Ideal Point Versus Attribute Models of Brand Preference: a Comparison of Predictive Validity

Bernard Dubois, Centre d'Enseignement Superieur des Affaires, Jouy en Josas, France
[ to cite ]:
Bernard Dubois (1975) ,"Ideal Point Versus Attribute Models of Brand Preference: a Comparison of Predictive Validity", in NA - Advances in Consumer Research Volume 02, eds. Mary Jane Schlinger, Ann Abor, MI : Association for Consumer Research, Pages: 321-334.

Advances in Consumer Research Volume 2, 1975      Pages 321-334

IDEAL POINT VERSUS ATTRIBUTE MODELS OF BRAND PREFERENCE: A COMPARISON OF PREDICTIVE VALIDITY

Bernard Dubois, Centre d'Enseignement Superieur des Affaires, Jouy en Josas, France

[The author expresses his thanks to Professor Charles W. Hofer of Northwestern University and Professor Susan P. Douglas of CESA for their helpful comments on an earlier draft. Financial support from the Foundation Nationale pour l'Enseignement de la Gestion des Entreprises, Paris, and the Graduate School of Management, Northwestern University is gratefully acknowledged.]

[Bernard Dubois is Assistant Professor of Marketing, Centre d'Enseignement Superieur des Affaires (CESA), Jouy en Josas, France.]

This paper reports and discusses the results of a comparative analysis of the predictive power of three classes of brand preference models (ideal point models, attribute models, and integrative models) in relation to an educational service. Results obtained at the group and individual levels demonstrate the superior performance of ideal point models, followed by integrative models. Also, the predictive value of structurally simple models is found to be comparable to that of more complex formulations. Finally, the increase in performance gained by giving explicit consideration to attribute weights is marginal. The paper discusses the conceptual, methodological, and managerial implications attached to these findings.

INTRODUCTION

The ultimate purpose of marketing activity is to influence buying behavior. To induce such behavior, marketers must understand how consumers generate their decisions and the role played by marketing stimuli in the decision-making process. Buying behavior, however, is so complex that the relationships existing between marketing stimuli and purchase activity are unlikely to be simple and direct ; consumers are under the influence of a host of mediating factors or situational variables such as social norms, word-of-mouth, financial status and the like

Confronted with this situation, marketers have often found it useful to turn their attention to various pre-purchase stages, hoping that for these stages it would be somewhat easier to assess and predict the performance of particular marketing actions. In deciding which stage would be most appropriate, they have generally considered two specific but somewhat contradictory criteria : the selected stage should not be too remote from the purchase act so as to preserve the connection with sales, and yet relatively "pure" in order to facilitate the identification of marketing stimulus influence.

Given these requirements, the stage of brand preference, defined as:

An affective stage of the individual which reflects his relative valuation of the alternative brands he considers when contemplating the purchase of a product of service has often been considered an appropriate compromise. Sufficiently close to the purpose act to take into account the competitive nature of the buying situation, it seems remote enough to be largely unaffected by situational factors which inevitably blur the impact of specific marketing stimuli.

As a result, marketers have expressed a special interest in models that would explain and predict consumers brand preferences. Among the various perspectives which have been adopted in the analysis of brand preferences, one of the most significant certainly is the information processing approach. More comprehensive than either subject or object-centered theories (that is, theories which exclusively focus on either consumer or product characteristics as explanatory variables), the information processing approach investigates the interaction existing between the subject and the object in terms of the mechanism whereby consumers evaluate the various alternatives available to them on the basis of a set of criteria reflecting their value orientations.

