Analysis of accident injury-severities using a correlated random parameters ordered probit approach with time variant covariates

Abstract

This paper employs a correlated random parameters ordered probit modeling framework to explore time-variant and time-invariant factors affecting injury-severity outcomes in single-vehicle accidents. The proposed approach extends traditional random parameters modeling, by accounting for possible correlations among the random parameters. On the basis of an unrestricted covariance matrix for the random parameters, the proposed framework can capture the combined effect of the unobserved factors – which are captured by the random parameters – on the injury-severity mechanism. The empirical analysis is based on traditional roadway-, traffic- and crash-specific information, and detailed weather and pavement surface disaggregate data, collected in the State of Washington, between 2011 and 2013. The results show that accident injury-severity outcomes are affected by a number of time-variant (ice thickness or water depth on pavement surface, sub-surface temperature) and time-invariant (roadway geometrics, and vehicle-, driver-, and collision-specific characteristics) factors, several of which result in statistically significant parameters – thus they have mixed effects on the injury-severity generation mechanism. The findings also present statistically significant correlation effects among the random parameters, which substantiates the appropriateness of the approach. The comparative assessment between the employed approach and its lower-order counterparts (i.e., fixed parameters, and uncorrelated random parameters ordered probit modeling approaches) shows that accounting for the unobserved heterogeneity interactions results not only in superior statistical performance (in terms of model’s fit, and explanatory and prediction performance) but also in less biased and more consistent parameter estimates.