Attribute Deficiency Segmentation:Measuring Unmet Wants

James H. Myers, Graduate Management Center, Claremont Graduate School
ABSTRACT - Three approaches are discussed that are designed to measure consumers' unmet wants/needs/expectations directly at the individual level: expectancy confirmation,, ideal point expectancy value model, ant value percept disparity. All three methods yield a vector of discrepancy scores for each respondent on each attribute, and these discrepancies (deficiencies) can be used to identify groups of respondents having similar profiles of unmet wants. The value percept disparity approach yielded 4 deficiency segments for a pet food product.
[ to cite ]:
James H. Myers (1988) ,"Attribute Deficiency Segmentation:Measuring Unmet Wants", in NA - Advances in Consumer Research Volume 15, eds. Micheal J. Houston, Provo, UT : Association for Consumer Research, Pages: 108-113.

Advances in Consumer Research Volume 15, 1988      Pages 108-113


James H. Myers, Graduate Management Center, Claremont Graduate School


Three approaches are discussed that are designed to measure consumers' unmet wants/needs/expectations directly at the individual level: expectancy confirmation,, ideal point expectancy value model, ant value percept disparity. All three methods yield a vector of discrepancy scores for each respondent on each attribute, and these discrepancies (deficiencies) can be used to identify groups of respondents having similar profiles of unmet wants. The value percept disparity approach yielded 4 deficiency segments for a pet food product.


Since the introduction of the market segmentation concept (Smith, 1956), a wide variety of approaches to segmenting markets have been proposed. Wilkie (1971) suggested that these various approaches could be classified into two major types: "empirical stream", based on characteristics of buyers or users, and "product stream", based on characteristics of the product or service itself, or on usage rates or patterns, or on important product attributes, or on situations or occasions of use. In a later review article, Wind (1978) used different terms in referring to this same dichotomy: "general customer characteristics" (for empirical stream) and "situation-specific customer characteristics" (for product stream).

This paper will present a new form of product stream segmentation that is based on perceived deficiencies in product/service attributes. Earlier work in attribute-related segmentation has focused on the relative importance of key attributes, in order to identify those the consumer wants most. But as Ries and Trout (1986) point out, "Knowing what the customer wants isn't too helpful if a dozen other companies are already serving the same customer's wants". This suggests that a more useful approach than simply measuring attribute importance might be to measure directly the extent of unmet needs in the most important product attributes. We will call these unmet needs "deficiencies". If these deficiencies can be measured at the individual level, consumers can be segmented into groups that have homogeneous patterns of attribute deficiencies. The resulting groups can be referred to as "deficiency segments". Marketing strategies aimed at reducing or eliminating sizeable deficiencies in important attributes for a target deficiency segment could be expected to have a meaningful impact on product/service/brand preference as well as purchase.


Researchers have used several approaches to measuring deficiencies in product/service attributes. Those most frequently used can be classified into one of three types: expectancy confirmation (EC), ideal point (IP), and value-percept disparity (VPD).

Expectancy Confirmation

Research on consumer satisfaction/dissatisfaction (CS/D) has been dominated by the measurement of discrepancies between consumer expectations and the perceived performances of a product/service. Some studies have measured these discrepancies at the level of overall or total performance only (Olshavsky & Miller, 1972; Anderson, 1973; Westbrook & Oliver, 1980; Cardozo, 1965; Cohen & Goldberg, 1970), while others have looked at several component attributes or features in addition (Pfaff, 1977; Morris, 1977; Hempel, 1977; Oliver & Linda, 1980).

Researchers have used a variety of approaches to measure expectancy confirmation at the specific attribute level. Trawick and Swan (1980) distinguish between inferred and perceived disconfirmation. Inferred disconfirmation is measured by asking buyers prior to usage to rate the level of each attribute they expect the product/service to have, and then to rate attribute performance after usage. The difference between before and after ratings constitutes inferred disconfirmation. In contrast, perceived disconfirmation obtains both of these measurements after product usage and therefore respondents are asked to recall their anticipated levels of attribute performance. To address some conceptual problems with these approaches, Oliver (1977, 1980a) used "better than expected - - - - - worse than expected" scales as measures of satisfaction and other post-exposure cognitions. Of course, some investigators have used only post-usage measures of satisfaction, but these do not offer any deficiency or discrepancy measures and therefore are not useful for purposes of the present study.

