Sigurd Villads Troye, Norwegian School of Economics and Business Administration, Bergen, Norway
ABSTRACT - In the present experimental study evoked set formation is conceptualized as an instance of categorization. The impact of the similarity and number of available alternatives on evoked set size and content is assessed. Generally it was found that subjects tended to evoke similar alternatives. The tendency to erroneously exclude "good" alternatives from the evoked set was weak and was related to number of available options, but not to the similarity of the alternatives.
[ to cite ]:
Sigurd Villads Troye (1984) ,"", in NA - Advances in Consumer Research Volume 11, eds. Thomas C. Kinnear, Provo, UT : Association for Consumer Research, Pages: 180-186.

Advances in Consumer Research Volume 11, 1984      Pages 180-186

EVOKED SET FORMATION AS A CATEGORIZATION PROCESS

Sigurd Villads Troye, Norwegian School of Economics and Business Administration, Bergen, Norway

[The author thanks Robert Spekman for helpful suggestions and comments on an earlier draft. The author also acknowledges the constructive contributions made by members of his Ph.d. committee (Chairman Jagdish N. Sheth).]

ABSTRACT -

In the present experimental study evoked set formation is conceptualized as an instance of categorization. The impact of the similarity and number of available alternatives on evoked set size and content is assessed. Generally it was found that subjects tended to evoke similar alternatives. The tendency to erroneously exclude "good" alternatives from the evoked set was weak and was related to number of available options, but not to the similarity of the alternatives.

INTRODUCTION

It is reasonable that the buyer, faced with the often perplexing multitude and complexity of brands in the market place, will seek to simplify the buying task. One possible strategy of simplifying the decision is for the buyer to limit the span of knowledge and attention to a subset of the available brands. This select subset of brands that become candidates to the purchase decision has been termed the buyer's evoked set (Howard and Sheth 1969). As there appears to be general agreement that brands included in the evoked set have a higher probability of being purchased, it becomes important to know not only how many alternatives the buyer tends to include in the evoked set, but also which alternatives are included. These two issues correspond to the two foci found in the literature on evoked set, one being the search for determinants of the size of evoked set, and the other the search for factors that determine the content of this set.

Evoked set formation is similar to other research areas such as brand loyalty, variety seeking (e.g. McAlister and Pessemier 1982) and categorization (Gutman 1981) in the sense that the concern is with how the consumer relates either concurrently or in successive periods - to a set of products. As the focus is on sets of items rather than individual choice alternatives a potentially useful approach appears to be to relate evoked set formation to properties of the set of available product stimuli such as number and heterogeneity. It is proposed that characteristics of the stimulus set offer interesting perspectives from a public policy, marketing/strategy as well as from a theoretical point of view. Therefore the issue that will be addressed in the present study is:

"To what extent is evoked set size and content affected by the similarity of the set of available alternatives?"

The aspects of evoked set content of concern here are the similarity, and desirability of the evoked alternatives. We will also examine how size is affected by the heterogeneity of the choice set. Size and similarity of the evoked alternatives are of interest from the perspective of the marketer given his concern about the competitive situation in the market place. A situation where a brand X tends to enter evoked sets of similar alternatives is strategically very different from a situation where the competing alternatives are different (Tversky 1972). Finally the desirability aspect of the evoked set members deserves attention as an indicator of the person's ability to make a "rational" selection of evoked set alternatives.

CONCEPTUAL FRAMEWORK AND RESEARCH HYPOTHESES

Although the correspondence among paradigms in studies of decision making and categorization has been noted in the literature (e.g. Lockheed 1980), evoked set formation has to our knowledge never been approached from this perspective. Lockheed (1980) underlines the very close analogy between decision making and categorization when stating that: "... a decision or a choice is a categorization ... A decision is a choice of some alternative just as any other categorization or response" (p. 149). By paraphrasing Lockheed, we assert that evoked set formation is a categorization. To the extent evoked set can be conceived of as the outcome of a categorization process, one should expect its formation to be guided by the same principles that govern categorization processes in general.

