# Determinants of Retail Patronage: an Examination of Alternative Models

ABSTRACT - Although site selection has always been a part of retailing decision making, the selection of new sites has become both more important, and challenging as retailers face the unique demands of today's complex environment. For example, the reduction in the number of new regional malls being constructed is in large part due to constraints within capital markets. Over stored major metropolitan areas have prompted retail chains to enter smaller communities as a means of maintaining growth. Other retailers such as K-Mart strive to improve sales growth through the productivity increases resulting from upgrading product presentation, expanding or contracting merchandise lines and assortments, and emphasizing store atmosphere and decor. The purpose of this paper is to identify the dimensions of retail patronage and to develop predictive models to explain the relative importance of location and store attribute variables in predicting purchase frequency:

##### Citation:

*
J. Patrick Kelly and Scott M. Smith (1983) ,"Determinants of Retail Patronage: an Examination of Alternative Models", in NA - Advances in Consumer Research Volume 10, eds. Richard P. Bagozzi and Alice M. Tybout, Ann Abor, MI : Association for Consumer Research, Pages: 345-350.
*

[This research was funded by a grant from the Skaggs Institute of Retail Management, Brigham Young University.]

Although site selection has always been a part of retailing decision making, the selection of new sites has become both more important, and challenging as retailers face the unique demands of today's complex environment. For example, the reduction in the number of new regional malls being constructed is in large part due to constraints within capital markets. Over stored major metropolitan areas have prompted retail chains to enter smaller communities as a means of maintaining growth. Other retailers such as K-Mart strive to improve sales growth through the productivity increases resulting from upgrading product presentation, expanding or contracting merchandise lines and assortments, and emphasizing store atmosphere and decor. The purpose of this paper is to identify the dimensions of retail patronage and to develop predictive models to explain the relative importance of location and store attribute variables in predicting purchase frequency:

INTRODUCTION

Two scenarios currently dominate retail growth strategy formulation. Each scenario focuses on location or store attribute dimensions. For example, Southland Corporation focuses on expansion of its 7-11 stores to new sites within existing markets such that shopper convenience and proximity are increased.

Alternatively, improvements in the quality and variety or merchandise, speed of service, and courtesy of employees within the existing stores could be improved.

Travel time or distance to a retail outlet is known to have a strong influence on the probability of purchasing at a giver. store location. The greater the distance, the lower the probability of purchasing. Using this approach, market potential could be determined by identifying patron concentration within a specific geographic area and which contained a specific probability of purchasing. This descriptive approach -to estimating market potential has been well researched (Huff, 1966; Huff and Blue, 1966; Nelson, 1958; Nakanishi and Cooper, 1974). However other attributes can mediate the willingness to purchase at a given store. The variety of merchandise at one location may cause patronage patterns to shift to more distant stores touting a greater variety. Store atmosphere and decor similarly influence purchasing behavior (Stanley and Sewall, 1976). It is the relative importance of the distance and store attribute dimensions describing alternative branch store locations and predicting purchase frequency that is at issue.

The purpose of this is to investigate and test as alternative predictors of purchase frequency: (1) a distance model; (9) a store attribute model; (3) a combined model of distance and store attribute dimensions, and; (4) a combined model employing orthogonal combinations (factor analysis dimensions) of the distance and store attribute dimensions that are derived from a principal components analysis.

To more fully describe these models, in model I, distance measures are entered into model I to determine how consistently distance measures predicted purchase frequency for patrons of each of six different stores. Model II is similar to model I excepting that store attribute dimensions were substituted for distance measures. Model III predicts purchase frequency using both the distance and store attribute dimensions. Model IV employs the results of a principal components 'actor analysis, where each of the distance and store attribute factors produced by the analysis contain common sets of variables which are correlated with some underlying dimension of retail patronage. The frequency of patronage is Predicted using these underlying dimensions.

METHODOLOGY

The focus of this study is a major metropolitan area within the western United States, Six branches of a a regional thrift store are located in close proximity to each other within this market. The maximum distance between any of the six stores was 15 miles, with each of the six stores strategically located along a major thoroughfare for easy customer access.

Approximately one hundred fifty patrons were surveyed in each of the six stores, resulting in a total of 880 completed interviews. Patrons reported a variety of information, including distance measures to the store from their place of residence (the reported measures were transferred into city-block and euclidean distance measures).

The city block metric measure is the most accurate measure of actual driving distance from home to the store. It is computed as the sum of vertical and horizontal distances within a cartesian plane. The uniformly rectangular city blocks that were characteristic of the community of study make the city block metric measure ideal for the problem at hand.

In contrast to the city block metric, the euclidean distance measure is in the form of straight line/shortest path distances from the store location to the residence. The euclidean distance measure is, in the literature. assumed analogous to perceptual distances.

