Similarities and Differences of Generalized Brand Attitudes, Behavioral Intentions, and Reported Behavior

Arch G. Woodside, University of South Carolina, [Associate Professor and Program Director of Marketing]
James D. Clokey, Jos. Schlitz Brewing Company, [Senior Research Manager, Business Research Department]
Joan M. Combes, University of South Carolina, [Candidate for Doctor of Philosophy in Business Administration]
[ to cite ]:
Arch G. Woodside, James D. Clokey, and Joan M. Combes (1975) ,"Similarities and Differences of Generalized Brand Attitudes, Behavioral Intentions, and Reported Behavior", in NA - Advances in Consumer Research Volume 02, eds. Mary Jane Schlinger, Ann Abor, MI : Association for Consumer Research, Pages: 335-344.

Advances in Consumer Research Volume 2, 1975      Pages 335-344

SIMILARITIES AND DIFFERENCES OF GENERALIZED BRAND ATTITUDES, BEHAVIORAL INTENTIONS, AND REPORTED BEHAVIOR

Arch G. Woodside, University of South Carolina, [Associate Professor and Program Director of Marketing]

James D. Clokey, Jos. Schlitz Brewing Company, [Senior Research Manager, Business Research Department]

Joan M. Combes, University of South Carolina, [Candidate for Doctor of Philosophy in Business Administration]

The first and third hypotheses of the study were supported: multi-brand/multi-attribute models more accurately classified consumer reported behavior than single brand/multi-attribute models and consumer beliefs and evaluations more accurately classified consumers' reported behavior than overall attitudes or brand intentions when complex cognitive structures existed for the consumer. The second hypothesis was not supported: normalization of belief and evaluation ratings increase correct classification of consumers' reported behavior versus the use of raw data. The theoretical and management needs for group versus individual analysis are discussed.

Attitude in consumer behavior has been defined in a multi-brand and multidimensional context: a cognitive state that, on a number of dimensions, reflects the extent to which the buyer prefers a brand in terms of the attributes of each brand in his evoked set in relation to other brands in the set (Howard and Sheth, 1969). However, most theoretical and operational definitions of attitude have been formulated for one brand at a time in isolation of other brands (Bhagat, Rajuy and Sheth, 1974; Sheth, 1974; Sheth, 1973; Harrell and Bennett, 1974: Bass and Talarzyk, 1972; Sheth and Talarzyk, 1972). Woodside and Clokey (1974) have formulated a set of multi-brand/multi-attribute (MBMA) models of attitude with the assumption that a consumer forms an attitude toward a brand based on attributes of that brand and attributes of other brands in the same product class and in the consumer's evoked set of brands. Clokey and Woodside (1974) empirically have tested and compared predictive and diagnostic capabilities of MBMA models with previously developed single-brand/ multi-attribute (SBMA) models; the predictive accuracies and explanatory powers of the MBMA models were dramatically greater than the SBMA models.

Bass and Talarzyk (1972), Bass and Wilkie (1973), and Wilkie and Pessemier (1973) have discussed potential difficulties of using cross-sectional analysis for predicting consumer attitudes because of scale level differences in belief and evaluation ratings between consumers and they recommend two methods as means to eliminate these difficulties: adjusting ratings for within-subject variance in responses (normalization) and individual-level analysis to correctly classify consumers by preferred brand. Bass and Wilkie (1973) conclude that "predictions of brand preference ranking for each respondent can be used to judge the predictive validity of cross-sectional analysis as compared to the theoretically preferred method of within-individual analysis. However, within-individual analysis is always the superior method of analysis when relative attitudes are used to predict preferences or choice, since this method does not imply respondent homogeneity, a very strong assumption necessarily implied by cross-sectional analysis."

While normalization of data for within-subject variance in rating attributes or intentions may be useful in increasing the levels of correctly classifying consumers by preferences, the theoretical preference of using within-individual instead of cross-sectional analysis is sometimes doubtful since it is important to recall "that homogenous behavior of subgroups of consumers (market segments) have always been of central concern to marketing scholars and practitioners. Therefore, finding groups that are 'homogenous enough' for cross-sectional analyses may be as important as improving the ways in which individual behavior may be more accurately modeled" (Wilkie and Pessemier, 1973). The very strong assumption implied by all types of cross-sectional analysis is theoretically and especially managerially useful for developing generalizations for subgroups of consumers, defining market segments, and planning marketing strategies, given that the cross-sectional analysis has produced substantially accurate predictions of brand preference or choice.