A central assumption in models of category formation is that characteristics of the set of stimuli out of which categories are formed, partly determine the categorization process. The focus on properties of stimuli sets as determinants of categorization in this stream of research is contrasted by the predominant use of person characteristics as explanatory variables in the evoked set tradition. Another assertion found in the categorization literature is that motivations and task instructions affect category formation (e.g. Glixman 1965, Bruner & al. 1967, Tversky and Gati 1980, Corbin 1980). This assumption is consistent with findings in the consumer literature with regard to information processing strategy (e.g. Bettman 1979, Johnson and Russo 1981).

The conceptual model upon which the empirical study is based (see Figure 1) reflects the notion that evoked set formation as a categorization process is affected by properties of the stimulus set such as number and similarity of the choice alternatives (arrow 1) and the person's motives for processing information (arrow 2). The present emphasis, however, centers on relation 1 and most of the findings to be reported will deal with the impact of the similarity of the available options.

FIGURE 1

FIGURE SHOWS CONCEPTUAL MODEL

One issue in the categorization literature that parallels the attitude formation issue, is the rule or strategy by which categories are formed (Medin and Smith 1977, Hayes-Roth and Hayes-Roth 1977, L. Smith 1981, Bruner & al. 1967). That is, what mechanism enables the consumer to form his evoked set. Several strategies are conceivable: One feasible strategy that has empirical support (Brisoux and Laroche 1980, Pras and Summers 1975, Park 1976, Lussier and Olshavsky 1979) is to evaluate each alternative separately using some conjunctive cut-off rule. A second method is to compare each alternative with some idealized, prototypical exemplar of the kind of product the buyer has decided is desirable ("prototype model", see Medin and Smith 1977). Third, the person may compare alternatives with specific exemplar(s) deemed to be desirable ("exemplar based model", see Medin and Smith 1977). The last two strategies mentioned involve the learning of concepts and the subsequent classification of items as these models assume that choice alternatives ("transfer items") are compared with existing cognitive representations (either prototypes or specific exemplars). A fourth strategy entails a "free classification (see Smith 1981) of alternatives based on overall similarity or identity on one or more dimensions.

The prototype and exemplar based models assume that object: are classified on the basis of their resemblance to the person's cognitive representation of the "typical" member of the category. In the case of buying decisions, the category becomes the "kind of product I want" and potentia candidates may be entered in the evoked set based on the degree to which they match the "ideal product". Such a process represents a plausible strategy which has several interesting implications. One implication is expressed in the following hypothesis:

H1: "The evoked alternatives - compared to the non-evoked ones - will be more similar to the person's "ideal" alternative".

One interesting issue is whether the similarity of the available alternatives will affect the degree to which the evoked options match the ideal. According to Shugan (1980) the degree of product differentiation may affect the "cost of thinking" and may therefore indirectly affect the ease with which a buyer achieves a match between the products in his evoked set and the ideal. There is limited evidence wit respect to the impact of product differentiation on "decision accuracy" (Jacoby & al. 1974, Staelin and Payne 1976, Malhotra & al. 1982). Findings with regard to the impact of number of choice alternatives are also mixed. Rather than stating specific hypotheses the two following research questions are therefore offered:

Q1: "To what extent does similarity of the available choice alternatives affect the match between the evoked options and the ideal?"

Q2: "To what extent is the tendency to erroneously exclude "good" alternatives from the evoked set affected by the similarity and number of the available options?"

Our next concern is the degree to which the decision maker limits the evoked set to similar options. To the extent pairwise similarity (Jain and Etgar 1975), distance in perceptual maps (Day, Schocker and Srivastava 1979), matching coefficients and attribute correlations (Srivastava 1981) are valid indicants of substitutability, one should expect evoked sets to be restricted to similar alternatives. The notion that evoked set will consist of similar options is indeed consistent with the assertion found in the categorization literature that similar or equivalent items will be classified together (Garner 1978, Anderson 1975, Tversky and Gati 1978, Smith 1981, Smith and Byron 1981, Gutman 1981). Nonetheless, pairwise similarity need not be the basis on which alternatives are included in the evoked set Two alternatives can be equally similar to a third item (for instance the "ideal") without being similar to each other. The degree to which evoked products are similar will probably depend therefore on the extent to which they share the attributes that make them similar to the "ideal product . Thus pairwise similarity may not be critical in order for two or more items to be entered in the evoked set. However, it appears reasonable that the likelihood that one option A becomes a member of the evoked set is higher if another item B - similar to A - is evoked. More formally stated, it is hypothesized that

H2: "The pairwise similarity of the evoked alternatives will be higher than the pairwise similarity of randomlY selected subsets".

