An Investigation of the Inclusion of the Explicit Ideal Point in the Multi-Attribute Attitude Model

Kenneth E. Miller, University of Utah [Assistant Professor of Marketing.]
ABSTRACT - Empirical investigations of alternative forms of multi-attribute attitude models have consistently found predictions to be higher with exclusion of the measured ideal point (a useful concept in brand positioning). The ability of the multi-attribute attitude model in predicting rank order preference was significantly improved when individually measured ideal point ratings were included selectively on the basis of whether or not the attributes were directional in nature.
[ to cite ]:
Kenneth E. Miller (1976) ,"An Investigation of the Inclusion of the Explicit Ideal Point in the Multi-Attribute Attitude Model", in NA - Advances in Consumer Research Volume 03, eds. Beverlee B. Anderson, Cincinnati, OH : Association for Consumer Research, Pages: 110-113.

Advances in Consumer Research Volume 3, 1976      Pages 110-113

AN INVESTIGATION OF THE INCLUSION OF THE EXPLICIT IDEAL POINT IN THE MULTI-ATTRIBUTE ATTITUDE MODEL

Kenneth E. Miller, University of Utah [Assistant Professor of Marketing.]

[The research reported in this paper was supported in part by the Division of Research, College of Administrative Science, The Ohio State University.]

ABSTRACT -

Empirical investigations of alternative forms of multi-attribute attitude models have consistently found predictions to be higher with exclusion of the measured ideal point (a useful concept in brand positioning). The ability of the multi-attribute attitude model in predicting rank order preference was significantly improved when individually measured ideal point ratings were included selectively on the basis of whether or not the attributes were directional in nature.

INTRODUCTION

Consumer attitude structure models offer a diagnostic tool for measuring brand image, and marketers employing these models can ascertain the position of their brand relative to competing brands. In addition, these models can furnish information concerning the ideal brand location in the cognitive product space as perceived by each consumer. The ideal brand location is that point in the product space which the consumer prefers over all others. With this information, marketers can ascertain the distance their brand is from the ideal brand and take steps to align their product more closely with the perceived ideal. In this way they can position their brand advantageously relative to competing brands.

The multi-attribute attitude model has been widely used by marketers. Wilkie and Pessemier (1973) have discussed it in the following general form:

EQUATION   (1)

where:

"j = attitude toward brand j.

Wi = importance of attribute i.

Bij = perceived amount of attribute i contained by brand j (beliefs).

Ii = ideal amount of attribute i (measured ideal point).

n = number of attributes.

K = parameter of the weighted Minkowski space (determines distance measure to be used in calculating the attitude score. Where K=1, city block distance is used, and where K=2, Euclidean distance measure is used).

However, there is a lack of agreement as to the form of the multi-attribute attitude model which best predicts consumer preference.

This paper examines two issues related to the use of the model. First is the question of whether the explicit ideal point should be included in the multi-attribute model. Previous research by Ginter (1972) and Winter (1972) found that the model performed significantly better in predicting preference when the measured ideal point was not included. The question of whether the measured ideal point should be selectively included in the model has not previously been reported.

The second issue was concerned with the performance of the conjunctive model in predicting rank order preference at the individual level of analysis in an optimization context. The conjunctive model simulates choice strategy by assuming that decision makers reject a brand if its perceived attribute performance is below a minimum cut-off on any attribute. Wright (1975) found that decision makers perceived a conjunctive model as a more likely optimizer than several linear compensatory attitude models.

THE DATA

The data were collected from a mail panel residing in Columbus, Ohio. The initial panel sample of 744 was randomly generated from a list of names in the Columbus telephone directory. [The final panel size was 446 as a result of experimental mortality. The demographic or attitudinal characteristics were not significantly different between the final panel and the drop-outs.] Over a period of three months, each respondent completed five questionnaires concerning eight Columbus fast food restaurant chains. During this period an extensive advertising and couponing campaign was undertaken by one of these fast food restaurants.

Responses were gathered from the mail panel on belief and importance components of the multi-attribute attitude model. Scale values ranged from one (very important or high) to six (very unimportant or low). This information was collected for the following seven attributes:

Taste of the food.

Speed of service.

Popularity with children.

Price.

Variety of menu.

Cleanliness.

Convenience.

