The Multi-Item, Multi-Stop Store Choice Process: an Exploratory Study

ABSTRACT - Preliminary evidence is presented that multi-item shopping trips involve a store choice process that differs significantly from the "independent store" choice process typically used in past studies of store patronage. Implications for future research are discussed.

Citation:

Melvin R. Crask and Richard W. Olshavsky (1983) ,"The Multi-Item, Multi-Stop Store Choice Process: an Exploratory Study", in NA - Advances in Consumer Research Volume 10, eds. Richard P. Bagozzi and Alice M. Tybout, Ann Abor, MI : Association for Consumer Research, Pages: 351-355.

Advances in Consumer Research Volume 10, 1983      Pages 351-355

THE MULTI-ITEM, MULTI-STOP STORE CHOICE PROCESS: AN EXPLORATORY STUDY

Melvin R. Crask, University of Georgia

Richard W. Olshavsky, Indiana University

ABSTRACT -

Preliminary evidence is presented that multi-item shopping trips involve a store choice process that differs significantly from the "independent store" choice process typically used in past studies of store patronage. Implications for future research are discussed.

INTRODUCTION

Most shopping trips are multi-item, multi-stop trips, yet store patronage studies have tended to focus exclusively on individual store choice. A large portion of patronage research can be categorized as store image studies (e.g., Nevin and Houston 1980; Pessemier 1980; Hirschman, Greenburg, and Robertson 1978; Doyle and Fenwick 1974; Lindquist 1974; Rich and Portis 1964; Fisk 1961). Like much of the brand attitude research, image studies concentrate on finding those attributes which the consumer views as important in store choice. Stores possessing important attributes or having images consistent with desired stores are always assumed to be more preferred and more patronized. The other major portion of patronage research attempts to explain store choice through post hoc analysis. Patronage is modeled using choice axioms that are most often adaptations of Reilly's Law of Retail Gravitation or of Huff's formulation for estimating probabilities of shopping center choice (e.g., Stanley and Seawell 1976; Huff 1962; Dodd 1950; Stouffer 1940). Some measure of distance from the consumer's home is used to reflect the "cost" of patronage and some measure of store size is used to reflect store "attraction." Patronage probabilities are then generated for each store or shopping area.

Neither type of study has been very predictive of store patronage and we hypothesize that this is due to the atypical choice process presented or modeled. In image studies the consumer is expected to patronize the store preferred due to its image. Gravity models also attempt to predict the store to be patronized. If all shopping trips were single stop, perhaps better prediction could be achieved (although similar brand choice models would seem to indicate that prediction of consumer behavior is difficult). Most shopping trips are not single-stop, however, and the effect of other stops to be made is not examined in either type of study.

It we could assume that the consumer makes independent choices, additional stops would have no influence. The consumer would choose the most preferred store for each stop. Other choice criteria are possible, though. The consumer might try to minimize the total distance travelled during the trip. This criterion may well lead to a choice of stores clustered together, even though none of these stores may be the most preferred. Another possibility would be that the consumer does choose the most ?referred store for one stop and the remaining stores are chosen due to their proximity to that store. Still another choice criterion might be that the consumer chooses a store because it allows the purchase of a large number of needed products, but this store might not be most preferred for any or these items purchased individually. Any of these planning rules could create store choices which would not be predicted using either of the approaches for studying patronage discussed earlier.

The exploratory study presented here seeks to examine such multi-stop store choice processes. Specifically, the following questions were addressed: 1) Is an independent store choice model used by consumers? 2) Does the store choice process change as the number of items within each category on the shopping list increases? 3) Does the store choice process change as the number of different categories on the shopping list increases?

Answers to these three questions are essential to furthering our ability to predict store patronage behavior.

METHOD

Subjects

A convenience sample of ten females was used in this study. Eight of these subjects worked as secretaries while the other two were retired nurses. All or these subjects had lived in the study area for a minimum or one year.

