On forecasting choice among dependent spatial alternatives

Most current models of spatial interaction are structured so that a change to an existing spatial alternative or the introduction of a new alternative will not change the ratio of flows from any constrained origin to any pair of unchanged destination alternatives. This recently identified weakness may cause minor problems when small changes occur in cases where the distribution of origins and destinations is relatively homogeneous. However, where facilities are not distributed uniformly and spatial substitution or agglomeration effects are observed to occur, this restrictive property, known as IIA (independence from irrelevant alternatives), can cause the models to produce counterintuitive results. In this paper this limitation is overcome using an information-theoretic approach, where additional information is provided on the entropy of aggregated alternatives which best capture any hierarchical elements in the spatial choice process. Suggestions are made on model structures suited to different forecasting applications, and the principles are illustrated by a small numerical example.