Revisiting spatial correlation in crash injury severity: a Bayesian generalized ordered probit model with Leroux conditional autoregressive prior

Abstract

To account for the spatial correlation of crashes that are in close proximity, this study proposes a Bayesian spatial generalized ordered probit (SGOP) model with Leroux conditional autoregressive (CAR) prior for crash severity analysis. Proposed model can accommodate the ordinal nature of injury severity and relax the assumption of monotonic effects of explanatory factors. Additionally, strength of spatial correlation is considered. Results indicate that the proposed SGOP model with Leroux CAR prior outperforms the conventional ordered probit model and SGOP model with intrinsic CAR. There is moderate spatial correlation for the crashes. Results indicate that factors including vehicle type, horizontal curvature, vertical grade, precipitation, visibility, traffic composition, day of the week, crash type, and response time of emergency medical service all affect the crash injury severity. Findings of this study can indicate the effective engineering countermeasures that can mitigate the risk of more severe crashes on the freeways.

KEYWORDS:

• Traffic crash
• injury severity
• generalized ordered probit model
• spatial correlation
• Leroux conditional autoregressive prior

Disclosure statement

No potential conflict of interest was reported by the author(s).

Additional information

Funding

This work was jointly supported by the International Science & Technology Cooperation Program of China [grant number 2017YFE0134500], the Natural Science Foundation of China [grant number 71801095], the Research Grants Council of Hong Kong [grant number 25203717], and the Fundamental Research Funds for the Central Universities [grant number 2020ZYGXZR007].