Estimating the impact of displays and other merchandising support on retail brand sales: partial pooling with examples
With the advent of scanning data, methodological issues have arisen, in particular as they relate to the reliability of parameter estimates in regression models. This study deals with the reliability of the coefficients of promotion-type dummy variables (e.g., display, leaflet, bonus pack,). Due to a lack of passthrough of trade deals to the end consumer, those coefficients can be typically unidentified or unstable when estimated at the store level and even at the chain level. Assuming that the individual-level coefficients of the dummy variables are draws from a common but arbitrary distribution, the authors suggest to "pool" the data for those variables (partial "pooling") across stores and across chains. They show with three real-life examples the increased reliability (with a correct sign) of the "pooled" coefficients as compared with the store-level or chain-level individual coefficients.
|Keywords||promotion, regression, scanning data|
|Persistent URL||dx.doi.org/1008043325970, hdl.handle.net/1765/2181|
|Series||ERIM Top-Core Articles|
|Journal||Marketing Letters: a journal of research in marketing|
Franses, Ph.H.B.F, Bemmaor, A.C, & Kippers, J. (1999). Estimating the impact of displays and other merchandising support on retail brand sales: partial pooling with examples. Marketing Letters: a journal of research in marketing, 87–100. doi:1008043325970