One may ask whether the tendency to evoke alternatives that are pairwise more similar than randomly selected options, vary with the similarity of the available alternatives. That is, when the options are either homogeneous or heterogeneous, does the person still try to evoke alternatives that - relatively speaking - are similar among themselves.

Q3: "Do the evoked alternatives tend to be more similar regardless of the homogeneity of the set of alternatives out of which the evoked set is composed?

The research questions advanced so far have all addressed aspects of evoked set content such as similarity and desirability of the evoked alternatives. The issue in the following hypothesis is the magnitude of evoked set. Belonax and Mittelstaedt (1978) expected and found evoked set size to be negatively related to attribute variance. One might question whether size of evoked set in general would be negatively related to product differentiation. However, assuming that alternatives are included in the evoked set due to their similarity to the ideal rather than their pairwise similarity, one should expect homogeneity of the choice set to affect the dispersion of evoked set magnitude. Thus, it appears reasonable that when all alternatives share a large number of dichotomous attributes (i.e. are highly similar) they will all tend to be either highly similar or highly dissimilar to the person's concept of his or her "ideal" brand. Therefore highly similar options will all tend to be acceptable or unacceptable. We can present the third hypothesis:

H3: "Persons exposed to highly similar alternatives will tend to either reject or include them all in the evoked set to a larger extent than persons who can choose between more heterogeneous alternatives. Therefore evoked set size in the case of highly similar alternatives will tend to be either very small or very large".

METHODOLOGY

Outline of Experimental Design

An outline of the experimental design is depicted i Figure 2. The similarity treatment has three levels: 1) low similarity (LoW), where there is a low degree of similarity between the alternatives to which the subjects were exposed, 2) high similarity (HIGH), and 3) segmented alternatives (SEG) where the choice alternatives are clustered into three segments with high within segment similarity and low between segment similarity. The subjects in the SEG-group were either exposed to 12 alternatives or 18 alternatives, while the subjects in the LOW and HIGH group were all exposed to 12 options. Most of the findings to be reported refer to the treatment conditions with only 12 stimulus alternatives.

FIGURE 2

FIGURE GIVES OUTLINE OF EXPERIMENTAL DESIGN IN THE EMPIRICAL STUDY

Population, Sample and Product Class

The sample consisted of students, mostly undergraduate, taking courses in the Summer and Fall semester 1981 from the Department of Business Administration at University at Normal-Bloomington. Each class had subjects assigned to at least three different sets of profiles. Questionnaires were handed out in class and the subjects were asked to fill them out at their convenience and return them at the next class-meeting. 310 questionnaires were administered and 203 usable questionnaires were returned. The findings to be reported are limited to the 170 subjects who were exposed to 12 product stimuli. Subjects were asked to assume that they were in the market for an apartment to rent and that they could choose among the apartments described on the front page of the questionnaire.

The use of student samples has been questioned on several grounds (e.g Ferber 1977, Park and Lessig 1977, Sheth 1970, Rosenthal and Rosnow 1969). As the present study is "theory" rather than "effects oriented" (Calder & al. 1981) we feel that the choice of sample is justified.

There were several reasons for choosing apartments as product stimuli. One advantage is that apartments are easily described in terms of dichotomous attributes which allowed the similarity treatment to conform to the format frequently used in studies of similarity and categorization (e.g. Tversky 1972, Tversky and Gati 1978, Smith 1981). Another argument centers on the concept of brand loyalty. As the evoked set literature demonstrates a relationship between brand loyalty and magnitude of evoked set (e.g. Gronhaug 1973/74, Gronhaug and Troye 1980), it was deemed to be desirable to use products where no established loyalty and preferences existed. Besides, it was felt that apartments are of importance and relevance to the population chosen. Finally, earlier studies on decision making (e.g. Payne 1976) have shown that apartments are feasible product stimuli

Presentation Format, Similarity Treatment and Manipulation Check

The selection of attributes was based on typical information provided in ads for rental apartments in the local newspaper and the list of attributes was revised after interviews with subjects in a pilot test. Apartment information was provided in matrix format with the 17 attributes column-wise and apartments row-wise. Absence of an attribute was indicated by a blank while "YES" indicated presence. Although exposure to the significate product (Howard and Sheth 1969) is important for making the final choice, the information provided may be appropriate for the stage of the buying process that one wanted to simulate in the present study. To familiarize the subjects to the presentation format, they were asked to rate their own residence in terms of the same set of attributes used to describe the stimulus apartments. Such a procedure was quite beneficial as some of the descriptors, like "walking distance" and "nice view" are highly subjective.

