Citation

Wang K, Bhat CR, Ye X. Transportation (Amst) 2023; 50(4): 1471-1495.

Copyright

(Copyright © 2023, Holtzbrinck Springer Nature Publishing Group)

DOI

10.1007/s11116-022-10284-x

PMID

unavailable

Abstract

Commute trips account for a large portion of travel demand in peak hours and significantly influence the operation of urban transportation systems. In this paper, we apply a fully flexible multinomial probit (MNP) model for the analysis of commute mode choice behavior, and compare this MNP model with more traditional discrete choice models, including the multinomial logit (MNL), the cross-nested logit (CNL), the heteroscedastic independent MNP (HI-MNP), and the homoscedastic non-independent MNP (HONI-MNP). The two-variate bivariate screening (TVBS) approach, a recently developed analytical evaluation for the multivariate normal cumulative distribution (MVNCD) function, is employed. The sample for analysis is drawn from a web-based travel survey conducted in Shanghai. Overall, from a data fit perspective at, both the disaggregate and aggregate levels, the MNP clearly outperforms all the other four models, underscoring the importance of considering both heteroscedasticity as well as correlated error terms when estimating mode choice models. Policy implications are also examined and discussed.

Language: en

Keywords

Cross-nested logit (CNL) model; Mode choice; Multinomial probit (MNP) model; Multivariate normal cumulative distribution (MVNCD); Two-variate bivariate screening (TVBS) approach