Although developed in so varied disciplines as attitude research, [For an example of an empirical study which relates attitude theory to brand preference, see Bass and Talarzyk (1969).] decision theory, [A review of some of the contributions made by decision theorists to the analysis of brand preference is available in Russ (1971).] micro-economics, [For a concise review of the contributions of micro-economics to the analysis of consumer behavior, see Nicosia (1966). A more detailed treatment of the underlying postulates is contained in Henderson and Quandt (1958).] and psychometrics, [See for example Coombs (1964).] information processing models of brand preference may be classified into two major categories depending upon their specific postulates regarding the nature of the evaluative process : (1) Ideal point models, which rest on the assumption that individuals develop an ideal representation of the product they consider for purchase and then evaluate existing alternatives on the basis of their overall proximity to the ideal brand ; and (2) attribute models which explicitly consider the desirability attached to each relevant attribute and explain preferences on the basis of the relative standing of alternatives along these attributes. More recently, a third class of models, integrative models, has been proposed as a compromise between the previous two categories. In these models, the ideal brand is decomposed into a set of ideal positions on each of the relevant attributes. Preferences are then understood in terms of attribute-specific deviations from the ideal.

While previous research has investigated the predictive power of each of these models considered separately, [For an analysis of the ideal point model, as applied to brand preferences, see Taylor (1967); for a comprehensive review of studies done on attribute and integrative models on a marketing context, see Pessemier and Wilkie (1972).] and while comparative research has been mostly limited to intra-rather than inter-class models, [Two exceptions however are Russ (1971) and Bass, Pessemier, and Lehmann (1972). Yet, the range of models investigated in these studies is not as comprehensive as the one presented in this research.] no researcher has yet tested on a single data base the relative performances of these three classes of models. Yet such a test is needed if one wants to discover which type of model is for a given type of product the best descriptor of consumers' brand preference structure. The objective of this research is to conduct and interpret such a compative analysis.

METHOD

The relative performances of alternative models can only be assessed in reference to one or several specific criteria. In the present study, brand preference models are compared on the basis of their predictive ability. Although by no means the only criterion (diagnostic power may be important too), it is undoubtedly the most important since few researchers would pay attention to a model with less than acceptable predictive power.

In order to test such power, a control set of preferences was obtained by direct measurement and the degree of convergence existing between these preferences and those predicted by each model was investigated and used as a measure of performance. Comparison was made both at the ordinal and interval levels so as to discriminate between various finer versions of the selected models.

The product class under investigation in this research is the American Master's degree program in the field of business administration. Accordingly, the "consumers" of the product are graduate business students and the various "brands" correspond to the various American graduate schools of business. The choice of this-specific environment was primarily guided by methodological considerations. It was hypothesized that the decision to enroll in a graduate business school was of sufficient perceived importance to facilitate the emergence and therefore measurability of significant brand preferences. A sample of one hundred students, all enrolled in the M.B.A program at Northwestern University were asked to describe their preferences for seven graduate business schools : Harvard, Wharton, Chicago, Northwestern, Stanford, Michigan, and Indiana. These institutions were found to be schools most frequently mentioned in a series of unaided recall tests conducted with the same population. [For further details, see Dubois (1973).] The fact that all respondents belonged to the same institution was not a problem since wide variation in preferences was observed.

Control Preferences

Rank-ordered preferences were obtained from respondents simply by asking them to indicate which school they would have selected first, second,... if admitted to all schools mentioned above. While indirect in nature, this approach to preference measurement is nevertheless adequate given the purposes of the present research. It is certainly true the fact of being enrolled in a specific program influences school preferences ; however, it equally affects school perceptions and images, that is, model input data,and therefore does not "contaminate" the measure of convergence provided enough variance in preferences is available, which was the case in this study. Interval measures of preference were obtained by means of the constant sum method (Meftessel, 1947 ; Comrey, 1950) and processed according to the Torgerson algorithm (1958) to generate an interval scale on which preference scores for each stimulus could be located.

Ideal point models

School preference predictions based on the ideal point model were obtained by means of the nonmetric multidimensional scaling procedure ; respondents were asked to compare similarity triads involving the seven schools listed above plus an eighth hypothetical Ideal. From these data, multidimensional maps were obtained in which the existing schools and the ideal were conjointly represented and in which distances between the ideal and each existing school could be interpreted as preferences.

