A Multiple Discriminant Analysis Approach to the Development of Retail Store Positioning

ABSTRACT - This paper applies multiple discriminant analysis to a large-scale consumer data base, to provide an empirical analysis of the factors underlying store choice behavior for men's fashions. These findings define retail store positions in a particular marketplace and pinpoint the major competitive dimensions there.


Lawrence J. Ring and Charles W. King (1978) ,"A Multiple Discriminant Analysis Approach to the Development of Retail Store Positioning", in NA - Advances in Consumer Research Volume 05, eds. Kent Hunt, Ann Abor, MI : Association for Consumer Research, Pages: 227-234.

Advances in Consumer Research Volume 5, 1978      Pages 227-234


Lawrence J. Ring, University of Virginia

Charles W. King, Purdue University

[This study was made possible by funds provided by several Toronto men's wear retailers who wish to remain anonymous, and by research grants to the author from the Colgate Darden Graduate School of Business Administration and the Krannert Graduate School of Management.]


This paper applies multiple discriminant analysis to a large-scale consumer data base, to provide an empirical analysis of the factors underlying store choice behavior for men's fashions. These findings define retail store positions in a particular marketplace and pinpoint the major competitive dimensions there.


Understanding the clothing and apparel shopping behavior of the contemporary consumer is vitally important to effective marketing in the fashion retail marketplace. Each retailing marketplace is characterized by a wide range of stores and associated strategies and operating methods. The sum total of those strategic and tactical actions taken by a given store or chain of stores is the "product" or "image" which that store presents to the consuming public.

In general, the retailer's objective has been to create a store image which will appeal to a particular segment of the population. It has been widely held [1], [2], [3], and [4] that store "image" is produced by the joint effect of a multiplicity of store or product attributes. The major objective is to determine whether consumers perceive major differences in store appeal when faced with a choice among several (10) competing chains in a given market place. The aim here is to provide an empirical analysis of the factors underlying store choice behavior with respect to men's fashions and to contrast the strengths and weaknesses of several different retailing institutions in the city of Toronto, Canada.


The data used in this study were gathered as part of a broader on-going research program in Toronto, Ontario, Canada. That program is the Toronto Retail Fashion Market Segmentation Research Program and has been described elsewhere [5] and [6]. The sample used in the analysis here was part of a specially recruited panel of husband and wife pairs in Toronto. A total of 1025 usable questionnaires were received from cooperating male panel members. The questionnaires were mailed to each panel member complete with standard self-administration instructions. Each panel member received a small gift for completing the questionnaire.

In attempting to profile the Toronto men's wear retail marketplace, it was first necessary to define a relevant set of retailers to present to respondents for their judgments across a variety of attitudinal measures.

The 1975 edition of the Toronto Yellow Pages lists over 200 separate retailers under the title "Men's Clothing and Furnishings - Retail" alone -- not including department and discount stores. The vast majority of these stores, however, are relatively small and are single-outlet proprietorships. Therefore, it was expected that a majority of the population would be unfamiliar with stores falling into the single-outlet proprietorship category, and thus unable to comment knowledgeably about their operations.

There are, on the other hand, a small set of stores which most consumers could reasonably be expected to recognize. Toronto's three major department stores and many of its multiple-outlet men's wear chain stores would appear to have sufficient market coverage for respondents to judge them.

In addition, the mass merchandiser (department stores and specialty chain) retailers as a group appear to hold a major (and growing) share of Canadian retail men's wear business, according to the most recent government figures. In 1974, department stores and chain stores obtained 58 per cent of the sales of men's wear in Canada and had maintained an average growth rate for the past five years of 12 per cent dollar sales per year. In contrast, independent stores' dollar sales in men's wear grew only 6 per cent over the same period. [Source: Statistics Canada Documents 63-002 and 63-005, 1974.]

For the purposes of this project, ten large chains were mentioned by name (an "any other store" response category was also included to allow respondents to choose stores other than those which were specifically mentioned in the questionnaire).

The detailed store-by-store analysis dealt with fashion retailing as it relates to the major department and major specialty chain stores which are currently actively servicing the Toronto men's wear market. Each of the chains included in the analysis was operating at least six retail outlets in Toronto in 1975. In addition, it was estimated that each of the chains was obtaining gross men's wear sales in Toronto of at least $5 million dollars per year.

