Traffic accident modeling: Statistical issues and safety applications

The success of road safety improvement programs and studies in reducing traffic accidents hinges upon the existence of reliable techniques for the estimation of road safety levels. Accident prediction models are rapidly becoming the preferred mathematical tools for estimating road safety. However, several statistical problems related to accident models need to be addressed. Additionally, there is potential for new model applications that constitute reliable methodologies for conducting certain existing safety studies that currently use earlier inaccurate techniques. The main objective of this thesis is to solve several statistical problems related to accident models and to provide new safety applications of these models.; This thesis presents a detailed discussion of several statistical problems related to accident models, namely the problems of model building, identification and removal of model outliers, and recalibration and testing of transferred models. The thesis devises statistical procedures for model building and conducting outlier analysis and presents sufficient evidence in defense of the validity of these procedures. On the issue of model transferability, the thesis presents a statistical method that can be used to test whether an accident model produces reliable safety estimates when transferred for use in different time periods and different regions of space. The thesis also proposes a new procedure for recalibrating transferred models to better suit local conditions.; The thesis presents three new safety applications of accident prediction models. The first is a methodology of ranking hazardous locations for priority of treatment. The thesis pinpoints effective ranking criteria and employs accident models in their measurement. The second application is a countermeasure-based procedure for identifying hazardous road locations. It calls for the use of pattern-specific accident models to identify locations that are hazardous with respect to accident patterns that can be targeted by known countermeasures. As such, it represents a cost-effective method of black spot treatment. The third application is a cross-sectional method for evaluating the safety benefits of road improvement measures prior to their implementation. This prior evaluation is necessary since alternative improvement measures are usually considered and must be economically evaluated to identify and implement the alternative possessing the highest benefit-cost ratio.