Consumer_store_choice_dynamics_An_analysis_of_the_.docx

《Consumer_store_choice_dynamics_An_analysis_of_the_.docx》由会员分享，可在线阅读，更多相关《Consumer_store_choice_dynamics_An_analysis_of_the_.docx（25页珍藏版）》请在淘文阁 - 分享文档赚钱的网站上搜索。

1、 See discussions, stats, and author profiles for this publication at: https:/ Consumer store choice dynamics: An analysis of the competitive market structure for grocery stores Article in Journal of Retailing July 2000 DOI: 10.1016/S0022-4359(00)00033-6 CITATIONS 115 READS 1,019 3 authors, including

2、: Peter T. L. Popkowski Popkowsk University of Alberta 53 PUBLICATIONS 884 CITATIONS Ashish Sinha University of Technology Sydney 29 PUBLICATIONS 377 CITATIONS SEE PROFILE SEE PROFILE All content following this page was uploaded by Peter T. L. Popkowski Popkowski Leszczyc on 14 Janu The user has req

3、uested enhancement of the downloaded file. Consumer Store Choice Dynamics: An Analysis of the Competitive Market Structure for Grocery Stores PETER T. L. POPKOWSKI LESZCZYC University of Alberta ASHISH SINHA University of Waikato HARRY J. P. TIMMERMANS Eindhoven University of Technology This study a

4、ims at formulating and testing a model of store choice dynamics to measure the effects of consumer characteristics on consumer grocery store choice and switching behavior. A dynamic hazard model is estimated to obtain an understanding of the components influencing consumer purchase timing, store cho

5、ice, and the competitive dynamics of retail competition. The hazard model is combined with an internal market structure analysis using a generalized factor analytic structure. We estimate a latent structure that is both store and store chain specific. This allows us to study store competition at the

6、 store chain level such as competition based on price such as EDLP versus a Hi-Lo pricing strategy and competition specific to a store due to differences in location. Competition in the retailing industry has reached dramatic dimensions. New retailing formats appear in the market increasingly more r

7、apidly. A focus on a particular aspect of the retail mix (e.g., service or price) means that retailers can compete on highly diverse dimensions. Scrambled merchandising and similar developments have implied that par- ticular retailers are now competing against retailers they did not compete with in

8、the past. Peter T. L. Popkowski Leszczyc is Associate Professor of Marketing, University of Alberta, Department of Marketing, Business Economics and Law, 4 30F Faculty of Business Building, Edmonton, Alberta, Canada T6G 2R6 (e-mail: ppopkowsgpu.srv.ualberta.ca). Ashish Sinha is Assistant Professor o

9、f Marketing, University of Waikato, Department of Marketing and International Management, Private Bag 3105, Hamilton, New Zealand (e-mail: asinhawaikato.ac.nz). Harry J. P. Timmermans is Professor of Urban Planning and Director European Institute of Retailing and Services Studies, Eindhoven Universi

10、ty of Technology, Faculty of Architecture, Building and Planning, P.O. Box 513, 5600 MB Eindhoven, The Netherlands (e-mail: eirassbwk.tue.nl). Journal of Retailing, Volume 76(3) pp. 323345, ISSN: 0022-4359 Copyright 2000 by New York University. All rights of reproduction in any form reserved. 323 32

11、4 Journal of Retailing Vol. 76, No. 3 2000 These trends can be observed in all segments of the retailing industry including the grocery industry, albeit perhaps in different form and intensity. As a result of these developments, consumers face a retail environment in constant flux. They continuously

12、 must decide to stay loyal, try out new formats, or use the complete system to obtain benefit from discounts on specific days or for specific items. Previous research has reported low store loyalty and significant store switching for grocery store purchases (Kau and Ehrenberg, 1984; Uncles and Hammo

13、nd, 1995; Popkowski Leszczyc and Timmermans, 1997). Given these findings, it is important to incorporate the store switching behavior in the study of consumer store choice. Furthermore, consumer reac- tions to a rapidly changing retail environment will additionally depend upon idiosyncratic preferen

14、ces and socio-economic characteristics that either allow or restrain them from pursuing some of the options. For example, active search requires a substantial amount of time that households working long hours may not have. For the retailer, the problem is how to cope with the increased competition i