Each of the store location and customer residence variables were recorded by cartesian coordinates on a gridded 40 mile by 45 mile map of the metropolitan area. These grit coordinates were used to determine travel distances to each store.

Patronage measures also included overall qualitative evaluations of the store, including evaluative ratings of quality and variety of merchandise available, courtesy and helpfulness of employees, price, speed of service, cleanliness, atmosphere and decor. One additional question asked for the customers' shopping frequency at the given store. Customers who visited the store once or more weekly were considered weekly shoppers. Those shopping once a month but less than 4 times were categorized as monthly shoppers and those shopping more than once a year but less than 12 times were categorized as yearly shoppers. A final category for shopping frequency was the first time shopper at each store.

By combining the response categories, the prediction of purchase frequency using both distance and attributes of a given retail outlet was conducted using two purchase frequency groups: those shopping once or more weekly, and those shopping less than four times per month.

First time shoppers were included in the latter group. for each of the four models tested, the weekly and less frequent shopper groups were differentiated using a two-way multiple discriminant analysis of the form:

D_{k} = d_{k1}z_{2} + d_{k2}z_{2} + ... + d_{kp}z_{p} (1)

where

D_{k} is the score derived from the kth discriminant function, where 1 < k < p-1, and p is the number of independent variables.

d_{kp} is the pth coefficient or discriminant weight obtained for the kth discriminant function.

z_{p} is the standardized value of the pth independent (Distance of store attribute) variable used in the analysis.

For the two group case, the discriminant weights dk are a linear trans formation or the unstandardized weights of a multiple regression analysis. For the standardized case, the discriminant weights are identical in both discriminant and multiple regression analyses.

The linear discriminant functions that produce the additive linear profile from distance and store attribute variables are first used to predict purchase frequency.

MODEL I: DISCRIMINANT AND CLASSIFICATION ANALYSIS RESULTS PURCHASE FREQUENCY BY DISTANCES

FINDINGS

Model I: Distance Analysis

The prediction of purchase frequency using the distance measures shows that shoppers who purchase weekly or more frequently, were correctly classified 70 - 80% of the time (see Table 1). It was further observed that the euclidean distances out performed the city block distances even though the city block distance is much more representative of actual driving patterns.

The standardized discriminant coefficients computed for the distance variables were observed to form a consistent pattern across both the aggregate analysis for the six stores, and the analysis of the individual stores. The standardized discriminant coefficients for the euclidean distances were observed to be near 1.00 for all stores and the aggregate analysis. Although the euclidean measure was most important in predicting purchase frequency, the city block measure showed little pattern across stores, and was of little consequence. It was entered in only-three of the six discriminant functions. Table s.x, the confusion matrix results does however show that overall, the distance function predicts high frequency purchase very well.

Model II: Store Attributes

In Model II, shopper's overall qualitative evaluations of the store were found to be related to purchase frequency (see Table 2). Evaluation included merchandise quality, variety and price, courtesy and speed of service, and the cleanliness and atmosphere and decor of the store.

The multivariate analysis conducted for the group of six stores showed that the quality of merchandise was the most important variable in predicting purchase frequency. Employee courtesy and store cleanliness were the next most important predictors, with store atmosphere and decor being the least important predictors.

The univariate analysis of the store attributes showed that merchandise quality, employee courtesy, and price were in general, the most important predictors of purchase frequency. These variables entered several of the six individual store models. The true disposition of this importance may however be questioned in that the standardized discriminant coefficients appear to be unstable in both magnitude and direction. This instability is evidenced in Table 6, where the percentage of shoppers correctly classified as weekly shoppers and less frequent shoppers is presented. In almost all cases, the store attribute measures did not predict weekly purchase frequency as well as did the distance measures alone. This finding is of great managerial importance in that patronage by the most frequent 'shopper best predicted by proximity, but not store attributes. Restated, tertiary markets are perhaps better appealed -o by a more precise development of the store attribute offering.

Model III: Combined Distance 2nd Attribute Measures

The multivariate and univariate discriminant models which predict purchase frequency using the combination of attribute and distance measures, are found in Table 3. The multivariate analysis conducted for the group of six stores shows that distance is more important in predicting high frequency of purchase than are any of the store evaluation attributes. This finding is consistent with the previous analyses where the importance of distance measures was consistently observed.

MODEL II: DISCRIMINANT AND CLASSIFICATION ANALYSIS RESULTS PURCHASE FREQUENCY BY STORE ATTRIBUTES

Model IV: Factor Analytic Model

The testing of model IV consists of a two-stage analysis. Stage one of the analysis refines the nine distance and store attribute measures into a smaller set of determinant factors or dimensions that predict store patronage. A set or five factors were identified which maintained approximately 90% of the total explanatory power of the original nine variables (Table 4). The score of each patron on each of the five factors were also obtained.

