【正文】 implementing their pricepromotions strategy, retail chain managers face the typical dilemma of “thinking globally, but acting locally.” In other words, they must plan their strategy, keeping in mind the global chainlevel impact of their promotions, to deliver on the mitments made to manufacturers. At the same time, managers need to make sure that the implementation of such strategy takes into account the fact that each store caters to a different market with different needs and responses to marketing programs. Moreover, the retail chain manager must consider not only how the promotion of a brand affects peting brands and total category sales, but also how it could affect sales in other categories.Our proposed model addresses these two important aspects of chainwide and storelevel crosscategory analysis. First, our propose factor regression model takes store differences and longitudinal market shifts into account, thereby providing the retail chain manager with unbiased global, chainlevel estimates. It also provides stable local estimates of crosscategory promotion effects at the store level. Second, while allowing this flexibility, our proposed model is parsimonious enough over existing alternatives, making it particularly useful for chainwide and storelevel crosscategory analysis. We apply the proposed model to storelevel data from one retail chain, paring it with several peting approaches, and demonstrate that it provides the best balance between flexibility and parsimony. Most importantly, we show that the proposed model provides useful insights regarding crosscategory effects at the chainlevel, for individual stores, and their patterns across stores.One possible solution is to use a randomcoefficients formulation, monly applied in consumer choice modeling to account for unobserved heterogeneity (cf. Manchandra et al. 1999). This would produce unbiased estimates of the average, chainlevel crossbrand and crosscategory effects, as well as storelevel estimates by taking advantage of the information available from all other stores. This “borrowing” of information is known to produce more reliable individuallevel estimates (., Blattberg and George 1991). Unfortunately, applying the usual randomcoefficients approach to crosscategory analysis would require a very large number of parameters to specify the multivariate distribution of the randomcoefficients across stores, as we will explain in more detail later. This makes traditional randomcoefficients models (either using a finite or continuous mixing distribution) often impractical for crosscategory brand level analysis as the number of brands and/or categories increases.The main purpose of this paper is to investigate crossbrand and crosscategory sales promotion effects both at the chain and store levels. Our intended contribution is two fold. First, we propose a new factorregression model that offers a viable, parsimonious and relatively simple alternative to the randomcoefficients models, which are widely used in the promotion response modeling (., Hanssens et al. 2001). This proposed model makes it possible to account for crosssectional and longitudinal variations in the regression coefficients, especially when the traditional randomcoefficientsregression model is not feasible due to a very large number of coefficients. Second, we also attempt to provide store managers with more insightful summaries regarding the patterns of crossbrand and crosscategory promotion effects across multiple stores. These cannot be fully obtained from a chainlevel aggregate model or an individual storelevel analysis. By providing a parsimonious way to account for variations in promotion crosselasticities across multiple stores and over time, this study can improve store managers’ understanding of crosscategory effects in category management。Literature on storelevel crosscategory promotion response modelingCompared to the growing literature on basket analysis using householdlevel scanner data, crosscategory promotion effects at store level are relatively underresearched. Among the first to tackle this problem are Walters and MacKenzie (1988), Walters (1991), and Mulhern and Leone (1991), who develop storelevel crosscategory sales response models using regression methods. Walters and MacKenzie (1988) use data from two stores (for purposes of validation) from a large supermarket chain to examine the impact of price promotions on store traffic, sales of promoted and nonpromoted products, and store performance with a structural equation approach. In their study, all crosscategory relationships are assumed to arise through store traffic and at the category level, rather than at the brand level, where managers are actually able to implement their promotion policies. Walters (1991) extends the Walters and MacKenzie’s (1988) study by considering two stores from peting retailers. The study finds that the pricing and promotion of brands in one category affect sales of brands in a plementary category. He also finds that discounting a brand in one store decreases sales of the same brand in another store, and decreases sales of the peting brands in other stores. Mulhern and Leone (1991) examine promotion effects on store profitability in the presence of demand interrelationships, using scanner data from two stores. Their findings confirm those of Walters (1991) within a store.Measuring the effect of a new point of sale system on the performance of drugstore operationsCelik ParkanComputers amp。 Operations Research 30 (2003) 729–744AbstractA new electronic point of sale (POS) system was deployed by a Hong Kong drugstorechain in eight of its drugstores as the first stage of a panywide introduction of POS automation. The management wanted to know if the new system had a signi1cant impact on the performances of the drugstores where the new POS system was deployed. We explain in this paper the putation of rati