Sheth (1969) has demonstrated that if a complex cognitive structure exists behind the unidimensional affect-type attitude, then evaluative beliefs are better predictors of behavior as compared to attitude; when the cognitive structure behind the attitude is itself simple, both the evaluative beliefs and attitude predicted behavior to the same extent.

Comparisons of the predictive accuracies of MBMA versus SBMA models, raw versus normalized ratings, and evaluative beliefs versus unidimensional affect type attitude (overall attitude) versus brand intention (BI) for prediction of reported behavior (RB) are examined in the present study. Specifically, the study focuses on the predictive and diagnostic merits of MBMA and SBMA models using raw and normalized ratings for consumers' beliefs that brands possess certain attributes, evaluations of these attributes, overall attitudes toward the brands, intentions to purchase the brands, and reported behavior with using the brands.

The hypotheses of the study are: (1) MBMA models more accurately classify consumer RB than SBMA models, (2) normalized ratings increase correct classification of consumers' RB, and (3) consumer beliefs and evaluations more accurately classify consumers' RB than overall attitudes or brand intentions when complex cognitive structures exists for the consumer.

METHOD

Data for testing the hypotheses were collected on a sample of 325 male household heads in the University of South Carolina's Consumer Panel. Families in the panel were selected on a quota basis of representative demographic characteristics of South Carolina

Beliefs for five brands of beer were collected. Four of these brands were selected because of their significant market share and the fifth because of its recent entry into the market.

Beliefs and evaluations of the product category were collected for 18 attributes. The 18 attributes were selected following factor analyses of data from previous brand perception and taste studies performed by the Jos. Schlitz Brewing Company. List of potential attributes for brand perceptions were developed from previous in-depth interviewing and taste studies of small, informal, consumer groups.

Belief was operationally defined by the following statement made to the respondents and followed by an example:

I would like you to tell me how well you thin'.; each phrase describes each of the brands listed below. Please do this by putting the letter shown for each brand on one of the spaces provided. Even if you haven't tried the brand, from what you may know or think about it, indicate how well you believe the phrase describes the brand.

EXAMPLE

Not for Young People    _:  _:  C:  D:   _:  AE:  B:   For Young People

This would indicate that you believe Brand B is more for young people than other brands. Notice that you can place more than one brand in one space. Please take your time in answering. [Different letters were actually used in study for the specific brands studied.]

Although it seems unlikely that a consumer would actually weight 18 attributes into his decision process, it is likely that different groups of consumers may use different sets of attributes as well as different brands to determine their brand choice. Belief scales were scored from -3 to +3.

Consumer evaluations for each attribute for beer were collected by use of the following scale scored -3 to +3:

I prefer a strong beer.

Strongly Disagree _:_:_:_:_:_:_: Strongly Agree

Overall attitude for each of the five brands of beer was collected using the following scale scored -3 to +3:

In general, I like very much.

Strongly Disagree _:_:_:_:_:_:_: Strongly Agree

Intention to purchase for each of the five brands was collected using the following scale scored -3 to +3:

I will buy some _______ during the next four weeks.

Very Unlikely _:_:_:_:_:_:_: Very Likely

Reported behavior was collected using the following question with 10 brands presented for possible response, plus an "other" category:

Overall, considering all the brands of beer that you drink at home, in someone else's home, in restaurants, bars, or taverns, which one brand do you drink most ("X" only one box below).

A number of different forms of expectancy-value models were examined in the research study with RB as the dependent variable:

EQUATION (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) and (12)

where:

Bij = belief i about brand j: that is, "how well you believe the phrase describes the brand."

ai = the evaluative aspect of Bij, that is, "I prefer a ____ beer."

"j = overall attitude toward brand j that is, "in general, I like J Brand K very much."

BIj = brand intention toward j that is, "I will buy some K during the next four weeks."

RBj = reported behavior toward j; that is, "which one brand to you drink most."

a = coefficient estimated by multivariate analysis, e.g., multiple discriminant analysis.

n = number of attributes

m = number of brands

Belief only expectancy models (1 - 4) were included in the models to be tested since some empirical evidence exists that the use of evaluations or "importances" may not substantially improve the classification accuracies of brand choice (Clokey and Woodside, 1974; Sheth, 1973; Bass and Wilkie, 1973). The use of belief x evaluative models may be more intuitively appealing since consumers may not always implicitly give evaluations when judging beliefs; consequently, the attributes which are salient (large |aij|) may be different for belief versus Bij ai models.

Only disaggregated forms of the expectancy models were formulated since all previous research comparisons of disaggregate with aggregate models have shown the disaggregate to be more accurate in predicting brand choice and richer in explaining what product attributes contribute to the consumers' preference and intention to purchase (Harrell and Bennett, 1974; Bhagat, Raju, and Sheth, 1974; Sheth, 1973).

