Analysis of Drivers’ Route Choice Behavior Considering Probabilistic Choice Set
Gang Xu,Tomio Miwa,Takayuki Morikawa
2013-01-01
Abstract:This paper aims to use the probability choice set (PCS) model, proposed by Manski in 1977 to analyze the drivers' en-route choice behaviors within the Toyota city's probe car data collected from the private vehicles in 2011. In a dense urban network, the number of feasible path for an OD pair might be infinite. Therefore, we always rely on some route generation technologies to create a paths set, and then we consider this set instead of the all feasible paths as the drivers' choice set. However, even in this created set, there are always including the paths which are not viable to the drivers. Therefore, in this paper, we analyze the drivers' route choice behavior based on the probability of individual's latent choice sets, furthermore, as the OD trips in this research are part of the whole trips for the drivers, We also consider the influences of drivers' en-route information to the probability of drivers' latent choice sets. The PCS model would be considered as a two-stage model including choice set formation and choice decision stages. In this research, the model used in the choice decision stage is multinomial logit (MNL) model, while in the choice set formation stage, it can be viewed as a process of examining the constraints. We assume that if a latent variable composed by path's and driver's characteristics exceeds a constraint's threshold value, then the constraint is satisfied. Therefore, the probability of a path to be in the driver's choice set is the probability of a path's latent variable exceeds the constraint's threshold value, this probability is calculated by the Binary Logit model. Within the increasing of the number of alternatives, the computation difficulties in choice formation stage would be exponential growth, therefore, the method proposed by Morikawa (1996) are applied here to solve this problem. In this research, we estimate models of the choice set formation and choice decision stages simultaneously using only the information of drivers' actual choices from probe data. Estimation results show that PCS model is remarkably better than MNL model in this research, and considering the en-route attributes in choice set formation stage of the PCS model could get a better estimation result than the one that do not.