Study Site

The stud site was a mid-western college city with a population or approximately 80,000. The retail structure of this town in terms of the number of stores, the geographic distribution of these stores, and the degree of clustering (malls, shopping strip, downtown, etc.) was judged to be typical of cities of this size. The city contains one large shopping mall, several smaller shopping centers, several shopping strips, a viable downtown area, and several free-standing stores.

Design

A two factor ANOVA design with repeated measures in all conditions was selected. Table 1 presents the design. The number of items within a product category ranged from one to four. The number of categories ranged from one to five. As shown in Table 1, the categories involved were food, health and beauty aids (HBA), clothing, household items, and services. These categories were chosen because they represent familiar products that could be purchased from many competing stores. Twenty conditions were formed by this design wherein each condition consisted of a fixed number of items and categories; the composition of each condition determined the composition of the shopping list presented to the subjects. Each subject served in each condition.

It should be noted at this point that the number of categories variable is confounded with the particular types of categories selected for use in this study. As will be discussed later, this feature of the design limits the generalizability of the results.

Procedure

The subjects were tested individually and each session lasted approximately 25 minutes. Once the subject was seated, she was handed the written instructions. The subject was told that the purpose of the study was to understand how she planned a shopping trip. A description of the actual task to be performed was also contained in the instructions. No cash payments were offered for participation. A local phone book was available to the subject for store reference.

Each subject was instructed to act as though she were actually going shopping for the items on the list. She was further asked to "think aloud" as she planned her trip and to state the store or stores that she would patronize for each item on the list and the order in which she would patronize each store.

After reading the instructions, one practice card (#3) was presented. Then the experimenter shuffled the cards, turned on the tape recorder, and handed the top card to the subject. A record was kept by the experimenter of: the order of cards presented, the time required to complete the task for each card, the number of omitted items, and the number of items mentioned but for which no specific store was mentioned.

TABLE 1

EXPERIMENTAL DESIGN

RESULTS

A manipulation check was performed on the data as a preliminary data analysis. The experiment had been designed to increase the complexity of the shopping itinerary by varying the number or items within each category to be purchased and the number of product categories into which these items fell. A manipulation check was necessary to insure this complexity had been achieved.

Two measures were used as manipulation checks: (1) the amount of time each subject took to provide the list or stores to be visited and (2) the number of stores to be visited. As the task became more complex, one would assume the subject would require a longer decision time. Likewise, one would assume more complex shopping tasks would necessitate selecting more stores to purchase the desired products. Table 7 presents these data.

The experiment had been designed to lend itself to a two-factor analysis of variance with four levels on the "item" factor and five levels on the "category" factor. Each person was treated as a block and a two- factor randomized block design with repeated measures was used for the analysis.

TABLE 2

AVERAGE DECISION TIME AND AVERAGE NUMBER OF STORES LISTED FOR EACH SHOPPING LIST PROVIDED

The results of the manipulation check on decision time revealed that the effect of increased items (F=35.6, d.f.=3,171) and increased categories (F=159.1. d.f.=4,171) were both significant at the .001 level. Furthermore, the interaction between the two factors was significant at the same level (F=9.4, d.f.=12,171). The blocking effect was significant at the .01 level (F=5.1, d.f.=9,171). While these results suggest an increase in complexity as intended, they also imply that the two factors are not independent. Examining the simple main effects of each factor it was found that the decision time increased as the number of product categories in creased, regardless of the number of items to be purchased within each category. On the other hand, decision time did not significantly vary as items within a category were added to the list until multiple product categories were also on the list. In essence, task complexity did increase as each factor was changed but in a curvilinear fashion; multiple-item/multiple-category shopping lists require much greater decision times.

Similar results were round when examining the number of stores the subjects selected for the shopping trip. Both main effects (ITEMS, F=44.5, d.f.=3,171; CATEGORIES, F=159.1. d.f.=4,171), the blocking effect (F=5.1, d.f.=9,171), as well as the interaction between the factors (F=9.41, d.f.=12,171), were significant. The simple main effects for each factor revealed the same curvilinear relationship which was found for decision times. The significant blocking effect found for both of these tests underscores the importance of controlling for this within-subject variance since both decision time and number of stores selected were round to vary across individuals within each treatment condition.

