A Comparative Analysis of Bayesian Methods for Identifying Highway Locations with High Proportions of 'Hot Spots'

摘要

The identification ofhighway system locations having safety problems, or "hot spots," is an importantfirst step in safety management. Most techniques for identifying such locations have utilized greaterthan expected crash :frequencies for making this determination. However, statistical techniquesutilizing higher-than-expected proportions of target crashes can also be used to identify such sites. Therecently released Highway Safety Manual (HSM) includes such a technique. This method identifiescrash prone locations by making inferences from a Bayesian posterior beta-binomial probabilitydistribution ofthe crash rate at each location. For this research, another Bayesian method is developedand compared to the current HSM methodology. Specifically, a hierarchical Bayesian logisticregression model is used to directly model individual site effects, including traffic volume. Sitespecific inferences can then be made from a Bayesian posterior model. For this method, a mixture ofthree normal distributions was used to estimate site effects, rather than a typical single normaldistribution. Direct comparison ofthese methodologies demonstrated that the hierarchical Bayesianmodel to be better suited for all distributions, but particularly for multimodal or sparsely distributeddata. Future areas of research are also identified.