Rank order preference was also obtained at each of the five waves of the questionnaire for each individual.

FORMS OF THE MODEL

Inclusion of the Ideal Point

There is considerable disagreement on the concepts, measurements, and analysis in marketing's use of the multi-attribute attitude model (Wilkie and Pessemier (1973)). One of the consistent empirical research findings is that the predictive performance of the model is significantly better when the measured ideal point is not included (Lehmann (1971), Ginter (1974), Winter (1972)). Exclusion of the ideal point results in the rather dubious assumption that more of an attribute is better than a lesser amount of the attribute: i.e., where EQUATION, then, holding W constant, the higher the value of B the higher (or more favorable) will be the attitude toward brand j. This model appears to be lacking where attributes are nondirectional in nature: e.g., variety of menu, price. Consumers do not desire maximum variety of menu or minimum price (as illustrated below). Attributes such as cleanliness and speed of service are directional in nature. Where attributes are non-directional (more of the attribute is not preferred to less), it is probable that the ideal point is not at the end point of the belief scale and exclusion of the ideal point from the model should result in poorer predictive performance of the multi-attribute model.

The measured ideal point model (IM) incorporates the ideal point as rated by the respondent. The distance metric used in this comparison of alternate forms of the multi-attribute model is K = 1, as use of the city block distance has been found to lead to significantly better predictions of consumer preference (Wilkie and Pessemier (1973)). Model IM has the form:

EQUATION   (2)

where: IMi = measured ideal point for attribute i.

The mean response to ideal brand ratings for each attribute at time period 1 are shown in Table 1.

TABLE 1

IDEAL BRAND RATINGS AT TIME PERIOD 1

From Table 1 it is seen that the mean response to each attribute is greater than one (where one is the assumed ideal point in the EQUATION model). The selected ideal point model (IS) incorporates measured ratings of the ideal point only where the attribute appears non-directional. From Table 1 the attributes variety of menu and price are non-directional in nature and the measured ideal points for these attributes, as given by each respondent, were used in Model IS. During the study, a promotional campaign was conducted by one of the fast food brands studied in order to change the popularity of the restaurant. The measured ideal point of popularity with children was also included in the model. For attributes taste of food, cleanliness, service, and convenience, the measured ideal point was not used. Respondents experience difficulty with the concept of an ideal brand (Ginter (1972)) which may lead to the lack of success of models which include the ideal point. However, selective inclusion of the ideal point combined with concurrent measurement of the ideal point and beliefs toward each brand was hypothesized to improve the predictive performance of the model.

For comparative purposes a model which did not incorporate the ideal point was tested. The model used was that found superior [Ginter (1974) compared alternative models to determine whether ideal points and attribute importance weights should be included.] by Ginter (1974). This model (IA) has the form:

EQUATION   (3)

where: IA = assumed ideal point (end point on the belief scale, i.e., 1).

Use of this model yields identical rank order attitude scores to the following model:

EQUATION   (4)

Conjunctive Model

The conjunctive model assumes that decision makers are satisfiers rather than maximizers. Wright (1975) found that decision makers perceive the conjunctive model as a more likely optimizer than several linear compensating attitude models. According to this approach, consumers will choose the brand which meets a satisfactory standard on all attributes. To obtain a rank ordering of brands under this approach, the satisfactory standard for any individual is increased so that brands are evaluated in terms of their minimum performance on each of the attributes. The formulation of this model (used by Pras (1973)) is:

Model CONJ

"j = n|Bij - Ii|max   (5)

where:

|Bij - Ii|max = maximum distance between perceived amount of each attribute and the measured ideal amount of each attribute.

n = number of attributes

THE HYPOTHESES

H1: Model IM does not outperform model IA in the prediction of rank order preference.

H2: Model IS does not outperform models IM and IA.

H3: Model CONJ does not outperform models IM, IS and IA.

THE ANALYSIS

For each respondent, at each of 4 [Rank order preference data was not collected at time period 1.] points in time and for each model, an attitude score for each brand was computed. The attitude scores obtained using each model were rank ordered and then correlated with the rank order preference measure. The mean Spearman rank correlation coefficients across respondents are shown in Table 2.

The Wilcoxon matched signs test was used to test for significant differences in the magnitude of correlations produced by each pair of models. A matrix of significance levels of differences is outlined in Table 3.

