Resumen
Regarding traffic safety, railway crossings are crucial points in the rail and road network. In Hungary the safety ranking of railway crossings has been performed since 1993 and the results have been used for selecting unsafe locations and for prioritizing safety measures. Prioritization is based on a complex point system taking into account accident data, traffic volumes and various traffic engineering aspects. Weights of various indicators were based on engineering judgement. Although this method has been renewed a few times, there is a need to update this method involving more thorough statistical methods. The objective of this paper is to calibrate a new model and to compare it with previous models. A sample of about 1700 railway crossings satisfying a few requirements (railway with public service, traffic volume data available) was set up. The first part of the paper describes data collection. Five years (2010?2014) of accident data, AADT (Annual Average Daily Traffic) for rail and road, speed limits and further data (road pavement width, control devices, crossing angle, alignment, number of tracks, sight distances) were gathered. The modeling was conducted in two steps. First, each variable was entered into the model alone to see which ones significantly affect accident frequency. As a next step, two-variable models were built where expected injury accident frequency is predicted by annual daily road traffic and annual daily rail traffic. The sample was also split up into subgroups based on the type of control devices in order to analyze their effect on safety. Models are proposed using the Generalized Linear Modeling approach (GLM) assuming a negative binomial error structure. The results give estimation of the impacts of explanatory factors on the safety of railway crossings. Annual daily road traffic and annual daily train traffic are significant predictors. A number of predictors such as crossing angle, track alignment, number of tracks and sight distances turned out to be not significant. As far as the type of control devices are concerned, as expected coefficients are decreasing, hence safety is increasing in the following order: passive crossings, flashing lights, flashing lights + half barriers. Based on the above results, the complex point system used by practitioners for safety ranking was simplified considerably leaving out variables which turned out to be not significant.