Segmentation of Complex Markets: Identification of Perceptual Points of View

Jonathan Gutman, University of Southern California
Thomas J. Reynolds, University of Texas, Dallas
Scott D. Alden, University of Southern California
ABSTRACT - This study demonstrates the use of sorting tasks in studying complex market structures. More importantly, a method for segmenting respondents into homogeneous subsets based upon their sorting data is shown. Group product spaces for respondents who used two and three categories to sort cereals were created. After segmenting respondents, product spaces were created for each subset of respondents at each level. Differences between the overall spaces and the segmented spaces are discussed.
[ to cite ]:
Jonathan Gutman, Thomas J. Reynolds, and Scott D. Alden (1982) ,"Segmentation of Complex Markets: Identification of Perceptual Points of View", in NA - Advances in Consumer Research Volume 09, eds. Andrew Mitchell, Ann Abor, MI : Association for Consumer Research, Pages: 392-397.

Advances in Consumer Research Volume 9, 1982      Pages 392-397


Jonathan Gutman, University of Southern California

Thomas J. Reynolds, University of Texas, Dallas

Scott D. Alden, University of Southern California


This study demonstrates the use of sorting tasks in studying complex market structures. More importantly, a method for segmenting respondents into homogeneous subsets based upon their sorting data is shown. Group product spaces for respondents who used two and three categories to sort cereals were created. After segmenting respondents, product spaces were created for each subset of respondents at each level. Differences between the overall spaces and the segmented spaces are discussed.


Research investigations to determine the structure of markets where a great many products exist are limited by respondents' knowledge of the vast array of products as well as their ability to cope with extremely lengthy judgment tasks (Rao and Katz, 1971); e.g., rating many products with respect to preference and other relevant dimensions Prime examples of this type of saturated market include cigarettes and cereals.

Market structure has been conceptualized as a multi-level hierarchical partitioning of items or brands (Rao and Sabavala, 1980) It is hierarchical because more broadly defined partitions at the top of the structure subsume more narrowly defined partitions at the bottom of the hierarchy In the studies of market structure, various researchers have defined product markets ranging from interbrand competition (Butler and Butler, 1970, 1971) to competition among widely divergent alternatives (Moran, 1973).

Knowledge of market structure can be developed from two perspectives: product based or demand based. Manufacturers, perhaps by tradition, often use obvious product based physical characteristics to create categories of products within a product class. In the case of cigarettes, a few critical physical dimensions come to mind--length, filter vs non-filter, amount of toxins, menthol vs regular--which permit the product category to be broken into relatively few combinations. It is posited that consumers' preferences, indeed even broad switching behavior, can then be positioned with respect to this manufacturer-oriented classification system.

For a market like cereals, though, the problem of a manufacturer developing a physical classification system could become much more involved given the great number of physical characteristics upon which products differ; e.g., types of grains or the possible combination of grains, sugar content, fiber content, hot vs. cold, nutritional value, or artificial vs. natural nutrition. Thus, attaining a total perspective of the cereal marketplace would involve constructing a tremendous number of categories reflecting the possible combinations of physical characteristics which might then be used as the template for studying market structure.

Comparing preferences or behavioral measures to these physical classifications may serve to help the marketer deal- more effectively with the conceptual framework within each of the segments in the marketplace. The most desirable alternative, though, would be to go beyond these physical classifications to develop a perceptual framework corresponding to the way in which consumers tend to organize brands; i.e., permitting the segmentation to include benefits or even higher order value orientations. Essentially what is desired is the identification of key grouping variables with respect to overall common perceptions. The ways in which consumers group and classify products need not be limited to physical characteristics and, indeed, most likely go beyond them. Thus, such an interpretation yields the structured demand-related components of the market from which a segmentation may then be developed. Rao and Sabavala (1980) have provided a partial listing of demand-based methods for market structure analysis; Day, Shocker, and Srivastava (1979) have provided a discussion of several of these approaches (see also, Gutman, 1981).

