Determinants of Choice Set Size: an Alternative Method For Measuring Evoked Sets

Thomas S. Gruca, University of Massachusetts at Amherst
ABSTRACT - An evoked set consists of those brands which the consumer is aware of and considers for purchase. It is proposed that actual choice data be used to investigate determinants of evoked set size for mature, frequently purchased products in stable markets. The measurement of choice sets is free of the problems of previous research: serial recall interference and limitations on processing capabilities. An empirical test of this assumption using a grocery product found that education and product importance had a positive association with choice set size. The finding with respect to education differs from previous studies and -is consistent the literature in variety seeking behavior.
[ to cite ]:
Thomas S. Gruca (1989) ,"Determinants of Choice Set Size: an Alternative Method For Measuring Evoked Sets", in NA - Advances in Consumer Research Volume 16, eds. Thomas K. Srull, Provo, UT : Association for Consumer Research, Pages: 515-521.

Advances in Consumer Research Volume 16, 1989      Pages 515-521

DETERMINANTS OF CHOICE SET SIZE: AN ALTERNATIVE METHOD FOR MEASURING EVOKED SETS

Thomas S. Gruca, University of Massachusetts at Amherst

ABSTRACT -

An evoked set consists of those brands which the consumer is aware of and considers for purchase. It is proposed that actual choice data be used to investigate determinants of evoked set size for mature, frequently purchased products in stable markets. The measurement of choice sets is free of the problems of previous research: serial recall interference and limitations on processing capabilities. An empirical test of this assumption using a grocery product found that education and product importance had a positive association with choice set size. The finding with respect to education differs from previous studies and -is consistent the literature in variety seeking behavior.

INTRODUCTION

An evoked set is a set of products of which a consumer is aware and which are considered for purchase (Howard and Sheth, 1969). Since the evoked set is a subset of the products available in the marketplace, the evoked set formation is seen as an information load reducing mechanism used by consumers to reduce the cognitive complexity of the brand choice process. The existence of evoked sets has been confirmed in the choice process for nondurables (grocery products - Campbell, 1969), consumer durables (cars - Gronhaug, 1974) and services (retailing - Spiggle and Sewall, 1987).

The demonstrated existence of evoked sets has important implications for marketers. It is assumed that a brand competes only with others in the evoked set of the consumer and not necessarily with all other brands in the market. Therefore, the size of the evoked set and the determinants of its size are important. In this paper, the determinants of evoked set size for a common food product, instant coffee, will be explored.

A number of previous studies have looked at the same problem with other frequently purchased products (Campbell, 1969; Jarvis and Wilcox, 1973; Brown and Wildt, 1987). For example, Campbell (1969) found that brand loyalty and importance of product class affect the size of the evoked set for grocery products (non-durables) whereas there are no effects due to demographic variables such as the age or education of the consumer. On the other hand, in studies of the choice process for consumer durables (Maddox, Gronhaug, Homans and May, 1978) the demographic characteristics of age and education of the purchaser do affect the size of the evoked set. This discrepancy is attributed to differences in the type of product studied (durable vs. non-durable). However, there are some limitations to the methodology used to study grocery products which may affect the validity of the measurement of evoked set size. The primary limitation is due to the use of unaided recall to measure evoked set size.

In this paper, new information about evoked set formation and new data sources will be used in an attempt to reconcile this discrepancy. Specifically, it is proposed that brand choice data be used to measure the size of the evoked set for mature, frequently purchased products in stable markets. In an empirical investigation, scanner panel data is used to determine the demographic determinants of choice set size. This methodology suffers from none of the limitations of previous studies of this type. The results can be used to cross-validate previous research findings and shed some light on the differences found between the Campbell (1969) and Maddox, et al. (1978) studies.

In the next section, an introduction to the research of evoked sets is presented. The past literature concerning the determinants of evoked set size is reviewed and critical discussion of the measurement methodology used in the past follows. A new methodology for measuring evoked set size is proposed and used in an empirical study. The results and discussion sections conclude the paper.

