Research Into Shopping Mall Choice Behavior

ABSTRACT - This paper explores some of the issues relevant to research into shopping mall choice behavior, including the measurement of patronage, situational specificity, and the level of aggregation in parameter estimation. Results of an exploratory study addressing these issues are presented. The findings suggest that shopping situation should be specified, that multiple indicators of patronage should be explored, and that parameters in a model of mall-choice behavior should initially be estimated separately for each shopping area under study.

Citation:

Roy D. Howell and Jerry D. Rogers (1981) ,"Research Into Shopping Mall Choice Behavior", in NA - Advances in Consumer Research Volume 08, eds. Kent B. Monroe, Ann Abor, MI : Association for Consumer Research, Pages: 671-676.

Advances in Consumer Research Volume 8, 1981      Pages 671-676

RESEARCH INTO SHOPPING MALL CHOICE BEHAVIOR

Roy D. Howell, University of Illinois-Urbana

Jerry D. Rogers, Southwest Missouri State University

ABSTRACT -

This paper explores some of the issues relevant to research into shopping mall choice behavior, including the measurement of patronage, situational specificity, and the level of aggregation in parameter estimation. Results of an exploratory study addressing these issues are presented. The findings suggest that shopping situation should be specified, that multiple indicators of patronage should be explored, and that parameters in a model of mall-choice behavior should initially be estimated separately for each shopping area under study.

INTRODUCTION

Consumer researchers have in recent years begun to focus more of their theory-building and empirical research efforts in the area of consumer patronage behavior (Darden 1980). Most of the research in this area has dealt with the store choice decision (Granbois 1977), as is proper and to be expected. Few retail outlets, however, exist as isolated entities. The synergistic effects of multiple retail establishments located in proximity to one-another, zoning ordinances which restrict locations in which retail businesses may operate, and the limited availability of "good" free-standing recall sites tend to encourage the clustering of retail trade into relatively compact areas. This clustering may be planned and formalized as a shopping center or shopping mall, may be unplanned, as in traditional downtown shopping areas, or may be formalized ex post facto, as in "revitalized" downtown malls. As such, the shopping area has become a legitimate object of research both in its own right and in its effect on the stores of which it is comprised.

As Bucklin (1967) has noted, the intra-urban shopping area, as a retail entity, lies somewhere between the individual store and the urban-entity itself on a scale of disaggregation. Research on shopping center preference and patronage has reflected this positioning, with theory and methodology being drawn from trading area theory, emphasizing the mass-distance relationship (Huff 1964), and from the store choice literature, which emphasizes store "image" and its components (Lindquist 1974-75). Stanley and Sewall (1977) demonstrate how these two approaches can be combined in store-choice research, as do Jain and Mahajan (1979). Recently, Nevin and Houston (1980) have applied a model containing both gravitational and image components to the area of shopping mall choice behavior.

Research into shopping center or shopping mall choice behavior cannot, however, be simply a straightforward extension of the trade area and store choice research streams, While some of the problems present in researching shopping mall preference and patronage are similar to those faced by the researcher investigating at the level of the retail store, others are unique. In particular, three issues confronting researchers investigating shopping mall choice behavior are addressed in this paper. The first of these issues concerns the choice and measurement of image components relative to shopping malls. The second concerns the measurement of patronage behavior. Third, the level of aggregation appropriate for parameter estimation in models of mall choice behavior is discussed.

These issues are discussed in the context of a study conducted in a medium-size midwestern SMSA. Personal interviews were conducted with two-hundred sixty middle to upper class females chosen from a two-stage area sample. Fifty-one respondents had not patronized any of the three shopping areas in the city, or had otherwise unusable questionnaires and were thus eliminated from further analysis. The respondents were asked to rate the adequacy of two enclosed shopping malls ("B" and "P") and the "revitalized" downtown area (D) on eighteen attributes (1-very unsatisfactory, 7-very satisfactory), as implied by the adequacy importance model (Mazis, Ahtola, and Klippel 1975). Also measured was the shoppers' distance from each of the three centers, and preference and patronage measures which are discussed below.

Both malls have approximately 270,000 sq. ft., with the downtown area relatively comparable in terms of number and types of stores. All have seven to nine stores carrying women's fashion items (the shopping situation specified in this research), and represent the only major shopping areas in or around the community.

