Laboratory Testing of Spurious Spatial Structure in Trip Distribution Models

Abstract In this paper we examine the risk of reporting spurious relationships in trip distribution models. We show how to make synthetic data sets that (by construction) are neutral with respect to clustering effects. We study a particular case with two non-interacting groups of jobs/workers. A competing destinations model is applied to 100 randomly drawn data sets of this type. Quite disturbingly, the loglikelihood ratio test reported a significant clustering effect in all of these data sets. This shows that statistical tests based on likelihood values may not be the right tool to examine the effect of such model extensions.