Three specific types of distances were tested corresponding to different assumptions regarding the shape of the utility curve implicitly used in the derivation of preferences (at the interval level) : simple euclidean distances, which correspond to a linear decrease in utility (Figure 1) ; exponential distances, which correspond to a sharp decrease in utility as soon as one moves away from the ideal (Figure 2) ; and parabolic distances, which correspond to an inverted U-shaped utility curve (Figure 3).

FIGURE 1

EUCLIDEAN FUNCTION

FIGURE 2

EXPONENTIAL FUNCTION

FIGURE 3

PARABOLIC FUNCTION

Although past research has mostly relied upon the euclidean model, several authors (Einhorn and Gonedes, 1971) have recently suggested that other forms of utility-functions may lead to better descriptions of preferential judgments.

Attribute models

Three major types of attribute models were considered in this research : the single-attribute model, the lexicographic model, and the expectancy=value model. The raw data used to test these models consisted of a set of semantic differential scales developed from Osgood's lists (Osgood, Suci, and Tannenbaum, 1957), an analysis of schools' catalogs and advertising literature,and a series of in-depth interviews conducted with Northwestern M. B. A. students. The original list of scales was then factor-analyzed and reduced to twenty attributes which represented the most important and reasonably independent dimensions. The final set is reproduced in Table 1. Accompanying instructions were borrowed from Osgood et al.(1957, 81). The only modification was that, in addition to completing scales for each school including the ideal, each respondent was asked to indicate the relative importance of each semantic attribute (on a four-point scale).

The single-attribute model. Predictions based upon the single attribute model according to which all brands are perceived and evaluated on the basis of a unique dimension, were obtained as follows : for each subject (or the average subject in case of group analysis), a search procedure was initiated to discover the attribute considered the most important (on the basis of weights provided by respondents). The position of the ideal school on this attribute determined which side was the positive one, and the location of the existing schools resulted in the preference structure.

TABLE 1

FINAL SET OF SEMANTIC SCALES

The lexicographic model - In its purest form, this model is a straightforward extension of the single-attribute model. According to the regular lexicographic model, the individual rank orders brands on the basis of their values on the attribute most important to him and, if two or more brands are tied on that dimension, breaks the tie by considering the second most important attribute and additional ones if necessary until a complete set of preferences is generated. Despite its simplicity, the regular lexicographic model has received some support from empirical studies of the consumer decision-making mechanism (Alexis, Haines, and Simon, 1968 ; Clarkson, 1963 ; Bettman, 1969 ; and Russ, 1970). Recently, a more sophisticated version of this model, the semi-order lexicographic structure, has been proposed in the literature and seems to perform better than the regular version (Russ, 1971). In this model, the second most important attribute is considered not only if the values obtained for two or more stimuli on the most important attribute are exactly identical but for all cases in which the differences between these values are considered non-significant or non-noticeable. In the present research, preference predictions were generated from both the regular and semi-order versions of the lexicographic model.

The expectancy-value model. As opposed to the lexicographic model, which is non-compensatory in nature, the expectancy-value model, which seems to enjoy some popularity among marketing researchers, allows low scores on certain attributes to be compensated by high scores on others. This is made possible by its algebraic structure which is of the form:

EQUATION

where

Yj = index of attractiveness attached to brand j

wi = weight attached to attribute i

Bj,i = value of brand j on attribute i.

As applied to the analysis of brand preferences, this model, which finds its roots in the work of attitude researchers such as Rosenberg (1956) and Fishbein (1967), has been found of good predictive value (Bass and Talarzyk, 1969 ; Hansen, 1969). Some dispute (Sheth and Talarzyk, 1972) exists, however, as to whether the two components of the model are equally important or in fact are both needed. In the present study, both unweighted and weighted formulations of the expectancy-value model were tested and compared on the basis of their predictive power. The full set of twenty scales was used in the derivation of preference indexes.

