Validating crash surrogates: a causal graphical framework and a SHRP2 case study of hard braking

Although there has been a variety of crash surrogates have been applied in traffic safety areas, formal discussions on how to validate crash surrogates for practical use remain limited. To fill this gap, this study seeks to propose a validation procedure suitable for traffic safety applications. To first establish key assumptions required for valid crash surrogates, this study reviews discussions on surrogate outcome from causal mediation, medical research, epidemiology clinical trials, and economics perspectives. A causally valid surrogate, whether full or partial, is a special type of mediator that lies on the causal pathway from the primary explanatory variable for a given research question (i.e., the "treatment") to the final crash outcome. Moreover, it captures an identifiable portion of the treatment's effect on the outcome, thereby enabling causal inference on the final crash outcome through the surrogate. To determine whether a candidate surrogate satisfies the above properties, the proposed validation procedure consists of three main components: (1) verifying the assumptions required for identifying the causal relationship between the candidate crash surrogate and the crash outcome; (2) quantifying the extent of surrogacy; and (3) evaluating the absence of a surrogate paradox and the comparability of the surrogate. A case study using SHRP2 Naturalistic Driving Study data is then presented to demonstrate the proposed procedure. Specifically, the case study examines whether hard braking is a causally valid surrogate for rear-end crash occurrence.