A MODIFIED LOGIT ROUTE CHOICE MODEL OVERCOMING PATH OVERLAPPING PROBLEMS. SPECIFICATION AND SOME CALIBRATION RESULTS FOR INTERURBAN NETWORKS

Random utility route choice models play a central role in stochastic assignment, to road networks. The usually adopted specifications for such models are Multinomial Logit (MNL) and Probit, both having pros and cons. In this paper a modified specification of the Logit model, named C-Logit, is proposed. The C-Logit overcomes the main shortcoming of MNL, ie unrealistic choice probabilities for paths sharing a number of links, while keeping a closed analytical structure allowing calibration on disaggregate data and efficient path flow computations when paths are explicitly enumerated. A behavioural interpretation of the C-Logit model is proposed by presenting it as a particular case of a wider class of implicit joint models of route perception-choice. In the second part of the paper a procedure for path enumeration is proposed, and MNL and C-Logit models are specified and calibrated from a sample of 1471 paths chosen by truck drivers on the Italian inter-city road network. The preliminary calibration results are promising and show significant improvements induced by both the commonality factor characterizing the C-Logit model and by label variables which can be seen as attributes explaining path perception. (A) For the covering abstract see IRRD 886400.