Number of Choice Alternatives and Number of Product Characteristics As Determinants of the Consumer's Choice of an Evluation Process Strategy

ABSTRACT - Various evaluation process models have been developed to quantitatively represent the mental algebra used by consumers in forming brand preferences. The specific variables underlying the individual's choice of evaluation strategy are not fully understood. This study examines the influence of two situational variables on the consumer's selection of an evaluation process strategy: (a) the number of choice alternatives and (b) the number of product characteristics involved in the evaluation task. Both variables were found to influence evaluation strategy choice. A dichotomy between compensatory additive utility models and noncompensatory lexicographic models emerged as the evaluation task's complexity was-experimentally manipulated.

Citation:

Russell G. Wahlers (1982) ,"Number of Choice Alternatives and Number of Product Characteristics As Determinants of the Consumer's Choice of an Evluation Process Strategy", in NA - Advances in Consumer Research Volume 09, eds. Andrew Mitchell, Ann Abor, MI : Association for Consumer Research, Pages: 544-549.

Advances in Consumer Research Volume 9, 1982      Pages 544-549

NUMBER OF CHOICE ALTERNATIVES AND NUMBER OF PRODUCT CHARACTERISTICS AS DETERMINANTS OF THE CONSUMER'S CHOICE OF AN EVLUATION PROCESS STRATEGY

Russell G. Wahlers, University of Notre Dame

ABSTRACT -

Various evaluation process models have been developed to quantitatively represent the mental algebra used by consumers in forming brand preferences. The specific variables underlying the individual's choice of evaluation strategy are not fully understood. This study examines the influence of two situational variables on the consumer's selection of an evaluation process strategy: (a) the number of choice alternatives and (b) the number of product characteristics involved in the evaluation task. Both variables were found to influence evaluation strategy choice. A dichotomy between compensatory additive utility models and noncompensatory lexicographic models emerged as the evaluation task's complexity was-experimentally manipulated.

INTRODUCTION

A great deal of research in marketing has investigated the manner in which consumers process information in arriving at brand evaluations. The various information integration methods used by consumers are cognitive processes which are neither observable nor directly measurable. Marketers have borrowed and adapted attitude measurement methodology from the various behavioral sciences and have developed a number of evaluation process models which offer alternative quantitative explanations of the mental algebra used by consumers in forming brand preferences. By examining the relative power of these evaluation models to mathematically explain consumers' ultimate preference patterns, researchers have made inferences about the nature of evaluation process behavior underlying the development of preferences.

While various evaluation process models representing alternative explanations of consumers' information processing behavior are found in the literature [See for example: William L. Wilkie and Edgar A. Pessemier, "Issues in Marketing's Use of Multiattribute Models,: Journal of Marketing Research, Vol. 10 (November 1973), pp. 428-41 and James F. Engel, Roger D. Blackwell, and David T. Kollat, Consumer Behavior, 2d ed. (Hinsdale, Ill.: The Dryden Press, 1978), pp. 391-406.], the expectancy-value model has become the dominant framework used by consumer researchers in the 1970's (Engel, Blackwell, and Kollat 1978, p. 39;). The popularity of this model implies a widespread acceptance of the specific information processing approach which the model was developed to operationally represent. Despite its research popularity, however, the universal applicability of the expectancy-value model in explaining consumers' evaluation process behavior across various types of brand evaluation situations has not been clearly demonstrated.

Engel et al (1978, p. 393) suggest that any of a number of evaluation process strategies will be used by the consumer depending on the circumstances of the evaluation task. They further contend that the specific variables underlying the consumer's choice of an evaluation strategy are not yet understood. Few published studies have systematically explored the role of selected situational variables on the consumer's choice of evaluation process approaches. Research by Bettman (1975) and Kakkar (1977) as well as Perreault and Russ (1977), comparing linear compensatory evaluation process models with nonlinear noncompensatory models of the lexicographic type, suggests that the power of lexicographic models may exceed that of linear compensatory models in explaining consumers' information processing strategies as the brand evaluation task becomes increasingly complex. There is reason to conjecture that the consumer's perception of the evaluation task's complexity increases as the number of choice alternatives and/or the number of product characteristics involved in the evaluation task increase.