To generate the apartment profiles corresponding to the three different similarity groups called for in the experimental design it was decided to use "profile correlation" as a proxy for similarity (Gregson 1975). In the case of dichotomous attributes profile correlation is interchangeable with other measures of similarity such as "matching coefficients" (Srivastava 1981) and "coattribute similarity" (Gregson 1975). To avoid any bias due to systematic preferential dominance it was decided to generate randomly more than one set of profiles for each similarity group. For this purpose the computer was used to create random vectors subject to a variance-covariance matrix given by the researcher. To obtain usable attribute-by-alternative matriCes the following procedure was employed: First one defined variance-covariance matrices corresponding to the three similarity groups, LOW, HIGH, and SEGmented (see Figure 2). For each similarity group three sets of apartment profiles were created. Second, the deviation scores produced by the computer program were used to generate {1, 0} scores by assigning 1 to positive scores and 0 to scores equal to or lower than zero. Third, a correlation analysis was run on the derived dichotomous scores to retain sets with correlations within the desired range. (.55 < r < 1.00 for Highly similar pairs and r < .25 otherwise). There are of course other ways of representing similarity but judging form the manipulation checks made the procedure chosen was quite successful. Several analyses were carried out using alternative measures of similarity and discrimination. All analyses showed that the subjective similarity measures to a significant degree did reflect the similarity treatment.

Measures

Evoked set membership was assessed by 5 different measures. Cronbach's alpha for a scale composed of the five measures was .88 which shows high degree of internal consistency. The results reported here are in terms of the alternatives designated as those the subjects "would consider" or "found acceptable". [The subjects were assigned to two different groups that differed in terms of the instruction task and the ordering of the evoked set measures. All results to be reported here are obtained using the first measure in both groups. Although the actual wording of the first evoked set measure differed between the two groups (i.e. "consider" vs. "acceptable") no significant group differences were observed with respect to the results reported.]

Pairwise similarity of the apartments was measured in the three following ways: 1) Subjects were asked to give 25 (out of 66 possible) pairwise similarity judgments of randomly selected pairs of alternatives to which they were exposed. A seven point scale was employed ranging from 1 ("very much similar") to 7 (not similar at all"). This measure is referred to as PERSIM. The two other measures are based on the degree of "attribute overlap". One of the measures, UNWSIM, does not reflect the varying salience or importance of the attributes. It assumes values ranging from 0, indicating perfect similarity, to n, where" is the number of attributes on which the two alternatives differ. The third measure, WSIM, is determined by the importance or salience of the attributes and reflects the notion that important or highly salient attributes may contribute more to overall similarity than less salient ones (Tversky and Gati 1978). The measure was constructed so that attributes judged to be "very undesirable" (score = 1) and attributes judged to be "very desirable" (score = 7) contributed equally to the similarity index. [EQUATION]

A measure of desirability was constructed by the profile or attribute vector of each apartment with the desirability ratings of the attributes and then summing over all attributes. This corresponds to the compensatory evaluation frequently encountered in the consumer literature. Measuring the desirability by means of a compensatory model does not imply the assumption that such an evaluation algorithm is used by the person when constructing the evoked set. Rather, it merely serves as a convenient instrument to obtain a proxy measure of the overall "goodness" or desirability of the options. By constructing such a measure instead of relying on overall ratings provided by the subjects, we lessen the possibility of "demand characteristics" (Orne 1969). Staelin and Payne (1976) employed a procedure analogous to the one used in this study as a measure of "decision accuracy". The DESIRability index was constructed so that the contribution from each attribute ranged from -3 ("very undesirable") to +3 ("very desirable").

RESULTS

H1 "The evoked alternatives - compared to the non-evoked ones - will be more similar to the person's "ideal" alternative.

