The Evaluation of Consumers' Product Assortments

Citation:

Erica van Herpen and Rik Pieters (1999) ,"The Evaluation of Consumers' Product Assortments", in E - European Advances in Consumer Research Volume 4, eds. Bernard Dubois, Tina M. Lowrey, and L. J. Shrum, Marc Vanhuele, Provo, UT : Association for Consumer Research, Pages: 89-96.

European Advances in Consumer Research Volume 4, 1999      Pages 89-96

THE EVALUATION OF CONSUMERS' PRODUCT ASSORTMENTS

Erica van Herpen, Tilburg University, The Netherlands

Rik Pieters, Tilburg University, The Netherlands

[We are thankful to Hoogenbosch Retail Group, E.I.M. and NIPO for their help in the data collection and for many stimulating discussions.]

SUMMARY

Consumers' product assortments are assortments of substitute products, such as c.d.'s, shoes or hats, that are owned by consumers themselves. Consumers can evaluate these assortments of products, and this study examines assortment evaluation by introducing assortment properties. Besides item evaluation, which has been used in previous research, size and variety evaluations can impact on assortment evaluation. Size of subgroups based on important attributes appears to be important as well. The properties do not directly affect evaluations, but are compared with individual ideal points.

INTRODUCTION

Consumers own various assortments of products, i.e. sets of items that satisfy a similar desire. Examples of such product assortments are sets of c.d.'s, books, trousers, shirts, or earrings. Products in an assortment have the same usage goal, but are imperfect substitutes of each other (Walsh 1995). In other words, they are distinct alternatives from the same product category. Interest in this area stems from an expected effect that products owned by a consumer have on his/her subsequent buying decisions, a relation that was already proposed by Alderson (1965). Green, Wind and Jain (1972) also argue that: "...The purchase of many products is conditioned, to some extent, by . what products she [the purchaser] currently has in inventory.".

An assortment is a specific type of product set. A set is any grouping of products or items, and an assortment is a product set in which the items come from the same product category. Although consumers' product assortments are claimed to be important for understanding consumer behavior, they have rarely been studied. There are some related but quite different areas of study, such as stockpiling behavior. Both stocks and assortments are sets of products from the same product category that are owned by a consumer. However, while stocks consist of items that have not yet been used, and which are perfect substitutes (e.g. stocks of sugar or paperclips), assortments consist of heterogeneous products, which have the same overall usage goal but different specific applications and which are used on and off. Assortments can exist for both durables and nondurables (e.g. assortments of soft drinks or biscuits), and although this study will focus on durable product assortments, it can be easily extended to nondurable assortments.

For assortments to have an effect on subsequent buying behavior, they need to be evaluated by their owners. The objective of this study is to develop a conceptual framework of the evaluation process of consumers' product assortments, and to provide a first test of it.

Evaluation of Consumers' Product Assortments

Consumers actively manage their assortments by supplementing, replacing, or removing products. To make these types of decisions, they need to evaluate the content and structure of the assortments. A negative assortment evaluation could, for instance, stimulate intentions to buy in the category. This raises the question of what assortment evaluations are based on.

Identifying Assortment Properties. Previous literature on set evaluation, primarily in the context of product bundling, considered the integration of product evaluations (e.g. Gaeth, Levin, Chakraborty & Levin 1990; Yadav 1994). Set evaluation here is derived from a (weighted) average of the item evaluations. So if a set contains items that are liked better, the total set will be liked better.

In research on product bundling, subjects have been typically presented with a set of fixed size, so effects of set size could not be examined. Yet, for consumers' assortments, differences between the number of products in each set are very important as they can influence overall set evaluations: more products may be better than less. This means that both the individual items and the size of the assortment might influence consumer evaluations of assortments.

Besides items and size, other properties of the assortment may also be influential. A set may not be valued highest when it contains duplicates of the most preferred item, but consumers are willing to trade in items for lesser preferred ones as this produces a more varied set. To clarify the influence of attributes and variety on set evaluation, an overview of a hypothetical assortment is presented in Table 1. The assortment, say for instance shirts, contains the items and their attributes and usage situations. The attributes vary from the concrete to the abstract (Johnson, Lehmann, Fornell & Horne 1992). Abstract attributes, such as fashionability or quality, need to be inferred from concrete attribute information, while concrete, perceptual, attributes such as size, are directly associated with the product (Bettman & Sujan 1987).