The research reported here was undertaken to examine the influence of two independent variables on the consumer's selection of an evaluation process strategy used in comparing alternative life insurance policies. The two variables are: (a) the number of choice alternatives and (b) the number of product characteristics involved in the evaluation task confronting the consumer. The investigation attempts to answer two basic questions:

(1) Is the consumer's choice of an evaluation process strategy influenced by the number of choice alternatives and/or the number of product characteristics involved in the evaluation task?

(2) Which type(s) of evaluation process strategy(ies) is (are) most appropriate in explaining the consumer's preference patterns given the different numbers of choice alternatives and product characteristics involved in the evaluation task?

These questions were addressed by examining the evaluation process explanatory power of four types of compensatory additive utility models and two types of noncompensatory lexicographic models as the numbers of choice alternatives and product characteristics involved in the evaluation task were experimentally manipulated.

The remainder of this paper proceeds by describing the six evaluation process models examined in the study. Next the research hypotheses are specified accompanied by a discussion of the methodology. After the analysis and conclusions are presented, the paper concludes with a discussion of the implications for the consumer researcher and the marketing practitioner.

SIX EVALUATION PROCESS MODELS

Six different evaluation process strategies are examined. These six strategies represent potential, alternative explanations of the subject's evaluative behavior. Each of the strategies is operationally represented in quantitative model form. The set of evaluation process models includes four compensatory additive utility models and two noncompensatory lexicographic frameworks:

(1) Expectancy-Value Model (EVM)

(2) Beliefs-Only Model (BOM)

(3) Variability-Weighted Expectancy-Value Model (VWEVM)

(4) Variability-Weighted Beliefs-Only Model (VWBOM)

(5) Strict Lexicographic Model (STLM)

(6) Satisficing Lexicographic Model (SFLM)

This list is not intended to be exhaustive but rather reflect a reasonably representative cross section of the basic multiattribute evaluation process explanations described in the marketing literature. Both additive utility and lexicographic evaluation models are included because of the marked differences between these conceptual approaches. In addition, earlier studies (Russ 1971; Perreault and Russ 1977; and Bettman and Kakkar 1977) have investigated differences among evaluation processes of the types used here.

Basic Expectancy-Value Model

Probably the best known of the compensatory additive utility evaluation process frameworks is the expectancy-value model which embodies the multiattribute attitude measurement approaches originally conceptualized by Fishbein (1963 and Rosenberg (1950). The model assumes that there are multiple evaluative criteria (or product attributes) against which each choice alternative is evaluated. It is further assumed that the consumer assigns varying weights of importance to each of the desired product attributes which are evaluated one at a time. In this evaluation, the consumer formulates perceptions or beliefs about the extent to which each desired attribute is exhibited by each choice alternative. The consumer's overall evaluation of < given choice alternative is based upon how well that item exhibits desired product attributes in addition to how important each of the attributes is to the consumer.

In symbolic notation, the basic expectancy-value model has the form:

EQUATION    (1)

where:

Ej = the consumer's evaluation of choice alternative j

Wi = the importance weight given to product characteristic i

Bij = the consumer's belief as to the extent to which characteristic i is offered by choice alternative i

n = the number of product characteristics used in the evaluation of the choice alternatives

The value of Ej is a measure of the decision maker's attitude toward choice alternative j. Given a set of m choice alternatives (j=l,...,m) the alternatives can be rank ordered in terms of decreasing Ej values to represent an implied preference ranking.

The above model is an additive utility framework in that the utility (desirability) of a given choice alternative i equal to the sum of the utilities of its parts. The frame work, moreover, is compensatory since a choice alternative's weakness on one characteristic may be compensated for by a strength on another characteristic (Green and Win, 1973, pp. 38-46).

Beliefs-Only Model

One of the issues addressed in the literature is whether o not the importance weight Wi adds to the explanatory power of the expectancy-value model (Cohen, Fishbein, and Ahtola 1972; Sheth 1973; and Sheth and Talarzyk 1972). Fishbein and Ajzen (1975) contend that the importance component is of conceptual importance. However, importance does not need to be measured as a separate model component because it is included in the polarity of the consumer's belief evaluation Bij. That is, the consumer's belief evaluation associated with a given attribute will tend to be significantly positive (desirable) or significantly negative (undesirable) only if that attribute has substantial importance to the consumer. Thus, the importance factor is an integral part of the belief evaluation and need not be measured separately. Tversky (1976) has proposed such a framework of the form:

EQUATION    (2)

wherein the concept of a product characteristic's importance to the consumer is considered an integral part of the Bi j belief component.

