A comparison of Euclidean Distance, Travel Times, and Network Distances in Location Choice Mixture Models

This article investigates the selection of a distance measure in location modeling. While in the empirical literature the choice usually boils down to picking one single measure, this research proposes a flexible approach in which several measures may be used in parallel to capture the surrounding economic landscape. This is intended to acknowledge that interactions between agents may take several forms, occurring through different channels and as such being based on different measures. The methodology is applied to the location choice of establishments in the Paris region, using a mixture of ”mono-distance” hurdle-Poisson models. Seven distance measures are considered: Euclidean distance, the travel times by car (for the peak and off-peak periods) and by public transit, and the corresponding network distances. For all the economic sectors considered, the mixture of hurdle-Poisson models performs significantly better than the “pure” mono-distance models. This corroborates that local spatial spillovers are indeed channeled by different means, hence best represented using several measures. The combination of peak and off-peak road travel times (slightly) outperforms other combinations including the Euclidean distance, supporting the choice of meaningful over more abstract measures in spatial econometric models. The distance measure most likely to capture local spatial spillovers varies depending on the economic sector examined, reflecting differences between sectors in operations and location choice criteria.