When considering the content and structure of an assortment, consumers could focus on items or on attributes. If consumers focus on items, mean item evaluations will influence assortment evaluations. A focus on attributes introduces other assortment properties, such as the size of product subgroups with certain attributes (e.g. number of small products), or for certain usage situations. Besides these subgroup evaluations, variety in the assortment can be important as well. Variety is related to the degree of similarity, which can exist at multiple levels: concrete or abstract attribute overlap, or usage. High concrete attribute overlap indicates that the products 'look alike', while high abstract attribute overlap indicates that the products have the same benefits. Usage related aspects of products do not necessarily match the physical properties, or product attributes (Lefkoff-Hagius & Mason 1993).

Several assortment properties have now been identified whose evaluation could influence overall assortment evaluation: mean item evaluation, assortment size, attribute and usage situation evaluations, and variety. Therefore, we propose the following hypothesis:

H1: Overall assortment evaluation is influenced by:

(a) mean item evaluation

(b) assortment size evaluation

(c) evaluation of the number of items with specific attributes

(d) evaluation of the number of items for usage situations

(e) assortment variety evaluation

TABLE 1

A HYPOTHETICAL PRODUCT ASSORTMENT

FIGURE 1

CONCEPTUAL FRAMEWORK OF ASSORTMENT EVALUATION

The relation between property evaluation and assortment evaluation. After having introduced assortment properties, their relation with assortment evaluation will now be discussed. In general, evaluations are not made in absolute terms, but by comparison to some standard or norm (Kahneman & Miller 1986). Assortment evaluations are assumed to be based on evaluations of the assortment properties, which result from a comparison between content and structure properties and norms for these properties. A conceptual framework of the evaluation process is presented in Figure 1. The role of norms in this framework is comparable to their role in the disconfirmation paradigm regarding consumer satisfaction. Experiences are compared with norms, and the resulting (dis)confirmation leads to satisfaction (Cadotte, Woodruff & Jenkins 1987).

In the satisfaction literature, the relation is assumed to be linear and positive, as negative disconfirmation leads to dissatisfaction, while positive disconfirmation leads to satisfaction. A similar relation can also be expected for item evaluations. If consumers like the items in an assortment (high item evaluation), the evaluation of the total assortment is likely to be high as well. For the other assortment properties consumers are likely to have ideal points for property levels as more is not always better. In this case deviations in both directions result in less satisfaction. First, consider the size of an assortment. When assortment size is lower than optimal, items fall short of the needs with respect to (expected) usage situations. On the other hand, when assortment size is high, space restrictions, budget considerations, and the possibility of fashion changes become important (Naddor 1961). A similar reasoning is applicable to the size of subgroups. This means that the hypothesized relationship in these cases is an inverted U. The same form of the relationship is assumed for variety evaluation. Irrespective of which concept of variety is used, very high variety between items may not be desirable as this means that there are few alternatives in the assortment in case of product breakdown. Very low item variety on the other hand means low differentiation between the items, which in case of attribute satiation is not optimal either.

H2: The relationship between property evaluation and overall assortment evaluation

(a) is positive and linear for mean item evaluation

(b) has the form of an inverted U for assortment size evaluation

(c) has the form of an inverted U for the evaluation of the number of items with specific attributes

(d) has the form of an inverted U for the evaluation of  the number of items for usage situations

(e) has the form of an inverted U for assortment variety evaluation

Property evaluation. Property evaluation is assumed to be based on norms and properties. As item evaluations have been discussed in previous literature, we will focus on the other assortment properties here. Assortment properties can be expected to have a direct positive effect on property evaluations; i.e. if an assortment has more items, people can be expected to evaluate it as having a large number of items. The ideals that consumers have regarding assortment properties may be related to product involvement. Involvement has received considerable interest from consumer behaviorists (Mittal & Lee 1989), and has been shown to influence the number of items that consumers buy (Tigert, Ring & King 1976). We posit that ideal assortment size, and subgroup sizes, is higher for high-involved consumers. In addition, we propose that high-involved consumers have higher ideals regarding variety. They are likely to be more sensitive to the fit of products with diverse usage situations. A consumer with a high norm will evaluate an assortment with a certain property less positively than a consumer with a low norm, and therefore a negative effect of involvement is hypothesized.

