Could we have predicted the increase in travel distances observed the last twenty hears with the gravity model of tip distribution?[Aurait-on pu prévoir l’allongement des distances des déplacements urbains observés ces vingt dernières années avec le modèle de distribution gravitaire ?]

Academic empirical research and planning practice face a common difficulty: the cost and availability of data. In order to evaluate infrastructure projects, it is common to use the gravity model to forecast origin-destination matrices. The calibration process uses cross-sectional data (number of trips and network level of service). In order to produce forecasts, the hypothesis is made that the parameters are stable over time. This study addresses the question of the robustness of the parameter stability hypothesis and the ability of the gravity model to explain the increase in travel distances. We calibrate gravity models using OD matrices constructed with data from the last three household travel surveys conducted in Lyon (1985, 1995 and 2006) and generalized time data from transport networks coded for the three dates. We then use the resulting parameters to estimate OD matrices at a later date and we compare the distances obtained from the "predicted" matrices with the observed distances. The results are contrary to initial intuition: the model parameters change, but the impact on OD matrices is not significant enough to completely invalidate the use of the assumption of stability in the reproduction of travel distances. Longer travel distances observed are mainly due to changes in input variables of the model, namely the evolution of the subsystem of locations, i.e. the location of the population and activities, and the evolution of the subsystem transport, i.e. network performance.