DEFINITION OF EVOKED SET

It is hypothesized that evoked sets form in response to the complex choice problem which arises when a large number of brands must be evaluated before a choice is made. In order to reduce the cognitive load of the choice process, the consumer considers only a subset of brands. As defined by Howard and Sheth (1969), there are two conditions-for inclusion of a product in a consumer's evoked set: 1. the consumer must be aware of the product and 2. there must be some intention to purchase the product. Since there is often a large number of brands available in a product class, consumers may be aware of only a subset of brands. It can also be assumed that of all the brands of which a consumer is aware, she may only consider a small number for purchase.

The awareness of a brand is typically measured by unaided or aided recall (recognition). The evoked set is determined by asking the consumer which brands would she consider for purchase or find acceptable for purchase. Myers (1979) suggests that alternative ways of measuring evoked sets are actually measuring different constructs. He conjectures that some brands which are "acceptable" may be marginal brands which are not actually "considered" in the routine choice process unless the preferred brands are not available. However, Troye (1984) found that there were no significant differences in the evoked sets when elicited with either question.

Although this research is concerned with the size of evoked sets, other research has focused on the formation and composition of evoked sets. One important example is Parkinson and Reilly (1978). Using both recall of evoked sets and rating of brands in the awareness and evoked set, they examined which decision rule best explains inclusion or exclusion from the evoked set. It was found that the unweighted linear compensatory or lexicographic evaluation models both explained their findings well. In the next section, the literature examining the determinants of evoked set size is reviewed.

DETERMINANTS AFFECTING THE SIZE OF EVOKED SETS

The determinants of the size of an evoked set has been a very active research topic. The types of factors studied can be broken down into two groups: 1. characteristics of consumers and 2. characteristics of products. In this paper, I will be concerned with those demographic factors which affect evoked set size. Some of the other characteristics of consumers studied were the ability to discriminate (Ostlund, 1973), self-confidence and familiarity with product class (Gronhaug, 1974). Some of the product characteristics studied in the past include: number of choice criteria (Belonax, 1979; Belonax and Middlestaedt, 1978) and level of abstractness of information processing (May and Homan, 1977). In the next section, I will describe the determinants of evoked set size used in previous research that will also be used in the empirical study.

Factors used in previous research:

The following paragraphs describe possible factors which could affect the size of evoked sets and the rationale behind the choice of these factors.

A. Number of brands that consumer is aware of - It has been shown that there are limitations to the amount of information that can be processed simultaneously in short-term memory (Miller, 1956). This limit is hypothesized to affect the number of products considered as well as the number of choice criteria used. It is expected that as the number of brands a subject is aware of increases beyond a certain limit, the evoked set size (those brands actually considered) will asymptotically reach some upper limit.

B. Importance of product category - The impact of the importance of a product category on evoked set size is based on Social Judgement Theory (Sherif, Sherif and Nebergall, 1965). The theory states that the number of acceptable positions on a social issue is inversely related to the level of involvement of an individual in that issue. Applying this theory to brand choice, with rising importance of a product category, we should observe smaller evoked set sizes. It is equally plausible, however, that increased category importance leads to more search and trial since the product class is so important to the consumer and a mistake would be costly. The importance of the product category may have a positive or negative influence on the evoked set size.

C. Education - May and Homans (1976) found that people who used concrete information processing in the brand choice process had smaller evoked sets. Since well-educated consumers have more experience with abstract thinking, it is assumed that such thinking will occur more frequently with increasing levels of education. The increased probability for abstract thinking will thus lead to larger evoked sets as the level of education increases.

D. Brand loyalty - Campbell (1969) and Ostlund (1973) conjecture that brand loyal customers use more concrete information processing than others. The use of concrete attributes to evaluate alternatives is expected to lead to smaller evoked sets since the range of acceptable brands is constrained by the specificity of the choice attributes. The choice attributes are assumed to be defined or at least affected by the preferred brand.

E. Age - It is assumed that the longer that a consumer has been using a product, that the more time she has had to form stable preference orderings (Howard and Sheth, 1969). This would lead to smaller evoked sets with increasing age. However, time in the market may also teach you that quality does not vary across brands and therefore, price (or another attribute) should guide brand choice.