TABLE 2

ATTITUDE PREFERENCE CORRELATION

Model IM outperformed model IA in two of the four comparisons of the models (p < .001 and < .08) with no difference found in the other comparisons. This result is inconsistent with that of Ginter (1974). The attributes used by Ginter, with a household cleaning product, were stain removing power, whitening power, sudsiness, mildness to clothes, mildness to skin, and pollution control. It appears that these attributes are more directional in nature than those utilized in this study. Hypothesis 1 was not rejected as significant differences occurred in only one of the four comparisons (p < .05).

TABLE 3

SIGNIFICANCE LEVELS OF INDIVIDUAL CORRELATION DIFFERENCES

Model IS outperformed model IA (p < .001) and model LM (p < .01 and p < .001). Hypothesis 2 was rejected.

The performance of model CONJ was quite poor. In fact, all other models tested outperformed model CONJ (p < .001). Hypothesis 3 was not rejected. Table 2 shows that the mean attitude-preference rank order correlation using this model varied from .588 to .619, whereas the selected ideal point model correlations varied from .745 to .775. Wright found that respondents viewed this model as if it was an optimization model. The conjunctive model used here performed poorly when used as if it was an optimization model.

CONCLUSION

The selected ideal point model outperformed all models evaluated. These were a model excluding the measured ideal point (i.e., the assumed ideal point model found superior in previous research) and the measured ideal point model. For this product category, fast food restaurants, the predictive ability of the multi-attribute attitude model was significantly improved when measured ideal point ratings were selectively included in the model, based on whether the attributes were directional in nature. Ideal point ratings were measured with belief ratings along attributes of the eight brands. The conjunctive model also performed poorly in an optimizing role when used to yield a rank order of brands which was correlated with rank order preference. There is a need to test these models with the criterion measure of probability of choice as use of the conjunctive model leads to the selection of perhaps several acceptable brands rather than a rank ordering of preferred brands.

The aggregate approach to the evaluation of consumer judgment strategies used in this research did not give insight into the distinct an important groups of "minority" consumers who exhibited use of alternative attitude models. For one group of consumers (20 percent of the sample), the conjunctive model yielded the highest correlation with rank order preference. Further research should be conducted using evaluation methods which isolate groups of consumers who use different judgment strategies.

The concept of the ideal point has proved quite useful in the interpretation of attitude structure models: e.g., multidimensional scaling. The concept of the ideal point has major relevance to the marketing manager, and this study has provided a circumstance and a model which used the explicit ideal point measure to increase the predictive performance of the multi-attribute attitude model.

REFERENCES

James R. Bettman, "To Add Importance or Not to Add Importance: That is the Question," Working Paper No. 5, Centre for Marketing Studies, University of California, Los Angeles, October, 1973.

James L. Ginter, "Attitude Change and Choice Behavior in New Product Introduction," Unpublished Doctoral Dissertation, Purdue University, 1972.

James L. Ginter, "An Experimental Investigation of Attitude Change and Choice of a New Brand," Journal of Marketing Research, 11(February, 1974),30-40.

Donald R. Lehmann, "Television Show Preference: Application of a Choice Model," Journal of Marketing Research, 8(February, 1971),47-55.

Bernard Pras, "Predictive Qualities of Linear and Non-linear Evaluation Process Models," Unpublished Doctoral Dissertation, Indiana University, 1973.

Jagdish N. Sheth and W. Wayne Talarzyk, "Perceived Instrumentality and Value Importance as Determinants of Attitudes," Journal of Marketing Research, 9(February, 1972),6-9.

William L. Wilkie and Edgar A. Pessemier, "Issues in Marketing's Use of Multi-Attribute Attitude Models," Journal of Marketing Research, 10(November, 1973),428-41.

William L. Wilkie and Rolf P. Weinreich, "Effects of the Number and Type of Attributes Included in an Attitude Model: More is Not Better," Proceedings, Third Annual Conference, Association for Consumer Research, 1972, 325-40.

Frederick W. Winter, "A Laboratory Experimental Study of the Dynamics of Attitude and Choice Behavior," Unpublished Doctoral Dissertation, Purdue University, 1972.

Peter Wright, "Consumer Choice Strategies: Simplifying vs. Optimizing," Journal of Marketing Research, 12 (February, 1975),60-7.

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