ATTRIBUTE MEASUREMENT

Nevin and Houston are essentially correct in stating that, "...little work has been done on the dimensions of shopping area image" (1980, p. 84), Most applications in mall-choice research have relied heavily on the store image/attitude dimensions described by Lindquist (1974-75). While little work has been done on those dimensions specific to mall choice, some agreement seems to be emerging in the literature on the store image dimensions applicable to the shopping mall.

Aggregating observations across shopping centers studied, Hauser and Koppelman (1979) found, using factor analysis, four dimensions from sixteen attributes, labeled variety, quality and satisfaction, value, and parking. Assessing the structure of sixteen attributes separately for each of the shopping centers in their study, Nevin and Houston (1980) found three dimensions (labeled assortment, facilities, and market posture) which are similar to the dimensions found by Hauser and Koppelman (1979). What is not clear, however, is the appropriateness of items drawn from the store image literature in mall choice research. Clearly, some store image items such as price/value, credit availability, and sales personnel are applicable to the mall choice process only to the degree that the center as an entity has been successful in establishing a cohesive and consistent overall image. Similarly, the mall has attributes which are not among those composing store image, much as special events/ exhibits, recreational value, and commons-area atmosphere. Additionally, the contribution of the various store images to the image of the mall, and vice versa, is a question which begs further research.

Also not firmly established is whether the dimensionality of the image/attitude items employed is consistent across centers, or whether consumers' perceptual space differs for each center studied. Hauser and Koppelman (1979) assume a "basic structure of perceptions" and perform a factor analysis across centers. Nevin and Houston (1980) perform separate factory analyses for each center and, using a test for factor congruency, indicate "a consistent factor structure for all five shopping areas.

TABLE 1

HYPOTHESIZED FACTOR STRUCTURE OF SHOPPING MALL ATTRIBUTES

In this study, confirmatory factor analysis (Joreskog 1971) was employed to assess both the hypothesized factor structure and the equality of factor structures for each of the centers. The hypothesized factor structure is presented in Table 1. When non-zero convariances are allowed between the factors (Y not diagonal), the structure in Table 1 provides an acceptable fit for each of the three centers (average c² = 137 with (171-51=120) d.f., p<.14) and for the pooled data. While the coefficients in L and specific variances are substantially similar for each group, the hypothesis that S₁ = S₂ = S₃ = S (aqua/ covariance matrices for each of the centers), based on the F-statistics derived from Box's M, must be rejected. While the hierarchical hypothesis testing procedures described by Joreskog (1971) for assessing the equality of factor structures of several groups is not strictly applicable in this situation (when we are dealing with one group measured on different sets of variables), we find that the hypothesis of invariant factor pattern cannot be rejected, while the hypothesis of equal specific variances is rejected at the .05 level.

It can be inferred from this that the five factor solution with a general factor pattern such as that in Table 1 is applicable to the attributes of each of the centers, but that the magnitude of the l in Table 1, and thus the variance of the variables not accounted for by the common factors, differs for the three centers. This further implies that the variance-covariance matrices of the five factors are not equal for each of the three centers.

In order to provide comparability of the coefficients in tests of the overall model, unit weights were employed such that summated scales were formed for each dimension of each center. Table 2 contains the reliability analysis of these and other measures used in this study. The alpha coefficients reported are consistent with the findings of the confirmatory factor analysis in that the attribute structures for B and P (the two enclosed malls) are more similar to each other than either is to the structure of Downtown. The reliabilities reported here compare favorably with those reported by Nevin and Houston (1980).

TABLE 2

SCALE ANALYSIS

Distance

The variety of different measures of proximity or distance have been applied in patronage research (Granbois 1977), While it is generally accepted that perceived as well as actual distance is important (Brunner and Mason 1968), many studies have relied on map distance. Others have used ratings of proximity (Gentry and Burns 1977-78) or drive time (Nevin and Houston 1980). In this study, the multiple indicator approach is employed in the measurement of distance in order to account for both perceptual variations and objective distances. One scale, labeled proximity, is composed of map distance measured in blocks and a subjective time estimate measured in minutes. The standardized item scale composed of these two measures exhibits acceptable reliability. To capture more fully the "convenience" aspect of the distance construct, respondents were asked to rate each of the centers with regard to its general availability, the degree of traffic congestion usually encountered in driving to the center and the general degree of difficulty they encountered in making a shopping trip to the center. These items form the scale labeled accessibility.