H3: Property evaluation is influenced by the property (positively) and involvement (negatively) for:

(a) assortment size evaluation

(b) evaluation of the number of items with specific attributes

(c) evaluation of the number of items for usage situations

(d) assortment variety evaluation

Figure 1 introduced a conceptual model of assortment evaluation, based on assortment properties. Although previous research generally used an integration of item evaluations, several other assortment properties were also identified. Size, subgroups based on attributes and usage situations, and variety can influence assortment evaluations as well. With an exception of mean item evaluations, ideal points are used for assortment property evaluations. After having introduced a conceptual model of consumers' product assortment evaluation, we offer an empirical exploration and first test of the hypotheses.

TABLE 2

WAVES OF THE STUDY

METHOD

A product category needs to be chosen that can provide meaningful data regarding consumers' assortments. Preferably this should be one in which many individual product differences occur, so that stockpiling of identical items is not likely. Since durable product assortments are expected to be more stable, durables are preferred over nondurables. We chose the product category of footwear. Duplication of products in relatively uncommon in this product category, as the prime characteristic of durables is that they can be used again, assuring product availability over multiple usage situations.

Sample and question timing

The data for the study were collected by a subsample of a computerized consumer panel (NIPO), that is representative of the Dutch population. Data collection has taken place over five consecutive weekends. As part of the study, subjects were asked to make photographs of their shoes. In the weekend of June 1 1996 (week number 2), panel members were asked for their willingness to photograph their shoes, to which 20.1% respond positively. Of these, 160 panel members were asked to photograph their shoes and to complete a computerized questionnaire. After excluding non-response (25), non-usable photos (29), and persons who bought shoes in the data collection period (25), the final sample size is 81. Table 2 provides on overview of the data collection period. As not all subjects have answered each question (either due to absence, or a "don't know" answer), the final numbers of subjects are provided in the table.

Advantages of letting subjects take photos of their shoes are that researchers can judge concrete attributes from these photos, which will lessen subjects' burden. In addition, the image that the photo represents remains detailed, insuring complete notation (Collier 1967). A small pretest among 30 different consumers was used to assure the ability of subjects to make clear and interpretable photos. Subjects used their own camera and film, for which they received the equivalence of US $15 as payment. They were instructed to photograph their shoes, up to a maximum of the 11 most frequently used shoes, and make one photo of the remaining shoes if they had more. They were asked not to include shoes that they wore only for sports, or shoes that they have not worn during the past year.

Measures

Subjects viewed questions and numbered answer categories on their computer screen. They typed in the number of their response, at which time the next question appeared on the screen.

Assortment evaluation. Overall evaluative judgment of the assortment was asked by the question "To what degree are you overall satisfied with the shoes that you own?". Answers were given on a five-point scale ranging from "totally not satisfied" to "very satisfied".

Assortment properties. Subjects were asked to report the total size of their shoe assortment. Subjects indicated which shoes they used in five important usage situations. Variety of the items in the assortment is judged from the photos by use of four judges. Judges directly coded the degree of similarity between the items in the assortment on a 5-point scale. Each assortment was coded by two judges, and correlation between these judges was .65 (significant at 1%). Item similarity was also measured by means of similarity calculations based on comparisons of item attributes. The judges coded concrete attributes [Coded attributes are: shoe type, primary color, secondary color, primary material, secondary material, shoe fastening, stitching, appearance, shoe height, openness of shoe, prints, accessories, heel height and type, sole height, and shoe nose.] with an average interjudge reliability of 0,88. Alternatively, similarity was measured by the overlap between abstract attributes or usage situations, both of which were provided by the subjects themselves. A score of zero or one is allocated to each attribute k depending on whether the two items are the same on this attribute; the scores are subsequently averaged to give the similarity coefficient between two items (Gower 1971; Everitt 1993, p.43). By averaging similarity coefficients for every possible combination of two items from an assortment, overall similarity in the assortment can be measured. This measure can be reformulated so that the only information needed is the number of items with a particular attribute level. In an assortment, if m items have a specific attribute k this equals m * (m-1)/2 item pairs which share the attribute. By summing over all attributes, and dividing by the total number of item combinations and the total number of attributes, assortment level measures of similarity can be obtained. The resulting measure will equal 1 when all attributes are common, and 0 when all attributes are distinct.

