Characterizing Heterogeneity among Merging Positions: Comparison Study between Random Parameter and Latent Class Accelerated Hazard Model

Abstract

This study aims to build an accurate merging position model by incorporating heterogeneity into the accelerated hazard model based on random parameter and latent class models. Three kinds of distribution (Weibull, log-normal, and log-logistic) are compared. The Weibull distribution is found to outperform the others. The latent class accelerated hazard model better captures the unobserved heterogeneity among drivers. The data used in this study can be naturally segmented into two classes. The large class consists of about 80% of drivers and the estimation results are similar to fixed parameter and random parameter models. The smaller one contains drivers that only consider the subject vehicle’s speed; these drivers merge rather late. Some of these drivers even merge into the adjacent main lane using the shoulder lane when they cannot find suitable gaps at the end of the auxiliary lane. Considering its good performance, the proposed latent class accelerated hazard model could also bring new insights into road design, such as the determination of the length of the acceleration lane or auxiliary lane.