Analyzing heterogeneity and unobserved structural effects in route-switching behavior under ATIS: a dynamic kernel logit formulation

Abstract

This paper focuses on modeling unobserved effects in route-switching dynamics under advanced traveler information systems (ATIS). The analysis explicitly accounts for the presence of heterogeneity in behavior and a general stochastic pattern for the unobservables. The dynamic kernel logit (DKL) framework (also referred to as dynamic mixed logit) is proposed and applied to model route-switching dynamics (with 55 repeated decisions per user), based on data from interactive simulator experiments. In contrast to the multinomial probit framework, the DKL is well-suited for calibrating dynamic travel behavior models with a large number of panel periods. To increase computational efficiency, the proposed formulation exploits a components of variance scheme to represent the correlation of error-terms (both within-day and day-to-day). The empirical results indicate that unobserved effects account significantly for the observed variability in route-switching behavior. Among the observed effects, users’ route-switching behavior is influenced by the nature, timeliness, and extent of real-time information, as also its quality. In addition, route switching is influenced by the level-of-service attributes on the alternative routes and users’ prior traffic experience. Among the unobserved effects, the results present evidence of considerable heterogeneity in route switching. The significance of experience variables, and the correlation of unobservables over time and within-day, indicate the presence of dynamic learning and adjustment processes in user behavior under ATIS. Although observed and unobserved preference and response heterogeneity are all significant, the largest improvement in model fit is achieved by incorporating observed heterogeneity followed by unobserved preference and response heterogeneity respectively. These findings have significant applications in route assignment models under information, design and evaluation of ATIS products and services, and assessment of various policy measures aimed at travel demand management.