Regardless of how they are measured, discrepancies between expected and observed performance at the attribute level can be used as inputs to clustering programs that identify relatively homogeneous groups of respondents based on their patterns of discrepancies. The general expectancy confirmation model is (following the notation of Swan and Martin, 1980):


where S= satisfaction; Ai= after-usage subjectively experienced attribute level; PREDi=expected value of the attribute level measured prior to usage; n= number of salient attributes. These differences produce a vector of difference scores for each respondent, and respondents can then be grouped using either hierarchical or partitioning clustering technologies. This same type of model could also be used for discrepancies between ideal and observed performance at the attribute level.

Ideal Point

The concept of ideal points was first mentioned by Coombs (1964, p. 8-9) and was later introduced into marketing as a part of multidimensional scaling (MDS) and related technologies (see Green and Carmone, 1968; Carroll, 1972; Coombs & Avrunim, 1977). These technologies construct idiosyncratic perceptual spaces based upon direct similarities judgments among objects such as brands of products/services. Then one or more ideal points (or vectors) for each individual is inserted into these spaces using a variety of algorithms (see Kamahura and Srivastava, 1986 and references). The notion of ideal points was then extended into multiattribute attitude models in the form of an "ideal-point" version of the general expectancy-value model. This model has the form:


where Ao= overall attitude toward an object (product/ service/ brand); Wik=importance weight assigned by individual i to attribute k; Bijk= individual i's rating of brand j on attribute k; Iik= the individual's ideal point on attribute k. If a single product or brand can be specified (e.g., brand preferred, brand bought last, brand used most often), then a vector of difference scores can be constructed for each individual based on the discrepancies between brand attribute ratings and ideal points. These vectors can then be input into a clustering algorithm in the same manner as for the expectancy confirmation discrepancies model discussed earlier.

Value-Percept Disparity

In a recent article, Westbrook & Reilly (1982) proposed an alternative to the expectancy confirmation model for measuring consumer satisfaction. It is based on the value-percept disparity model first proposed by Locke (1967, 1969) in the context of measuring job satisfaction This model proposes that satisfaction is based on the discrepancy between perceptions (beliefs) of an object/action and a person's values, stated in terms of needs, wants or desires. The greater the value-percept disparity, the lower the satisfaction, and vice versa.

Translating this model into the marketing context, Westbrook and Reilly (p. 257) state, " What is expected in a product, however, may or may not correspond to what is wanted or desired in that product." They suggest that satisfaction may be more a function of performance relative to aspirations than to expectations. Comparisons of these two models using LISREL failed to confirm the superiority of the value-percept disparity model as compared to expectancy confirmation. However, their study asked respondents to list as many of their "needs" (rather than wants or desires) in an automobile as they could, and then to rate the extent to which their own automobiles met these needs, using a 7-point semantic differential scale anchored with "Provides for less than my needs" (7) and "Provides exactly what I need" (1).

Myers (1976, 1977) took a different approach to operationalizing the value-percept discrepancy model. In the context of searching for new product/service ideas, he argued that what people "want", or would like to have, is more important than what they "expect" from products or services now on the market. Focusing on the latter tends to restrict respondents to thinking only within the existing array of products and services rather than on new ones, or on major improvements in existing ones, that they would greatly prefer. The primary purpose of his study was to measure the extent of deficiencies in each of dozens of cleaning product attributes. He asked respondents to rate the last product actually used in terms of both how much they recalled wanting each attribute and how much they actually got that attribute from the product they used (using a 4-step equal-interval verbal scale). If this approach were used to measure consumer overall satisfaction with a particular product/service, the model would have the following form:


where DS= dissatisfaction; Wik= wanted rating of individual i on attribute k (similar to importance ratings); Gijk= got rating of individual i on attribute k for product/brand j (similar to beliefs ratings). The greater the discrepancies between wants and gots, weighted by the importance of each attribute, the greater the consumer's overall dissatisfaction with a particular product/brand.

Using this approach, it is possible to probe for respondents' potential interest in product benefit/features that are not offered by any existing brands in a category (e.g., an ingredient in canned dog food that kills fleas and ticks systemically, or a contraceptive). This makes it easily possible to include several "mini concept test" statements among the usual assortment of product attribute descriptors rated by respondents. At the same time, deficiencies can also be measured in each of dozens of existing features/benefits/ imagery attributes for products/ services now on the market.

