Attitudinal Influence on Retail Patronage Behavior

ABSTRACT - Traditional research on retail patronage behavior which relies on the gravitational models has long ignored attitudinal factors from its design. The purpose of this study is to demonstrate that retail patronage behavior may be significantly explained by shopper attitude scale scores that are derived from paired comparison data and incorporated into a gravitational formulation.

Citation:

Masao Nakanishi (1976) ,"Attitudinal Influence on Retail Patronage Behavior", in NA - Advances in Consumer Research Volume 03, eds. Beverlee B. Anderson, Cincinnati, OH : Association for Consumer Research, Pages: 24-29.

Advances in Consumer Research Volume 3, 1976      Pages 24-29

ATTITUDINAL INFLUENCE ON RETAIL PATRONAGE BEHAVIOR

Masao Nakanishi, University of California, Los Angeles

ABSTRACT -

Traditional research on retail patronage behavior which relies on the gravitational models has long ignored attitudinal factors from its design. The purpose of this study is to demonstrate that retail patronage behavior may be significantly explained by shopper attitude scale scores that are derived from paired comparison data and incorporated into a gravitational formulation.

INTRODUCTION

In recent years much of the effort in consumer research has tended to focus on products and brands, while retail store patronage behavior has not managed to generate the research it perhaps deserves. Traditionally, research into retail patronage behavior has used models which are analogous to the model of gravitation in physics (Reilly, 1929; Huff, 1962). These approaches typically use shopping center (or store) size as the surrogate measure of center (or store) "attraction," and physical distance or travel time as the surrogate measure of "resistance" which shoppers must surmount in reaching their destination.

While these gravitational models have performed reasonably well in predicting the spatial behavior of shoppers, the validity of using physical or functional distance measures and selling space as explanatory variables has come under criticism. It has been argued (Mittelstaedt, et al., 1974) that cognitive (or cognized) distance reflects more closely both the physical and evaluative dimensions which influence shopper behavior. The use of selling floor space to represent retail "attraction" has been questioned by others. Kotler (1971) argues that different images, accessibilities, and prices of retail centers create a basis for differential store preference. LaLonde (1962) tested an extension of the gravitational model which incorporates various factors and found that individual store size per se does not have the great influence claimed on drawing power. Nakanishi and Yamanaka (1975) also found that shopping center size and travel time together do not fully account for the drawing power of retail centers.

The purpose of this study is to examine the relationship between retail store patronage and consumer attitudes on such attributes as accessibility, merchandise quality and store atmosphere. The scale values derived from paired comparison data are used to test whether these variables account for store choice in a shopping trip.

METHOD

Data

The location selected for data collection was a small community in a suburban area of Osaka, Japan. A survey of the patronage patterns of two superstores (combination of discount house and supermarket) was conducted during spring of 1973. Store A with a gross selling space of 35,000 sq. feet had opened four years prior to the study, and had been enjoying a dominant position among retail stores in the community. Store B had opened a year prior to the study within 650 yards of Store A and with a gross selling space nearly one and one-half times that of Store A.

A cluster sample of 400 housewives were selected from the residential area surrounding the two superstores. The area was divided into 14 subareas based on natural boundaries and distance from the stores. See Figure 1. From these l4 areas 80 interviewing locations were selected and five interviews conducted per cluster. Interviews were conducted in the homes and each respondent (housewife) was asked, among other questions, how often she shopped at each superstore per month. In addition each respondent was asked to compare the two stores on each of the following attributes:

1. ease of travel and access to the store

2. the quality of merchandise [It may be noted that there was some difference in the wording of item 2 merchandise categories. The wording was "sense (or taste)" for Clothing, "freshness" for Food, and "quality" for Sundries.]

3. quantity and assortment of merchandise

4. lowest price

5. demeanor of store personnel

6. store atmosphere (lighting, cleanliness, etc.)

7. feeling of comfort and ease while shopping.

Each of these items (except l) was repeated for three major merchandise categories: Clothing, Food, and Sundries. From the shopping frequency data, each store's "share of shopping trips" figure was computed for each merchandise category within each area. The responses for items i through 7 above were summarized in each subarea by calculating the proportion of respondents who rated Store A higher than Store B with regard to each attribute for each merchandise category.

Model

Although the usual gravitational model does not accommodate the type of data obtained in this study (i.e., paired comparisons), a scaling model has been developed to integrate paired-comparison data into a gravitation-al-type model (Nakanishi & Cooper, 1975). There are three basic assumptions to this scaling model.

