodology, the hypothesis of predictive validity could not be rejected.
The present study attempted to test both multicomponent and single component models of attitude on the same set of data examining convergent and predictive validities of each. Attitudes and actual behaviors with respect to personal participation in political activities of college students were investigated. The attitude object was chosen because it lies somewhere between the extreme afforded by Norman's (1975) study of attitudes toward volunteering as a subject in an experiment and Ostrom's (1969) and Fishbein and Ajzen's (1974) investigation of religious attitudes. It is thus probably representative of many common, every day attitudes. That is, attitudes toward personal participation in political activities, while not experiencing quite the degree of classical and operant conditioning as religious attitudes, nevertheless are shaped by some instrumental and cognitive learning processes and hence have the potential for achieving distinct affective and cognitive dimensions (cf., Converse, 1964, 1970; Jennings and Niemi, 1968; Sigel, 1970). The attitude object also lends itself to the study of unobtrusive and self-reported behavioral criteria.
The first hypothesis that will be tested in the present research is that convergent validity will be achieved for both the single-component and affective-cognitive models of attitude. The rationale for this prediction is based on the expected impact of learning experiences on attitude formation. It is maintained that specific antecedents in the form of cognitive and observational learning are likely to have occurred throughout life prior to the performance of actual participation in political activities and the measurement of attitudes. As research indicates, the learning of political attitudes occurs through family socialization, formal education, and cultural indoctrination and role modeling in the mass media (Hess and Torney, 1965; Hyman, 1959; Jennings and Niemi, 1968, Sigel, 1970). It is claimed that these conditions are necessary for the formation of separate affective and cognitive reactions toward political participation.
The second hypothesis that will be examined is that the single component model will not predict actual and self-reported political participation even though all attitude measures are highly correlated with all behavior measures. In essence, it is proposed that the single-component model confounds the effects of affect and cognitions on behavior by treating both as parallel measures of attitude. This confounding can be seen at an intuitive level by noting that the pattern of correlations across attitude and behavior measures is nonuniform. The maximum likelihood ratio chi-square statistic can be used to explicitly test this hypothesis.
Finally, as a third hypothesis, it is proposed that cognitions will be the primary predictors of behavior and that affect will not significantly predict behavior when the multicomponent model of attitude is examined. The causal model provides a multivariate statistical test of this hypothesis.
The reason for this prediction is based on the attitude object and assumed antecedents to attitude formation. Although learning and socialization experiences can be expected to have occurred to a degree sufficient for the development of cognitive and effective reactions toward political participation, the cognitive component receives the most numerous and powerful influences during the first 18 years of life. The educational indoctrination experienced in the classroom is likely to be heavily oriented toward cognitive learning. Because actual voting behavior is unlikely to have occurred, except for mock elections within the secondary school setting, the degree of classical and operant conditioning was probably quite minimal for the sample studied. Affect toward political participation is therefore hypothesized to consist of relatively low levels of evaluative reactions, and the connection between these and behavior--when tested along with cognitions in a multivariate context--is expected to be negligible or at least less forceful than cognitions. This hypothesis should be contrasted to the study of religious attitudes performed by Bagozzi and Burnkrant (1979), where it was found that both affect and cognitions predicted behavioral intentions and self-reported behaviors.
To Test the hypotheses, a reanalysis was performed of data originally reported in Tittle and Hill (1967). Tittle and Hill examined the attitudes and behaviors of 301 students with regard to personal participation in student political activities. Five attitude measures were taken: (1) Thurstone successive-interval, (2) semantic differential, (3) summated-rating (Likert), (4) Guttman, and (5) self-rating scales. The five behavioral criteria included: (1) a measure of past voting behavior unobtrusively obtained from voting records, (2) each respondent's self-report of past voting behavior, (3) a Likert index comprised of the sum of responses to items tapping frequency of engagement in ten student political activities, (4) a Guttman scale measuring responses to eight self-reports of behavior, and (5) a modification of the standard Woodward-Roper index of political participation. Only the actual and self-reported behaviors will be examined here.
Method of Analysis
Convergent Validity. To test for convergent validity, the analysis of covariance structures model developed by Bagozzi (1978) was used. Briefly, the null hypothesis of convergence is:
where x is a vector of k single-component or l multicomponent measures, A is a vector of components, L is a matrix of parameters relating each attitude measure to its respective component, and x is a vector of unique scores (i.e., errors in measurement). The c2 goodness-of-fit test yielded by LISREL (Joreskog and van Thillo, 1972) tests the null hypothesis against the alternative hypothesis that the variance-covariance matrix of observations is any positive definite matrix.