RESULTS

A total of 308 questionnaires were returned (95% response rate). A total of 92 (30%) of these respondents reported that they did not drink beer. A total of 112 (36%) respondents completed all sections to the questionnaire. Another 104 (34%) returned partially completed questionnaires due to lack of experience with the new brand and low usage in drinking beer. Analysis of the results is based on the 112 respondents completing all sections of the questionnaire.

Data from the study were analyzed by stepwise multiple discriminant analysis (MDA) since the dependent variable (RB) was nominal and the independent variables can be assumed to be interval. This procedure reduced the problem of covariance in the independent variables used in the models tested, i.e., the correlation coefficients between the indePendent variables entered in initial steps (up to 15 steps) of the MDA functions were rarely statistically significant.

A total of 9 consumer groups by brand choice were found among the 112 respondents. Correct classifications for RBj for the 12 models tested are shown in Table 1. The SBMA percentages shown in Table 1 are averages of correct classifications for brand choice groups for Brands K, L, M, and N, the four largest brand groups. MDA was performed for SBMA models for belief data for each of these four brands, i.e., 4 INDA's were run for each SBMA model based on the B scores for each brand and the correct classification for the respective brand groups was recorded. The correct classifications for the MBMA models are across all 9 consumer groups by brand choice. The correct classifications would be slightly higher for the MBMA models shown in Table 1 if only the 4 major consumer groups by brand choice were used to calculate correct classifications. Therefore the results in Table 1 are biased in favor of the SBMA models.

The MBMA models substantially out perform the SBMA models by 10% or greater increases in correct classifications in all cases, e.g., MBMA for equation 7 and 10 variables had 66.7% correct classification of consumers into brand groups while SBES for equation 5 and 10 variables had 51.3%. These results confirm the first hypothesis.

TABLE 1

ANALYSIS OF CORRECT CLASSIFICATIONS OF MODELS TESTED FOR 5 AND 10 VARIABLES

TABLE 2

PERCENT CORRECT CLASSIFICATION BY BRAND GROUP FOR FIVE VARIABLES

However, the discriminant analyses computed in this study used all the data to estimate the discriminant coefficients because of the relatively small sample sizes and the desire to obtain the most stable estimates of the coefficients available. This procedure produces an upward bias in the percentage correctly classified. The modified V2 validation procedure suggested by Frank, Massy, and Morrisson (1965) was applied to the data. The data for each respondent were randomly assigned into the brand groups while the sizes of the groups were kept equal to their actual values. MDA was performed on these data. This procedure was performed twice for each model. The average correct classifications after random assignment were substantially below actual results (14% less or greater) in all comparisons.

The second hypothesis was not supported by the analyses. The normalized models did not substantially increase or decrease the percentages of correct classifications compared with the models using the raw data. Evidently, similar use of the possible scaled responses were used by the respondents for Bij, ai, and BI scores. The normalized model for overall attitude actually produced a decrease in correct classification compared with the raw data model (35.7% versus 50.0% correctly classified, respectively).

The use of BI, raw data, scores produced the highest percentages of correct classification (55.4%) when 5 variables were entered into the MDA functions. To test the third hypothesis, the two consumer brand groups with the most accurate correct classifications were compared with the two brand groups with the least accurate correct classifications for Brands K, L, M, and N for the BI and overall attitude, raw score models versus the MBMA, Bij ai, raw data model (equation 7). Equation 7 should produce more accurate classifications of consumers by brand choice than equations 9 or 11 for those brands classified relatively well. Results are shown in Table 2.

The third hypothesis is confirmed as shown in Table 2. Brand N drinkers was the group most accurately classified by the BI model (100%) and least accurately classified by the MBMA, Bij ai, model (23.8%). Brand K was the most accurately classified by the BI model. Accuracy decreases for the MBMA, Bij ai, model as accuracy increases for the BI model across the four brands. The same finding occurs for the MBMA, Bij ai, model versus the overall attitude model.

The belief only (Bij) versus the belief times evaluation models (Bij ai) were somewhat more accurate in all cases when 10 variables were entered into the MDA functions. The two types of models performed similarly with 5 variables. The greater accuracy found for belief only models has occurred in previous research and suggests the need for comparisons of differences in the saliencies of variables in forming both types of disaggregate models. The multiplicative models may be more useful for diagnostic reasons even though they are somewhat out performed in accuracy by the belief only models.

