The Relations Among Attribute and Importance Components of Rosenberg-Fishbein Type Attitude Model: an Empirical Investigation

Citation:

Reza Moinpour and Douglas L. MacLachlan (1971) ,"The Relations Among Attribute and Importance Components of Rosenberg-Fishbein Type Attitude Model: an Empirical Investigation", in SV - Proceedings of the Second Annual Conference of the Association for Consumer Research, eds. David M. Gardner, College Park, MD : Association for Consumer Research, Pages: 365-375.

Proceedings of the Second Annual Conference of the Association for Consumer Research, 1971     Pages 365-375

THE RELATIONS AMONG ATTRIBUTE AND IMPORTANCE COMPONENTS OF ROSENBERG-FISHBEIN TYPE ATTITUDE MODEL: AN EMPIRICAL INVESTIGATION

Reza Moinpour, University of Washington

Douglas L. MacLachlan, University of Washington

[Assistant Professors of Marketing, Graduate School of Business Administration, University of Washington, Seattle, Washington 98195.]

Various models of attitude have recently appeared in the marketing literature (e.g. models of Osgood & Tannenbaum, 1955; Rosenberg, 1956; Rokeach & Rothman, 1965; and Fishbein, 1967). The work of Rosenberg, Fishbein and others has indicated strongly that an individual's attitude toward any object is a function of his evaluative beliefs about that object (Fishbein, 1967).

Rosenberg and Fishbein models, in particular, have been used to examine the underlying structure of attitudes and their relations to affect and behavioral intention (brand preference) and behavior (brand choice) . Sheth (1970) investigated the theoretical links among beliefs, affect, behavioral intention and behavior regarding brands of a convenience good product. He found that evaluative beliefs (information about a brand on a set of relevant characteristics), when used separately in a multiple regression analysis, provided good predictors of both affect (general like or dislike of a brand) and behavioral intention (verbal expression of intent to buy the brand); however, they proved to be better predictors of affect than of behavioral intention. In addition, when evaluative beliefs were summed, substantially lower association resulted between attitude and behavior. Sheth and Talarzyk (1970), in their study of the cognitive structure of consumer attitudes, applied the functional relationship of Rosenberg's theory to brand preference and attitude data. The analysis consisted of regressions on the two attitude components (i.e., perceived instrumentality and value importance) which were treated, in a summed form, both by themselves and when multiplied together. They noted that of the two components of the model, the perceived instrumentality factor was a better predictor of preference (a measure of attitude). Moreover, generally lower associations were obtained when the two factors were multiplied (perceived instrumentalities weighted by value importances). Cohen and Houston (1971) examined the composite structure of Rosenberg-Fishbein attitude construct. They evaluated the use of attribute possession and importance scores in the prediction of consumer behavior toward brands by treating the components (by themselves and multiplied together, both separately and in a summed form) as predictor variables in a multiple regression analysis. The results indicated that: (a) attribute importance scores were poor predictors of behavior toward brands, and (b) inclusion of the attribute importance scores did not improve the behavioral prediction of the attribute possession scores.

It is generally suggested, based on Rosenberg-Fishbein approach, that attitude toward a product is a function of the sum of perceived attributes (perceived instrumentality or strength of belief aspect) weighted as to their importance (value importance or evaluative aspect) possessed by that product. However, this multiplicative relationship as well as the extent of contributions by these two components in determining a consumer's attitude toward a product remains an important and unresolved issue in attitude research and the area of consumer behavior. Further investigation of this issue should precede any attempt to test the attitude-behavior relationships as the next linkage along a hierarchical sequence of effects (i.e., cognitive, affective, and conative dimensions).

RESEARCH OBJECTIVE

The objective of this study is to examine the relative importance of the two components of Rosenberg-Fishbein type model in determining a buyer's attitude. The model for this research can be represented as follows:

EQUATION

Where:

"_x = a subject's attitude toward a particular product or brand x

W_i = the importance or weight of attribute i

Bi_x = the product's satisfaction score on attribute i; subject's belief about attribute i for product x

n = the number of product attributes.

