Modeling Transfers in Multimodal Trips

Abstract

Transfers are essential parts of multimodal trips. A detailed description of the transfer process identifies time, costs, and effort related to the different stages of the transfer process. Since inclusion of multiple transfer attributes generally leads to high correlations between parameter estimates, most travel choice models only distinguish a small number of transfer attributes. When attributes are highly correlated, it is difficult to establish their impact on the travel choice process, especially when more advanced random utility models are used. By analyzing path-size logit models that differ in the way the transfer process is accounted for, combinations of transfer attributes that best reflect the role of transfers in the travel choice process are established. The analysis uses revealed-preference data on interurban multimodal train trip making in the Netherlands. Apart from log likelihood values, correlations between different classes of transfer attributes are used to evaluate the different models. It is shown that it is better to exclude the number of transfers from the utility specification because its inclusion results in high correlations with other transfer attributes. Correlations are explained from the different stages in the transfer process to which transfer attributes relate. In addition, it is demonstrated that transfers can alternatively be accounted for by representing the transfer process in travel choice models at a high detail level and thereby prevent high correlations. It turns out that the best model contains mode indicators, railway station indicators, parking costs, transfer walking times, and transfer waiting times.