Optimal zoning systems for spatial interaction models

The design of zoning systems for spatial interaction models is a major problem which affects both the interpretation and acceptability of these models. This paper demonstrates that zoning-system effects on parameter values and model performances are nontrivial, and that their magnitude is far larger than was previously thought likely. An approach which is most appropriate in an applied context, where there is also the problem of poor model performance, is to identify a zoning system which will approximately optimise model performance. The paper gives details of how this may be achieved. This method is demonstrated by a series of empirical studies. Finally, there is a brief discussion of the general implications for spatial model building.