Predictive Validity. To test the hypothesis that attitude is a significant predictor of actual and self-re-ported behavior, the causal modeling methodology developed by Bagozzi (1980) was used. The null hypothesis is:
where equation (2) expresses the predictive relationship between attitude, A, and behavior, B (represented as a); equation (3) represents the measurement relations between A and B on the one hand and their operationalizations on the other hand; x is a vector of attitude measure; y is a vector of behavioral criteria; z is a vector of errors in equations; w is a vector of errors in measurement for B; and the remaining parameters are as defined earlier. Again, LISREL can be used to rest this null hypothesis.
Single-Component Model. Application of the model of equation (1) and LISREL to the data shows that the hypothesis of convergent validity can not be rejected for the single-component attitude model, as predicted (c2 = 9.05, d.f. = 5, p .11). Thus, even though the semantic differential and self-rating scales represent evaluative content, while the Thurstone, Likert, and Guttman scales indicate cognitive content, all five converge to measure one, single underlying attitude construct.
Multicomponent Model. The effective/cognitive multicomponent model consists of one cognitive component measured by the Likert and Guttman scales, one cognitive component measured by the Thurstone scale, and two separate effective components measured by the semantic differential and self-rating scales, respectively. This was the most parsimonious model to fit the data. Application of the model of equation (1) and LISREL to the data shows that the hypothesis of convergent validity can not be rejected, as hypothesized (c2 = 5.72, d.f. = 8, p .33).
Single-Component Model. Application of equations (2) and (3) and LISREL to the data (with A specified as a single construct measured with five scales and the two behavioral measures as criteria) yields c2 = 45.88, d.f. = 13, and p .00. Thus, if affect and cognitions are treated as congeneric measures of a single attitude construct, then one must reject the hypothesis that attitudes predict behavior, as proposed.
Multicomponent Model. Because the single component model treats affect and cognitions as equivalent, undifferentiated measures, it confounds the true relation between attitudes and behavior. To test the null hypothesis implied by equations (2) and (3) when affect and cognitions are modeled as distinct, yet interrelated, components of attitude, LISREL was used to determine the multivariate impact of the separate components on behavior. The results of this test are shown in Figure i where the value of the chi-square statistic (i.e., c2 = 18.53, d.f. = 10, p .05) indicates a borderline fit. However, when the four nonsignificant paths from the attitude components to the behavior measures are constrained to equal zero, the model produces an acceptable fit (i.e., c2 = 20.35, d.f. = 14, p .13). The parameter estimates relating attitudes to behavior generally support the hypothesis that cognitions, and not affect, are the primary predictors of behavior, as posited. The cognitive construct measured by the Likert and Guttman indices predicts behavior very well, as proposed (i.e., a1 = 0.444, t = 5.32; a2 = 0.489, t = 5.79). Similarly, as hypothesized, affect as measured by the semantic differential does not predict either behavior criterion significantly (i.e., a5 = 0.053, t = 0.81; a6 = 0.000, t = 0.00), nor does affect measured by the self-report measure predict actual behavior significantly (i.e., a7 = 0.029, t = 0.49). Contrary to predictions, the Thurstone measure of cognitions did not predict actual behavior (i.e., a3 = 0.024, t = 0.38) and was negatively related to self-reported behavior (a4 = -0.128 t = 2.03). Also, the self-report measure of affect significantly predicted the self-reported behavior measure (a8 = 0.158, t = 2.69). It should be noted, however, that the paths from the cognitive construct measured by the Likert and Guttman indices (i.e., a1 and a2) are about three to three and one half times as large as a4 and a8. Hence, most of the impact from attitudes to behavior is due to cognitions, as hypothesized. The negative sign on a4 is most likely the result of some multicollinearity among exogenous variables. However, the absolute value of this effect is relatively low.
DISCUSSION AND CONCLUSIONS
The distinction between the single component and multi-component models of attitude is not a simple, either-or question. Rather, it appears that either conceptualization may have utility, depending on the conditions associated with the formation of attitude and the attitude object itself.
