Highway design standards and operational characteristics in relation to levels of safety on interurban roads in Israel

This study deals primarily with the levels of safety of the main interurban road network in Israel and covers 70% of the total interurban roads, with a total of 8910 police-reported accidents in the years 1990 to 1992. It presents a macro approach to the relation between road safety and various road categories as reflected in the overall geometric and traffic operational features and explores the effect of other individual geometric features on safety. The safety performance of the various road categories are investigated on an comparative basis. After an extensive literature review, advanced statistical tools are applied to handle the randomness in the occurrences of accidents and to explore the relations between safety and other explanatory variables. The Classification and Regression Tree (CART) analysis, which generates a binary tree structure from the data showing the criteria (variables) for each split and giving pictorial representation of the data, is used to illuminate the relation between the variables and road accidents as a measure of safety. The CART results portray the importance of ADT and the need for stratification of ADT in accident modelling. It also indicates probable power function models and highlights the effects of shoulder type among single carriageways. The occurrence and number of road accidents are discrete random events which are probabilistic in nature and have non-negative integer values; thus after an extensive literature review multi-variate multiplicative Poisson regression models with log links are fitted to the data set using the Generalized Linear Interactive Modeling (GLIM) package. The results emphasize the relative safer nature of freeways as compared to the conventional single and dual carriageway roads and show the relation between between average daily traffic (ADT) and safety is curvilinear meaning that as ADT increases and congestion sets in, the incidence of accidents decreases. For the fitted models the effects of ADT were found to be most important. In the case of models for single carriage roads, other explanatory variables, namely free flow speed, type of shoulders and junction frequency showed some significance. The Poisson models developed can be used as a basis for the identification of accident blackspots and in the evaluation of safety treatments.