Satisfied that the experiment provided the complexity variation desired, attention was turned to investigating whether the choice process used by each subject varied as task complexity changed. To test this, the choice process used by each subject was classified as being an "independent store" choice process or some "other" choice process for each treatment condition tested. [The verbal data collected here cannot reveal insights concerning the detailed processes of the various choice processes identified here. This is because the subjects were not specifically instructed to articulate their choice processes at this level. For instance, we cannot say if the independent store choice process was additive, conjunctive, lexicographic or any other process. Further, even if the subjects had been instructed to articulate processes at this level we doubt that they could have complied with this request, except in a retrospective manner, because the actual choice process probably occurred some time in the distant past. Our subjects were largely revealing their habitual behaviors. Our use of the term choice process therefore refers to the types of "planning" processes that were exposed by these verbal data.]

Store Choice Process

The results from a single rater who categorized all subject/treatment combinations with respect to type of choice process will be reported here since our initial concern was to achieve consistency across ratings. To try to offset any bias towards categorizing a decision process as more complex than it really was, the rater was instructed to classify the choice process as being "independent store" if he was unsure of the actual process. While this may tend to overstate the true occurrence of independent store choice decisions, it does serve to provide a more conservative test of the existence of any "other" choice process. Of the 200 choices classified, only ;1.5 percent were identified as independent store. Thus, it would seem that the independent store choice process does not adequately represent the majority or the situations examined.

To test whether the use of alternative choice processes increased as the number of items within a category increased or as more product categories were included, the choice process used was transformed into a dichotomous variable and tested using the same ANOVA design as was previously discussed. Only the category effect was found significant (F=48.0, d.f.=4,171, a =.001). As the number of product categories increased, the subject tended to move away from an independent store choice model. Here it must be emphasized, however, that the number of categories is confounded with the specific categories selected for this study.

Other Choice Processes

Figure 9 presents a classification scheme developed for ali of the store choice processes that were revealed in the data and the letter assigned when coding each type. Table 3 summarizes the distribution of choice process types across all conditions. The pattern revealed b the ANOVA, that subjects tended to shift to more complex choice processes as the number of categories increased, is presented in detail in Table 3. The total occurrence of each type of choice process was as follows: A-83, B-5, C-1, D-12, E-1, A/B-9, A/C-43, A/D-21, A/E-1, C/D-2, C/E-7, "other"-15. [A second rater also categorized all subject/treatment combinations and agreed with 144 out of 200 (72%) of the classifications: the differences between raters largely concerned the specific type of "other" choice process involved.] An explanation and example of each type of choice process follows:

FIGURE 2

TYPES OF STORE CHOICE PROCESSES

Type A is the "independent store" choice process. An example is provided by Subject 1 in condition 4.

"I go to Krogers to pick out bread, cheese, soft drinks and ground beef."

Type B is the "one stop shopper-store" choice process. According to this choice process all items on the list are obtained at a single store. Type A and Type B are distinguished in that Type B arises only when multiple categories are involved and when the subject does not patronize the preferred store within each category (as determined from responses in other conditions involving only one category). An example is provided by Subject 10 in condition 8

"Here again I would shop at the Marsh supermarket or these things; soft drinks, ground beef, eggs. bread, perfume/cologne, bath soap, prescription drugs and headache remedy.

Type C is the "one stop shopper-cluster" choice process. According to this choice process all items on the list are obtained at a single cluster. [A problem of defining the term cluster was encountered in assigning models to subject's responses. In general, a cluster refers to a group of stores that are geographically localized; e.g., downtown stores, stores within a shopping strip, or a mall. A mall is a physically distinct cluster yet there are usually stores adjacent to the mall property. Whether to include these stores and the mall in one cluster is open to debate. Because the subjects generally made this distinction, we concluded these stores from the cluster. Including these near-by-stores within the cluster, in some cases. would change the classification of the planning process inferred from the data, but the overall conclusions of the study would remain the same.] An example is provided by Subject 10 in condition 14.

