Document details

This work is part of a dissertation that addresses the supermarket site assessment problem. We propose a 3-steps method for stores' site evaluation. (The 1st step yields the constitution of analogue groups of existent supermarkets, using a clustering procedure. On the 2nd step we use classification trees to classify new stores into specific analogue groups. Finally, on the 3rd step, we build a linear regression model to forecast new sites’ sales, based on several predictor variables, including dummy variables referred to analogue groups). In order to deal with demographic and competition data related to each supermarket, we use neighborhood delimitation techniques. Three alternative delimitation techniques and two aggregation procedures are compared. Results are evaluated based on the proportion of sales turnover variance that the alternative predictors are able to explain. (As a result, we select one aggregation procedure, although we conclude that none of the delimitation models: shortest path polygons and multiplicative weighted Voronoi diagrams, first and second order, present similar performance). Finally, we compare the relative importance of spatial data predictors in site assessment evaluation, using Dominance Analysis. As a result, the relevance of spatial analysis predictors clearly emerges being only dominated by the "trade area".