Variability-Weighted Expectancy-Value Model

Bettman (1974) contends that the inclusion of importance weights enhances the model's explanatory power when the product characteristics are significantly variable in importance to the decision maker. Perreault and Russ (1977) concur and add that a product characteristic may also be more instrumental in the evaluation task i, the belief evaluations Bij along the particular characteristic i are substantially varied across all choice alternatives j=l,...,m. They, therefore, developed an operational measure for each product characteristic's variability such that:

EQUATION    (3)

where:

Vi = the variability of product characteristic i

BiU = the highest belief rating among all choice alternatives against product characteristic i

BiL = the lowest belief rating among all choice alternatives against characteristic i

Bi = the average belief rating of a consumer among all choice alternatives against product characteristic i

This measure of product characteristic variability Vi was incorporated by Perreault and Russ (1977, p. 424) into the basic expectancy-value model resulting in the following variability-weighted expectancy value framework:

EQUATION    (4)

Variability-Weighted Beliefs-Only Model

The influence of product characteristic variability can also be incorporated into the previously described beliefs-only framework resulting in a variability-weighted beliefs-only model of the form:

EQUATION    (5)

This formulation accommodates both Fishbein's position that the product characteristic's importance weights need not be measured separately and Bettman's view that product characteristic importance is influenced by the consumer's polar range of belief evaluations across all choice alternatives being compared.

Strict Lexicographic Model

The strict lexicographic model suggests that in evaluating choice alternatives the consumer perceptually ranks the alternatives based on the extent to which each alternative possesses the single most important product characteristic. If more than one choice alternative exhibit the same utilities along the most heavily weighted characteristic, a preference tie initially exists. In this instance, the alternatives are further evaluated against the next most important characteristic until the tie is broken (Perreault and Russ 1977, p. 424). Strictly speaking, only the single most heavily weighted characteristic is salient in determining preference is there are no perceived ties along that attribute. Product characteristics of lesser importance become salient only as tie breakers. The lexicographic evaluation process is noncompensatory in that a choice alternative's deficiency on an important characteristic cannot be offset by a strength on another characteristic of lesser importance. It is also assumed that separate and independent utility functions exist for each characteristic. Thus the model cannot be mathematically represented in terms of a single utility index.

Symbolically, let Zj denote a given choice alternative where j=l,...,m. Further interpret "> " to mean is preferred to and "~" is indifferent to. Assume that the characteristics are ordered such that i=1 is the most important characteristic, i=2 the second most important characteristic, and so forth for all" characteristics. Then for any two choice alternatives, say x and Y. Zx>Zy implies that there exists an ith characteristic such that biX>biy and for all k<i (i.e., for all characteristics more important than characteristic i) bkX~bky. It follows that if Zx~zy then biX~biy for all i.

Satisficing Lexicographic Model

Simon (1955) has proposed a satisficing evaluation model suggesting that the decision maker often selects the first choice alternative encountered which exceeds some minimum acceptable utility level on each characteristic involved in the evaluation task. Perreault and Russ (1977, p. 426) have proposed an evaluation model which incorporates this satisficing concept with the lexicographic ordering sequence previously discussed. In this satisficing lexicographic model, choice alternatives are ordered lexicographically along each product characteristic taken in order of decreasing importance weights. However, a minimum level of acceptability exists on each characteristic which serves to disqualify choice alternatives with out-of-tolerance evaluation ratings. In effect this framework represents a lexicographically ordered conjunctive model (Kotler 1976. P.903.

The preference ordering rules associated with the strict lexicographic model apply to the satisficing lexicographic model with one exception. Let S=(s1, s2...., sn) represent a vector of minimally satisfactory product characteristic values defined over all i=1,...,n characteristics. Thus for m choice alternatives, Z1, Z2, ... Zm implies that there exists an ith characteristic such that bi1>bi2>...,bim and all ratings bij>si for j=1,...,m. In general, Zj is satisfactory if and only if bij>si where the ith product characteristic is the basis for the ordering sequence.

