A Recursive Logit Model with Non-Link-Additive Attributes in a Multimodal Network

Abstract

Recursive logit (RL) models can predict mode and path choices at the same time by modeling them as a sequence of link choices in a multimodal network. However, RL models can incorporate link-additive attributes only, constraining their applicability by restricting the variable selection. Therefore, this study proposes a methodology to include non-link-additive attributes to the RL model to analyze and predict users' intermodal path and mode choices on a multimodal network. This study proposes a link-additive approximation method that approximates a non-link-additive path attribute into a corresponding link attribute that holds the link-additivity. The methodology is applied to the actual network and trip data in Seoul, Korea, with two non-link-additive attributes: transit fare and transfer penalty. The result shows that including those non-link-additive attributes in the RL model improves both the goodness-of-fit and accuracy of the model.