Inertia effects of past behavior in modal shift behavior: Interactions, variations and implications for demand estimation

Abstract

This paper focuses on empirically investigating the inertia effects of past behavior in modal shift behavior and contributes to the current state of the art by three aspects. Firstly, this study introduces and tests the potential influences of the inertia effect of past behavior on the traveler’s preferences regarding level-of-service (LOS) variables, besides the impacts of inertia effect on the preference for the frequently used transport mode in the past. Secondly, the mode- specific inertia effects are investigated to distinguish the differences in the inertia effects for different transport modes based on posterior individual-specific parameter estimations. Thirdly, the factors contributing to the heterogeneity of inertia effects including demographics and travel contexts are quantitatively examined. A joint random parameter logit model using a revealed and stated preference survey regarding commuting behavior is employed to unravel the three aspects. Five different model specifications with a demand estimation analysis are utilized to test the additive values of the proposed features in terms of model fits, estimation results, and predicting travel demands in various contexts. The results reveal significant interactions of inertia terms with LOS variables indicating the influences of past behavior on travelers’ evaluations on attributes of their previous choice. The mean values and variances of inertia effects for different transport modes are significantly and substantially distinct. For instance, the inertia effect of frequently using car is substantially positive representing strong stickiness to car, while the inertia effect of using metro has large variance among travelers and mostly appears as dispositions to change. Besides, the effects of personal characteristics and travel contexts on the magnitude of the inertia effects of different transport modes are identified as well. Incorporating the interactions and mode-specific inertia effects can remarkably improve the model performance. The demand estimation will be biased if they are neglected.