SITUATIONAL SPECIFICITY

In this Study respondents were asked to rate the attributes of the centers and to report their patronage of each of the centers only in the contact of shopping for women's fashion clothing items for themselves or members of their family. In this respect the study is different from most reported research in this area, wherein shopping situation has not been specified,

In store choice research, the scores included as competitors are often highly similar in the distribution of types of merchandise carried by the scores (e.g. department stores are usually analyzed vis-a-vis other department stores, supermarkets vis-a-vis other supermarkets, etc.). In such research, global measures of patronage such as number of visits over a given time -period, average frequency of visit, last store visited, and other measures of this type are acceptable to the extent that the type of merchandise carried by the stores under consideration is similar. This is not an unacceptable assumption. The distribution of types of stores and thus merchandise line availability in shopping centers is subject to great variability, however, and this variability in merchandise distribution may substantially affect the number of dollars spent, frequency of visit, and other possible measures of patronage, as well as influencing the relationships between image/attitude variables, mass/distance variables, and patronage.

For example, one of the shopping areas in this study (B) is anchored by a large supermarket in addition to a national chain department store. Similarly, the presence of a large discount-drug store distinguishes center P. Observation tends to suggest that such diversity of store types among centers is the rule rather than the exception. In the case where the researcher does not specify the shopping situation, a center such as B, anchored by a supermarket, will show a higher average frequency of visit, etc., than will a center without a supermarket, with this difference quite independent of overall preference or image. Similarly, the relationship between distance and patronage is likely to be larger for a center with a large area devoted to convenience goods and frequently purchased merchandise.

This phenomenon may account for the findings of such studies as Gentry and Burns (1977-78), where proximity is found to out-perform image/attitude variables in explaining frequency of visits.

Specifically, as discussed in the brand choice area by Warshaw (1980), the probability that an individual shops at area j, P(s_j), is in fact dependent on the probability that the individual encounters and shops under condition i, denoted P(S_i). The probability of shopping at area j is thus

P(s_j) = S_iP(S_i)P(s_j|S_i)    (1)

Since P(S_i) is not equal for all i (e.g., shopping for food is likely to occur more frequently than shopping for fashion clothing), nor for all individuals (e.g., some individuals shop for food more often than others), and P(s_j | S_i) is not equal for all j given the distribution of merchandise available at the j centers in addition to perceptual factors, attempts to explain the frequency of individuals' shopping trips to j shopping areas, as in Nevin and Houston (1980) and Hauser and Koppelman (1979) are likely to meet with limited success.

By specifying a particular shopping situation, we are dealing with only one S_i. While individual differences in P(S_i) remain, the analysis is not contaminated by merchandise distribution causes of variation in P(s_j | S_i). It is suggested that, in research in this area, the researcher either (1) specify the shopping situation(s), or (2) include variables in the model which represent the number of stores (or square footage) devoted to various types of merchandise in each center.

MEASURING PATRONAGE

Much of the research in both store and mall-choice behavior has utilized assures of affect, preference, or behavioral intentions as dependent variables. While this is useful, predicting and explaining behavior is a more rigorous and preferred task. The question is however, just what behavior or behaviors constitute patronage? A variety of measures have been utilized as operational measures of patronage (see Granbois 1977). The most popular seems to be frequency of visit. Nevin and Houston (1980), for example, utilize shopping frequency measured on a six-point scale ranging from "less than once a year" to "once a week". Hauser and Koppelman (1980) utilize self reported frequency of visits, while Gentry and Burns (1977-78) also utilize frequency of use.

It seems that "patronage" implies more than simply frequently visiting a shopping center. Is the individual who shops frequently, but seldom buys, a "patron" to the degree as one who not only shops frequently but also buys a large quantity of merchandise at the center? What about the individual who shops infrequently but makes large volume purchases? These considerations, along with the increased measurement reliability obtainable through the use of multiple measures, suggest the use of several indicators in the measurement of patronage.

As can be noted in Table 2, four measures of patronage were combined as a (standardized item) summated scale in this study, exhibiting reasonably good reliabilities. Purchases, frequency of visits, and dollars spent (all over the three months prior to the study) and number of weeks since last purchase were reported by respondents for each of the three centers under study. As there were no strong reasons, a priori, that any of the measures were more important than the others, equal weight was given to each. Alternatively, managerial or theoretical considerations may indicate unequal weighing, or the first principle component could be utilized. (Since the items are measured in different units and thus have unequal variances, each was standardized before summation).