Q1 "To what extent does similarity of the available choice alternatives affect the match between the evoked options and the "ideal"?"

Rather than relying on the subjects' description of the ideal apartment, we constructed the ideal exemplar based on the person's desirability ratings of the various attributes listed. Thus for each attribute deemed to be desirable (value 5-7) the "desirable" or "ideal" apartment earned the value of 1 which implied presence of the desired attribute. Similarly, for each undesirable attribute and each attribute to which the subject was indifferent (value 1-4) the "ideal" was given a zero (i.e. absence of attribute). The ideal apartment generated this way was then compared with the evoked and non-evoked apartments in terms of the UNWSIM and WSIM similarity measures to test the first hypothesis and answer the first research question. As the findings in Table 1 show the first hypotheses is supported.

Additional analysis supports our findings that the evoked options matched the ideal better than non-evoked ones. Subjects described also an existing "favorite" department. A similar analysis was done for the self-designated ideal apartment as for the "inferred" ideal. The same pattern was found although the differences between members and nonmembers of the evoked set in terms of their similarity to the ideal was less pronounced.

TABLE 1

TABLE SHOWS SIMILARITY OF MEMBERS AND NON-MEMBERS OF THE EVOKED SET TO THE "IDEAL" ALTERNATIVE

Q2: "To what extent is the tendency to erroneously exclude "good" alternatives form the evoked set affected by the similarity of and number of the available options?"

TO assess the degree to which subjects tended to erroneously exclude "good" alternatives, the following procedure was employed: For each subject the DESIRability score was used to identify the "best" and the "worst" member of the evoked set. The non-members that had DESIRability scores higher than the "best" member of the evoked set was counted. This can be conceived of as a measure of the subject's tendency to commit a "TYPE II" error (i.e. erroneously exclude "good" alternatives from their choice). Similarly, we counted the number of non-included apartment that were poorer than the "worst" evoked apartment. As should be expected (see Table 2) the number of "good" apartments that were excluded is small, and the number of non-included "bad" alternatives is large. The subjects appear to commit very few "TYPE II" errors. This is particularly true among subjects in the Low group.

TABLE 2

TABLE SHOWS NUMBER OF NON-INCLUDED ALTERNATIVES THAT ARE "BETTER" AND "WORSE" THAN ALTERNATIVES THAT ARE INCLUDED IN THE EVOKED SET

The tendency to commit "TYPE II" error does not appear to be affected by the similarity of the available alternatives. However, in the SEG18 group, where number of available apartments was 18, subjects commit TYPE II errors with almost 6 times greater frequency. Contrasting the number of erroneously excluded apartment in the SEG group with the corresponding number in the 12-apartment/groups combined, the difference is significant (T= 1.94 p<.05). Thus the "accuracy" of the evoked set selection only appears to be adversely affected by the number and not by the degree of homogeneity of the choice alternatives. As the number of available alternatives was not allowed to vary in the LOW and HIGH group the possible number-by-similarity interaction effect cannot be assessed.

H2: "The pairwise similarity of the evoked alternatives will be higher than the pairwise similarity of randomly selected subsets".

Q3: "Do the evoked alternatives tend to be more similar regardless of the homogeneity of the set of alternatives out of which the evoked set is composed".

Pairs from the evoked set and randomly selected pairs were compared in terms of all three similarity measures, (i.e. the UNWSIM, WSIM, and the PERSIM measure). The findings given in Table 3 support hypothesis 2. Pairs from the evoked set are significantly more similar than randomly selected pairs. However, the results suggest that pairwise similarity is not a sufficient condition in order for alternatives to be classified in the evoked set. As the results in the HIGH group show, the subjects were able to constrain their evoked set to a small subset of alternatives although this subset was not more similar than what one should expect by chance. Another finding supports the hypothesis that evoked pairs will tend to similar: 34 of the 70 subjects in the SEG group limited their evoked set to only one of three available homogeneous clusters.

TABLE 3

MEAN PAIRWISE SIMILARITY OF EVOKED AND RANDOMLY SELECTED ALTERNATIVES

H3: "Persons exposed to highly similar alternatives will tend to either reject or include them all in the evoked set to a larger extent than persons who can choose between more heterogeneous alternatives. Therefore evoked set size in the case of highly similar alternatives will tend to be either very small or very large."