1. The shopper's attitude toward a store may be measured along several attitude dimensions (or attributes). Let X_kij be the underlying attitude measure of a shopper from area i toward store j along the k^th attribute dimension. Assume -4 < X_kij < 4.

2. The probability that a shopper from area i rates store j over store j' in a pair-wise comparison along the k the attribute dimension is given by

EQUATION

3.The probability that a shopper from area i chooses store j in a shopping trip is given by

FIGURE 1

SIMPLIFIED MAP OF THE STUDY AREA

EQUATION   (1)

where exp(^.) is an exponential function, m is the number of stores, q is the number of attributes, and a_k is the parameter of shopper sensitivity toward the k^th attribute.

Note that the last assumption--equation (1)--is an ex- tension of the traditional gravitational model, incorporating attitude measures as explanatory variables.

The three assumptions permit one to use the paired comparison data for estimating the parameters of the model. First, assumptions 1 and 2 lead to the estimates of the attitude scale score assigned to store j by shoppers in area i with respect to attribute k. Let P_(k)ijj' be the observed proportion of respondents in area i who rate store j over j' with respect to the k^th attribute. P_(k)ijj' = 1/2 if j = j', by assumption 2. The desired attribute scale score is given by

EQUATION     (2)

It must be pointed out that p_(k)ij is an estimate of g_k(X_kij - X_ki), where X_ki is the mean of X_kij. Next, from assumptions 2 and 3, the following regression equation is derived.

EQUATION   (3)

See the Appendix for derivation of (2) and (3). This last regression model was fitted to and the b_k's were estimated for each merchandise category (Clothing, Food, or Sundries) separately.

RESULTS

Table 1 gives the regression results based on model (3). The regression coefficients are given in the standardized form (i.e., "beta coefficient") to take account of the non-equality of variances among the attribute scale estimates. [Since estimated attribute scale score, p_(k)ij = g_k(X_kij - X_ki.), regression model (3) gives the estimate of b_k = a_k/g_k, but not that of a_k per se. (See the Appendix for details.) But, when standardized, the regression estimates of b_k becomes equivalent to that of a_k. To show this, let a_k^* and b_k^* be the standardized regression coefficient of a_k and b_k, respectively. Also let s_p(k), s_x(k), and s_p be the standard deviation of p_(k)ij, X_kij and p_ij respectively. Then, EQUATION] The share of shopping trips for clothing items is most strongly affected by the atmosphere of store, but negatively affected by store personnel and merchandise quality. The ease of travel and access has a relatively minor influence on the share of shopping trips for clothing items. For food items, the share of shopping trips is most significantly affected by the ease of travel and access and the lowness of price, but negatively affected by the atmosphere of the store. For sundries, the ease of travel and access again is the most significant attribute and other factors have virtually no effect. These findings are also supported by the regression coefficients of the "best" equation for each merchandise category, which are formed by deleting non-significant variables from the full equation.

At this point, some counter-intuitive results, such as the negative coefficients for personnel and merchandise quality for clothing items, may be questioned. One may suspect that the results are unreliable due to multicollinearity, since scale scores for different attribute dimensions may be highly correlated with each other, owing to a "halo" effect Indeed, the p_(k)ij's are highly correlated with each other, with product-moment correlation coefficients ranging as high as .95. In order to remove multicollinearity from the data, the attribute scale scores were factor-analyzed. Since there were slight differences in wording of the questions between merchandise categories, scale scores for each merchandise category were factor-analyzed separately. Because the purpose of analysis is to produce orthogonal dimensions that are parsimonious, the co-variance matrix, rather than the correlation matrix, was factored. Resulting factors were Varimax-rotated. For all merchandise categories, two factors were sufficient to cumulatively explain more than 90% of over-all variance. Table 2 gives the factor analytic results.

The pattern of factor loadings are highly similar across merchandise categories. Factor 2 for clothing and food and Factor 1 for Sundries are clearly related to the ease of travel and access. Factor 1 for clothing and food represent generally the impression of the store and merchandise, while Factor 2 for sundries is associated with the impression of the store and personnel. Factor scores for these two factors are then substituted for raw scale scores in regression model (3).