F. Income - A risk averse consumer should have a smaller evoked set than others since an acceptable brand may be repeatedly purchased in order to avoid a "mistake" by purchasing an unknown, possibly unsatisfactory brand. However, the larger the income of the customer, the lower the risk of making a "mistake" since the unsatisfactory product can be discarded with lower penalty. The higher income allows a consumer to sample a wider variety of brands with, all else equal, less risk.

There are two types of products studied in previous research, durables and non-durables. From Table 1, it is clear that the various demographic variables have differential effect on the evoked set size depending on the type of product studied. For example, both age and income affect the size of the evoked set when the product being considered is a consumer durable (car). However, these variables have no affect on the evoked set size for grocery products like toothpaste and laundry detergent. The constructs being measured in the different studies are not too different since brand loyalty impacts the evoked set size over both types of products. Product importance also affects the evoked set size for both types of products, but in opposite ways.

The question then arises as to the reason for these discrepancies. Are they due to the product types chosen (durables vs. non-durables) or could they be due to measurement errors? In the remainder of the paper, new insights into the evoked set formation process and new data sources will be used to try to answer this question. In the next section, a critical review of the previously used methodology will be presented. A description of the proposed methodology will follow.

TABLE 1

TABLE OF PREVIOUS RESULTS

CRITICAL REVIEW OF PREVIOUS METHODOLOGY

The review will focus on the methodology used in the Campbell (1969) study since its results are the source of the apparent anomaly. In the study, housewives were asked which brands they would consider before making a purchase in the category being studied.

This question would yield a relevant measure of the evoked set if the housewife made up a shopping list at home from which she never deviated once in the grocery store. However, if the researcher wants to know which products were considered in the store, the woman might not have access to that information. As Nisbett and Wilson (1977) point out, consumers may have little access to higher mental processes. This may include the narrowing process from the brands she is aware of and the evoked set. In other words, this question constructs evoked sets from retrospective self-reports without controlling for time since the choice was made or exposure to other stimuli like advertising or point of purchase displays.

Assuming for the moment that the consumer does know which brands were considered, can she accurately recall and report them to the interviewer? Since the housewife is asked to recall the products that she would consider, they must be either recalled serially or as some type of set. If recalled serially, we know that the recall of the first brand can inhibit recall of subsequent brands. Alba and Chattopadhyay (1985) found that in the case of non-cued recall, the items recalled first interfered with the ability to recall subsequent items. The output interference might lead to the reporting of smaller evoked sets than are actually used. In either case of serial or set recall, it has been shown that the short-term memory stores can hold only 7 + 2 chunks of information (Miller, 1956). A number of authors of evoked set studies point out that the limit to the size of evoked sets reported is also about seven. They conclude that this is a further confirmation of Miller's Magic Number. It is assumed that since consumers can only report up to seven alternatives that only about seven are actually evaluated. This is an incomplete observation since the limits on cognitive processing by consumers may also limit the ability to report. Consequently, the consumer may evaluate and consider more than seven alternatives but is unable to recall and report these alternatives.

Since serial interference and the limits on cognitive processing can limit the size of the evoked set reported, it is proposed that other measures be used to determine the size of evoked sets. It is proposed in this paper that actual choice behavior can be directly used to measure the evoked set. The next section will present those conditions under which the choice set can be used as a proxy for the evoked set.

PROPOSED METHODOLOGY

In this study, the actual purchases by a household will be used to measure the size of the evoked set for that household. Although this procedure will not be appropriate for all situations, it can be argued that for the product being studied, instant coffee, there are conditions which make purchase data an excellent indicator of the evoked set. These conditions include a mature product in a stable environment (without product introductions or deletions).