The third hypothesis was tested by comparing the distribution of the evoked alternatives in the LOW and HIGH similarity group. The distribution in the two groups is shown in the Table below: The results appear to support this hypothesis. That is, when subjects were exposed to highly similar alternatives 8 (or 21% of the subjects in this group) designated only one or zero alternatives in their evoked set, yet the comparable number in the LOW group was only 2 (or 3.5% of the persons in this group). At the other extreme there was one subject in the HIGH group who designated all twelve alternatives as members of their evoked set, and none in the LOW similarity who did the same. Thus, subjects exposed to highly similar alternatives were found to reject or accept all to larger extent than subjects facing more heterogenous options.

TABLE 4

TABLE SHOWS DISTRIBUTION EVOKED SET SIZE IN THE LOW AND THE HIGH GROUP (X2=24.9, P <.005)

DISCUSSION AND CONCLUSIONS

The results indicate that the subjects were quite rational in their evoked set formation: The evoked options tended to match the "ideal", and while the worst options typically were discarded, few good alternatives were erroneously excluded from the evoked set. Thus "TYPE II" errors at this stage cannot account for low decision accuracy in the final choice. Moreover, number rather than similarity of the available choice alternatives seemed to explain variations in the tendency to commit "TYPE II" errors. It is difficult to assess the degree to which these findings are attributable to the choice of subject population and information format. As student subjects tend to be better information processors (e.g. Capon and Burke 1980) and matrix information format may facilitate information processing (see Bettman 1979) it is conceivable that the high degree of "decision accuracy" found across the similarity groups may be particular to the methodological approach chosen.

The results indicate that evoked options tend to be more similar than randomly selected pairs of alternatives. This conclusion does not imply, however, that similarity is the basis on which options are entered in the evoked set. When the available alternatives were all highly homogeneous, subjects were still able to constrain their evoked sets to subsets that did not differ from randomly selected subsets in terms of pairwise similarity. Additional analyses showed that about half of the subjects who were exposed to segmented alternatives, composed their evoked set of alternatives from more than one homogeneous cluster. As reported elsewhere (Troye 1983) subjects with heterogeneous evoked sets were found to differ from those with homogeneous evoked sets in terms of the degree to which they discriminated between alternatives. Subjects with mixed evoked sets were, however, no less accurate in their similarity Judgments than subjects who evoked similar alternatives. Furthermore, they did not erroneously exclude "good" options from the evoked set more frequently than subjects with homogeneous evoked sets, but appeared more willing to include "bad" alternatives.

Additional analyses (Troye 1983) also showed that similarity of the evoked apartments was positively related to the subjects' buying experience (as measured by number of products costing more than US $200 that were purchased over the last 3 years). As the similarity of evoked alternatives appears important from a marketing strategy perspective, (Tversky 1972) future research should focus on the determinants of the homogeneity of evoked set.

The findings indicate that when all available alternatives were highly similar, each had a lower probability of entering the evoked set. Subjects exposed to highly similar apartments had more frequently small evoked sets compared to subjects who could choose between more varied options. Analyses showed (Troye 1983) that the smaller evoked set size in the HIGH group compared to the LOW similarity (mean 3.92 and 4.50 respectively) could not be attributed to the desirability of the available options as non-included alternatives in the HIGH group typically were more desirable (as measured by the DESIR-index).

In addition to the unresolved issues stated above, it is believed that future research should devote attention to issues like the following: What are the characteristics of consumers who have varied evoked sets? How are they different from consumers who evoke similar options? How does the similarity of the evoked set change over the decision process? Does the similarity of the evoked set increase as time of purchase approaches?

APPENDIX

TABLE A1

TABLE SHOWS RESULTS OF ANALYSES OF GROUP MEAN DIFFERENCES USING ONEWAY-ANOVA FOR VARIABLES USED AS MANIPULATION CHECKS

TABLE A2

TABLE SHOWS CORRELATION COEFFICIENTS BETWEEN "OBJECTIVE" MEASURES OF SIMILARITY AND PERCEIVED SIMILARITY. ALL CORRELATIONS ARE SIGNIFICANT BEYOND THE .01 LEVEL

REFERENCES

Anderson, N.H., (1975), "On the Role of Context Effects in Psychophysical Judgment" in Psychological Review, Vol. 82, No. 6, 462-482.