Perceptual segmentation typically involves a methodology from which distances between products can be derived either directly or by inference. These distances are then the input for multidimensional scaling or cluster programs or a combination of the two. Obviously, the construction of direct distances is forestalled in the case of markets that have a great number of products or brands In markets that also have a tremendous number of basic classifications, obtaining the relation among possible categories becomes virtually impossible--particularly if benefits are included. Thus, to gain insight into the perceptual nature of a complex market, a data gathering methodology is required that effectively deals with the practical limitation imposed by large stimulus sets as well as one that permits quantitative analysis that leads to direct interpretations.


The purpose of this study is to demonstrate the efficacy of single sorting methods for studying product markets. Their ease of use, quickness, and the relation between the demands of the task and the abilities of respondents to structure stimuli make sorting tasks potentially viable methods for studying large data sets

Other researchers (most notably Rao and Katz, 1971; and Bourgeois, Haines, and Summers, 1980) have used sorting tasks to study large numbers of brands to determine how markets are structured and how different analytical techniques recover those structures. The present study replicates some aspects of the Bourgeois, Haines, and Summers study in that it employs a sorting task. It differs, though, in two respects: (1) it uses a different measure of item similarity (to be discussed in the methodology section); and (2) it develops a measure of person similarity from the sorting task that can be used to segment respondents, thus giving a clearer picture of market structure than would be otherwise obtained.

Due to the complexity of the market, cereals were chosen for experimental study. To investigate the performance of the task, comparisons of the group spaces before and after segmenting respondents for the similarity of their sorts was made It was expected that the resulting product space for the subgroups would yield information that could not be derived from the aggregate spaces, thereby leading to a more informative perceptual segmentation of the Product category


As part of a larger experiment (Gutman, 1980), respondents were given 18 breakfast cereals to sort (each one on a slip of paper) into as many categories as they felt they needed. They were instructed to place cereaLs into groups such that cereals in the same groups were similar and thus, different from cereals in other groups. The respondents recorded the code letters of the cereals by group on their questionnaires

Subjects were students in introductory psychology classes at the University of Southern California All respondents spoke English as their native language, thereby reducing language as a source of bias and increasing the likelihood of familiarity with all the cereals. They participated in the experiment as part of their course requirements. The sorting task was administered to two hundred respondents in five groups. After omitting respondents who were not familiar with all the cereals 170 respondents remained.


Number of Categories Used In Sorting Cereals

A summary of the number of categories used in sorting the cereals is shown in Table I. As can be seen in the table, most respondents used between two and six categories. Since the primary purpose of this study is to demonstrate the feasibility and ultimate value of segmenting respondents based upon their sorts, only the two- and three-group respondent's data will be subjected to analysis





Intercereal Distances

Table 2 shows the data for the twenty respondents who used two groups to sort the cereals. The 1's and 2's are merely nominal labels given to the groups and imply no ordering. Following Rosenberg and Kim (1975), a cereal-by-cereal matrix of product distances was formed by calculating a disagreement score (sij) for each pair of cereals by counting the number of respondents in a group who put the two cereals (i and j) into two different categories A distance measure, (aij), was defined as follows:

EQUATION, where K is the number of cereals.

"The aij measure is essentially the s-measure squared plus a measure of the degree to which two terms do not co-occur indirectly. An indirect co-occurrence for two terms i and j refers to an instance in which i and k co-occur in one respondent's sorting, j and k occur in another respondent's sorting, but i and j do not co-occur in either sorting" (Rosenberg and Kim, p. 492).

Rosenberg, Nelson, and Vivekananthan (1968) use an analogy to explain an indirect co-occurrence. "Consider a social distance measure between two persons which is inversely related to how often they interact directly with each other and how often they interact with the same other individuals, whether or not they do so on the same occasions" (p 285). Therefore, two people are more similar to each other if they have the same friends than if they don't. And, two products are more similar to each other if they are judged to be similar to the same products than if they are not.