Traditionally, it is suggested that a consumer's attitude toward any product is a function of his beliefs about the product in terms of product attributes and the importances of these attributes; the consumer prefers the product toward which he expresses the more favorable attitude. The aim of this research is to show that the buyer's attitude toward a product can be determined mainly from the attribute scores; the importance criteria is not a major contributor. In other words, the saliences (importance or weight) of the various product attributes are inherent in the attribute scores.

The central focus of this study is to examine the underlying cognitive structure of attitudes by evaluating relative contributions of product attribute and importance scores in determining consumer attitude (affect or preference) toward a product. Previous studies in this area have considered only a group as the experimental unit and thus have been limited in their scope of analysis to regression technique. The theory is meant to depict an individual rather than a group construct. Generalizing the individual model across subjects leads to response circumstances that are confounded and difficult to isolate (the response sets are constant within but not across subjects). The individual analysis is conceptually the proper one to use and the high correlation values presented in a later section of this study clearly argue for this conclusion. This research presents a stronger case by: (a) considering both the individual and a group as the unit of observation (Wicker, 1969), (b) considering the components by themselves and multiplied together, both separately and in a summed form, and (c) bringing to bear upon the problem various techniques of correlation, regression, and multidimensional scaling (MDS)

DATA FOR THE EXPERIMENT

The sample space consisted of 40 housewives selected randomly from Columbus, Ohio. The following information for 9 brands of headache and pain remedies were collected:

Section 1. Each housewife was asked to rate each brand on 9 attributes, [The choice of attributes was influenced by a previous self-medication project. See J. F. Engel, D. A. Knapp, & D. E. Knapp, "The Decision-Making Process in Self-Medication and its Relation to Community Health," The Ohio State University. 1967 (mimeographed).] using a 6-point scale ranging from 1--satisfactory to 6--unsatisfactory. The list of brands and attributes appear in Table 1.

Section 2. Each respondent was also asked to evaluate these attributes in terms of importance on a 6-point scale, ranging from 1--important to 6-unimportant.

Section 3. The respondent was asked to rate the 9 brands in terms of preference, using a 10-point scale ranging from 1--most prefer to 10--least prefer.

Section 4. The participant was also asked to rank the same 9 brands in order of preference from 1 to 9.

TABLE 1

LIST OF BRANDS AND CHARACTERISTICS

ANALYSIS AND RESULTS

Correlation Analysis

For the individual analysis, two attitude scores were calculated for each brand for each individual using the following formulations:

EQUATION; attitude as a function of weighted attribute scores

EQUATION; attitude as a function of unweighted attribute scores

Preference rankings were obtained for each respondent (on the basis of the more favorable the attitude the more preferred the brand) from these attitude scores (Talarzyk & Moinpour, 1970). These derived preference rankings were compared to the stated preference rankings via the Spearman rank difference correlation coefficient. As indicated in Table 2, the addition of weights has not improved the results. In fact, the unweighted attribute scores appear to be better determinants of subjects' attitudes toward products. Measures of the sum of weighted attribute scores (suggested by the Rosenberg-Fishbein model) were expected to correlate highly with the stated preference measures. They are both indicants of the same notion, attitude. It is important to note, however, that when weights were deleted, higher correlations between measures of the sum of attribute scores and stated preferences resulted. Sheth and Talarzyk (1970) reached similar conclusions via a different approach (the regression technique); the weights (value importance), in their study, in fact suppressed the predictive power of attribute scores (perceived instrumentality) in determining attitude (preference). Our findings add credence to Sheth and Talarzyk study whose results were suspect because of low values of the coefficients of determination.

TABLE 2

SUMMARY OF CORRELATIONS FOR TWO MODELS

Regression Analysis

For the group analysis, we used two methods: multiple regression and multidimensional scaling.