The regression results using factor scores as independent variables are given in Table 3. The regression coefficients are again standardized. The factor that represents the ease of travel and access (Factor 2 for clothing and food and Factor 1 for sundries) is more important than the other factor in determining the share of shopping trips for food and sundries, but not for clothing. The impression on the store and merchandise is the most significant factor for clothing, but relatively less important for food and sundries. These results generally confirm prior expectations on shopper behavior. The explanatory power of the factor scores, measured by R², is high for food and sundries, but there is a significant reduction in the case of clothing (from R² = .902 to .708). Judging from the high percentage of raw score variance explained by the two factors, the significance of the negative regression coefficients for the merchandise quality and store personnel attributes for clothing in the raw scores regression results is probably spurious.

TABLE 1

RAW SCORE REGRESSION RESULTS

TABLE 2

FACTOR ANALYSIS RESULTS

SUMMARY AND DISCUSSION

This study was conducted to see if attitudinal factors may effectively be brought into the traditional gravitational model of shopper behavior. The predictive ability of estimated attribute scales scores (generated from paired comparisons) were generally high, explaining more than 90% of the variance in the share of shopping trips figures for different subareas. As expected, the ease of travel and access is the prime consideration in choosing a store for a shopping trip for food and sundries; this attribute is less important for choosing a store for clothing items. For other attribute dimensions, the conclusion is less clear. Regression results using raw attribute scores indicate that store atmosphere is important for the choice of store for clothing items and that price is important for food items. But the negative coefficients for merchandise quality and store personnel in the case of clothing and that for store atmosphere in the case of food are intuitively unappealing. Regression results using factor scores show that the factor representing the impression of store and merchandise results in a positive coefficient for all merchandise categories and a particularly significant one for clothing.

TABLE 3

FACTOR SCORE REGRESSION RESULTS

Thus the study clearly demonstrated the usefulness of attribute scale scores from paired comparison data in predicting the drawing power of stores, but there are some unresolved issues. First, in order for the researcher to be able to determine relative importance of attributes, it is necessary to select attribute dimensions which are as orthogonal as possible to avoid the multi-collinearity problem. How one should phrase paired comparison questions to maintain the orthogonality of scale scores has not become clear. Second, to the store manager, it is not enough to know that his store suffers from a poor image; he needs to know how a poor image may be changed. It is therefore necessary to relate each attribute dimension to a set of store and location characteristics. Of course, some of the attributes are not under the control of the store manager. A more global change in the image of the entire chain may be called for. Finally, other methods for measuring shopper attitudes toward stores should be explored. For example, one may ask each respondent to rate each store on interval-scaled attitude scales. This individual-level approach, however, suffers from the fact that individual attitude scores will somehow have to be aggregated for each area since the gravitational formulation requires that the scale score for each store be determined for each area. The effect of such aggregation has not been investigated so far. The techniques used in this study, aside from its ease of administration, handles the aggregation problem in a natural manner.

APPENDIX

REFERENCES

Huff, David L., "Probabilistic Analysis of Consumer Spatial Behavior," W.S. Decker, ed., Emerging Concepts in Marketing. Chicago: American Marketing Association, 1962, 443-61.

Kotler, Philip, Marketing Decision Making: A Model Building Approach. New York: Holt, Rinehart and Winston, 1971, 316-20.

LaLonde, Bernard J., "Differentials in Supermarket Drawing Power," Marketing and Transportation paper 11. East Lansing: Bureau of Business and Economic Research, Michigan State University, 1962.

Mittelstaedt, Robert, William W. Curtis, Sanford L. Grossbert and Robert D. Rogers, "Psychological and Evaluative Dimensions of Cognized Distance in an Urban Shopping Environment," in Combined Proceedings. Chicago: American Marketing Association, 1974, 190-3.

Nakanishi, Masao, and Lee G. Cooper, "Parameter Estimation for a Multiplicative Interaction Model--Least Squares Approach," Journal of Marketing Research, 11 (August 1974), 303-11.

Nakanishi, Masao, and Lee G. Cooper, "A Scaling Approach to the Synthesis of Choice Models and Multiattribute Attitude Models" Working Paper (in preparation). Los Angeles, 1975.

Nakanishi, Masao and Hitoshi Yamanaka, "Measurement of Drawing Power of Retail Centers: Regression Analysis," Working Paper Series No. 30. Los Angeles: Center for Marketing Studies, University of California, Los Angeles, 1975.

Reilly, William J., "Methods for the Study of Retail Relationship," University of Texas Bulletin, No. 2944 (November 22, 1929).

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