"Okay, I think that with the place I live I would go to Highland Village to purchase these things because it is only three miles from my home. Plus the variety of stores there. I usually shop at the IGA there where I would purchase the eggs and the milk and then I would go to the 3D store and I would buy the comb and the hairbrush, the shoes, the earrings and the bedsheets."

Type D is the "contingent store" choice process. According to this choice process one item or a category or items determine a particular store, then patronage of that store determines choice of the stores for the remaining items on the list. The distinction between Type D and Type C is that Type C involves the choice of an entire cluster whereas Type D involves the choice of a single store first and then other stores are chosen. An example is provided by Subject 7 in condition X

"I would go to Osco to get a prescription and comb and bath soap and then to Eisner's to get soft drinks, bread, and eggs.

Type E is the "contingent cluster" choice process. According to this choice process one item (or a category of items) determines a particular cluster, then patronage of that store determines choice of the stores for the remaining items on the list. An example is provided by Subject 4--in condition 12.

"Well I'm going out to the mall. I will stop at Kroger's for all my groceries--milk, bread, ground beef, eggs. I will skip the comb and hair brush, go to the pharmacy at the Mall for my prescription drugs and the headache remedy. Go to Target for the deodorant. The earrings I will probably buy at Blocks. And I don't know where I would buy the shoes. skirt, or the blouse. depends on where...

TABLE 3

THE DISTRIBUTION OF STORE CHOICE PROCESSES

Hybrid types of choice processes refer to those processes that involve some combination of these basic types of choice processes (e.g., A/C, A/B, A/D/E). Hybrid types were also classified as "other" choice processes.

DISCUSSION

This exploratory study provides preliminary evidence that the store choice process used is contingent upon the number or categories on the shopping, list. When only one category is involved, the traditional "independent score" choice model appears to be appropriate. But when the number of categories is greater than one, consumers tend to switch to other types of choice processes. The "ocher" types or choice processes refer to processes whereby stores or clusters are selected that can provide all or the items (categories) on the list or processes whereby the choice or a store is contingent upon the choice o t another store or cluster. Or, "other" choice process can refer to some combination of these processes. Here the independent store choice model may be used for one item, but is not used to choose stores for all items (categories) on the list.

These "other" choice processes are merely suggestive. This study is preliminary to any construction of a model of the actual choice processes used. Understanding the nature or these processes, however, must precede model construction to prevent improper conceptualization.

For example, if store choice is contingent upon prior stores visited, independent store choice models such as suggested by Pessemier (1980) cannot capture such dependencies. While one might argue that the "convenience" component of the model incorporate such dependencies, this is not possible. Since consumers are asked only about a single store's attributes, the convenience component will either be as it relates to convenience from the home or an "average" convenience from other stores. lo correctLy capture the dependency, such a model would have to be iterative, where the image is redefined after each stop on the trip.

The issue of dependent-choice decisions has been explored in product choice. Farquhar and Rao (1976) and McAlister (1979), for example, examine how a second brand or item within a product category may be contingent upon the first brand or item chosen. Although their contingency models may provide a starting point for modeling the s core choice processes discovered, two problems may occur if they are adapted here.

First, they are additive utility models. Research is i necessary to determine if such a choice rule is operative instead of some non-compensatory rule. Second, the choice situation examined is not identical. While some shopping trips may involve choosing a second store of the same type (which parallels the choice situation examined in these dependent choice models), other trips involve stopping at different store. types. This choice situation is not modeled in these multiple-item, single-category approaches.

Green and Devita (1974, 1975) have looked at contingent choice in a multiple-category context; specifically, how the choice of a dessert is dependent upon entee selection. The complementary nature Of the products they examined may restrict the usefulness of their approach to the store-choice decisions investigated here. In addition, since their model is based upon conjoint anal sis, the same additivity issues discussed above may apply. Olshavsky and Acito (1980) also found that, although the additive conjoint analysis model could accurately reproduce their subjects' rankings, the choice rules used by the subjects were often non-compensatory. Thus, managerial implications derived from the conjoint analysis would be inaccurate.