METHODOLOGY

In the study, the power of the four additive utility models and the two lexicographic models to explain subjects' preference patterns was investigated as the numbers of choice alternatives and product characteristics involved in the evaluation task were systematically manipulated. By investigating the explanatory power of these selected models, inferences are made regarding the evaluation strategies used by the subjects under study.

The Evaluation Task

A 12 x 7 matrix of potential evaluative information was developed involving alternative life insurance policies. The matrix contained information on 7 policy attributes for 12 policy alternatives. The set of policy attributes was based on the results of a pilot study and consisted of:

(1) Policy type (term vs. permanent)

(2) Annual premium ($)

(3) Long-run average annual net cost ($)

(4) 20-year cash value ($)

(5) Dividend policy (participating vs. nonparticipating)

(6) Size/scope of insurance firm's operation (description)

(7) Firm's sale/service methods (description)

Using this information as an information pool, an algorithm was developed to generate questionnaires containing information on a controlled number of life insurance policy alternatives (choice alternatives) and a controlled number of policy attributes (product characteristics). In all, nine separate questionnaire versions were used - each reflecting a unique combination of 4, 8, or 12 policy alternatives and 3, 5, or 7 policy characteristics. Each questionnaire presented evaluative information to the subject in a format similar to the product ratings summaries featured in Consumer Reports magazine.

A judgment sample of 351 undergraduate business students distributed among five Midwestern universities was used in the study. Each subject was given a questionnaire containing information on a controlled number of life insurance policies and a controlled number of policy attributes. A brief discussion of questionnaire terminology was conducted. Subjects were then asked to evaluate the information provided by the questionnaire and rank order the life insurance policies in terms of decreasing preference. In addition to this ranking task, subjects were asked to assign importance weights to the various policy characteristics, rate each policy alternative's characteristics on a desirability scale, and identify a minimum acceptability limit on each policy characteristic contained in the questionnaire.

The Dependent Variable

Using the decision rules associated with the six evaluation process models applied to each subject's choice alternatives-by-product characteristics ratings and importance weights, six hypothetical insurance policy preference rankings were generated. Each of the six model-generated preference rankings represented an alternate explanation for the given subject's evaluation behavior. The preference ranking generated by each model was then correlated with the subject's original preference statement. Spearman's Rho was used as the measure of rank correlation. For each subject, the correlation coefficient computed for each of the six evaluation models expressed the power of the given model to explain the subject's stated preference ranking of choice alternatives. This correlation coefficient associated with each model served as the dependent variable and is subsequently referred to as the given model's explanatory power. For each of the 351 subjects in the study, a row vector of six correlation coefficients was computed representing the explanatory power of each evaluation process model in reconstructing the subject's stated preference ranking.

Hypotheses

The following hypotheses were tested. Both focus on the explanatory power of each evaluation process model as the criterion of interest.

Hypothesis 1: The number of choice alternatives and/or the lumber of product characteristics involved in the evaluation task exert no influence on the explanatory power of :he six evaluation process models under study.

Hypothesis 2: There are no pairwise differences in explanatory power between evaluation process models - given each f the nine evaluation task combinations of controlled numbers of the choice alternatives and product characteristics.

The first hypothesis was intended to test the general significance of the influence of the two independent variables In the explanatory power of the models. The second hypothesis was aimed at providing insight into the relative appropriateness of each model as a possible explanation of he subjects' evaluation process approaches.

ANALYSIS AND RESULTS

The mean explanatory power coefficients for each evaluation process model was computed. These explanatory values are shown in Table 1 broken down by the nine experimental treatment combinations representing the different evaluation tasks. Several interesting explanatory value patterns emerged.

As shown in the lower-right-hand marginal totals cell of Table 1, the model exhibiting the highest degree of overall explanatory power was the strict lexicographic model (.737) while the variability-weighted beliefs-only model (.495) was the poorest performer within the set of models over all 351 respondents. Further, the explanatory power of each of the two lexicographic models was found to exceed the performance of all of the additive utility frameworks. Among the four additive utility models,the expectancy-value model exhibited the highest explanatory power (.591) - but well below that of the lexicographic models. This general pattern of the evaluation models' explanatory power values is consistent with the findings of Perreault and Russ (1977, p. 428) in their earlier study.