EQUATION

m_kl = number of items with attribute level l for attribute k

K = total number of attributes included

n = total assortment size

Hoch, Bradlow and Wansink (1998) propose a different, but related measure for variety. Their measure is based on dissimilarities between pairs of items. They introduce, among others, the Hamming distance, which is perfectly negatively correlated with the SIM measure, after adjusting for the number of item pairs present (=n (n-1)/2).

Assortment property evaluation. Item evaluations were constructed by a composite measure consisting of the sum of attribute evaluation multiplied by its importance. This was done for three abstract attributes: quality, comfort and fashionability. By averaging the item evaluations, the mean item evaluation of the assortment was obtained. [Item evaluations can also be constructed without taking attribute importance into account, by averaging attribute evaluations. The resulting average item evaluation of the assortment has a correlation of .887 with the measure including attribute importances, and leads to similar conclusions regarding the evaluation model.] Attribute evaluations were measured by having subjects sort the numbered shoe photos into three piles, representing low, mediate and high levels of the attribute, for instance low quality, neither low nor high quality, and high quality. Subjects were asked to type in the numbers of the photos in each of the piles. Attribute importance was acquired by letting subjects choose between pairs of the abstract attributes and price. A prime advantage of this method is that it reduces halo effects, in which all aspects are indicated as being important (Green, Carmone & Smith 1989). As an example, subjects would be asked to choose between "(1) My shoes have to be of good quality, even if this means that they are not completely in fashion" and "(2) It is important for me that my shoes are in fashion, even if the quality is a little less". The four attributes that were used for the paired comparisons are assumed to be located on one latent dimension that is the same for each person. Individual subjects have their own ideal points, and the scale is folded in this point to determine the attribute importances by their distance to this ideal point.

An evaluation measure for assortment size was included by the question: "When you give a close look to the shoes you own, then the number of shoes you own is .". Answers were provided on a five-point scale, featuring: (1) too high, (2) high, but not too high, (3) exactly enough, (4) low, but not too low, (5) too low. A "no response" option was also included. In a similar fashion, and with identical answer categories, questions were asked related to the evaluation of the number of items with high levels for the attributes and the evaluation of the number of items for important usage situations, for instance "When you give a close look to the shoes you own, then the number of high quality shoes you own is.". Regarding assortment variety, the question was formulated as "When you give a close look to the shoes you own, then the number of different kind of shoes that you own is.".

Involvement. Product involvement was assessed on a 4 item scale with response alternatives: (1) totally disagree, (2) disagree, (3) neither disagree nor agree, (4) agree, and (5) totally agree. Four statements were used for this construct, an example being "I am very involved with shoes". Items are similar to the ones used by Mittal and Lee (1989). Cronbach's alpha equals .79.

Analysis method

Regression analysis will be used to test the hypotheses. Several of the independent variables involve evaluations of subgroups of the assortment, which will partly overlap. Also, the hypothesized U-curved effects lead to the inclusion of squared terms in the regression model. Therefore, multicollinearity between explanatory variables is likely to occur, which may cause unreliable regression estimates. Therefore, we will not focus on the regression estimates, but on alternative nested models for testing hypotheses (Dougherty 1992).