Measuring Deficiencies

Despite major differences in the primary objectives of the three models discussed above, they all can be used to measure discrepancies between perceived actual product/service performance and some relevant reference point, at the individual level and for each product/service attribute separately. These discrepancies can be considered alternative ways of measuring expectancies/needs/wants that are not being met to the extent desired by a respondent; that is, deficiencies. Segmentation based on these deficiencies is a way, and it may be is the most direct way, of locating people that would respond to specific changes in any element of the marketing mix.


The purpose of the present study was to explore the potential value of deficiency segmentation using the general category of consumer packaged goods as a vehicle. Thus, any one of the three major attribute deficiency measurement approaches discussed above could have been used. Since the primary objective of the present study was to develop customer-based ideas for new types of pet food products, the value-percept disparity model was selected. This approach gave respondents the opportunity to indicate their unmet wants or desires in each of a large number of attributes that could be used to describe the benefits or ingredients in products/brands in the pet food category, including some benefits/ingredients that were not available in any existing commercial products.

The Sample

A convenience sample of 302 respondents was intercepted in shopping malls in 5 cities nationwide. Each respondent was qualified as a pet owner who had the primary responsibility for deciding what their pet would eat for main meals. A number of questions were first asked about each pet (e.g., type or breed, size, age, sex, health condition), followed by previous purchases of different types and brands of pet foods. This was followed by questions about awareness of the various brands and types, feeding patterns, purpose of ownership, and the like.


Each respondent was then presented a list of 54 attributes that could be used to describe both existing and potentially desirable characteristics of the particular type of pet food of interest (e.g., contained lots of fiber, extremely high quality ingredients, a well-known brand, was very crunchy). Respondents were asked to remember the last time they served this general type of food to their pet and to indicate how much they both wanted and got each of the attributes at that occasion. Ratings were given on a 10-point scale (10=extremely; 1=not at all).

The value-percept disparity was measured by subtracting got from wanted ratings, yielding a vector of 54 discrepancy scores for each respondent. Of course, discrepancies could be either positive or negative (negative in cases where a respondent got more of a particular attribute than he/she wanted.) Negative deficiencies are considered by this writer as being equally important and meaningful as positive deficiencies. Negative deficiencies can mean one of two things: 1) the respondent get an attribute more than he/she wanted it (e..g., sweetness, thickness, aroma, thrills, excitement) and did not appreciate this, or 2) the respondent got more of a desirable attribute than he/she wanted simply because he/she did not want it very much (e.g., great tasting food, nutrition, styling). In either case, it is the pattern of deficiencies that defines segments, and these patterns should be based on negative as well as positive deficiencies. Of course, the various attribute deficiencies should be related by the underlying constructs they are measuring.

Vectors of discrepancies were input into a k-means (disjoint partitioning) clustering algorithm, and solutions were obtained for 2,3,4,5, and 6 clusters.


In any kind of multivariate segmentation analysis, the most difficult problem is that of deciding how many segments exist. This might be done on either a theoretical (a priori) or an empirical (a posteriori) basis. A priori specification of segment numbers should be done only when based on a solid theoretical foundation, or at least on prior investigation designed to provide some form of market structure related to the variables of interest. However, most segmentation studies conducted by business firms do not meet either of these conditions, and of course this is the primary reason for conducting such a study.

Since one of the objectives of the present study was to see how many (if any) deficiency segments might exist, an a posteriori determination of segment membership was necessary. Ideally, the major packaged software clustering programs would provide an apriori estimate as to the number of clusters that are likely to exist in any data set. None of them appear to offer such an option, and also there is no general agreement as to the best solution to this problem (see Everitt, 1979, 1980). This is true for both hierarchical and partition clustering. However, a recent simulation study (Milligan & Cooper, 1985) using 4 hierarchical clustering methods provides some guidance for researchers using this technique.

Therefore, an investigator must use some combination of the following criteria for deciding a posteriori how many clusters best describe the data: 1) trends in F-ratios of the differences among means of the various clusters, 2) balance among the numbers of cases in the clusters, 3) patterns among the discrepancies in a cluster, in terms of logical consistency, 4) size of a particular cluster having a deficiency pattern of interest, and 5) number (and size) of segments that could be addressed by a business firm with finite resources.


