Graphical Models, Surrogate Endpoints, and Crash Modification Factors
In road safety, a commonly-used measure of treatment effectiveness is the crash modification factor, and here we explore the possibility of using surrogates to estimate crash modification factors. As in other situations where observational data are used to estimate causal effects it is necessary to leverage background causal knowledge with observational results. When the background knowledge is such that a crash-generating mechanism can be represented by a directed acyclic graph the connectivity structure of the graph can be used to identify candidate surrogates. The modification factor associated with a safety-related improvement can then, in principle, be estimated from knowledge of how the improvement affects the surrogates, together with information on how the surrogates are distributed in the population of crashes.
About the speaker
Gary Davis is a Professor at the University of Minnesota and holds the Richard P. Braun/CTS Chair in Transportation. He received Bachelor degrees in Philosophy and Psychology from Eastern Washington University and Masters and PhD degrees in Transportation Engineering from the University of Washington.
His research interests include causal inference and impact assessment in traffic safety, the application of Bayesian statistical methods in traffic and transportation engineering, and practical astronomy in 18th-19th century exploratory surveying.