This calculation of indirect co-occurrences has the potential to create a continuum of scores with many more separate data points than the number of respondents The distance measure suggested by Rosenberg, Nelson, and Vivekananthan seems to have much potential in deriving distance measures from free-sorting data. As yet, it has not had much use in marketing research studies

Market Structure For Cereals

The cereal distance data so derived were submitted to a multidimensional scaling program (Young, 1972) and to a hierarchical clustering program (UCLA Health Sciences Computing Facility, 1977). The resulting two-dimensional maps with cereal clusters overlaid upon them are shown for the two- and three-category respondents in Figures 1 and 2 (the low Kruskal stress values of 03 and .05 suggest that two dimensions are adequate to represent the task). In looking at the two figures, it can readily be seen that the two-category map yielded a "kids" cereal cluster in the upper right and what appears to be a "health" cereal cluster in the lower left, with the remainder of the cereals in between The three-category respondent map seems to break the "health" cereals as shown in the two-category map into another segment termed "natural "

Since each map has one more group in it than respondents used in their sorts, it is obvious that different types of sorts are being combined in the aggregate distance measure. Therefore, it appears advisable to segment the respondents into homogeneous groups to reflect the true nature of their sorts.



Segmentation of Respondents

To segment the respondents, one must translate the concept of categorical membership derived from sorting to operational terms Stated most simply, a category is formed whenever two objects are grouped together (Mervis and Rosch, 1981). This concept can be directly converted into a distance measure by counting the number of times any two respondents put the same two cereals in the same category If all 153 pairs (18 x 17/2) of the 18 cereals were arrayed in a vector, each respondent could receive a "1" for each pair of cereals that they placed in the same category and a "0" if they were not categorized together. Two respondents' vectors of scores on these pairs could be crossmultiplied to yield a measure of "strong matches" (1,1's). These could then be summed to yield a similarity index across respondents.

However, a respondent with a large number of cereals in one group and a small number in the other(s) would be similar to most other respondents Therefore, it becomes necessary to divide this sum by the number of l's in each person's vector and to take the smallest fraction as the index of interperson similarity.

For example, Persons 1 and 2 in Table 2 have 63 l,l's-instances where they have put two cereals in the same categories. However, Person 1 has more potential for matches with any random sort as he has put 15 cereals in one category md 3 in the other category (15 x 14/2) + 3 x 2/2 = 111). The calculation for Person 2 is (13 x 12/2 + 5 x 4/2 = 88). Since 63/111 = .567 and 63/88 = .716, the number used to represent the similarity of Person 1 and 2's sorts should be .567 as shown for cell 2,1 in Table 3. (Table 3 contains the lowest of each set of respondent's fractional indexes).

Market Structure by Respondent Segment

The interpersonal similarity data were submitted to the hierarchical clustering program. The score of 100 for respondents 1 and 14, 3 and 4, and 9 and 18 indicates those respondents had identical sorts

Nine respondents (1, 14, 119 5, 3, 4, 10, 16, and 13) and six respondents (6. 17. 9. 18. 7. and 12) were defined as the two major respondent segments. Respondents 15, 8, 2, 19, and 20 were omitted from either segment. The same process for generating interstimuli distances as was used with the aggregate data was used for each segment. The resulting cereal maps are shown in Figure 3 for the two-category respondents and in Figure 4 for the three-category respondent segments.





As can be seen, the basic difference between the two maps for the two-category respondents is in the latitude of the "kids" cereal group. The map shown in Part A of Figure 3 shows a narrowly defined group with only the "clearly" defined children's cereals as members (Cap'n Crunch, Sugar Frosted Flakes, and Sugar Crisp). The map on the right (Part B) adds Corn Flakes, Rice Krispies, Cheerios, and to a much lesser extent, Raisin Bran and Wheaties to this set.

For the three-category respondents, the "kids" cereal group is virtually identical to those for each of the segments for the two-category respondents, however, the adult group is divided differently. For the subgroup with the narrowly defined "kids" cereal category (1 =9), the adult group is divided into "regular adult" vs. "natural" cereals (see Part A of Figure 4). For the subset with the broadly defined "kids" cereals (n=7), the adult cereals are divided into "natural" and "health" categories (see Part B of Figure 4)






This study has demonstrated the usefulness of sorting tasks and how data gathered using this methodology can be converted to distance data. While this in itself may not be new, it is hoped that the possibilities for the segmentation of complex markets will increase the use of these techniques.