If attempts are undertaken at present to model the store choice processes we discovered, researchers may run into the same situation found by Olshavsky and Acito--high predictive accuracy but poor conceptual accuracy. Further research is needed to improve our understanding or the decision process before modeling begins. Protocol anaLysis may reveal decision rules which cannot easily be modeled using existing approaches. The computer simulation modeling approach used by Olshavsky and Acito is a good modeling possibility, however.

The effect of the limitations encountered in our study should also be addressed by future research. As previously noted, the design or this study confounds the "number of categories" variable with the specific categories used. Therefore, we cannot say to what extent these results are specific to the categories used and the manner in which the number of categories was increased. It remains for further research to test the generalizability of these findings.

Problems relating to the identification of strategies due to the ambiguous nature of the definition of a "cluster" were also noted. Other limitations of this study arise from the small size and convenience nature of the sample. Also the distribution of the stores, malls, and clusters in the study site man have influenced the results. However, we judged our study site to be quite typical in terms of both the number of stores and their distribution for cities of this size.

Finally, further research is needed to determine the composition of actual shopping lists. For example, since the amount of improvement over the prediction using an independent choice model depends upon the number of multiple-stop trips taken by consumers, it is necessary to determine the number of such trips taken. It should be noted, however, that, except for our extreme condition (i.e. ,4 items and; categories), all of our subjects found the lists presented to be quite "reasonable."

REFERENCES

Dodd, Samual C. (1950), "The Interactance Hypothesis: A Gravity Model Fitting Physical Masses and Human Groups," American Sociological Review, 15, 9, 245-56.

Doyle, Peter and Fenwick. Ian (1974), "How Store Image Affects Shopping Habits in Grocery Chains," Journal of Retailing, 50, 4, 39-52.

Farquhar, Peter H. and Rao, Vithala R. (1976), "A Balance Model for Evaluating Subsets of Multi-Attributed Items," Management Science, 5, 528-39.

Fisk, George (1961), "A Conceptual Model for Studying Customer Image," Journal of Retailing, 37, 4, 1-8.

Green, Paul E. and Devita, Michael T. (1974), "A Complementary Model of Consumer Utility for Item Collections," Journal of Consumer Research, 3, 56-67.

Green, Paul E. (1975), "An Interactive Model of Consumer Utility," Journal of Consumer Research, 7, 146-153.

Hirschman, Elizabeth C., Greenburg, Barnett and Robertson, Dan H. (1978), "The Intermarket Reliability of Retail Image Research: An Empirical Investigation," Journal of Retailing, 54, 1, 3-13.

Huff, David L. (1969), A Retail Location Model, Los Angeles: University or California Press.

Lindquist, Jay D. (1974), "Meaning of Image," Journal of Retailing, 50, 4, 29-38.

McAlister, Leigh (1979), "Choosing Multiple Items from a Product Class," Journal of Consumer Research, 4, 213-224.

Nevin, John R. and Houston, Michael J. (1980), "Image As A Component of Attraction to Intraurban Shopping Areas," Journal of Retailing, 36, 1, 77-93.

Olshavsky, Richard W. and Acito, Franklin (1980), "An Information Processing Probe into Conjoint Analysis," Decision Sciences, 3, 451-470.

Pessemier, Edgar A. (1980), "Store Image and Positioning," journal of Retailing. 56, 1, 94-101.

Rich, Stuart N. and Portis, B. D. (1964), "The 'Imageries' of Department Stores," Journal of Marketing, 98, 9, 10-5.

Stanley, Thomas J. and Seawell, Murphy A. (19,6), "Image Inputs to a Probabilistic Model: Predicting Retail Potential," Journal of Marketing, 40, 3, 48-53.

Stouffer, Steven A. (1940), "Intervening Opportunities: A Theory Relating Mobility and Distance," American Sociological Review, 5, 6, 845-67.

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