First, using ordinary least squares multiple regression, we regressed the stated preference rankings for each brand on the weighted and unweighted attribute scores and on the weights themselves for all individuals (n = 40). That is, for each brand we fit the following three models:

EQUATION

Where:

X_ui = the nine unweighted attribute scores

X_wi = the nine weighted attribute scores wi

X_vi = the weights themselves

The adjusted R2 values for these regressions are given in Table 3. (We present the adjusted coefficients of determination to allow comparison with other studies.) For seven of the nine brands, the unweighted attribute scores explained more variation in the preference rankings than did the weighted attribute scores. Additionally2, the unweighted attribute regressions resulted in more significant adjusted R values (in number and size) than the weighted attribute regressions. We infer from these results that the weights tend to dampen the influence of the attribute scores on the preference rankings for the brands. [The same regressions were run with preference ratings as dependent variables rather than preference rankings. These regressions provided similar results. Although their adjusted R2 values were often higher than those given in Table 3, they are less meaningful because of the low dispersion of dependent variable values.] As we anticipated, the last column of Table 3 indicates that the weights themselves explain essentially none of the variation in the preference rankings. We should point out that our regression results for the unweighted case are in agreement with those reported previously by Sheth (1970); the disaggregative use of the attribute scores (perceived instrumentality) enhanced the prediction of attitudes (preference). A direct comparison cannot be made with Cohen and Houston (1971) report due to a difference in the criterion variable (past behavior in their study, preference in ours). However, the implications of both studies are similar regarding the "weighting" hypothesis. That is, the inclusion of weights do not add to the predictive power of the attribute scores.

TABLE 3

SUMMARY OF REGRESSION RESULTS

Multidimensional Scaling Technique

As the next step, inter-brand proximity measures were computed using both weighted and unweighted absolute value distance formula (Green, Maheshwari, & Rao, 1968; Neidell & Teach, 1969; and Moinpour, 1970):

EQUATION

Where:

d_ij = distance between stimulus (brand) i and j

x_ik and x_jk = attribute scores for brands i and j on attribute k

W_k = the weight (importance) of attribute k; W_k = 1 if attributes are weighted equally

n = the number of product attributes

The derived dissimilarities thus obtained were averaged over the forty respondents; they were subsequently rank ordered and submitted to TORSCA-9 (Young, 1968). The configurations for both cases (weighted and unweighted) depict the stimulus objects (nine brands) as points in a Euclidean-space of two dimensions [While the three dimensional solution is more robust, for purposes of illustration, the two dimensional configuration of the MDS solution is decided upon.] (see Figures 1 and 2). Labels of "buffer" and "strength" were assigned to first and second axes respectively. [Factor analysis of the attributes (using BMDO3M General Factor Analysis) pointed out 2 factors which accounted for 81% of total variance. Variables with large factor -loadings on factor 1 were "few side effects" and "easy to take," and those contributing to factor 2 were "speed of relief" and "extra strength." The attribute scores were also correlated with the coordinates of the points (brands) in both configurations. Results of correlation and factor analysis confirmed the labels of "buffer" and strength.] These obtained MDS solutions are expressions of the average respondent's perception of the nine brands of headache and pain remedies. It was assumed that all respondents perceived the brands along a common set of dimensions. The relative position of brands, in these perceptual maps, were determined by their psychological distances, brands clustered together were perceived to be more similar than those far apart. Four clusters are clearly evident in both configurations: Alka Seltzer and Bromoseltzer; Vanquish, Excedrin and Empirin; Anacin and Bufferin; and Bayer and Rexall aspirin. As shown in Table 4, the rankings of the interpoint distances of these configurations corresponded closely to the dissimilarities of the original data. The stress values (for two dimensional space) were judged as good to acceptable.

Both the Spearman correlation coefficients and the stress were generally stronger for the unweighted-attribute case. An examination of the two MDS solutions (weighted and unweighted) revealed that the configuration had remained invariant over both weighted and unweighted attributes (Moinpour, 1970). In particular, both profiles exhibited similar composition with regards to the four previously described clusters. It should be noted, however, that the configuration invariance is partially the result of averaging over subjects. It is therefore concluded that consumers do in fact take into account the saliences of different product characteristics when evaluating various products in terms of these characteristics. In other words, the scale values of the attributes reflect the saliences of these attributes as well.

