Citation

Cheng W, Gill GS, Zhou J, Ensch JL, Kwong J, Jia X. J. Transp. Saf. Secur. 2020; 12(5): 628-652.

Copyright

(Copyright © 2020, Southeastern Transportation Center, and Beijing Jiaotong University, Publisher Informa - Taylor and Francis Group)

DOI

10.1080/19439962.2018.1525631

PMID

unavailable

Abstract

The central issue for successful implementation of a multimodal approach is the development of appropriate crash frequency models that can jointly estimate the crash risk of different mode users. This study proposed two multivariate spatial-temporal models to analyze 6 years of modal crash data from counties in California: one with fixed time trend applied to all modes, the other with mode-varying time trend coefficients. These models were compared with three other multivariate models from past studies. The major objective was to examine the benefits of the newly proposed models that have substantially increased computational cost, because both dimensions of time and space, and their interaction, are considered. The model fitness results from penalized criterion revealed that the proposed models significantly improved the model fitting by pooling strength from neighboring sites, consideration of time trend, as well as their interactions. The model with mode-varying time coefficients was observed to be superior in terms of prediction accuracy. The posterior deviance was observed to be the governing factor for overall fit and consistently showed a strong positive correlation between model fit and prediction accuracy. Overall, the study is anticipated to enhance the understanding of safety impacts from the interaction of various modes.

Language: en

Keywords

county; interaction; multimodal; multivariate; prediction accuracy; spatiotemporal