An Empirical Study of Some Spatial Interaction Models

This paper attempts to focus attention away from a mathematical and theoretical perspective to one where empirical and statistical aspects can assume greater significance. The need for this change of emphasis is demonstrated by the results of an empirical study of a selection of nine spatial-interaction models. Not only do the theory-based models fail to achieve significantly better levels of performance than some empirically derived models, but both types of model fail to achieve satisfactory levels of descriptive performance. These models are shown to be sensitive to the choice of calibration procedure and various kinds of errors in the constant and independent variables. It is suggested that the use of these models should always be accompanied by estimates of confidence limits, and that there is an urgent need for the development of new model designs which can at least provide accurate descriptions of observed reality.