Integrative models

Preference predictions obtained from integrative models were generated as follows : Instead of simply using the ideal school's position on attributes as an indicator of which side was favorable, the specific location of the ideal school was explicitly taken into consideration. The main difference between predictions based on expectancy-value and integrative models was that in the latter deviations from the ideal instead of raw scores were used.

The most general functional form of the integrative model can be represented as follows :

EQUATION

where

Yj = distance from existing brand j to the ideal brand (measure of preference)

wi = weight attached to attribute i

Bj,i = value of brand j on attribute i

Bi = value of the ideal brand on attribute i

r = Minkowski order (r > 1)

n = number of attributes

Thus far, this model has received limited testing, and the only two studies which have analyzed it (Lehmann, 1971 ; Bass, Pessemier, and Lehmann, 1972) have generated disappointing results. Yet, the model has high intuitive appeal. In this study, four types of predictions were obtained from the model depending upon whether weights were or not considered and whether euclidean (r=2) or city-block (r=l) distances were used.

EMPIRICAL FINDINGS

Data will be analyzed both at the group and individual levels. The essential reason for presenting individual results is theoretical : All brand preference models considered in the present study have been proposed to explain individual rather than group choice behavior. Yet, managers of educational programs may express limited interest in data displayed at the level of each student, given that they are not generally in a position to tailor their offering to individual needs. For this reason, results will be presented first at the group level. Of course, the assumption of homogeneity of preference will be made in order to justify the aggregation process. For example, in testing ideal point models, only group ideal points, as derived from aggregated data, will be considered. This assumption will be relaxed when individual results are shown.

Group level results

Tables 2 and 3 exhibit the overall brand preference structure revealed at the group level. It is against the values presented in these tables that Preference model predictions were tested.

TABLE 2

OVERALL BRAND PREFERENCE STRUCTURE : ORDINAL LEVEL

TABLE 3

OVERALL BRAND PREFERENCE STRUCTURE : INTERVAL LEVEL

It should be noted that the results presented in Tables 2 and 3 are perfectly consistent thereby enhancing the validity of the constant-sum method as a preference measurement instrument.

Table 4 exhibits, in summary form, the comparative performances of all structural models considered in the present research. Four main conclusions may be derived from an analysis of this table :

1. The ideal point model performs consistently better than all other formulations. This is especially true for the exponential version of the model which exhibits an impressive predictive power (almost 90 % of variance explained).

2. The supremacy of ideal point formulations is further confirmed by the fact that the city-block version of the integrative model is found to be superior in predictive ability to the expectancy-value model, while the only difference between the two structures is that explicit consideration is given to ideal positions on semantic attributes in the former model.

TABLE 4

COMPARATIVE ANALYSIS OF PREDICTIVE POWER OF SELECTED BRAND PREFERENCE MODELS : AGGREGATE LEVEL

3. Students seem to need only a few dimensions to generate their complete set of preferences. This statement is suggested by the good performance of structurally simple models such as the single-attribute and lexicographic models. Conversely, when many attributes are considered, as in the expectancy-value and integrative models, it seems that a considerable amount of "noise" is added, either due to redundant or irrelevant information (or both). As a result, the predictive power of these models decreases.

4. Finally, weights seem to add little, if anything, to the predictive power of unweighted models.

Individual level results

Table 5 presents, in summary form, the respective performances of the various models tested in this research. Only ordinal correlatiOns are displayed in this table given that, at the group level, ordinal and interval coefficients were found to be highly congruent. Also, only one version of the lexicographic model is considered since little difference was observed between the various versions tested at the group level.