TABLE 2

DETERMINANTS OF CHOICE SET SIZE

Awareness and the Size of Evoked Sets

The first condition for using purchase data concerns the effect of awareness of brands on the size of reported evoked sets. Although Jarvis and Wilcox (1973) found an insignificant relationship, a more recent study (Brown and Wildt, 1987) found a significant relationship between awareness and evoked set size. Furthermore it has been shown that there is situational awareness of a brand based on other brands recalled (Alba and Chattopadhyay, 1985). Since one of the mediating factors of including a brand in an evoked set is the consumer being aware of it, any awareness bias will be removed if consumers are made aware of all brands.

It is assumed that a consumer will be aware of all brands available in the market when brand choices are made in the supermarket. So long as all brands are available (in stock) in every store, the set of alternatives is clearly defined and stable. In this case, the outcome of the choice process is based on stimulus-based processing (Lynch and Srull, 1982) rather than recall-based processing. Recall-based processing is subject to the limitations proposed by Miller (1956) and serial recall interference effects (Alba and Chattopadhyay, 1985) discussed above. Since store brands are not a factor in this market (instant coffee) and no product introductions or deletions occurred in the period of study, the set of brands was stable across time and outlet.

Howard's Theory of Concept Learning and Evoked Sets

The second condition concerns the maturity of the product category. Applying Howard's (1977) theory of Concept Learning, May (1979) conjectured that the composition of the evoked set would change as consumers became familiar with the product category. For a new product category, the consumer has little information on products or the criteria to he used in the choice process. In this Extensive Problem Solving stage, evoked sets are assumed to consist of untried brands only. As the consumer gains experience with the product category through trial of various brands, the choice criteria become more well-defined. In this Limited Problem Solving stage, evoked sets are assumed to consist of both tried and untried brands. Finally, the consumer reaches a stage of Routine Response Behavior in which she has formed a predisposition towards a set of brands. In this stage, the evoked set will consist only of tried brands (as long as there are no new introductions). The prediction of May (1979) with respect to Routine Response Behavior has been confirmed by a study of Canadian beer drinkers by Brisoux and Larroche (1980).

In the case of instant coffee, sales of the product category have been flat or falling in the years previous to the study. It can be safely assumed that instant coffee is a mature product. May (1979) further assumed that in the maturity stage of the product life cycle, most consumers are in the Routine Response Behavior Stage. Therefore, for mature products, the choice set (set of brands purchased over time) will be the same as the evoked set (set of brands considered for purchase). So if we can observe the choices of a consumer over time, we can use this set as the evoked set.

Inert, Inept and Evoked Sets

At this point there can be some question whether the choice set is a true reflection of the evoked set. Since the choice set can include brands which may be subsequently judged as inadequate, it may be a larger set than the actual evoked set. In order to better understand the role of trial and evoked set formation, we turn to the evoked set taxonomy developed by Naryana and Markin (1975).

The entire set of brands available in the marketplace is called the "total set". The set of brands of which the consumer is aware is termed the "awareness set". The awareness set is subdivided into three sets based on the consumers evaluation of the individual brand. A brand for which the consumer has a positive evaluation is considered in the "evoked set". A product-with a negative evaluation is considered in the "inept set". Those products for which the consumer has made no evaluation are considered in the "inert set".

If we view the choice set of the consumer over the entire product life cycle, it consists of products with favorable and unfavorable evaluations (evoked and inert sets). The unfavorable tag however will be decided only after trial. How can we be sure that only favorably evaluated brands are included in the choice set (placing it also the evoked set)?

In the maturity stage of the product life cycle, evoked sets are composed of only tried brands (Brisoux and Larroche, 1980). We can therefore conclude that a brand would not be in the evoked set if it had been found to be unacceptable (and hence in the inept set).

Another objection to the assumption that only acceptable brands will be purchased may be the possibility that brands could move over time from the inept to the evoked set. This possibility is discussed by Naryana and Markin (1975). However, recent work on the formation of evoked sets (Sutton, 1987) has shown that new products are more likely to be included in an established evoked set than previously rejected brands for which new positive information is available. Although a brand may move from the inert set to the evoked set, it is assumed that such "sampling" behavior is most prevalent in the earlier stages of the product life cycle. The choice set of a product in the maturity stage can thus be safely assumed to be the same as the evoked set.