Inspection of the 5 clustering solutions using these criteria led to the selection of the 4-cluster solution as offering the best opportunity for company implementation. All of the 54 attributes differed among the 4 clusters at the p <.01 level of significance or beyond except 3. (These were at the .06, .11, and .53 levels respectively.) The relative sizes of the 4 clusters are shown in Table 1. The reason why Cluster 2 showed such homogeneity (in terms of average distance of all cases from the centroid) was because respondents in this segment had deficiencies that were about the same as for the total sample on all but 5 of the 54 attributes.



Cluster 1 was of particular interest because }3 nearly all deficiencies were at least slightly higher than for the total sample, 2) 2 of the highest deficiencies were among attributes that distinguished most among the clusters (with highest F-ratios), 3) the 4 attributes that distinguished most among the clusters (including the 2 mentioned earlier) all related to the same product aspect texture [This, as well as all statements in Table 2, are disguised, since this is a highly competitive market and the company is now working on product improvements based upon the actual desired attributes for this deficiency segment. However, all numbers in both Tables 1 and 2 are correct, and the disguised texture attributes were taken from the same study.]. Taken as a group, these attributes were logically consistent, best separated the 4 clusters, had above-average deficiencies among members of Cluster 1, and were among the most important attributes driving customer satisfaction. These attributes and their respective F-ratios (reflecting mean differences among clusters) are shown in Table 2.

This segment also showed some clear differences from the others in terms of both "people demographics" (size, age, reason for purchase). All of these data gave a very clear profile of the wants and characteristics of Deficiency Segment 1. The company is now actively working on a product modification aimed specifically at the deficiencies perceived by this segment.




This study has shown that segmentation based on measured deficiencies in important product attributes is feasible and can lead to meaningful clusters of respondents who share similar patterns of unmet needs or desires. These deficiencies can be measured at the individual consumer level in 3 ways: expectancy confirmation, ideal-point expectancy value, and value-percept disparity. The latter was selected for this study and was operationalized in a manner suitable for uncovering need-gaps in the market that could lead to improvements in existing pet food products or to entirely new product concepts.

While it might have been desirable to measure wants prior to the last time a particular pet food was served (rather than asking respondents to recall these wants later), such an approach requires two survey waves. The cost of this would be considered prohibitive in most commercial studies. However, most owners feed their pets a single main meal each day, and wants should be relatively stable from one day to the next in this context. Also, the fact that Trawick and Swan (1980) have shown that inferred and perceived disconfirmation were highly related (r= .85) in a study of fabric cleaners lends credence to the approach used in this study. This finding supports that from an earlier study by Oliver (1979).

There is also the problem of reliability when measuring wants and gots at different points in time. Oliver (1977, 1980) has criticized this approach on the grounds that expectation and disconfirmation scores are not independent of each other (there is often a slight negative correlation). To circumvent this problem he recommended a single scale administered only after product use (Worse than expected Better than expected). Prakash and Lounsberry (1982) proposed that the problem of the correlation of expectation and discrepancy scores could be due to very low reliabilities of the before - after difference scores. In a study of their own they found these reliability coefficients to be .46 for fast food hamburger restaurants and .19 for beer. All of this suggests that measuring all discrepancies only after product/service use, as in the present study, may well be superior to obtaining separate before-after ratings.

It is important to note that this study did not attempt to compare the effectiveness of the three types of discrepancy scores (EC, IP, VPD) in terms of explaining customer satisfaction or identifying deficiencies. Even though all 3 types look similar, they tend to be used for different purposes and in different settings, but all of them can yield deficiency scores. Instead, the present study proposed a new approach to segmentation based on using only VPD discrepancy scores as a measure of unmet wants and needs. This approach appears to be especially appropriate for the objective of searching for new product/service ideas, but it is clearly applicable for other objectives as well.

Comparisons with Conventional Ideal Point Approaches

What do the discrepancy score approach discussed in this paper add to the conventional uses of an ideal point? Note that there are currently two such uses: 1) ideal point or vectors of importance overlaid on a perceptual map, 2) the ideal-point expectancy value model.

In the case of a perceptual map constructed at the individual level (using MDS or a related technology), attributes are usually represented by vectors rather than points (if they are represented at all). This makes it difficult to calculate attribute discrepancy distances directly. Also, these distances would have to be inferred based on a comparison of the distance of the ideal product/service from a specific attribute versus the distance of the same attribute from the brand used last or most often. This writer is not familiar with this or any other approach to measuring attribute discrepancies directly from perceptual maps that have been reported in the literature. Instead, discrepancies are usually measured in terms of the distance of each brand, taken as a whole, to the ideal brand.

