Spatial interaction modelling in retail planning practice: the need for robust statistical methods

Some recent developments in 'gravity' modelling of shopping behaviour are reviewed, in the light of Breheny's comments on their increasing level of use in commercial forecasting of retail activity. It is shown that the well-known singly constrained 'retail potential' model can be transformed into a Poisson regression model. This allows much more flexible specification of models, and simpler calibration procedures using generalised linear modelling software. These arguments are illustrated in analyses of shopping diary records from a British city. This approach is shown to be consistent with regression-based forecasting methods which are already used in commercial retail planning practice.