Causal inference in transportation safety studies: Comparison of potential outcomes and causal diagrams

摘要

The research questions that motivate transportation safety studies are causalin nature. Safety researchers typically use observational data to answer suchquestions, but often without appropriate causal inference methodology. Thefield of causal inference presents several modeling frameworks for probingempirical data to assess causal relations. This paper focuses on exploring theapplicability of two such modeling frameworks---Causal Diagrams and PotentialOutcomes---for a specific transportation safety problem. The causal effects ofpavement marking retroreflectivity on safety of a road segment were estimated.More specifically, the results based on three different implementations ofthese frameworks on a real data set were compared: Inverse Propensity ScoreWeighting with regression adjustment and Propensity Score Matching withregression adjustment versus Causal Bayesian Network. The effect of increasedpavement marking retroreflectivity was generally found to reduce theprobability of target nighttime crashes. However, we found that the magnitudeof the causal effects estimated are sensitive to the method used and to theassumptions being violated.