RESULTS

A general overview with respect to assortment properties is provided in Table 3. The composite measure of item evaluation requires the calculation of attribute importances by determining an underlying unidimensional scale. To accomplish this we perform unfolding using the program UNFOLD (van Blokland-Vogelesang 1990). Three subjects provided intransitive choices, leaving a total of 71 subjects. UNFOLD examines all possible unidimensional orderings and determines the best scale as the one with the minimum number of inversions from individuals' rankings. The scale that was selected has the highest number of perfect fitted individual ranges (54 out of 71), and a c2 of 0.16 (df=2, p=.923), which is an almost perfect fit. The final scale is given in Figure 2, with the four attributes, and six ideal points which are presented by the number of subjects between brackets. For instance, the four subjects with the ideal point located most to the left, find price the most important, followed by comfort, quality, and finally fashionability. The distances between ideal points and attributes need to be transformed into relative attribute importances. Large distances represent low importance, and therefore we apply: (TOT-A)/(2"TOT) where TOT=total distance from ideal point to the three attributes, and A=distance ideal point-attribute.

TABLE 3

OVERVIEW OF SHOE ASSORTMENT PROPERTIES

FIGURE 2

UNIDIMENSIONAL SCALE FOR ATTRIBUTE IMPORTANCES

Hypothesis testing

Our hypotheses were tested by stepwise introduction of variables into the model. Table 4 presents the relevant models, and their regression results. The first model introduces mean item evaluation into the regression model. This is the basic model, in which assortments are evaluated higher if they consist of higher evaluated items. The regression coefficient for mean item evaluation is indeed positive, consistent with hypothesis 2a, and significant at p=.06.

Assortment size evaluation is first added, and it significantly improves the model, which is in support of the hypothesis 1b (models 1 versus 2 in Table 4). Comparing models in which the squared term is present and absent (models 2 versus 3) shows a significant curvilinear effect for size evaluation. Figure 3 indicates that an inverted U shape between the two variables can be found, as was hypothesized (2b).

As indicated in Table 4 by comparing models 1, 4 and 5, adding abstract attribute evaluations and squared terms of these significantly adds to the basic model with only mean item evaluations, confirming hypothesis 1c. Non-linear effects are found, and for quality and comfort this is an inverted U effect, as was hypothesized. For fashionability no inverted U curve could be found, as no respondents answered 'high' or 'too high' to this question. Assortment size evaluations no longer add to the model once attribute evaluations are introduced. This means that attribute evaluations and size evaluations offer a different explanation of the same variance component in overall assortment evaluations. This may be partly due to question wording, as the subjects were asked to evaluate the number of items with specific attributes. In further extension models, size evaluations are no longer incorporated. Evaluations of the number of items for specific usage situations and their squared terms are added to the model, and provide additional explanatory power (models 7 through 9). The proposed inverted U curve of hypothesis 2d is not present.

Table 4 shows that variety evaluations add to the model, consistent with hypothesis 1e, also after mean item and attribute evaluations are included. Comparison of models 7 and 8 shows a curvilinear effect of variety evaluations at 10% significance. Figure 4 reveals the hypothesized inverted U relation between variety evaluations and assortment evaluations.

Model 12 is the final regression model of assortment evaluation, and figures 3 and 4 are based on the estimates of this model.

Property evaluation. Property evaluation is hypothesized to be based on the property level and involvement in hypothesis 3. With respect to total assortment size, the number of shoes indeed has a significant positive effect on assortment size evaluation, but involvement does not have a significant effect (total regression model: F2/77=7.1; p-value=.00). For the evaluation of the number of items with attributes, only the model for quality is significant (F2/77=6.8; p-value=.00), due to a positive effect of involvement. None of the models for the evaluations of items for important usage situations are significant. Variety evaluation can be based on item similarities with respect to concrete and abstract attributes, and usage situations. Correlations between these SIM measures are not significantly larger than 0, indicating that they measure different aspects. Coded similarity based on the photos has a significant correlation of .69 with concrete attribute overlap, indicating that important attributes were included, and that the overlap measures are valid. A regression model including the three SIM measures and involvement as explanatory variables for variety evaluation was not significant (F4/72=0.7; p-value=.57).