Stage two of the analysis evaluates the ability of the five patronage variable factors to predict purchase frequency. Table 5 presents the results of the discriminant analysis. The factor dimension modeL did not predict as well as the combined distance and attribute model. However, given the aggregate nature of the factors, prediction might be better than is expected. It is important to note that model IV evaluates the dimensions of patronage not specific patronage measures.

In summary of the results of the four models, it is apparent from Table 6 that the combination of distance and attitude measures leads to an interesting pattern of results. The combination distance and attitude measures increase the total percentage of correct classification over that observed for distance or store attribute models alone. This increase in percentage of correct classification is, however, due to an increase in the correct classification of the less frequent purchaser. That is, the distance model alone predicts the patronage of the high frequency shopper who shops the store on a weekly or more often basis. However, the combined distance and store attribute model is superior for predicting patronage of the shoppers who visits the store less frequently.

Implications and Summary

The dialog evaluating the relative merits of distance and attribute models in making site selection decisions is not new to the marketing literature.

The focal point of this study is to assess the relative value of distance measures as compared to psychological measures in predicting purchase frequency. The findings of the study indicate that the distance model predicts frequent patronage better than does the store attribute model, or a combined distance and store attribute model. This finding is of considerable importance in that it suggests that distance between the store and place of residence is the most critical. factor explaining high frequency shopping behavior. Expressed in terms of the basic classification of shopping behavior, minimal shopping distance is responsible for patronage of the convenience and emergency good shopping within a given store. This finding is both intuitive, appealing, and important in that the superiority of distance measures to image measures derived from attitudes about store attributes has also been demonstrated. The second major finding of importance was that less frequent shopping behavior was best predicted using a combination distance and store attribute model. This combination model decreased the correct classification of the high purchase frequency shopper.

These results suggest each store should identify its strengths and weaknesses. Attempts should be made to improve upon the weaknesses and to use the strengths in promotional efforts where possible. If more frequent shoppers can be identified as rating a store higher on certain attributes this suggests the attributes may have influenced shopping frequency. Each retail store should then be concerned with those factors which are used to predict shopping frequency.

The importance of each dimension was further evaluated using factor analysis. This approach was used to predict each dimension importance when combined with all the other dimensions. For all stores, distance was again the best predictor of patronage by customers. Distance was followed by merchandise quality and variety, and finally by courtesy and speed of service. The other dimensions were not significant in the equation for all stores. Distance was most important for stores #2, 3, 4, and 5. For store #1 courtesy and speed of service were most important and in store #6 customers indicated merchandise quality and variety were most important.

Distance appears to be the single most important dimension in predicting store purchasing frequency in the study described in this paper. But because merchandise quality and variety, employee characteristics such as courtesy and speed of service, as well as atmosphere and decor and prices explain half of the variance in patronage frequency, these dimensions

VARIMAX ROTATED RACTOR LOADINGS

PERCENT CORRECT CLASSIFICATION COMPARISON OF DISCRIMINANT MODELS

REFERENCES

D. Huff, "A Programmed Solution for Approximating an Optimal Retail Location, Land Economics, (August, 1966)

D. Huff and L. Blue, "A Programmed Solution for Estimating Retail Sales Potentials," (Lawrence, Kansas, Center for Regional Studies, University of Kansas, 1966).

M. Nakanishi and L. Cooper, "Parameter Estimation for a Multiplicative Competitive -Interaction Model-Least Squares Approach," Journal or Marketing Research, (August, 1974).

R. Nelson, The Selection of Retail Locations, (New York, F.W. Dodge Corporation, 1958).

T. Stanley and M. Sewall, "Image Inputs to a Probabilistic Model: Predicting Retail Potential," Journal of Marketing, (July, 1976).

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

##### Authors

J. Patrick Kelly, Brigham Young University

Scott M. Smith, Brigham Young University

##### Volume

NA - Advances in Consumer Research Volume 10 | 1983

##### Share Proceeding

## Featured papers

See More#### Featured

### Unexpected-Framing Effect: Impact of Framing a Product Benefit as Unexpected on Product Desire

Monica Wadhwa, INSEAD, Singapore

Christine Kim, Hong Kong University of Science and Technology

Amitava Chattopadhyay, INSEAD, Singapore

Wenbo Wang, Hong Kong University of Science and Technology

#### Featured

### Consumers’ Trust in Algorithms

Noah Castelo, Columbia University, USA

Maarten Bos, Disney Research

Donald Lehmann, Columbia University, USA

#### Featured

### Increasing Tax Salience Alters Investment Behavior

Abigail Sussman, University of Chicago, USA

Daniel Egan, Betterment

Sam Swift, Bowery Farming