What was not examined in great detail were all the remaining smaller chains and the hundreds of small independent fashion specialty stores. Table 1 represents a comparison of consumers' perceptions of the sample set of chains described above with their perceptions of the aggregated remainder of the stores servicing the Toronto men's wear market.

This comparison reveals that the 10 specific sample chains studied here were well known and held, in the aggregate, dominant positions on all of the dimensions of retail patronage which were measured in the research. On the critical dimensions of "last store shopped" and "store shopped most often", the 10 sample chains were named by approximately 70 per cent of the respondents.


The initial analysis pass involved the computation of frequencies of mention across the sixteen measures of patronage which were listed in Table 1. The simple frequency analysis was useful for building a baseline statement of individual chain strengths and weaknesses.

Following the computation of frequencies, cross-classification analysis was performed on the determinants of patronage and other key variable sets using the share of shoppers, a surrogate for market share, as the dependent variable. The share of shoppers has historically been measured by market researchers in a variety of ways. In this project, two measures were employed:

1. Which store was visited the last time the respondent bought an article of men's wear?

2. Which store was shopped most often for men's wear?

The purpose of the cross-classification analysis was to pinpoint the strengths and weaknesses of each of the chains as perceived by their own customers for each of the patronage dimensions.

The cross-classification analysis was also useful in the development of decision rules for the grouping of the chains together for further analysis.

The final step in the analysis of the determinants of patronage in the Toronto retail men's wear market involved the construction of the "product space," that is, a spatial representation of the relative positioning of the ten chains included in this survey in Toronto.

This spatial representation was constructed through the use of multiple discriminant analysis. The mathematics of the multiple discriminant method are developed briefly later in the paper.

The objective here was somewhat strategic in nature. The technique permits the identification of potential opportunities in the existing retailing milieu. And, pragmatically, this technique can be useful in discovering directions in which any or all of the existing store images may be modified for greater sales and profitability.


Univariate cross-classification analysis was used to determine the relative importance of the patronage dimensions to each of the chains covered in this project. For each chain, the cross-classification tables were restructured so that the diagonal values from the cross-classification of each patronage determinant by "store last shopped" were ordered from highest to lowest per cent of mentions. A summary of the diagonals is presented in Table 2. For example, for Chain A [Chain names have been disguised for this analysis to protect the proprietary interest of those firms who sponsored the research. Chains A, B, and C are department stores; Chains D and E are discounter/mass merchandisers; and Chains F, G, H, I, and J are men's wear specialty chains.], ordering the data from Table 2 highest to lowest, among those who "last shopped" for men's wear at this chain ...

69 per cent said Chain A was "easiest to get to from home"

36 per cent said Chain A had the "best value for the money"

33 per cent said Chain A was the "best for conservative, everyday men's wear"

32 per cent said Chain A had the "most knowledgeable, helpful salesclerks"

27 per cent said Chain A had the "largest overall merchandise assortment/selection"

24 per cent said Chain A had the "most exciting merchandise display" 22 per cent said Chain A had the "lowest prices"

19 per cent said Chain A had the "best advertising"

16 per cent said Chain A was "best for current, up-to-date men's wear"

15 per cent said Chain A was "best for very latest, most fashionable men's wear"

12 per cent said Chain A had the "highest quality"

Assuming that those dimensions receiving the highest share of mentions for each chain (in this case, Chain A) were, in fact, the most important determinants of retail patronage for that chain, this ordered set of diagonals represented the "retail image" of each chain as seen by the chain's customers.





Looking at Chain A, for example, the highest percentage was 69 per cent, and it relates to location. This suggests that Chain A's location(s) was a major reason for shopping at this chain. Of lesser importance were "value", "conservative men's wear", and so forth. However, only on location did this chain appear particularly strong.

The univariate analysis of patronage determinants suggested that there were four kinds of chains serving the men's apparel market in Toronto. They were:

1. The discounter/mass merchandisers which appeared to be drawing their customers primarily by their low prices, but also on good value, location, and conservative everyday men's wear.

2. The department stores, which were obtaining their shoppers through good value and conservative, everyday men's wear also, but in addition because of their advertising and large overall merchandise displays.

3. The mid-range fashion specialty chains, which were drawing primarily on value, current, up-to-date men's wear, knowledgeable, helpful salesclerks, and exciting merchandise displays.