TABLE 4

IMPACT OF PROPERTY EVALUATIONS ON ASSORTMENT EVALUATION

FIGURE 3

ASSORTMENT SIZE EVALUATION EFFECTS

FIGURE 4

ASSORTMENT VARIETY EVALUATION EFFECTS

CONCLUSION AND DISCUSSION

Consumers' product assortments have been introduced as sets of heterogeneous items, which have the same overall usage goal but different specific applications. Assortment evaluation is assumed to be based on property evaluations, which entails more than just an integration of item evaluations. The present study provides an initial empirical exploration of assortment evaluation.

The five assortment properties of (1) mean item, (2) size, (3) attributes, (4) variety, and (5) usage situation were shown to be related to assortment evaluations, consistent with our first hypothesis. The addition of attribute evaluations to the model eliminated the need for assortment size evaluations. Therefore, size evaluations split into attributes (resulting in evaluations of the number of items with an attribute) outperformed overall size evaluations. The regression models have shown that item evaluations, evaluation of subgroup sizes, and variety evaluations all add to the explanation of assortment evaluations.

The form of the relationship was hypothesized as positive linear for mean item evaluation, and inverted U for the other assortment properties. The positive relation between mean item evaluation and assortment evaluation has indeed been found, as well as inverted U curves for assortment size evaluations, attribute evaluations, and variety evaluations. No inverted U curves were found for the usage situations, which is an unexpected result that we can not easily explain. Possibly, upper boundaries such as space restrictions and budget considerations are not operating in this case.

Property evaluations seem to be made through comparisons with ideal points, but the exact process is not clear from the present study. Size evaluations were shown to be based on the objective size of the assortment, but an attempt to introduce norms by including involvement failed. As several evaluations show a clear inverted U with assortment evaluations, the proposition of ideal points remains.

Discussion. Previous research regarding set evaluation has examined the integration of item evaluations (e.g. Yadav, 1994), or attribute satiation (e.g. McAlister, 1979). To date, no studies have focused on the use of different assortment properties in set evaluation. Although the process of integration is left for future research, this study shows that item, subgroup and variety evaluations all are important for assortment evaluation, and have an effect independent of each other. This means that to explain assortment evaluation, item evaluations alone are insufficient, and the structure between the items needs to be taken into consideration. This structure can be measured in the sense of subgroups or variety. Another important point from the study is that property evaluations are based on individual ideal points. Not assortment properties as such determine assortment evaluation, but the perception of these properties by the consumer. The present study shows that individual perceptions are important, and we propose that the process will be influenced by property norms.

Limitations. There are several limitations of the study. No direct investigation of the evaluation process was undertaken. Instead, tentative conclusions were reached, based on consumer assortment and property evaluations. Future research could use direct process measures to further investigate the evaluation process. In this study, the assortment evaluation questions focused on the number of items with specific attributes or usage situations. Subjects were asked to indicate their evaluation of the number of items with high quality, or the number of items that can be used for parties, and so on. Evaluation of assortments needs not focus on number of items alone, however, and more qualitative aspects might also be important for assortment satisfaction.

Future research. The present study offers an exploration of the evaluation process regarding consumers' product assortments, an area that has received little research attention. Future research should take into consideration the different properties that can influence assortment satisfaction, and the concept of ideal points. Property evaluations that incorporate consumers' ideal points have been shown to relate to assortment satisfaction. The relationship with objective assortment properties is less clear, due to these individual ideal points. Regression of objective assortment properties, such as counts of items in subgroups, would not have provided with an explanation of assortment satisfaction in the present study.

Future research should also be directed towards the consequences of assortment (dis)satisfaction. How does it influence buying intentions and purchases in the product category? To what extent do people consider their current product assortment when they make new purchases?

Extensions of the present study can also be found in the area of retail assortments. The exact process of property evaluation is likely to differ for retail assortments, but retail assortment evaluations may very likely be based on the same properties as are consumers' product evaluations. As relatively little is known about retail assortment evaluation processes, this warrants further research in this area. Knowledge of consumers' assortments can also provide assistance in building a retail assortment strategy. This is not to say that retail assortment need to match consumers' assortments. For instance, retail assortments could focus on specific parts of consumers' assortments (e.g. related to a specific usage situation).