PREDICTIVE VALIDITY OF THE MULTICOMPONENT MODEL
The results of this study and a number of others suggest that the antecedents of attitude, the nature of the attitude object, and/or the degree of correspondence between attitude and behavior interact complexly to determine the structure and content of attitude. The first area where such factors appear to influence attitude can be seen in the evidence for convergence. Although the presence of high correlations among alternative measures of attitude would appear to indicate convergence (cf., Fishbein and Ajzen, 1974), such zero-order relations can be misleading, masking the true structure of attitude experienced by the respondents. For example, Bagozzi and Burnkrant (1979) found that the affective-cognitive model of attitude, but not the single component model, achieved convergent validity. Apparently the affective and cognitive components had attained a high degree of crystallization or discrimination in addition to showing high within component convergence. Such a structure is perhaps common for well-formed attitudes toward objects such as religion where separate antecedents in the form of classical conditioning, instrumental learning, and cognitive learning are operative throughout much of life.
In the present study, however, convergent validity was achieved for both the single component and affective-cognitive models. Given that the attitude object was political participation, it is likely that much less classical conditioning and less instrumental learning occurred for the respondents compared to more central attitude objects such as religion. Under the conditions surrounding the formation of attitudes toward political participation, one might not expect an attitude structure exhibiting fully distinct components. Yet, because some instrumental and cognitive learning processes probably occurred through family, school, and general cultural socialization, it can be expected that both affective and cognitive reactions had formed. These learning experiences would be consistent with the attainment of convergent validity for both the single component and affective-cognitive models. For still less central attitudes, such as those toward a particular brand-name in a product class, it might be expected that the affective and cognitive dimensions would not achieve differentiation, and thus convergence would be obtained for the single component model only.
Overall, then a continuum might be hypothesized beginning with less central or surface attitudes wherein the affective and cognitive dimensions possess little if any unique variance but rather converge to a single unidimensional attitude construct and ending with central, well-formed attitudes produced by distinct antecedents and resulting in separate affective and cognitive components that demonstrate convergence for only the multicomponent conceptualization of attitude. Somewhere in between these extremes lies attitudes such as those toward political participation where it is impossible to discriminate, based on convergent validity, between the single component and multicomponent models.
In the present study, it was found that cognitions are the major predictors of actual and self-reported behavior, while affect appears to function largely as a parallel response not strongly related to behavior. Apparently, although separate evaluations and cognitions exist with respect to the attitude object (as indicated by the convergent validity hypotheses), only the latter influences behavior. Affective reactions may have formed through classical and operant conditioning experiences involved in family socialization and peer group relations. However, these objective experiences would not likely have been tied very directly to behavior because subjects had not performed political participation behaviors to any extent prior to college. The prevalent cognitive learning experiences of the classroom and other explicit consideration of political anticipation were likely to have been more closely tied to behavior for these subjects. This suggests that the cognitive component would be the stronger moderator guiding behavior. On the other hand, when attitude structure shows a clear crystallization between affect and cognitions, such as with attitudes toward religion, both appear to predict behavior with affect being the stronger predictor (Bagozzi and Burnkrant, 1979).
It would appear that the attitude-behavior relationship is a complex one contingent on the circumstances. For attitude objects exhibiting primarily cognitive learning experiences, it might be sufficient to analyze only the influence of the cognitive component on behavior. For objects involving strongly crystallized attitudinal components due to extensive prior classical, operant, and cognitive learning experiences, it appears that both components need to be considered, since they contribute independently to the prediction of behavior (c.f., Bagozzi and Burnkrant, 1979). This research suggests that for attitudinal objects relevant to consumer behavior researchers, the efficacy of the single component versus multicomponent model depends on the nature of the object and prior learning experiences involving that object. Objects involving only cognitive or observational learning (such as attitudes toward political participation in young adults) are probably indicative of strong cognitive-behavior relationships and weak affective-behavior relationships. Other objects in which classical and operant conditioning experiences are more prevalent may exhibit stronger affective-behavior connections. More central objects involving extensive classical, operant, and cognitive learning experiences are likely to require consideration of both cognitive and affective components.
Bagozzi, R. P. (1978). "The Construct Validity of the Affective, Behavioral, and Cognitive Components of Attitude by Analysis of Covariance Structures," Multivariate Behavioral Research, 13, 9-31.
Bagozzi, R. P. (1980) Causal Models in Marketing, New York: John Wiley & Sons.
Bagozzi, R. P. and Burnkrant, R. E. (1979). "Attitude Organization and the Attitude-Behavior Relationship," Journal of Personality and Social Psychology, 37, 913-929.
Bagozzi, R. P., Tybout, A.M., Craig, C. S. and Sternthal, B (1979). "The Construct Validity of the Tripartite Classification of Attitudes," Journal of Marketing Research, 16, 88-95.
Campbell, D. T. (1960). "Recommendations for APA Test Standards Regarding Construct, Trait, or Discriminant Validity," American Psychologist, 15, 546-553.
