Evaluating traffic safety network screening: an initial framework utilizing the hierarchical Bayesian philosophy
Highway crashes result in over 40,000 deaths per year (500,000 worldwide). Their impact on the national economy is estimated at more than 230 billion dollars. Highway safety is the top priority of the United States Department of Transportation (US DOT). Funds dedicated to the problem are expected to increase substantially.;Highway safety is a multidisciplinary issue. An important tool is the safety improvement candidate location (SICL) list. SICL lists list high crash locations for potential mitigation. SICL lists are developed using crash data. Crash frequency, rate, or loss is used to rank the worst locations. Classical statistical techniques are applied. In some cases, simple frequency analyses are used to draw attention to "problem" locations.;Simple ranked lists suffer from methodological and practical limitations. Chief among these is the inability to identify "sites with promise", sites where mitigation has the best chance of success. Agencies representing engineering and enforcement generally examine top sites prior to resource dedication. This is resource intensive and efforts of different safety interests are often not well coordinated.;For over 20 years, empirical Bayesian (EB) has been proposed to address these limitations. EB identifies sites where mitigation might be most effective, increases estimate confidence, and provides information on relative site safety. EB is being widely implemented at the national level. State and local agencies continue SILL development based on long-standing procedures.;EB allows decision makers to more reliably estimate the crash reduction potential at specific sites. However, EB requires development of safety performance functions for road type classes. The technique also requires a priori development of accident modification factors. These requirements add significant expense.;Powerful computers and advanced statistical sampling techniques allow hierarchical Bayesian statistics to be applied to highway safety. Hierarchical Bayesian eliminates the need for a priori functions and factors. This approach can readily incorporate additional information. It can also explicitly identify important relationships between causal factors and safety performance. The approach uses data to define results, based on an analyst-specified level of uncertainty. This dissertation discusses SICL list development and evaluates the potential of Bayesian statistics to improve their utility.