Creating homogeneous subsets of respondents clearly demonstrates the difference in the ways in which consumers structure the large number of entries that comprise many product classes. Furthermore, the sorting task shown allows respondents to determine the number of categories they feel are needed to encompass the relevant differences in the market. Additionally, the relations between consumers using different number of categories can be ascertained.

In this study, the key distinction was the size of the "kids" cereal category. Respondents made varying distinctions among the remaining cereals with those respondents using more categories dividing the "adult" cereals into finer groupings.

To marketers, critical distinctions relate to usage rate or circumstances related to consumption. The nature of consumers categorical systems reveals to marketers the basis upon which they make judgments about items in the product class. If consumers have a highly inclusive definition of "kids" cereals, and this is a "don't use" category, this information is critical to marketers. Similarly, if the "adult" cereal category has many sub-categories based on many narrow distinctions, a much tighter product-positioning is called for than if the adult category was divided into few subcategories.

Lastly, the hierarchical nature of market structure suggests that demand-based systems using consumer-supplied categories can tell marketers how narrow distinctions at the bottom of the hierarchy are related to broad distinctions at the top of the hierarchy. These connections can tell marketers more than either element alone does.


Bourgeois, Jacques C., Haines, George H. Jr., and Summers, Montrose S. (1979), "Defining an Industry," Market Measurement and Analysis Proceedings of ORSA/Tims Special Interest Conference, (eds.) D. B. Mongomery and D. R. Wittink, (Cambridge. MA: Marketing Science Institute, pp. 120-133).

Butler, B. Jr., and Butler, D. H. (1970 and 1971), "Hendro-dynamics: Fundamental Laws of Consumer Dynamics," Croton-on-Hudson, New York: Hendry Corporation, Chapter 1 and Chapter 2.

Day, George S., Shocker, Allen D., and Srivastava, Rajendra K. (1979), "Consumer-Oriented Approaches to Identifying Product Markets," Journal of Marketing, 43, pp. 8-19.

Gutman, Jonathan (1981), "A Means-End Model for Facilitating Analysis of Product Markets Based on Consumer Judgment," Advances in Consumer Research, (ed.) Rent B. Monroe, Vol, VIII,(Ann Arbor: Association for Consumer Research, pp. 116-121).

Gutman, Jonathan (1980), "Equivalence Range in Categorization," in Advances in Consumer Research, (ed.) Jerry C. Olson, Vol. VII,(Ann Arbor: Association for Consumer Research, pp. 411-416).

Mervis, Carolyn B., and Rosch, Eleanor (1981), "Categorization of Natural Objects," in Annual Review of Psychology, (eds.) Mark R. Rosenzweig and Lyman W. Porter, Vol. 32, (Palo Alto: Annual Reviews, Inc., pp. 89-115).

Moran, W. R. (April 1973), "Why New Products Fail," Journal of Advertising Research, 13, pp. 5-13.

Rao, Vithala R., and Katz, Ralph (November 1971), Alternative Multidimensional Scaling Methods for Large Stimulus Sets," Journal of Marketing Research, Vol. VIII, pp. 488-494.

Rao, Vithala R., and Sabavala, Darios Jal (1980), "Methods of Market Structure/Partitioning Analysis Using Panel Data," Market Measurement And Analysis. Proceedings of the ORSA/Tims Special Interest Conference, (eds.) D. B. Montgomery and D. R. Wittink,(Cambridge, MA: Marketing Science Institute, pp. 439-454).

Rosenberg, Seymour, and Kim, Moonja Park (October 1975), "The Method of Sorting as a Data-Gathering Procedure in Multivariate Research," Multivariate Behavioral Research, 10, pp. 485-502.

Rosenberg, Seymour, Nelson, Carnot, and Vivekananthan, P. S. (1968), "A Multidimensional Approach to the Structure of Personality Impressions," Journal of Personality and Social Psychology, Vol. 9, Number 4, pp. 283-294.

UCLA Health Sciences Computing Facility (1977), Biomedical Computer Programs P-Series, (eds.) W. J. Dixon and M. B. Brown, (Los Angeles: University of California).

Young, Forrest W. (1972), POLYCON Users Manual. A Fortran IV Program for Polynomial Conjoint Analysis, (Chapel Hill: L. L. Thustone Psychometric Laboratory, Report No. 104.