Abstract Manager

Crash-based safety analyses present challenges, especially for rare types of crashes, so surrogate safety assessment has been proposed as an alternative approach. This approach uses “near crashes” or “near misses,” generally derived from video observation, as a surrogate for actual crashes. The idea is that these near miss events are most likely associated with crash occurrence and, therefore, are strong indicators of safety. However, there are some differences in using this method to estimate vehicle-to-pedestrian vs vehicle-to-vehicle crashes. In this presentation, which is part of my PhD dissertation, by comparing the findings of two studies, I aim to demonstrate how the same method can be influenced by the type of traffic conflict. This exploration utilized a comprehensive database of pedestrian and left-turn opposed conflicts, encompassing five years of crash data and 24-hour video-derived traffic conflicts. Severe conflicts were identified and classified based on their safety index. Following this classification, models were developed using mixed- and fixed-effects linear regression models and generalized linear models to relate crashes to severe conflicts and to combinations with other variables. Based on the results, conclusions were drawn on the differences in spatial heterogeneity, prediction accuracy, and severity prediction between vehicle-to-pedestrian and vehicle-to-vehicle crash estimation models.