In the next section, the variables to be measured will be discussed. The source of the purchase data and statistical tests to be performed will follow. The subsequent sections will contain the results and a discussion of the results.

VARIABLE DEFINITIONS AND DATA SOURCES

The variables to be used in this study are similar to those used in other studies. Demographic information on age, income, and education are supplied by the households. The importance of the product class will be measured by the total purchases made over the entire time period. It would be preferable to divide the total purchases by income to measure the riskiness of an individual purchase. However, income is reported in ranges which precludes this type of measure.

The database used for this study is the IRI Academic Research Database of coffee purchases. The data consists of all ground and instant coffee purchases over a two-year period by several hundred households. There are two cities represented in the data: Marion, IN and Pittsfield, MA. The data is aggregated across both markets since the set of available products for both products is the same.

Only the instant coffee purchases will be used since it has been suggested that ground and instant coffee serve different uses rather than being used as substitutes (Urban, Johnson and Hauser, 1984). Furthermore, only those households with at least 10 purchases will be included in the study. If data from a longer period of time were available, fewer families would be left out. However, a researcher should observe at least 10 purchases in order to make sure that the entire purchasing pattern has been observed (Blattberg, Buesing and Sen, 1980).

Brand loyalty will be modeled as an index of disloyalty based on the distribution of purchases across the choice set. If a household is totally without loyalty to any brand, it is assumed that the purchasing distribution will be uniform across all brands in the choice set. The index of disloyalty will be the geometric mean of the choice probabilities. This is given by X = [n*(pipi)] + (1/n) where n is the size of the choice set and pi is the probability of choosing choice i. The index i varies from 1 to n. The term (1/n) is introduced to eliminate any bias due to the size of the choice set.

Since much of the demographic data is reported in ordered ranges, the proper statistic to detect association for these variables is Kendall's Tau. The program used to calculate this statistic will be from the BMDP series of statistical software. The other statistic to be reported is the product-moment correlation.

RESULTS

Since the incidence of store brands in these markets is very small, they were eliminated from the analysis. Using a rule of thumb from previous work using scanner panel data, only those brands accounting for at least 1% of total panel purchases are included in the analysis. There were 11 brands which satisfied this criterion: High Point Decaffeinated, Taster's Choice Regular and Decaf, Sanka Regular and Freeze Dried, Nestle Regular and Decaf, Brim Decaf, Maxwell House Regular, Maxim Freeze Dried, and Folger's Regular. The different package sizes are considered to be the same brand. All were available for the entire period of study and there were no new product introductions or product deletions.

There may be some question whether caffeinated and decaffeinated brands should be considered as part of the same market. Without benefit of survey data on uses, I will defer to previous research in market structuring which put both regular and decaf brands in the same submarket (Fraser and Bradford, 1983; Grover and Srinivasan, 1987).

DISCUSSION

The results have shown that product importance is positively related to choice set size which validates the results of the auto studies by Gronhaug and others but differs from the results of Jarvis and Wilcox (1973). This is due to the different ways importance is measured. In the Jarvis and Wilcox (1973) study, product importance was measured relative to other product categories. In the auto studies, product category importance was absolute. Therefore, the results of this study confirm the findings in the consumer durable studies. This is important since it indicates that consumers have different levels of importance for low involvement products which subsequently affects their choice behavior.

However, as measured by the disloyalty index, brand loyalty has no effect on the choice set size. This could be due to the measure used or reporting bias. It is also interesting that education is positively related to choice set size but income is not. The lack of influence of income on choice set size is consistent with the conjecture that risk is not an important factor in the choice of frequently purchased goods. The association with education confirms the observation from the variety seeking behavior literature that the better educated consumer often becomes bored with the same choices and may experiment with new brands (McAlister and Pessemier, 1982). Finally, age had a small but insignificant relationship with choice set size. For this data set, the range of ages was very skewed towards the over 55 group with over 50% of the reported male and female ages falling in that range. This is to be expected based on the product category studied and may have affected the results.