Belonax, J.J. and Mittlestaedt, R.A., (1978), "Evoked Set Size as a Function of Number of Choice Criteria and Information Variability" in Advances in Consumer Research, Vol. 5, editor H.K. Hunt, Association for Consumer Research, 48-51.

Bettman, J.R., (1979), AN INFORMATION PROCESSING THEORY OF CONSUMER CHOICE, Reading, MASS: Addison-Wesley.

Brisoux, J.E. and Larroche, M., (1981), "Evoked Set Formation and Composition: An Empirical Investigation Under a Routinized Response Behavior Situation" in Advances in Consumer Research: Vol. 8, editor Rent M. Monroe, Ann Arbor, MI: Association for Consumer Research. 357-61.

Bruner, J.S., Goodnow, J.J. and Austin, G.A., (1967), A STUDY OF THINKING, John Wiley, New York.

Calder, B.J., Phillips, L.W. and Tybout, A.M., (1981), "Designing Research for Application" in Journal of Consume Research, Vol. 8, Sept.

Capon, N. and Burke, M., (1980), Individual, Product Class and Task-Related Factors in Consumer Information Processing in Journal of Consumer Research, Vol. 7, (Dec.), p. 314-26.

Corbin, R.M., (1980), "Decisions that Might Not Get Made" in Cognitive Process in Choice and Decision Behavior, editor: T.S. Wallsten, Hillsdale, N.J.: Lawrence Erlbaum Assoc. Inc.

Day, G.S., Schocker, A.D. and Srivastava, R.K., (1979), "Consumer-Oriented Approaches to Identifying Product-Markets" in Journal of Marketing, Vol. 43, p. 8-19.

Enis, B., Cox, R. and Stafford, J., (1972), "Students as Subjects in Consumer Behavior Experiments, in Journal of Marketing Research, Vol. 9, 72-84.

Ferber, R., SELECTED ASPECTS OF CONSUMER BEHAVIOR, National Science Foundation, Washington, DC, USA, (1976).

Garner, U.R., (1978), "Aspects of a Stimulus: Features, Dimensions, and Configurations" in Cognition and Categorization, editors: E. Rosch and B.B. Lloyd, Hillsdale, N.J. Lawrence Erlbaum Associates, Publ.

Glixman, A.F., (1965), "Categorizing Behavior as a Function of Meaning Domain" in Journal of Personality and Social Psychology, Vol. 2, No. 3, 370-7.

Gregson, R.A.M., (1975), PSYCHOMETRICS OF SIMILARITY, New York, Academic Press, Inc.

Gronhaug, R., (1973), "Some Factors Influencing the Size of the Buyer's Evoked Set" in European Journal of Marketing , 8 (Winter), 232-41.

Gronhaug, K. and Troye, S.V., (1980), "Exploring the Content of Evoked Set in Marketing in the 90's: Changes and Challenges, editors: Bagozzi & al. Chicago: American Marketing Association.

Gutman, J., (1981), "A Means-End Model for Facilitating Analyses of Product Markets Based on Consumer Judgment" in Advances in Consumer Research: Vol. 8, Editor Kent B. Monroe, Arlington, W. Virginia: Association for Consumer Research, 116-121.

Hayes-Roth, F., Hayes-Roth, B., "Concept Learning and Recognition and Classification of Exemplars" in Journal of Verbal Learning and Learning Behavior, Vol. 16, (3) (1977), 321-38.

Howard, J. and Sheth, J.N., (1969), THE THEORY OF BUYER BEHAVIOR. New York: John Wiley and Sons, Inc.

Jacoby, J. and Speller, D.E. and Kohn, C.A., (1974), "Brand Choice Behavior as a Function, of Information Load" in Journal of Marketing Research, Vol. 11, (June), 63-9.

Jain, A.K. and Etgar, E., (1975), "How to Improve Antitrust Policies with Marketing Research Tools" in Combined Proceedings of the Marketing Association, editor E. M. Magge, Chicago: American Marketing Association, 72-5.

Johnson, E. and Russo, J., (1981), "Product Familiarity and Learning New Information" in Advances Consumer Research: Vol. 8, editor K.B. Monroe, Ann Arbor: Association for Consumer Research, 151-55.