On the basis of the data exhibited in this table, it can be easily seen that the conclusions developed at the aggregate level also hold for individuals. The ideal point model is still found to be the most powerful model, at least from a predictive standpoint, its superiority being even more clearly established than previously. Then, adopting the number of correlation coefficients greater than .90 as a measure of predictive power, one successively finds : the single-attribute and lexicographic models which exhibit fairly similar performances ; the city-block version of the integrative model ; and, finally, the two versions of the expectancy-value model and the remaining formulations of the integrative model. As noted at the group level, it seems that weights do not generate additional predictive power ant that a large number of attributes is somewhat detrimental to overall performance.

TABLE 5

COMPARATIVE ANALYSIS OF PREDICTIVE POWER OF SELECTED BRAND PREFERENCE MODELS : INDIVIDUAL LEVEL

DISCUSSION AND CONCLUSIONS

There are several important conceptual, methodological, and managerial implications attached to the above findings. The superiority of the ideal brand model leads one to conclude that students do form an image of the ideal school they would like to attend and evaluate existing schools on the basis of their proximity to the ideal. Furthermore, their ideal image seems rather precise since even slight deviations strongly affect preferences, as indicated by the superior performance of the exponential formulation over the euclidean and parabolic models. A possible explanation of this phenomenon involves an examination of the nature of the service investigated in this research. The decision to enroll in a graduate business school represents a substantial financial and psychological commitment. As a result, the level of risk involved in such a decision is likely to be high as well as the need for information. For these reasons, students take the time to develop judgment criteria which lead them to 2 conceptualization of the ideal brand. The extent to which the results obtained in this research are generalizable to other similar types of products (of the high risk-high investment type) is unknown but, if confirmed by additional studies, these results would mean that a firm involved in the marketing of such a good would be well advised to analyze in detail the characteristics of the ideal brand, as perceived by its consumers, as well as the perceived characteristics of its own product and then design and implement an appropriate strategy for reducing the existing gap. In so doing, the firm may act at several levels of analysis including that of market segments. In this study, however, it was found (from an analysis of standard deviations attached to semantic profiles) that there was no more intersubject difference regarding the perception of the ideal brand than there was concerning existing schools. An undifferentiated strategy would therefore seem appropriate.

The second finding, that of the good relative performance of simple models,suggests that there is a rather severe limit on the amount of information that students can or wish to process before they establish their preferences. The obvious implication it that school administrators probably need not develop extensive lists of attributes unless they want to analyze specific dimensions (which they may sometimes need as a guide to brand positioning or advertising theme development,for example). What seems more important is to identify those few dimensions (actual and perceived) that really count, and carefully appraise the position of their school on each of these dimensions.

The last finding (limited impact of attribute weights) is in keeping with recent results obtained in the marketing literature (Sheth and Talarzyk, 1972 ; Moinpour and MacLahlan, 1971 ; Churchill, 1972 ; Scott and Bennett, 1971 ; Sheth, 1973). Along with Sheth and Talarzyk (1972), one may argue that the relative importance of the semantic scales is implicitly taken into account in the formulation of belief statements. In other words, a student will not assign an extreme position to a school unless he feels the scale is relatively important. Conversely, for unimportant scales, he will mainly rely upon intermediate positions. Sheth and Talarzyk have supported their augmentation by showing that the belief and weight components were substantially correlated, which violates one of the fundamental assumptions of the expectancy-value model. For practical purposes, it would therefore appear that an explicit consideration of these weights is generally superfluous.

REFERENCES

Alexis, M., Haines, G. H., Jr., and Simon, L. Consumer information processing: the case of women's clothing. In R. L. King (Ed.), Marketing and the new science of planning. Chicago: American Marketing Association Conference Proceedings, 1958.

Bass, F. M., Pessemier, E. A., and Lehmann, D. R. An experimental study of the relationships between attitudes, brand preferences, and choice. Behavioral Science, 1972, 17, 532-41.

Bass, F. M., and Talarzyk, W. W. A study of attitude theory and brand preference. In P. R. McDonald (Ed.), Marketing involvement in societY and the economy. Chicago: American Marketing Association 1969.

