Bayesian approaches for calibrating the mixed C-logit model of truck route choice

Abstract

This study examines logit models applied to the truck route choice problem with data from the Dallas metropolitan area. Instead of assuming a fixed coefficient of a variable in the conventional multinomial logit model, the proposed model assumes a certain probability distribution for each coefficient, typically called the mixed C-logit, in an attempt to better reflect the preference heterogeneity. Three Bayesian approaches with different hierarchy levels are introduced and are solved by the mean-field variational inference with the implementation of the block coordinate algorithm. The associated models are tested on two subnetworks in two scenarios, the first of which has toll alternatives while the other does not. It is found that all the three proposed models notably outperform the conventional multinomial logit model, which conforms to the behavior indicated in the simulation test. Generally, our study finds that travel time is the most significant factor considered in truckers’ route choices in both scenarios. The relative importance of attributes in affecting truckers’ route choices differs between scenarios. In Scenario 1, travel time dominates other attributes. However, in Scenario 2, with a less dense network than in Scenario 1, it is found that using a route that entirely consists of state or interstate highway segments is as essential as using a route with a short travel time for most drivers. Additionally, the truck drivers’ preference for roadway delay and network density is found to vary widely in the numerical test. In contrast, their preference for travel time and roadway designation is relatively consistent.