Since the methodology used in this study is free from the reporting and recall biases of previous work, the discrepancy between the results of studies of durable and non-durable goods can be addressed. It seems clear once the results are presented that different choice processes are at work during the different choice tasks. The risk reducing mechanisms of brand loyalty and product experience (represented by age) would be more important in the high involvement purchase of a durable like an auto. The larger search (and trial) implied by the association with product category in this study would be expected when the cost of trial is low. Finally, the association with education can be explained using the variety seeking literature. The differences between the results of the Campbell (1969) and this study can be attributed to what appears to be a reduction in measurement bias due to full product awareness using the measures proposed in this work. The differences between these results and the automobile studies (Gronhaug, 1974; Maddox, et al., 1978) can be attributed to the different choice tasks faced by the consumer.

Although other researchers always call for more research into the process of evoked set formation, the current study indicates that the properties of choice sets are equally interesting. In the product optimization literature, it is assumed that the choice set size is stable with entry. If a new brand is chosen by a consumer, it will displace another from the choice set. The discontinuities introduced by this assumption make the optimization problem virtually intractable (Sudharshan, May and Shocker, 1987) Also, the choice set is assumed to be composed of the brands closest to the ideal point of the consumer in some multi-attribute space spanned by choice attributes. Can this assumption be used to generate joint-space maps from purchase data? These questions will be come more important to the marketer as product concept optimization programs become more widely accepted and scanner panel data become more available.

CONCLUSIONS

It is proposed that actual choice data be used to investigate the determinants of evoked set size. The use of choice sets as evoked sets is free of the problems of previous research, namely serial recall interference and limitations on processing capabilities. In the case of a mature product in a stable market, it was argued that the choice set would approximate the evoked set well. This assumption was used to cross-validate previous research in the grocery product area. Previous results of studies using consumer non-durables differed greatly from those studies involving consumer durables. The demographic variables of age, income and education and derived measures of product importance and brand loyalty were used as determinants of choice set size. It was found that education and product importance had a positive association with choice set size. The finding with respect to education differs from previous studies using grocery products and is supported by work in the area of variety seeking behavior.

REFERENCES

Alba, Joseph W. and Amitava Chattopadhyay (1985), "Effects of Context and Part-Category Cues on Recall of Competing Brands," Journal of Marketing Research, 22 (August), pp. 340-49.

Belonax, Joseph J. (1978), "Decision Rule Uncertainty, Evoked Set Size and Task Difficulty as a Function of Number of Choice Criteria and Information Variability," in Advances in Consumer Research: Vol. 6, editor William L. Wilkie, Association for Consumer Research, pp. 232-35.

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

Blattberg, Robert C., Thomas Buesing and Subrata K. Sen (1980),"Segmentation Strategies For New National Brands," Journal of Marketing, 44 (Fall), 59-78.

Brisoux, J.E. and M. Larrouche (1981), "Evoked Set Size and Composition: An Empirical Investigation Under a Routinized Response Behavior Situation," in Advances in Consumer Research: Vol. 8, editor Kent M. Monroe, Association for Consumer Research, pp. 357-61.

Brown, Juanita J. and Albert R. Wildt (1987), "Factors Influencing Evoked Set Size," in 1987 Proceeding of AMA Summer Educators' Conference, p. 221.

Campbell, Brian M. (1969), 'The Existence and Determinants of Evoked Set in Brand Choice Behavior," unpublished doctoral dissertation, Columbia University.

Fraser, Cynthia and John W. Bradford (1983), "Competitive Market Structure Analysis: Principal Partitioning of Revealed Substitutabilities," Journal of Consumer Research, 10 (June), pp. 15-30.

Gronhaug, Kjell (1973), "Some Factors Influencing the Size of the Buyer's Evoked Set," European Journal of Marketing, 8 (Winter), pp. 23241.

Gronhaug, Kjell and Sigurd V. Troye (1980), "Exploring the Content of Evoked Set in Car Buying," in Marketing in the 90's: Changes and Challenges, editors Bagozzi, et al., Chicago: American Marketing Association, pp. 143-147.