Although neither the summated scale approach nor the weighing schemes mentioned above may be optimum in an empirical sense, the measurement of patronage is not an empirical question. In applied, managerially oriented research each of several measures of patronage may be utilized separately as a dependent variable, with the impact of image, distance and other variables estimated for each. For the development of theory in the area of patronage behavior, whether store or shopping mall patronage is concerned, a consistent operationalization of the patronage construct should be defined, and this operationalization should be theoretically richer and operationally more meaningful and reliable than simple frequency of visit.

ESTIMATION

Yet another issue to be addressed in mall-choice research is the specification of and estimation of a model. Inherent in the model estimation issue are decisions concerning the form of the dependent variable(s), the level of assertion at which the parameters are to be estimated, and the statistical method to be employed. Considerations concerning each of these areas and the approach employed in this study are discussed in turn below.

Dependent Variables

Patronage, however defined, or measured, is the compelling choice as an ultimate dependent variable in either store or mall centered research. However, as discussed earlier, individual differences exist in P(S_i), the probability of encountering and shopping under various situations. The literature on shopping and prepurchase information search clearly documents one rather obvious conclusion: That some people do much more of it than others. Darden's patronage model (1980) suggests numerous factors which may influence the absolute amount of shopping engaged in by an individual. Demographic, life style, life cycle, amd socioeconomic variables affect the amount of shopping the consumer will engage in by affecting the length and composition of the individual's "shopping list", or "need queue", and thus the probability and frequency of encountering various shopping situations. Similarly Darden (1980) and Howell (1979) have shown that individuals differ in their shopping specific life styles, or shopping orientations. Shopping orientations such as enjoyment of shopping, recreational shopping, and shopping proneness affect the predisposition to shop and thus the amount of shopping done by consumers encountering a given shopping situation.

The point of this discussion is to indicate that unless one has a very rich data set containing measures such as those described above, attempts to model patronage, as such, will generally be less than satisfactory. In the absence of variables which explore the amount of shopping done by the individual, it would seem that the appropriate ultimate dependent variable should be patronage share rather than any absolute measure of patronage.

Also, list applications in the mall patronage (and store patronage) literature measure either affect, preference intentions, or behavior, in addition to cognitive attitudinal (image) variables and external facilitators (distance). The somewhat lower ability of image measures to explain patronage when included with distance measures can be to some degree accounted for by the failure to model the linkage of these through preference, as is suggested in the behavioral and brand choice literature (see Reibstein 1978), although Nevin and Houston (1980) were actually more successful in explaining frequency of visit than in explaining affect or intentions.

Preference (Y_lj) is measured in this study on a 1--6 scale ranging from least preferred to most preferred.

Similarly, it is not clear whether distance/proximity, however measured, is related to behavior through preference or whether it is a facilitator which moderates the affect of preference on behavior. Hauser and Koppelman (1980) address this issue directly by holding availability/accessibility constant. Nevin and Houston's (1980) results seem to indicate that distance is relatively more powerful in explaining intentions and behavior than in explaining affect.

These considerations lead to the testing of a model of the general (simplified) form:

DIAGRAM

wherein perceptions of each center affect preference for that center, which in turn affects patronage share. Proximity and accessibility are hypothesized as affecting both preference and patronage share.

Model Specification

Given the concept of determinant attributes, and Luce's choice axiom, it does not necessarily follow that having a high (low) perception of the adequacy of a center on a particular attribute will lead to higher (lower) preference or patronage share for that center, since all of the competing centers could be perceived as being adequate (inadequate) (Bearden 1977).

Thus the adequacy of each center relative to the perceived adequacy of the other competing centers should prove a better predictor of patronage share than absolute level of adequacy. Similarly absolute proximity or accessibility should not be as effective as relative proximity or accessibility in explaining preference or patronage share.

It follows directly from the above discussion that a preferred model in this situation would be a multiplicative competitive interaction model, which has been often employed in retailing research (Stanley and Sewall 1976; Jain and Mahajan 1979; Hansen and Weinberg 1979). To estimate the MCI model through least squares techniques, however requires, in this case, that each center have a non-zero patronage share for each individual. When this is not the case, either (1) individuals who have not patronized all three centers can be deleted, or as is more often the case, (2) individuals can be aggregated by area (Stanley and Sewall 1976; Jain and Mahajan 1979), The first solution may, in many cases, result in an unacceptable and potentially biasing loss of data (Jain and Mahajan 1979), while the second approach presupposes a homogeneity of perceptions within each area. In the absence of substantial empirical evidence to the contrary, this would seem to be a somewhat unrealistic assumption.