In the case of the ideal-point expectancy value model, earlier in this paper this model was represented as one of the 3 alternative ways to measure attribute deficiencies directly. However, this model does not appear to have been used for this purpose in any studies reported in the literature, even though it could have been. Actually, very little of the research on multiattribute models has involved the use of a scaled ideal point for each attribute. One reason might be that some investigators have found it hard for respondents to specify any ideal point other than at the top of a scale. It may be that "wants" are considered by respondents to be more meaningful and easier to scale than "ideal" points, since the former are measures of intensity of feeling rather than some unrealistic, often unattainable point of perfection. The present study did not investigate these issues, however.

Deficiency Spaces

An extension of the work reported here would involve the construction of "deficiency spaces" based on the vectors of want-got discrepancy scores for each respondent. These would be input to a discriminant analysis program, which would produce some number of discriminant functions and canonical discriminant vectors. Each of the major brands could be positioned in terms of these functions, and the results could be represented in 2 or 3 dimensional space showing both brands and attributes. The result would be a deficiency space that looks very similar to a perceptual space created from brand descriptive (beliefs) ratings using discriminant analysis, but it would show the most prominent deficiencies associated with each of the major brands. A deficiency space might reveal a very different picture than a perceptual space, since there need be no real relationship between brand ratings and brand deficiencies at the attribute level. It might also be much more actionable.

It is important to note the constraints placed upon deficiency scores by the magnitude of beliefs (i.e., got) ratings: the maximum possible deficiency score depends upon the degree of the beliefs rating. Thus, a belief rating of 8 on a 1& point scale could result in a maximum positive deficiency score of 2, whereas a belief rating of I might have a deficiency of up to 9 (such a score is occasionally obtained in practice). Therefore, there should be somewhat negative correlations between beliefs ratings and deficiency scores across all respondents and brands.


Anderson, Ralph E (1913) "Consumer Dissatisfaction: The Effect of Disconfirmed Expectancy on Perceived Product Performance," Journal of Marketing Research, 10(Feb) 38-44.

Cardozo, Richard N. (1965) "An Experimental Study of Customer Effort, Expectations, and Satisfaction," Journal of Marketing Research, 2(August), 244-249.

Carroll, I. Douglas (1972), "Individual Differences and --Multidimensional Scaling," in Multidimensional Scaling: Theory and Applications in the Behavioral -. Sciences, eds. Shepard, R.N. Romney, A.K., and Merlove, S.; New York: Seminar Press, 105-155.

Cohen, Joel B. and Goldberg, Marvin E. (1970), "The Dissonance Model in Post-Decision Product Evaluation," Journal of Marketing Research, 7(August), 315-321.

Coombs, Cyde H. (1967), A Theory of Data, John Wiley & Sons, Inc., New York, N.Y.

Everitt, B.S. (1979), "Unresolved Problems in Cluster Analysis", Biometrics, 35,169-181.

Everitt, B.S. (1980), Cluster Analysis, 2nd Edition, London: Heineman Educational Books Ltd.

Green, Paul E. and Carmone, Frank I. (1970) Multidimensional Scaling and Related Techniques in Marketing Analysis, Allyn and Bacon, Inc., Boston, Mass.

Hempel, Donald J., "Consumer Satisfaction with the Home Buying Process: Conceptualization and Measurement" in Conceptualization and Measurement of Consumer Satisfaction and Dissatisfaction, H. Keith Hunt, ed., Cambridge, Mass: Proceeding of Conference conducted by Marketing Science Institute with support of National Science Foundation, 275-299.

Locke, Edwin A. (1967), "Relationship of Success and Expectation to Affect on Goal Seeking Tasks," Journal of Personality and Social Psychology, 7(2), 125-134.

Locke, Edwin A. (1969), "What is Job Satisfaction," Organizational Behavior and Human Performance, 4, 309-336.

Milligan, Glenn W. and Martha C. Cooper, "An Examination of Procedures for Determining the Number of Clusters in a Data Set", Psychometrika, V. 50, No. 2,159-179.

Morris, Earl W. (1976) "A Normative Deficit Approach to Consumer Satisfaction" in Conceptualization and Measurement of Consumer Satisfaction and Dissatisfaction, H. Keith Hunt, ed., Cambridge, Mass: Proceeding of Conference conducted by Marketing Science Institute with support of National Science Foundation, 240-274.