15、n light of the dynamics of consumer shopping behavior. Should retailers invest in loyal consumers and not worry too much about the customer who is cherry-picking the market? Or, should one try to aggressively attract new customers? Or perhaps should they try to capture a substantial share of the swi

16、tching population of shoppers? To make better informed decisions on this issue, retailers need to know more about the timing of shopping trips, store choice, and switching behavior of consumers, together with those factors that influence this relationship, to develop appropriate strategies. Hence, a

17、ccording to this framework, it is pertinent to know the magnitude of store loyal/store switching behavior, the nature of the competitive structure in their market and how it is changing, and to be aware of any differences in these regards between consumer segments. The dynamic store choice decision

18、can be conceptualised as a problem of deciding where and when to shop. The first decision is the traditional store location choice problem. The second is the shopping trip incidence problem relating to the timing of shopping trips and implies information about intershopping trip times. Information o

19、n a sequence of shopping trip events yields information about the number or percentage of consumers choosing the same store on subsequent shopping trips (repeat shopping or store loyalty). Transitions between stores on successive shopping trips provide measures of store-switching behavior. These two

20、 choice processes are, of course, interrelated. Store choice is dependent on the timing of shopping trips, as consumers may go to a smaller local store for short fill-in trips and go to a larger store for regular shopping trips (Kahn and Schmittlein, 1989). Also, store choice and shopping trip timin

21、g decisions tend to differ for individuals and house- holds as a result of personal differences, household composition, and activity patterns (Popkowski Leszczyc and Timmermans, 1997; Kim and Park, 1997). Most previous research has focused only on the timing or the store choice decision. Furthermore

22、, the majority of research studying store choice behavior has applied cross- sectional data. To the extent that prior research has considered the dynamics of store choice, it has been limited by the assumptions made. For example, the dynamic Markov model (Burnett, 1973) is based on the assumptions t

23、hat the average number of shopping trips is the same in each successive, equal-length time period, and that the transition matrix is time-invariant. Hence, store choice probabilities are constant over time. The NBD and Dirichlet models, which have been applied to store choice (see, e.g., Kau and Con

24、sumer Store Choice Dynamics 325 Ehrenberg, 1984; Wrigley and Dunn, 1984, 1985) combine purchase timing and store choice. However, they employ the assumption that shopping trips are made in equal time periods, and that the number of purchases at a particular store by a single consumer or household in

25、 successive equal time periods is independent. This research further limited consideration of shopping trips to separate product classes. Shopping trips for all other purchases are not included in the analysis. To overcome the shortcomings of previous research, we propose a dynamic model of store ch

26、oice, a hazard model, where store choice is dependent on the timing of shopping trips. The hazard model is a Semi-Markov or Markov Renewal process (Howard, 1964). Store choice is modeled by a discrete-state space consisting of the choice set of stores, and intershopping time is incorporated by allow

27、ing the time between shopping trips (the transition rates) to be a random variable after some distribution function. Hazard or transition rates are estimated for transitions between all stores, including repeat shopping trips and switching between different stores (e.g., Kalbfleisch and Prentice, 19

28、80). In terms of modeling consumer store choice dynamics, the use of hazard functions is appropriate in the sense that the hazard rate, defined in this context as the rate at which a new shopping trip is made at some time t, given that the consumer has not shopped until time t, can be linked to cova

29、riates that indicate how and to what degree these influence consumer store choice dynamics and the competitive structure of the retail market. In particular, hazard models allow one to derive the effects of explanatory variables on intershopping timing and store choice from scanner panel data. This

30、model specification has been used to study brand switching behavior (Vilcassim and Jain, 1991; Gonul and Srinivasan, 1993; Popkowski Leszczyc and Bass, 1998; and Chintagunta, 1998), but applications to store choice are lacking. We employ this hazard model to estimate competition both at the store an

31、d store chain level by combining it with an internal market structure analysis. We estimate a generalized factor analytic structure model (Sinha, 2000). This has a latent structure with an idiosyn- cratic component that is store chain specific and a common factor that is both store and store-chain s

32、pecific. This structure allows us to study store competition at the store-chain level (e.g., competition based on price, such as EDLP vs. a Hi-Lo pricing strategy) and competition specific to a store (e.g., due to differences in location). Furthermore, the idiosyncratic component estimates the uniqu

33、e unobserved components of store chains. This is an extension of the model by Chintagunta (1998). Different from Chintagunta (1998), we specify a factor analytic structure on the self-transitions and the shape parameters of the hazard model. This provides a more general and parsimonious model. We be

