Use of Propensity Score Matching Method and Hybrid Bayesian Method to Estimate Crash Modification Factors of Signal Installation

Abstract

In recent years increased attention has been given to predicting the effects of roadway improvements on traffic safety. Tools have been developed in cooperation with FHWA and TRB that attempt to predict crash experience, and they require estimates of crash modification factors (CMFs) to produce predicted reduction benefits. The tools make use of empirical Bayes statistics, which currently require that crash data be overdispersed. The main purpose of this paper is to illustrate two alternative methods of estimating CMFs that can be applied whether or not the crash data are overdispersed. The first combines the hierarchical Bayes model with a model that allows for temporal changes in the covariates. The second uses an estimated propensity score to match a group of locations that had a countermeasure installed with a comparison group that was not treated. Both methods were used to compute estimates for the CMFs associated with signalizing a set of nonrural intersections. Neither method found evidence of an overall reduction or an increase in crashes following signalization. However, use of the first method indicated that right-angle crashes decreased while rear-ending crashes increased, as has been observed in other studies.