Grover, Rajiv and V. Srinivasan, " A Simultaneous Approach to Market Segmentation and Market Structuring," Journal of Marketing Research, 24 (May), pp. 139-153.

Howard, John A. (1977), Consumer Behavior: Application of Theory, New York: McGraw-Hill.

Howard, John A. and Jagdish H. Sheth (1969), The Theory of Buyer Behavior, New York: John Wiley and Sons, Inc.

Jarvis, Lance P. and Wilcox, James B. (1973), "Evoked Set Size - Some Theoretical Foundations and Empirical Evidence," in Combined Proceedings, editor Thomas V. Greer, Chicago: American Marketing Association, pp. 236-40.

Lynch, John G. Jr. and Thomas K. Srull (1982), "Memory and Attitudinal Factors in Consumer Choice: Concepts and Research Methods," Journal of Consumer Research, 9 (June), pp. 18-37.

Maddox, R. Neil, Kjell Gronhaug, Richard E. Homans, and Fredrick E. May (1978), "Correlates of Information Gathering and Evoked Set Size for New Automobile Purchasers in Norway and the U.S.," in Advances in Consumer Research: Vol. 5, editor H. Keith Hunt, Chicago: Association for Consumer Research, pp. 167-70.

May, Fredrick E. (1978), "Evoked Set Formation and Composition: The Learning and Information Processing Hypotheses," in Advances in Consumer Research: Vol. 6, editor William L. Wilkie, Association for Consumer Research, pp. 222-26.

May, Fredrick E. and Richard E. Homans (1976), "Evoked Set Size and the Level of Information Processing in Product Comprehension and Choice Criteria," in Advances in Consumer Research: Vol 4., editor William D. Perreault Jr., Urbana, Ill.: Association for Consumer Research, pp. 172-175.

McAlister, Leigh and Edgar A. Pessemier (1982), "Variety Seeking Behavior: An Interdisciplinary Review," Journal of Consumer Research, 9 (December), pp. 311-322.

Miller, George A. (1956), 'The Magic Number Seven, Plus or Minus Two: Some Limits on Our Capacity for Processing Information," The Psychological Review, 63 (March), pp. 81-97.

Myers, James H. (1979), "Methodological Issues in Evoked Set Formation and Composition,", in Advances in Consumer Research: Vol 4., editor William D. Perreault Jr., Urbana, Ill.: Association for Consumer Research, pp. 236-37.

Naryana, Chem L. and Rom J. Markin Jr. (1975), "Consumer Behavior and Product Performance: An Alternative Conceptualization," Journal of Marketing, 39 (October), pp. 1-6.

Ostlund, Lyman E. (1973), "Evoked Set Size: Some Empirical Results," in Combined Proceedings, editor Thomas V. Greer, Chicago: American Marketing Association, pp. 226-30.

Parkinson, Thomas L. and Michael Reilly (1979), "An Information Processing Approach to Evoked Set Formulation," in Advances in Consumer Research: Vol. 6, editor William L. Wilkie, Association for Consumer Research,-pp. 227-31.

Sherif, C.W., M. Sherif and R.E. Nebergall (1965), Attitude and Attitude Change, Philadelphia: W.B. Saunders Company

Spiggle, Susan and Murphy A. Sewall (1987), "A Choice Set Model of Retail Selection," Journal of Marketing, 51 (April), pp. 97-111.

Sudharshan, D., Jerrold H. May, and Allan D. Shocker (1987), "A Simulation Comparison of Methods for New Product Location," Marketing Science, 6 (Spring), pp. 182-201.

Sutton, Robert J. (1987), "Using Empirical Data to Investigate the Likelihood of Brands Being Admitted or Readmitted into an Established Evoked Set," Journal of the Academy of Marketing Science, 15 (Fall), pp. 82-89.

Troye, Sigurd V. (1984), "Evoked Set Formation as a Categorization Process," in Proceedings of AMA Summer Educators' Conference, Chicago: American Marketing Association, pp. 180-85.

Urban, G. L., P. L. Johnson and J. R. Hauser (1984), "Testing Competitive Market Structures," Marketing Science, 3 (Spring), 83-112.

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