Campbell, D. T. and Fiske, D. W. (1959). "Convergent and Discriminant Validation by the Multitrait-Multimethod Matrix," Psychological Bulletin, 56, 81-105.
Converse, P. E. (1964). "The Nature of Belief Systems in Mass Publics," in D. E. Apter, (ed.), Ideology and Discontent, New York: Free Press, 206-261.
Converse, P. E. (1970). "Attitudes and Nonattitudes: Continuation of a Dialogue," in E. R. Tufte, (ed.), The Quantitative Analysis of Social Problems, Reading, Mass.: Addison-Wesley, 169-189.
Eagley, A.H. and Himmelfarb, S. (1974). "Current Trends in Attitude Theory and Research," in S. Himmelfarb and A.H. Eagley, (eds.), Reading in Attitude Change, New York: John Wiley & Sons, 594-643.
Eagley, A.H. and Himmelfarb, S. (1978). "Attitudes and Opinions," in M. R. Rosenweig and L. W. Porter (eds.), Annual Review of Psychology, Vol. 29, Palo Alto, California: Annual Reviews, 517-554.
Fishbein, M. and Ajzen, I. (1974). "Attitudes Towards Objects as Predictors of Single and Multiple Behavior Criteria," Psychological Review, 81, 59-74.
Fishbein, M. and Ajzen, I. (1975). Belief, Attitude, Intention and Behavior: An Introduction to Theory and Research, Reading, Mass.: Addison-Wesley.
Greenwald, A. G. (1968). "On Defining Attitude and Attitude Theory," in A. G. Greenwald, T. C. Brock, and T. M. Ostrom (eds.), Psychological Foundations of Attitudes, New York: Academic Press. 361-388.
Hess, R. D. and Torney, J. V. (1965). The Development of Basic Attitudes and Values Toward Government and Citizenship During the Elementary School Years, Part I. Cooperative Research Project No. 1078, Washington, D.C.: U.S. Office of Education.
Hyman, H. (1959). Political Socialization, New York: Free Press of Glencoe.
Insko, C.A. and Schopler, J. (1967). "Triadic Consistency: A Statement of Affective-Cognitive-Conative Consistency," Psychological Review, 74, 361-376.
Jennings, M. K. and Niemi, R. G. (1968). "The Transmission of Political Values from Parent to Child," American Political Science Review, 62, 169-184.
Joreskog, K. G. (1969). "A General Approach to Confirmatory Maximum Likelihood Factor Analysis," Psychometrika, 34, 183-202.
J÷reskog, K. G. and van Thillo, M. "LISREL: A General Computer Program for Estimating a Linear Structural Equation System Involving Multiple Indicators of Unmeasured Variables," Princeton, NJ: Educational Testing Service.
Kothandapani, V. (1971). "Validation of Feeling, Belief, and Intention to Act as Three Components of Attitude and Their Contribution to Prediction of Contraceptive Behavior,'' Journal of Personality and Social Psychology, 19, 321-333.
Norman, R. (1975). "Affective-Cognitive Consistency, Attitudes, Conformity, and Behavior," Journal of Personality and Social Psychology, 32, 83-91.
Ostrom, T. M. (1969). "The Relationship Between the Affective, Behavioral, and Cognitive Components of Attitude," Journal of Experimental Social Psychology, 5, 12-30.
Rosenberg, M. J. and Hovland, C.I. (1960). "Cognitive, Affective, and Behavioral Components of Attitudes," in M. J. Rosenberg, C.I. Hovland, W. J. McGuire, R. P. Abelson, and J. W. Brehm, (eds.), Attitude Organization and Change: An Analysis of Consistency Among Attitude Components, New Haven: Yale University Press.
Schuman, H. and Johnson, M. P. (1976). "Attitudes and Behavior," in A. Inkeles, J. Coleman, and N. Smelser, (eds.) Annual Review of Sociology, Vol. 2, Palo Alto, California.
Sigel, R. (1970). Learning About Politics, New York: Random House.
Thurstone, L. L. (1931). "The Measurement of Social Attitudes," Journal of Abnormal and Social Psychology, 26, 249-269.
Tittle, C. R. and Hill, R. J. (1967). "Attitude Measurement and Prediction of Behavior: An Evaluation of Conditions and Measurement Techniques," Sociometry, 30, 199-213.
Woodmansee, J. J. and Cook. S. W. (1967). "Dimensions of Verbal Racial Attitudes: Their Identification and Measurement," Journal of Personality and Social Psychology, 7, 240-250.