Model Misspecification and Spatial Structure in Spatial-Interaction Models

Abstract

Concern has been expressed about the effects of spatial structure on parameter estimates from spatial-interaction models. The problem is essentially one of model misspecification. With a correctly specified model assumed, in which destination attraction depends on whether it is near to an origin or not, the consequences of using a misspecified model are examined. Explicit expressions for bias in the parameter estimates are derived; these are complex, but depend on terms that can be clearly interpreted in terms of aspects of spatial structure, such as scale, compactness, shape, remoteness of destinations, etc. Some simple special cases show how, with misspecified models, estimates from different systems will almost certainly differ. Extensions of the analysis and problems of estimation and interpretation are discussed.