Injury, Fatal, and Property Damage Accident Models for Highway Corridors

Abstract

The relationship between fatal, injury, and property damage accident frequencies and traffic volume, segment length (SL), and vehicle miles traveled (VMT) is explored here. For this purpose, a generalized linear regression modeling framework was applied. Poisson, negative binomial, Gaussian, and log-normal distributions were evaluated in terms of their ability to model accident frequency for an interstate highway corridor in Colorado. Results indicate that, for injury, property damage, and total accidents, the Poisson regression with log-transformed predictors performed significantly better than the Poisson regression with linear predictors. For fatal accidents, the log-normal regression performed better than other regression models examined. However, the Poisson model performed better in terms of estimating accidents for the corridor. Models based on data disaggregated by median SL and annual average daily traffic (AADT) or VMT performed better than aggregated models. In addition, the SL-AADT–based models estimated a smaller difference between observed and expected total number of accidents for the corridor than the VMT-based model. SL and traffic volume were significant for injury and property damage accidents. For fatal accidents, an exposure-based predictor VMT was significant across all segment groups.