Even more interesting, however, was the pattern of explanatory values among the experimental treatment cells in Table 1. For the simplest evaluation task comprised of 4 choice alternatives and 3 product characteristics, each of the additive utility models exhibited better explanatory performance than either of the lexicographic models. However, two patterns emerged in treatment cells characterized by increasing numbers of choice alternatives and/or characteristics. Not only was the explanatory power of each additive utility model found to decrease, but the performance of the two lexicographic models was observed to improve. This suggested that the subjects' choices of evaluation strategies tended to change from the compensatory additive utility frameworks to the noncompensatory lexicographic strategies as the evaluation task was complicated by increasing numbers of choice alternatives and/or characteristics.

Multivariate analysis of variance was used to test the significance of the number of choice alternatives and number of product characteristics effects (Hypothesis 1). The experimental design framework was that shown in Table 1. The 39 x 6 subjects-by-model explanatory values matrix within each treatment cell served as the multivariate criterion. The MANOVA results are summarized in Table 2. Both of the experimental effects were found to be highly significant as well as their interaction. Hence, Hypothesis 1 was rejected. These findings suggested that the respondents' selections of evaluation process strategies were singularly and jointly influenced by the number of choice alternatives and the number of product characteristics involved in the evaluation task.

The second research question addressed the appropriateness of the evaluation process strategies (represented by the models) in explaining the subjects' preference patterns given the different numbers of choice alternatives and product characteristics involved in the evaluation task. Hypothesis 2 posited no pairwise differences in explanatory power values between models within each of the nine experimental treatment cells.

When six models were examined in each treatment cell, difference in explanatory power between models was investigated for each of the fifteen possible model pairs. The mean explanatory power coefficient associated with each model was used as the basis for between model comparisons. The differences between the mean explanatory coefficients of the models representing each possible pair were computes and these differences were tested for significance using the Newman-Keuls procedure. These test results are shown in Table 3. Several interesting patterns emerged from this analysis.

For the simplest evaluation task (4 choice alternatives and 3 product characteristics) the expectancy value model was the best performer in explaining subjects' preference patterns (Table 1). However, based on the Newman-Keuls comparisons (Table 3), no significant differences were found between the expectancy-value model and the other additive utility frameworks with the exception of the variability-weighted beliefs-only model. The strict lexicographic model was found to differ from the expectancy-value framework but not from the other additive utility models.

In the case of the most complex task (12 alternatives and 7 characteristics), the explanatory performance of each lexicographic model significantly exceeded that of each additive utility model. Further, there were no pairwise differences identified between the two lexicographic models nor between any of the combinations of the additive utility frameworks.

For the experimental treatment cells involving moderately complex combinations of alternatives and characteristics, no significant differences were observed between the two lexicographic models, and few differences were found between pairs of the additive utility models.

To summarize the results of the Newman-Keuls tests, few differences existed between pairs of the additive utility models, and almost no differences existed between the lexicographic frameworks across the nine experimental treatment cells. It appeared that a general dichotomy between additive utility and lexicographic models emerged from the analysis. That is, in selecting a model to explain the subjects' evaluation processes, choosing either an additive utility or lexicographic type of model appeared to be more important than selecting a specific variation of either general type. Regarding Hypothesis 2, any pairwise model comparison marked with an "X" in Table 3 indicates a rejection of the null hypothesis of no between model difference.

CONCLUSIONS AND IMPLICATIONS

The study examined the influence of the numbers of choice alternatives and product characteristics involved in the evaluation task on the consumer's selection of an evaluation process strategy. Six different evaluation models were investigated using life insurance as the product class of interest. It was determined that both independent variables were singularly and jointly significant in affecting subjects' evaluation strategies in comparing life insurance policies. For the simplest evaluation task the additive utility models outperformed the lexicographic frameworks. However, as the number of either or both of the situational variables were increased, respondents' evaluation patterns were best explained by either of the two lexicographic models under study. Thus, subjects exhibited an increasing tendency to employ a lexicographic evaluation strategy as the evaluation task became increasingly complex. Further the Newman-Keuls testing of pairwise model differences suggested that the respondents' evaluation strategies could be at best only generally characterized as either a non-compensatory lexicographic process or a compensatory additive utility process.