Examples of the additional diagnostic capabilities of the MBMA models compared with the SBMA attribute models have been provided elsewhere (Clokey and Woodside, 1974). The data shown in Table 3 are examples of the type of information analyzed from MBMA, Bij ai, models (equation 7).

Variable 5 for three different brands (1 of 18 attributes) was the most salient variable in discriminating consumer groups along with variable 18 ( for Brand N) and variable 17 (for Brand L). Note that the highest mean for a particular variable and brand is for the brand drinkers of that brand, e.g., X = 6.59 for brand K drinkers for variable 5 for brand K which is greater than any other mean for brand group K and for variable 5(K). Thus, attributes related to the brand drunk appear to be salient for the consumer of that brand. Other attributes for other brands also appear to be salient, e.g., the standardized beta (b*) for variable 17 for brand L was discriminating not only for brand L drinkers but also for brand N. In fact, among the five variables used in calculating the MDA functions, 17(L) was more salient than all other variables in discriminating the brand N group.

TABLE 3

MEANS (X), STANDARD DEVIATIONS (S), STANDARDIZED BETAS, AND CONFUSION MATRIX FOR MBMA, BIJ AI, MODEL FOR FIVE VARIABLES ENTERED INTO THE MDA FUNCTIONS

The confusion matrix shown in Table 3 presents the correct and incorrect classifications produced by the discriminant functions. The off-diagonal percentages provide indications of degree of associations between brands, e.g., brands K, M, and N appear to be associated with other brands while brand L drinkers are somewhat likely to be classified as brand K or M drinkers. Potential brand switching behavior and reasons for such behavior may be indicated by confusion matrixes and b*.

IMPLICATIONS

Methods of answering important managerial and theoretical questions are suggested by the use of MBMA models of consumer attitudes and brand choice. What attributes of brand X are used by brand X consumers in choosing brand X? What attributes of brand Y are used by brand X consumers in choosing brand X? What attributes of brand X are used by brand Y consumers in choosing brand Y? What attributes of brand Y are used by consumers of brand Y in choosing brand Y? What brand consumer groups used attributes of brand X in choosing their brand? What brand consumer groups do not use attributes of brand X in choosing their brand? MBMA models provide an approach to defining competing brands for a particular brand based upon consumer beliefs and evaluations, as well as the attributes salient for specific brands among such consumers.

Normalization may not necessarily increase classification accuracies or the variation explained in data. However, the question of whether or not the use of raw data suppresses the significance of the results of analysis should be answered. Some consumers (e.g., light users) may tend to systematically respond to scaled items differently than others. Further research on this subject is needed.

The need for obtaining consumers' beliefs and evaluations about a brand as well as overall attitude and intention has been supported; such beliefs and evaluations may be especially relevant when complex cognitive structures exist for the consumer in his brand choice.

REFERENCES

Bass, F. M. & Wilkie, W. L. A comparative analysis of attitude predictions of brand preference. Journal of Marketing Research, 1973, 19, 262-269.

Bass, F. M. & Talarzyk, W. W. An attitude model for the study of brand preferences. Journal of Marketing Research, 1972, 9, 93-96.

Bhagat, R. S., Raju, P. S. & Sheth, J. N. The prediction of consumer buying intentions: a comparative study of the predictive efficacy of two attitudinal models. Paper presented at the 82nd Annual Convention of the American Psychological Association, August - September 1974.

Clokey, J. D. & Woodside, A. G. Multi-brand/multi-attribute versus single-brand/multi-attribute attitude models: Some results. Combined Proceedings, American Marketing Association, 1974, forthcoming.

Frank, R. E., Massy, W. F., & Morrisson, D. G. Bias in multiple discriminate analysis. Journal of Marketing Research, 1965, 2, 250-258.

Harrell, G. D. & Bennett, P. D. An evaluation of the expectancy value model of attitude measurement for physician prescribing behavior. Journal of Marketing Research, 1974, 11, 269-278.

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

Sheth, J. N. Attitude as a function of evaluative beliefs. PaPer presented at the American Marketing Association Consumer Workshop, Columbus, Ohio, August 1969.

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

Sheth, J. N. & Talarzyk, W. W. Perceived instrumentality and value importance as determinants of attitudes. Journal of Marketing Research, 1972, 9, 462-465.

Wilkie, W. C. & Pessemier, E. A. Issues in marketing's use of multi-attribute attitude models. Journal of Marketing Research, 1973, 10, 428-441.

Woodside, A. G. & Clokey, J. D. Multi-attribute/multi-brand attitude model for the study of consumer brand choice. Journal of Advertising Research, 1974, 14 (5), 33-40.

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