Investigating the predictive capabilities of discrete choice models in the presence of spatial effects

Fully spatial treatments of discrete choice models are difficult to specify and can be computationally intense. Several "ad hoc" procedures have been proposed in the land use literature to mitigate the potential estimation problems caused by spatial effects. We investigate the consequences of different spatial effects for several of these non-fully-spatial approaches using Monte Carlo simulation of a binary choice and a real world example of land use on the island of Sumatra in Indonesia. We find that there are few effects on categorical prediction and that inclusion of spatially lagged explanatory variables is the most effective procedure. Copyright (c) 2009 the author(s). Journal compilation (c) 2009 RSAI.