FIGURE 1

AVERAGE INDIRECT DISSIMILARITY CONFIGURATION - (UNWEIGHTED ATTRIBUTES); STRESS = .096

FIGURE 2

AVERAGE INDIRECT DISSIMILARITY CONFIGURATION - (WEIGHTED ATTRIBUTES): STRESS = .069

TABLE 4

SUMMARY OF MULTIDIMENSIONAL SCALING RESULTS

SUMMARY AND DISCUSSION

The results of individual analysis (correlation) and group analyses (regression and MDS) of this study make it quite convincing that the major component of Rosenberg-Fishbein type model in determining a consumer's attitude toward a brand is the consumer's perception of the brand's possession of attributes. The inclusion of the importance (salience or weight) of product attributes does not improve, and often suppresses, the result. It is surmised that the attribute importance component of the model can best serve as a selection criterion: if an attribute is judged important, it ought to be included and vice versa (on individual basis). This is intuitively logical since for a given individual an unimportant attribute could not have any relevance to determining that consumer's attitude toward a brand.

REFERENCES

Cohen, J. B., & Houston, M. "The Structure of Consumer Attitude: The Use of Attribute Possession and Importance Scores." Mimeographed, University of Illinois, January 1971.

Fishbein, M. "A Behavior Theory Approach to the Relations Between Beliefs About an Object and the Attitude Toward the Object." In M. Fishbein, (Ed.), Readings in Attitude Theory and Measurement. New York: Wiley, 1967, pp. 389-400.

Green, P. E., Maheshwari, A., & Rao, V. "Dimensional Interpretation and Configuration Invariance in Multidimensional Scaling: An Empirical Study." Mimeographed, University of Pennsylvania, 1968.

Moinpour, R. "An Empirical Investigation of Multidimensional Scaling and Unfolding Techniques to Predict Brand Purchasing Behavior." Unpublished Doctoral Dissertation, The Ohio State University, 1970.

Neidell, L. A., & Teach, R. D. "Preference and Perceptual Mapping of a Convenience Good." Paper presented at the meeting of the American Marketing Association, Cincinnati, August 1969.

Osgood, C. E. & Tannenbaum, P. H. "The Principle of Congruity in the Prediction of Attitude Change." Psychological Review, 1955, 62, 42-55.

Rokeach, M., & Rothman, G. "The Principle of Belief Congruence and the Congruity Principle on Models of Cognitive Interpretation." Psychological Review, 1965, 72, 128-42.

Rosenberg, M. J. "Cognitive Structure and Attitudinal Affect." Journal of Abnormal and Social Psychology, 1956, 53, 367-72.

Sheth, J. N. "An Investigation of Relationships Among Evaluative Beliefs, Affect, Behavioral Intention, and Behavior." Mimeographed, University of Illinois, April 1970.

Sheth, J. N., & Talarzyk, W. W. "Relative Contribution of Perceived Instrumentality and Value Importance Components in Determining Attitudes." Paper presented at the meeting of the American Marketing Association, Boston, August 1970.

Talarzyk, W. W., & Moinpour, R. "Comparison of an Attitude Model and Coombsian Unfolding Analysis for the Prediction of Individual Brand Preference." Paper presented at the "Attitude Research and Consumer Behavior" Workshop, University of Illinois, December 1970.

Wicker, A. W. "Attitudes vs. Actions: The Relationship of Verbal and Overt Behavioral Responses to Attitude Objects." Journal of Social Issues, 1969, 25, 41-78.

Young, F. W. "TORSCA-9, An IBM 360/75 Fortran 4 Program for Nonmetric Multidimensional Scaling." Journal of Marketing Research, 1968, 5, 319-21.

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