Locating a supermarket using a locally calibrated Huff model

Abstract

The Huff model is one of the most frequently used models in the field of retail distribution. Traditionally, parameters reflecting the effect of size and distance on determining the customers’ purchase probabilities in this model have been assumed constant along the study area. Applying some transformations on the Huff model formulation, these parameters can be calculated by means of ordinary least squares (OLS). In this paper, we used a local regression model, the geographically weighted regression model, instead of the usual global OLS model, with the aim of considering spatial nonstationarity in the model parameters. The estimated capture for a store was calculated by replacing global parameters with local ones. We present an application in which parameters showed spatial nonstationarity. The location of a new store was analysed too. We conclude that, for this case, the local model fits better than the global one. Moreover, the local model can provide individual information about customer preferences that global models ignore.