Level of Aggregation

Also, the MCI model assumes that the parameters which relate each of the predictors in the model to the dependent variable are the same for all objects (centers) in the market. It seems that this should be treated as an empirical question rather than as an assumption.

The MCI is not unique in its assumption of the applicability of pooling across stores or centers in the parameter estimation process. The use of discriminant function analysis has similar implications. Other techniques, such as least squares regression, can be estimated either across centers or for each center individually. Both approaches are evident in the literature, although the issue is seldom explicitly addressed (Hauser and Koppelman 1980; Nevin and Houston 1980).

An examination of the regression coefficients estimated separately for each center under consideration by Nevin and Houston (1080) seems to indicate that an assumption of equal coefficients would be unwarranted, although they do not test for this explicitly. In particular, the coefficients they obtain for the downtown area in their study seem to be substantially different from those they obtain for the four mall-type shopping centers.

If the parameters relating the predictors to preference and patronage share are (statistically) equal for each of the centers, aggregation is preferred, as the increase in the number of observations used to estimate each parameter will yield estimates with lower variance. However, inappropriate aggregation can lead to biased parameter of estimates. The approach taken in this study is to estimate at the lowest level of aggregation possible; those centers with equal coefficients can be subsequently pooled and reestimated.

Based on the above considerations, the model estimated in this research is

EQUATION   (2)  and    (3)

As such, this model, though cross-sectional, bears some resemblance to the market share model proposed by Houston and Weiss (1974). As a recursive model, each equation can be estimated using ordinary least-squares methods. However, since equations (2) and (3) are estimated separately for each center (j), important information contained in each error term (e_1j and e_2j) can be brought to bear on the estimation process.

Consider the three estimates of equation (2) estimating preference for each of the three centers. To the extent that the model fails to account for an individual's preference for shopping center "B", it probably fails similarly to account for that individual's preference for centers "P" and "D" due to factors influencing that individual's preference not included in the model. Similarly, to the extent that the error term for an individual in one of the three preference equations is small, error terms for the other two equations are likely to be small, since the model has captured to a large extent the variables salient to that individual's preference function. Thus, positive covariance is expected between the j error vectors of equation (2).

Since the three error vectors of equation (3) represent mis-estimation of dependent variables whose sum is constrained aqua/ to one, overestimation of patronage share for one center may lead to underestimation of patronage share in the other equations (although 02j* is not constrained to equal 1.0), As a result, negative covariance is expected among the three error vectors estimated through equation 2. In situations where there is nonzero covariance among the error terms and the variables in the equations are not all equal, the technique developed by Zellner, Seemingly Unrelated Regressions (SUR), can produce estimates with smaller variance than OLS estimates (Johnston 1972), and is thus employed in estimating equations (2) and (3).

RESULTS

Table 3 contains the standardized coefficients corresponding to the parameters to be estimated in equations (2) and (3). Tests for the equality of regression coefficients for the three centers reject the hypothesis of equality at or below the .05 level (Johnston 1972); the estimation of an aggregate model would he inappropriate.

TABLE 3

REGRESSION RESULTS (STANDARDIZED)

TABLE 4

CORRELATION AMONG ERROR VECTORS

Generally, the model performs acceptably. In particular, it can be noted that accessibility is a more important predictor of preference for the downtown center than for the other three centers. It can also be noted that proximity is a good predictor of neither preference nor patronage when included in a model including perception of accessibility. Also. accessibility generally affects both preference and patronage, indicating that it acts both as a component of preference and as a facilitator in addition to its affect on preference.

As expected, there is a negative covariance of errors across the three models of patronage share. Unexpectedly, however, the contemporaneous error covariance matrix, S, for the equations predicting preference contains both negative and positive elements. In particular, as the model overestimates an individual's preference for center B, it underestimates preference for center D (Downtown). Thus individuals in this sample who prefer "B" do not prefer Downtown and vise versa, while the errors for "B" and "P" positively co-vary. This seems to indicate a "preference map" dimension with mall type centers at one extreme and downtown at the other, which is independent of the predictors included in this model.