Consistent with prior research found in the marketing literature, this study supports the notion that consumers are highly adaptive to their task environment. That is to say the subjects of this experiment appeared to adjust their evaluation strategies in response to the nature of the evaluation tasks to which they were subjected. The study's findings also support the contention that consumers may utilize a lexicographic evaluation approach as a simplifying strategy when overloaded with evaluative information.

TABLE 1

MEAN EXPLANATORY POWER OF EVALUATION PROCESS MODELS BY EXPERIMENTAL TREATMENT

TABLE 2

SUMMARY OF MULTIVARIATE ANALYSIS OF VARIANCE FOR NUMBER OF CHOICE ALTERNATIVES AND NUMBER OF PRODUCT CHARACTERISTICS EFFECTS

TABLE 3

BETWEEN MODEL COMPARISONS OF THE MEAN EXPLANATORY POWER

The study has several implications. From a consumer research perspective, more research attention needs to be paid to lexicographic evaluation processes - particularly in modeling the information integration behavior of consumers confronted by highly complex evaluation tasks. From a marketing management perspective, knowledge concerning the manner in which the consumer processes information in making brand evaluations will aid in formulating marketing communication strategies. For simple evaluations in which the consumer may tend to process information one brand at a time across all product characteristics, information should be presented to the consumer on all salient characteristics For highly complex evaluation tasks (i.e., life insurance) rn which the consumer may tend to process information lexicographically, it is critical that the brand be promoted on the basis of the characteristics that are most highly salient. The most heavily weighted product attribute is often the sole determinant of brand choice in lexicographic evaluations.

More research is needed to explore the influence of other variables connected with task complexity on the decision maker's choice of an evaluation strategy. Such task situational variables might include varying degrees of time pressure, product type, ego involvement, and perceived risk inherent in the brand choice problem.

REFERENCES

Bettman, James R. (1977), "To Add Importance or Not to Add Importance: That is the Question," in Advances in Consumer Research. 1 (eds.) Scott Ward and Peter Wright, Urbana, IL: Association for Consumer Research, 291-301.

Bettman, James R. and Kakkar, Pradeep (March 1977), "Effects of Information Presentation Format on Consumer Information Acquisition Strategies," Journal of Consumer Research 3, 233-40.

Cohen, Joel B., Fishbein, Martin, and Ahtola, O. T. (November 1972), "The Nature and Uses of Expectancy-Value Models in Consumer Attitude Research," Journal of Marketing Research. 9, 456-60.

Engel, James F., Blackwell, Roger D., and Kollat, David T. (1978), Consumer Behavior, 2d ed. (Hinsdale, IL: The Dryden Press).

Fishbein, Martin (1963), "An Investigation of the Relationship Between Beliefs About an Object and the Attitude Toward That Object," Human Relations, 16. 233-40.

Fishbein, Martin and Ajzen, Icek (1975), Belief Attitude Intention and Behavior: An Introduction to Theory and Research (Reading, MA: Addison-Wesley).

Green, Paul E., and Wind, Yoram (1973), Multiattribute Decision in Marketing: A Measurement Approach (Hinsdale, IL: The Dryden Press.

Kotler, Philip (1976), Marketing Management: Analysis, Planning, and Control, 3d ed. (Englewood Cliffs, NJ: Prentice-Hall, Inc.).

Perreault, William D. Jr., and Russ, Frederick A. (1977), "Comparing Multi-Attribute Evaluation Process Models, Behavioral Sciences, 22, 423-31.

Rosenberg, Milton J. (1950), "Cognitive Structure and Attitudinal Effect," Journal of Abnormal Psychology, 53, 367-72.

Russ, Frederick A. (1971), "Evaluation Process Models and the Prediction of Preference," in Proceedings of Second Annual Conference (ed.) D. Gardner, Association for Consumer Research, 256-61.

Sheth, Jagdish N. (February 1973), "Brand Profiles from Beliefs and Importance," Journal of Advertising Research, 13, 37-42.

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

Simon, H. A. (1955), "A Behavioral Model of Rational Choice," Quarterly Journal of Economics, 69, 99-118.

Tversky, Amos (1967), "Additivity, Utility, and SubJective Probability," Journal of Mathematical Psychology, 4, 175-202.

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

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