A Cohort-Based Data Structure Design for Analyzing Crash Risk Using Naturalistic Driving Data

Although naturalistic driving studies (NDS) have become more prevalent in recent years, many challenges remain in analyzing the data. One challenge is inclusion of exposure in modeling crash risk. While this is a potential strength of NDS, comparatively few studies have emphasized exposure-based analyses. A second challenge is the formulation of analysis methods that include driver attributes, event attributes, and driving environment in a structured formulation. A third challenge is the formulation of baseline hazard to frequently accompany the identification of NDS "events" (e.g. crashes, near crashes and/or safety critical events). This paper reports on a cohort-based data structure design to address these three challenges. Collision warning alert frequency data from University of Michigan Transportation Institute (UMTRI)’s Roadway Departure and Curve Warning System (RDCW) Field Operation Test (FOT) are used to demonstrate this approach. The paper concludes with a discussion of applications which include crash and other NDS-observed events, including potential applications to road safety management through the development of enhanced safety performance functions.