Bettman, J. R. Behavioral simulation models in marketing systems. Unpublished doctoral dissertation, Yale University, 1969.

Churchill, G. A. Jr. Linear attitude models: A study of predictive ability. Journal of Marketing Research, 1972, 9, 423-6.

Clarkson, G. P. E. A. A model of the trust investment process. In E. A. Feigenbaum and J. Feldman (Eds.), Computers and thought. New York: McGraw-Hill 1963.

Comrey, A. L. A proposed method for absolute ratio scaling. Psychometrika, 1950, 15, 317-25.

Coombs, C. H. A theory of data. New-York: Wiley and Sons 1964.

Dubois, B. A comparative analysis of brand image and brand preference models and measurement methods. Unpublished doctoral dissertation Northwestern University, 1973.

Einhorn, H. J., and Gonedes, H. J. An exponential discrepancy model for attitude evaluation. Behavioral Science, 1971, 16 (4), 33-9.

Fishbein, M. A. 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 and Sons. 1967. Pp. 230-40.

Hansen, F. Consumer choice behavior: An experimental approach. Journal of Marketing Research, 1969, 6, 436-43.

Henderson, M. J. and Quandt, R. E. Microeconomic theory. New York: McGraw Hill. 1958.

Lehmann, D. R. Television show preference: Application of a choice model. Journal of Marketing Research. 1971, 8, 47-55.

Meftessel, M. A. A proposal for quantitative reporting of comparative Judgments. Journal of Psychology. 1947, 24, 229-35.

Moinpour, R., and MacLachlan, D. L. The relationships among attribute and importance components of Rosenberg-Fishbein type attitude model: An empirical investigation. In D. M. Gardner (Ed.), Proceedings of the second annual conference. College Park, Maryland: Association for Consumer Research, 1971. Pp. 365-75.

Nicosia, F. M. Consumer decision processes. Englewood Cliffs, N. J.: Prentice-Hall. 1966.

Osgood, C., Suci, G. J., and Tannenbaum, P. H. The measurement of meaning Urbana, Ill.: University of Illinois Press, 1957.

Pessemier, E. A., and Wilkie, W. L. Multi-attribute choice theory: A review and analysis. Paper presented at the ACR/AMA workshop on buyer/consumer information processing, University of Chicago, November, 1972.

Rosenberg, M. J. Cognitive structure and attitudinal affect. Journal of Abnormal and Social Psychology, 1956, 53, 367-72.

Russ, F. A. Consumer evaluation of alternative product models. Paper presented at the American Marketing Association Fall Educators Conference, Boston, 1970.

Russ, F. A. Evaluation process models and the prediction of preference. In D. M. Gardner (Ed.), Proceedings of the second annual conference. College Park, Maryland: Association for Consumer Research, 1971. Pp. 256-61.

Scott, J. E. and Bennett, P. D. Cognitive models of attitude structure: "Value importance" is important. Ln F. C. Allvine (Ed.), Combined proceedings. Chicago: American Marketing Association, 1971. Pp 256-61.

Sheth, J. N. Brand profiles from beliefs and importances. Journal of Advertising Research, 1973, 13, 37-43.

Sheth, J. N. and Talarzyk, W. W. Perceived instrumentality and value importance as determinants of attitudes. Journal of Marketing Research, 1972 9. 6-9.

Taylor, J. R. An empirical evaluation of the application of unfolding theory to market segmentation. In R. Moyer (Ed.), Changing marketing systems. Chicago: American Marketing Association, 1967. Pp. 288-96.

Torgerson, W. S. Theory and methods of scaling. New York: Wiley and Sons, 1958.

Wilkie, W. L., and Weinreich, R. P. Effects of the number and type of attributes included in an attitude model: More is not better. In M. Venkatesan (Ed.), Proceedings of the third annual conference. Iowa City, Iowa: Association for Consumer Research, 1972. Pp. 325-40.

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