Publication Details
Comparison of crash modification factors for engineering treatments estimated by before - after Empirical Bayes methods and Propensity Score methods
Type: Paper
Subtype: Final report
Author(s): Lan, Bo; Srinivasan, Raghavan
Publisher: Southeastern Transportation Center
Publication Date: Dec-2017
Address: Knoxville, TN
Abstract: Cross-sectional and the empirical Bayes (EB) before-after are the two most common methods for estimating crash modificationfactors (CMFs). The EB before-after method has now been accepted as one way of addressing the potential bias due to RTM.However, the EB requires the before and after periods data and they may not be available. In those cases, researchers rely on cross-sectional studies to develop CMFs. One of the primary challenges of cross-sectional studies is it cannot address confounding issue,thus the estimated CMFs may be biased and unreliable. The propensity score (PS) methods along with cross-sectional regression models is one of the methods that can be used toremove the confounding effects of such factors if they are measured in the data. Though the propensity score methods are widelyused in epidemiology and other studies, there are only a few studies using the propensity score methods in CMF derivations intransportation safety. The intent of this study is to evaluate and compare the performance of cross-sectional regression models that make use ofpropensity scores with the results from the EB and traditional cross-sectional methods. The cross-sectional method that make useof various propensity score methods were explored in this study. These methods were evaluated and compared with the traditionalcross-sectional and the EB methods using two carefully selected simulated datasets. It was found the optimal propensity scoredistance (PSD) matching with maximum variable ratio of 5 performs best using the two datasets. It correctly identifies the trueCMFs in the two datasets while the EB and the traditional cross-sectional methods failed.