CONCLUSIONS

It is anticipated that this paper has raised more questions than it has answered conclusively. This is intended however, because the issues of attribute and patronage measurement, situational specificity, and model specification and estimation must be specifically and empirically addressed if the investigation of mall-choice behavior is to prove fruitful.

Other issues remain which are not addressed in this paper. One such issue is the relationship between the retail store and the shopping area. Surely this relationship is reciprocal, with store image and preference affecting the image of and preference for the shopping area and center in which it is located, while the image, location, and preference for a shopping area affects the stores located there. The "special store" variable included by Nevis and Houston (1980) is a step in this direction; more work in this area is required.

Similarly, the model tested in this paper presupposes a one-way relationship between perceptions, preference, and behavior. However perceptions and preference are measurable at a point in time, while behavior is measurable only over a given (past) tine period in cross-sectional research. This, in addition to cognitive consistency theories, might indicate a reciprocal relationship between behavior and perceptions. Research is currently underway to test this hypothesis.

REFERENCES

Bearden, William O. (1977), "Determinant Attributes of Store Patronage: Downtown Versus Outlying Shopping Centers," Journal of Retailing, 53, 29-38.

Brunner, James A. and Mason, John L. (196B), "The Influence of Driving Time Upon Shopping Center Preferences," Journal of Marketing, 32, 57-61.

Bucklin, Louis P. (1967), "The Concept of Mass in Intra-Urban Shopping," Journal of Marketing, 34, 32-36.

Darden, William R. (1980), "A Model of Consumer Patronage Behavior," in Ronald W. Stampfl and Elizabeth Hirschman, eds., Competitive Structure in Retail Markets: The Department Store Perspective, American Marketing Association, 43-52.

Gentry, James W. and Burns, Alvin C. (1977-78), "How Important are Evaluative Criteria in Shopping Center Patronage?," Journal of Retailing, 53, 73-85.

Granbois, Donald H. (1977), "Shopping Behavior and Preferences," in Selected Aspects of Consumer Behavior, National Science Foundation, Washington D.C., 259-298.

Hansen, Michael H. and Weinberg, Charles B. (1979), "Retail Market Share in a Competitive Market," Journal of Retailing, 55, 37-46.

Hauser, John R. and Koppelman, Frank S. (1979), "Alternative Perceptual Mapping Techniques: Relative Accuracy and Usefulness," Journal of Marketing Research, 16, 495-506.

Houston, Franklin S. and Weiss, Doyle L. (1974), "An Analysis of Competitive Market Behavior," Journal of Marketing Research, 11, 151-155.

Huff, David L. (1964), "Defining and Estimating a Trade Area," Journal of Marketing, 28, 34-38.

Jain, Arun K. and Mahajan, Vijay (1979), "Evaluating the Competitive Environment in Retailing Using Multiplicative Competitive Interaction Model," in Jagdish Sheth, ed., Research in Marketing - Vol. II, JAI Press, 217-235.

Johnston, J. (1972), Econometric Methods, New York: McGraw-Hill, 238-241.

Joreskog, K. C. (1971), "Simultaneous Factor Analysis in Several Populations," Psychometrica, 30, 4, 409-426.

Kunkel, John R. and Berry, Leonard L. (1968), "A Behavioral Conception of Retail Image," Journal of Marketing, 32, 21-27.

Lindquist, Jay D. (1974-75), "Meaning of Image," Journal of Retailing, 50, 29-37.

Mahajan, Vijay, Jain, Arun K., Bergier, Michel (1977), "Parameter Estimation in Marketing Models in the Presence of Multicollinearity: An Application of Ridge Regression," Journal of Marketing Research, 14, 586-591.

Mazis, Michael B., Ahtola, Olli T., and Klippel, R. Eugene (1975), "A Comparison of Four Multi-Attribute Models in the Prediction of Consumer Attitudes," Journal of Consumer Research, 2, 1, 38-52.

Nevin, John R. and Houston, Michael J. (1980), "Image as a Component of Attraction to Intraurban Shopping Areas," Journal of Retailing, 56, 77-93,

Reibstein, David J. (1978), "The Prediction of Individual Probabilities of Brand Choice," Journal of Consumer Research, 5, 163-168.

Stanley, Thomas J. and Sewall, Murphy A. (1976), "Image Inputs to a Probabilistic Model," Journal of Marketing, 40, 48-53.

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