34、gin by discussing the relevant literature on store choice dynamics. Next, we discuss the methodology and our models properties. Then, the data are briefly discussed, followed by a general summary of estimation results. Finally, we provide conclusions and areas of future research. LITERATURE REVIEW D

35、ynamic models of store choice behavior have received relatively limited attention in the literature. There are a number of articles that have used aggregate level supermarket 326 Journal of Retailing Vol. 76, No. 3 2000 TABLE 1 Summary of Characteristics of Models of Store Choice Model Attributes Ma

36、rkov Model Dirichlet Market Structure Nested Logit Hazard Model Multiple Shopping Trips Yes Yes No Yes Yes State Dependence Yes No No No Yes Repeat Shoppers and Switchers Yes No Yes No Yes Time Varying Probabilities No No No No Yes Store Choice and Timing No No No No Yes Exogenous Variables No No No

37、 Yes Yes Right Censoring No No No No Yes Heterogeneity No Yes Yes Yes Yes scanners to study the effectiveness of marketing mix variables on store sales and store substitution. Normally, weekly sales levels for brands within specific product categories are related to marketing mix variables. For exam

38、ple, Hoch et al. (1994), Wittink et al. (1987), Kumar and Leone (1988), and Blattberg and Wisniewski (1987) have investigated the efficacy of individual brand promotions on store choice and store sales. Although these studies provide valuable information about the effects of marketing mix variables

39、on store sales, aggregate store data do not reveal individual store switching behavior. To study store switching behavior, we need individual level data. Table 1 provides an overview of relevant models using individual level analyses. We consider models of store choice (the logit and Dirichlet model

40、s) and models that estimate transition matrices (market structure models, Markov models, and the hazard model) and distinguish between repeat shoppers and switchers. Models of Store Choice Models like the well-known NBD and Dirichlet models have been applied to store choice (e.g., Kau and Ehrenberg,

41、 1984; Wrigley and Dunn, 1984, 1985; Uncles and Ehrenberg, 1988). In the NBD model, the number of purchases at a store in successive equal time periods is assumed to be independent and follow a Poisson distribution. This leads to an exponential purchase timing distribution. There is no duration depe

42、ndence and the probability of a shopping trip remains constant over time. In particular, the assumption of the absence of any short-term trend in the aggregate sales levels at a given store seems unrealistic and hence has been criticized; empirical evidence suggests that the propensity of consumers

43、to make shopping trips varies with the day of the week. Wrigley and Dunn (1984) suggest a multivariate extension including store choice, the Dirichlet model. The Dirichlet model combines purchase timing and store choice. The purchase timing component of the model is based on the same assumptions as

44、the negative binomial model. The choice component is based on the assumption that each consumer has a particular probability of purchasing a given product at a particular store that remains Consumer Store Choice Dynamics 327 the same over the periods analyzed. The probabilities for a given store fol

45、low a beta distribution, whereas they are assumed to follow a Dirichlet distribution for all stores. The final and most limiting assumption is that the store choice and purchase timing compo- nents are independent. Again, this is a very limiting assumption. One would expect that the choice behavior

46、of heavy users is different from that of light users and that differences would be expected between regular versus fill-in trips. Although the NBD and Dirichlet models provide good descriptions of market behavior under equilibrium conditions, they do not provide information about the underlying caus

47、al variables explaining shopping behavior.1 The logit model has been used to study the effects of exogenous variables on store choice (Bell and Lattin, 1998; Fotheringham, 1988). Bell and Lattin (1998) have studied consumer shopping behavior and the effect of pricing format (Hi-Lo vs. EDLP) on store

48、 choice. They estimated a simple logit model for store choice, where choice is a function of distance, whether a purchase has been made at a store during the initialization period, of feature advertising for a store, and of the household expectations about basket attractiveness at stores. The expect

49、ed basket attrac- tiveness, conditional on small versus large basket size shoppers, is calibrated using a nested logit model of brand choice and purchase incidence.2 They concluded that large basket shoppers prefer EDLP stores, whereas small basket shoppers prefer to shop at stores using a Hi-Lo pricing strategy. Sinha (2000) estimated a factor analytic nested logit model that combines spatial interaction models with internal market structure analysis. Store choice is estimated as a t