4. The high fashion specialty chains, which performed most strongly among their own shoppers on salesclerk help, high quality, very latest, most fashionable and current, up-to-date men's wear, and exciting display.

The question that remains revolves around what is the joint interaction of all the patronage dimensions taken as a whole? Therefore, the final step in the analysis of the structure of the Toronto retail men's wear market involved the construction of the "product space"; that is, a spatial representation of the relative positioning of the ten chains included in this survey in Toronto.


The objective in this exercise was somewhat strategic in nature. Perceptual mapping can be used as a tool for identifying potential opportunities for both the new men's wear retailer and the currently operating retailers. In particular, this technique can be useful to the existing retailer in discovering directions in which to modify his image to produce the greatest sales gain.

In recent years the construction of geometric spatial models of particular markets has received a great deal of attention in market segmentation analysis. And, a number of techniques and combinations of techniques have been employed in constructing these so-called "product" or "brand spaces". Among the most popular tools have been discriminant analysis [9] and [10], and factor analysis [11].

The data gathered for this analysis in the project was obtained through use of the associative techniques. Simply speaking, each respondent was asked to associate one chain with each of the following questions:

1. Which store is the easiest to get to from home?

2. Which store has the lowest prices?

3. Which store has the highest prices?

4. Which store has the most knowledgeable, helpful salesclerks?

5. Which store has the highest quality men's fashions?

6. Which store has the lowest quality men's fashions?

7. Which store gives you the best value for the money in men's fashions?

8. Which store gives you the worst value for the money in men's fashions?

9. At which store do you find the most exciting merchandise display of men's fashions?

10.Which store has thc best fashion advertising?

11. Which store has the largest overall merchandise assortment/selection in men's fashions?

12. Which store is best for conservative, everyday men's wear?

13. Which store is best for current, up-to-date men's wear?

14. Which store is the best for the very latest, most fashionable men's wear?

So, essentially each respondent picked "1 of n" (n=11) stores on each of the above dimensions. However, in the course of choosing one chain for each dimension and therefore ranking that chain as the best (or the worst, the largest, and so forth), each consumer assigned a rating to all the stores. That is, in the dummy variable sense, a "1" was assigned to the chosen store and a "0" was assigned to each of the other stores.

As a result, each respondent has rated each chain a "1" or a "0" on each dimension. This, in effect, constitutes approximately 1000 observations across the eleven store alternatives. Given this data, multiple discriminant analysis was chosen as the technique best suited for constructing a spatial model of the Toronto men's wear market.

In the Toronto data, respondents rated a set of eleven men's wear chains on fourteen determinants of patronage. The fourteen dimensions have been considered as the reasons taken together which account for the ways in which Toronto shoppers think men's wear chains differ from one another. The objective is to determine how these "reasons" discriminate among the set of chains.


The results of the discriminant analysis were very consistent with the univariate analysis results presented earlier in this paper. Table 3 reports the structure for the eight most significant discriminant functions and the canonical correlations between the 14 determinants of patronage and the 11 group (chain) membership variables.

The structure coefficients in this table suggest that the first eight discriminant functions appear to be measuring:

1. Low quality/low price (+) versus high price (-)

2. Conservative men's wear with large assortment and advertising (+) versus very fashionable men's wear with high quality and high price (-)

3. High price, large assortment, best advertising (-)

4. Location and high price (-)

5. Worst value and location (+)

6. Very latest, most fashionable merchandise (+)

7. Low price (+) versus low quality (-)

8. Best value for the money (+)

Table 4 indicates that the first two functions accounted for approximately 83 per cent of the discrimination among the 11 Toronto chains. Additional interpretation can be obtained by locating the 11 group (chain) centroids on the discriminant functions.

Figure 1 displays both the group centroids and the determinants of patronage on the first two discriminant functions, and Table 6 presents the locations of the group centroids on the first eight discriminant functions. Here, and in Table 3, the discriminant functions have been standardized in order to adjust for the variability in the original variables and to therefore accurately report the relative importance of the variables.

Figure 1 shows the perceptual map of the Toronto men's wear market in two dimensional form. The location of each store is indicated on these two dimensions. The horizontal dimension seems to contrast low quality and low price on the right with high price on the left. The vertical dimension appears to suggest conservative men's wear with a large assortment and advertising at the top in contrast to very fashionable, high quality, high priced men's wear at the bottom.