Khera, I. and Benson, J., (1970), "Are Students Really Poor Substitutes for Businessmen in Behavioral Research?" in Journal of Marketing Research, Vol.7, 529-32.

Lockheed, G.R., (1980), "Know, Then Decide" in Cognitive Processes in Choice and Decision Behavior, editor T.S. Wallsten, Hillsdale, N.J.: Lawrence Erlbaum Associates, Inc.

Lussier, D.A. and Olshavsky, R. (1979), "Task Complexity and Contingent Processing in Brand Choice" in Journal of Consumer Research, Vol. 6 (September), 154-65.

Malhotra, N.K., Jain, A.K. and Lagakos, S.W., (1982), "The Information Overload Controversy. An Alternative Viewpoint" in Journal of Marketing, Vol. 42, Spring, p. 27-37.

McAlister, L. and Pessemier, E.A., (1982), "Variety Seeking Behavior: An Interdisciplinary Review in Journal of Consumer Research, Vol. 9, (Dec.), 311-28.

Medin, D.L. and Smith, E.E., (1981), "Strategy and Classification in Journal of Experimental Psychology: Human Perception and Performance, Vol. 7, No. 4, 221-253.

Orne, M.T., (1969), "Demand Characteristics and the Concept of Quasi-controls" in Artifact in Behavioral Research, editors: R. Rosenthal and R.L. Rosnov, New York: Academic Press. Inc.

Park, C.W., (1976), "The Effects of Individual and Situation-Related Factors on Consumer Selection of Judgmental Models" in Journal of Marketing Research, Vol. 13, May, 144-51.

Park, C.W. and Lessig, V.P., (1977), "Students and Housewives: Differences in Susceptibility to Reference Groups in Journal of Consumer Research, Vol. 4, 102-110.

Payne, J. (1976), "Task Complexity and Contingent Processing in Decision Making: An Information Search and Protocol Analysis", Organizational Behavior and Human Performance 16, 366-87.

Pras, B. and Summers, J., (1975), "A Comparison of Linear and Nonlinear Evaluation Process Models" in Journal of Marketing Research, Vol. 12 (August), 276-81.

Rosenthal, R. and Rosnow, R.L., 1969), ARTIFACT IN BEHAVIORAL RESEARCH, Academic Press, USA.

Sheth, J.N., (1970), "Are There Real differences in Dissonance Reduction Behavior between Students and Housewives?" in Journal of Marketing Research, Vol. 7, 243-245.

Shugan, S.M., (1980), "The Cost of Thinking" in Journal of Consumer Behavior, Vol. 7 (September), 99-111.

Smith, L.B., (1981), Importance of the Overall Similarity of Objects for Adults' and Children's Classification in Journal of Experimental Psychology: Human Perception and Performance, Vol. 7, No. 4, 811-24.

Smith, J.D. and Byron, J., (1981), "Individual Differences in the Classification of Stimuli by Dimensions" in Journal or Experimental Psychology: Human Perception and Performance, Vol. 7, No. 5, 1132-45.

Srivastava, R.K., (1981), "Usage-Situational Influences on Perception of Product Markets: Theoretical and Empirical Issues", in Advances in Consumer Research, Vol. 8, editor K.B. Monroe, Ann Arbor: Association for Consumer Research, 106-11

Staelin, R. and Payne, J.W., (19?6), "Studies of the Information Seeking Behavior of Consumers" in COGNITION AND SOCIAL BEHAVIOR, editors: J.S. Carroll and J.W. Payne, Hillsdale, N.J.: Lawrence Erlbaum Associates, Publishers.

Troye, S.V., (1983), EVOKED SET FORMATION FROM THE PERSPECTIVE OF DECISION MAKING AS AN INSTANCE OF CATEGORIZATION, unpublished ph.d. dissertation, University of Illinois at Urbana-Champaign.

Tversky, A., (1972), "Elimination by Aspects: A Theory of Choice" in Psychological Review, 281-99.

Tversky, A. and Gati, I., (1978), "Studies of Similarity" in Cognition and Categorization, editors E. Rosch and B.B. Lloyd, Hillsdale, N.J.: Lawrence Erlbaum Associates, Publishers 81-98.

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