Transferability analysis of Logit models of gap acceptance behaviour

Gap-acceptance behaviour at unsignalized intersections has been extensively studied in the field of traffic theory and engineering using various methods. Among several approaches that have been adopted for modelling this type of behaviour, Logit models appear to be especially suitable as they allow to analyse the effect of several factors on gap acceptance. In this paper, we study in particular the issue of transferability of such models across different intersections, which is a specific instance of spatial model transferability. The problem of model transferability has been studied mostly with reference to travel demand models (for example, mode choice models). Our transferability analysis is based on experimental observations collected at two stop-controlled intersections, and is conducted by specifying and estimating Logit models on one intersection (called estimation context) and then transferring the estimated models to the other intersection (called application context). More specifically, the effectiveness of the full model transfer (direct transfer) has been considered first. In order to obtain some indications about the sample size in the application context necessary for an effective model transfer, an analysis of the performance of different transferred models (for varying dimension of the sample dataset used for updating the previously estimated models) has been carried out. Both transferring methods and methods for evaluating model transferability described in the literature have been considered. The primary objective of the present work is to evaluate, with specific reference to gap-acceptance Logit models, the possibility of performing a successful transfer of such models from an original (estimation) context to a different (application) context using the parameters of the original model and a “small” sample of gap-acceptance data in the application context. This is clearly a very important issue in practical applications of gap-acceptance models.