The mean rating of each chain on each determinant can be determined by erecting perpendiculars from each centroid to each "determinant vector." For example, Chain H is viewed as the chain with the highest prices, followed by Chain I, "others", Chain G, Chain A, Chain J, Chain F, Chain B, Chain C, Chain E, and Chain D.

Figure 2 displays the territorial map of the Toronto retail men's wear market. In this figure, all space on the graph has been classified according to the store centroid to which it is closest. The territorial map clearly suggest that there are three major chain-type divisions in the Toronto market, the department stores, the specialty stores, and the discounters.

The discriminant analysis indicated that in Toronto the major competitive dimensions of the discounter/mass merchandisers were low quality, low price, and worst value for the money. The fashion specialty chains did best on high price, most fashionable men's wear, high quality, current men's wear, exciting merchandise display, and knowledgeable, helpful salesclerks. The department stores showed strongest on best assortment/selection, best advertising, best conservative everyday men's wear, best location, and best value for the money.














Leonard J. Berry, "The Components of Department Store Image: A Theoretical and Empirical Analysis", Journal of Retailing, 45 (Spring, 1969), 3-20.

Irving Burstiner, "A Three-Way Mirror: Comparative Images of the Clienteles of Macy's, Bloomingdale's, and Korvette's." Journal of Retailing, 50 (Spring, 1974), 24-37.

Theodore Clerenger, Jr., Gilbert A. Lazier, and Margaret Clark, "Measurement of Corporate Images by the Semantic Differential." Journal of Marketing Research, 2 (February, 1965), 80-82.

Robert J. Dolan, "Market Segmentation Via Alternative Discriminant Procedures," in The 1975 Combined Proceedings of the American Marketing Association, ed. by E. M. Mazze, (The American Marketing Association, 1975), 132-136.

Richard M. Johnson, "Market Segmentation: A Strategic Management Tool." Journal of Marketing Research, 8 (February, 1971), 13-18.

Charles W. King and Lawrence J. Ring, "The 1975 Toronto Male Fashion Research Project: A Descriptive Overview." Institute for Research in the Behavioral, Economic, and Management Sciences, Purdue University, No. 526. (Lafayette, Indiana: Krannert Graduate School of Industrial Administration, Purdue University, September, 1975).

Joseph B. Kruskal, "Nonmetric Multidimensional Scaling: A Numerical Method." Psychometrika, 9 (June, 1964), 115-29.

Eleanor G. May, "Management Applications of Retail Image Research", Marketing Science Institute Working Paper. The Marketing Science Institute, Cambridge, Massachusetts, 1973).

Lawrence J. Ring, Douglas J. Tigert, and Charles W. King, "Fashion Adoption, Retail Image, and Lifestyle: An Integrated Research Program", a paper presented before the 83rd Annual Convention of the American Psychological Association, Division 23, Chicago, Illinois, (September, 1975).

Ricardo L. Singson, "Multidimensional Scaling Analysis of Store Image and Shopping Behavior." Journal of Retailing, 50 (Summer, 1975), 38-52.

Douglas J. Tigert, The Changing Structure of Retailing in Europe and North America: Challenges and Opportunities, a monograph. (The University of Toronto Press, January, 1974).



Lawrence J. Ring, University of Virginia
Charles W. King, Purdue University


NA - Advances in Consumer Research Volume 05 | 1978

Share Proceeding

Featured papers

See More


Nostalgiacising: A Performative Theory of Nostalgic Consumption

Ela Veresiu, York University, Canada
Ana Babic Rosario, University of Denver
Thomas Derek Robinson, City University of London, UK

Read More


Gaze Reflects Loss Aversion

Feng Sheng, University of Pennsylvania, USA
Arjun Ramakrishnan, University of Pennsylvania, USA
Darsol Seok, University of Pennsylvania, USA
Puti Cen, University of Pennsylvania, USA
Michael Platt, University of Pennsylvania, USA

Read More


G2. You Run When Time Flies: Time Metaphors Affect Inferences from the Speed of Time

N. Alican Mecit, HEC Paris, France
L. J. Shrum, HEC Paris, France
tina m. lowrey, HEC Paris, France

Read More

Engage with Us

Becoming an Association for Consumer Research member is simple. Membership in ACR is relatively inexpensive, but brings significant benefits to its members.