Myers, James H. and Tauber, Edward (1977), Market Structure Analysis, American Marketing Association, Chicago, II.

Myers, James H. (1976), "Benefit Structure Analysis: A New Tool for Product Planning," Journal of Marketing, 40(October) 23-32.

Oliver, Richard L. (1980), "A Cognitive Model of the Antecedents and Consequences of Satisfaction Decisions," Journal of Marketing Research, Vol. XVII, 460-9.

Oliver, Richard L. (1977), "Effect of Expectation and Disconfirmation on Post-exposure Product Evaluations: An Alternative Interpretation," Journal of Applied Psychology, 62 (August), 480-6.

Oliver, Richard (1979), "Product Satisfaction as a Function of Prior Expectation and Subsequent Disconfirmation: New Evidence," in New Dimensions of Consumer Satisfaction and Complaining Behavior, Ralph L. Day and H. Keith Hunt, eds., Bloomington, Ind: School of Business, Indiana University.

Oliver, Richard L. (1980), "Theoretical Bases of Consumer Satisfaction Research: Review, Critique, and Future Direction," in Theoretical Developments in Marketing, Charles W. Lamb, Jr. and Patrick M. Dunne, cd., Chicago: AMA Proceedings Series, 206-210.

Oliver, Richard L., and Linda, Gerald (1980), "Effect of Satisfaction and Its Antecedents on Consumer Preference and Intention," in Advances in Consumer Research, Vol. XIII, Kent B. Monroe, ed., Ann Arbor. Association for Consumer Research, 88-93.

Olshavsky, Richard N. and Miller, John A. (1972), "Consumer Expectations, Product Performance, and Perceived Product Quality," Journal of Marketing Research, 9(Feb), 19-21.

Prakash, Ved, and Lounsbury, John W. (1982), "A Reliability Problem in the Measurement of Disconfirmation of Expectations," in Advances in Consumer Research, Vol. X, Richard P. Bagozzi and Alice M. Tybout, ed., Ann Arbor Association for Consumer Research 244-249.

Pfaff, Martin, (1976), "The Index of Consumer Satisfaction: Measurement Problems and Opportunities," in Conceptualization and Measurement of Consumer Satisfaction and Dissatisfaction, H. Keith Hunt, ed., Cambridge, Mass: Proceeding of Conference conducted by Marketing Science Institute with support of National Science Foundation, 36-71.

Ries, at and Trout, Jack (1986), Marketing Warfare, McGraw-Hill Book Company, New York, N.Y.

Smith, Wendell (1956), "Product Differentiation and Market Segmentation as Alternative Marketing Strategies," Journal of Marketing, 21(July), 3-8.

Swan, John E., and Martin, Warren S. (1980), "Testing Comparison Level and Predictive Expectations Models of Satisfaction," in Advances in Consumer Research, Vol. VIII, Kent B. Monroe, ed., Ann Arbor Association for Consumer Research, 77-82.

Trawick, Fredrick, and Swan, John E. (1980), "Inferred and Perceived Disconfirmation," in Marketing In the 80's, Richard P. Bagozzi, et al, ed., Chicago: Educators' Conference Proceedings, Series No. 46, 97-100.

Westbrook, Robert A. and Oliver, Richard L. (1980), "Developing Better Measures of Consumer Satisfaction: Some Preliminary Results," in Advances in Consumer Research, Vol. VIII, Kent B. Monroe, ed., Ann Arbor Association for Consumer Research, 94-99.

Westbrook, Robert A. and Reilly, Michael D. (1982), "Value-Percept Disparity: An Alternative to the Disconfirmation of Expectations Theory of Consumer Satisfaction," in Advances in Consumer Research, Vol. X, Richard P. Bagozzi and Alice M. Tybout, ed., Ann Arbor: Association for Consumer Research, 256-261.

Wilkie, William L. (1971), "Market Segmentation Research: A Conceptual Analysis, "Institute for Research in the Behavioral, Economic and Management Sciences, Lafayette, Indiana: Paper No. 324, (September).

Wilkie, William L. and Pessemier, Edgar A. (1973), "Issues in Marketing's Use of Multi-Attribute Attitude Models," Journal of Marketing Research, 10(Nov) 428-441.

Wind, Yoram (1978), "Issues and Advances in Segmentation Research," Journal of Marketing Research, Vol. XV(August) 317-337.