Individual and Population Learning in Route Choice Behavior of Drivers: Experiment and Model

This paper carries out four sessions experiment upon drivers' route choice behavior, which aims to study drivers' decision rules under complete information and incomplete information, Historical (H) and No Historical (NH) conditions. With the theory of learning in games as a tool, this study compares experimental data across sessions, and proceeds to construct a population-level model and an individual-level model of the drivers' choice behavior based on a recursive and updated belief set. The paper also measures the influence of various information conditions and drivers' previous experience on their route choice behavior. It's found that learning is a prevalent phenomenon in the process of drivers' decision-making, and the effect of learning remarkably depends on payoff differences between alternative routes, even when a driver may not have a clear idea of travel time of other options. It has been shown by experiment that experience and information exert their important influence on drivers' choice behavior, although they fail to accelerate convergence to network equilibrium. A driver will become more sensitive to payoff differences when he gains experience and holds complete information of alternative routes. The experienced drivers can more accurately perceive traffic conditions, consequently the frequency of switch between routes is greatly reduced. There exist considerable differences in the effect of complete information among drivers due to their different learning rules. This paper classifies drivers in terms of their different learning rules, and explores the influence of a variety of information and traffic conditions on drivers' learning behavior.