A second possible extension is to include consumer interactions. There are many situations in which assortments are not owned by single consumers, but rather by a household as a whole (e.g. videotapes). Individual preferences of different household members will influence the content and structure of such an assortment.

The empirical investigation only provided with a first understanding of what may be happening when people make assortment evaluations. This gives an indication of possible underlying processes, but these processes warrant further investigation. Especially when it comes to property evaluations, the underlying processes are not clear in the present results. Interesting issues are not only the identification of these processes, but also possible individual differences, and potential influential factors.

REFERENCES

Alderson, Wroe, 1965, Dynamic Marketing Behavior; A Functionalist Theory of Marketing, Homewood: Richard D. Irwin, Inc.

Bettman, James R., and Mita Sujan, 1987, "Effects of Framing on Evaluation of Comparable and Noncomparable Alternatives by Expert and Novice Consumers", Journal of Consumer Research, 14 (September), 141-154

Blokland-Vogelesang, Rian van, 1990, Unfolding and Group Consensus Ranking for Individual Preferences, The Hague: Cip-Data Koninklijke Bibliotheek

Cadotte, Ernest R., Robert B. Woodruff, and Roger L. Jenkins, 1987, "Expectations and Norms in Models of Consumer Satisfaction", Journal of Marketing Research, 14 (August), 305-314

Collier, John Jr., 1967, Visual Antropology: Photography as a Research Method, New York: Holt, Rinehart and Winston

Dougherty, Christopher, 1992, Introduction to Econometrics, Oxford: Oxford University Press

Everitt, Brian S., 1993, Cluster Analysis, New York: Halsted Press

Gaeth, Gary J., Irwin P. Levin, Goutam Chakrabory, and Aron M. Levin, 1990, "Consumer Evaluation of Multi-Product Bundles: An Information Integration Analysis", Marketing Letters, 2 (1), 47-57

Gower, J.C., 1971, "A General Coefficient of Similarity and Some of its Properties", Biometrics, 27, 857-871

Green, Paul E., Frank J. Carmone, Jr., and Scott M. Smith, 1989, Multidimensional Scaling; Concepts and Applications, Boston: Allyn and Bacon

Green, Paul E., Yoram Wind, and Arun K. Jain, 1972, "Preference Measurement of Item Collections", Journal of Marketing Research, 9, 371 -377

Hoch, Stephen J., Eric T. Bradlow, and Brian Wansink, 1998, "The Variety of an Assortment", paper presented at the ACR conference

Johnson, Michael D., Donald R. Lehmann, Claes Fornell, and Daniel R. Horne, 1992, "Attribute Abstraction, Feature-Dimensionality, and the Scaling of Product Similarities", International Journal of Research in Marketing, 9, 131-147

Kahneman, Daniel, and Dale T. Miller, 1986, "Norm Theory: Comparing Reality to Its Alternatives", Psychological Review, 93 (2), 136-153

Lefkoff-Hagius, Roxanne, and Charlotte H. Mason, 1993, "Characteristic, Beneficial and Image Attributes in Consumer Judgments of Similarity and Preference", Journal of Consumer Research, 20, 100-110

McAlister, Leigh, 1979, "Choosing Multiple Items from a Product Class", Journal of Consumer Research, 6 (December), 213-224

Mittal, Banwari, and Myung-Soo Lee, 1989, "A Causal Model of Consumer Involvement", Journal of Economic Psychology, 10, 363-389

Naddor, E., 1961, "Evaluation of Inventory Control", Proceedings of the Second National Conference on Operational Research, London: English Universities Press Ltd

Tigert, Douglas J., Lawrence J. Ring, and Charles W. King, 1976, "Fashion Involvement and Buying Behavior: A Methodological Study", Advances in Consumer Research, 3, 46-52

Walsh, John W., 1995, "Flexibility in Consumer Purchasing for Uncertain Future Tastes", Marketing Science, 14 (2), 148-165

Yadav, Manjit S., 1994, "How Buyers Evaluate Product Bundles: A Model of Anchoring and Adjustment", Journal of Consumer Research, 21 (September), 342-353

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