Surrogate-based Transferability of Crash Modification Factors

Abstract

Crash modification factors (CMF), commonly used to measure roadway treatments’ safety effectiveness, are critical inputs to Highway Safety Manual (HSM). Although the transferability issue of CMF has been a hot topic ever since HSM first came out, most research has tended to focus on geographic transferability. Existing CMFs are statistical summaries of conditions prevailing on North American roads in the past few decades. As automated vehicles (AV) gradually increase in market share and improve in capability, an emerging issue regarding CMF’s transferability is the transfer of CMFs estimated in current situation to future situation where AVs are present. This study aimed to develop novel methods, using surrogate measures, for estimating CMF associated with roadway treatments and how existing estimates of CMFs might be transferred to situation with AVs present. When the crash-generating process can be represented by a directed acyclic graph (DAG), under certain conditions, a set of surrogates that support the transfer of a CMF estimated in one situation to a different situation can be identified based on the connectivity features of the DAG. This allows one to learn about the relation between crash outcome and surrogates in one situation, learn about the relation between surrogates and the countermeasure of interest in a different situation, and combine these relations to transport CMF from one situation to another. Analytical results were developed for three simplified but plausible examples and illustrated by computational examples to explicate this claim. Besides, asymptotic theories and uncertainty quantification methods for the resulting CMF estimators were established to support the practical application of this approach. Finally, suggestions for further study were provided. This study is an effort to answer a broader question of how past roadway safety research accomplishments might be leveraged with a newer condition, an environment with AVs, to support future decision-making.