Analysis of Accident Risk at Railroad Grade Crossing

Increasing number and more severed traffic accidents are observed at railroad grade crossings. These accidents have caused significant impacts both from personal and societal perspectives. The factors that cause traffic accidents at a railroad grade crossing can be attributed to either railway or highway related characteristics, or both. They may also be caused by driveri¦s aggressive behaviors or incautious actions. These factors have different degrees of impact on the potential risk in terms of accident likelihood and severity at a typical railroad grade crossing. The purposes of this research are to compare the capabilities of various count data models in identifying the risk level at a railroad grade crossing, and to evaluate the model performance in the prediction of accident risk at a railroad grade crossing. In the present study, we have evaluated three statistical models based on the historical accident data and railway/highway traffic and geometric characteristics in Taiwan. They include Poisson, negative binomial, and zero-inflated Poisson (ZIP) models. The empirical study indicates that the ZIP based models generally outperform the other two models. Moreover, traffic exposure related variables (e.g., number of daily trains and AADT) are generally found to have different degrees of impact on accident risk. The model evaluation results may provide the government agencies with beneficial information in identifying the hazardous crossings and preparing a safety improvement program accordingly.