The 2010 Highway Safety Manual (HSM) provides methods for predicting the number of motor vehicle crashes on various roadway facilities (AASHTO, 2010). However, it includes only a simplistic method for predicting the number of bicycle-related crashes. The research team investigated the bicycle-specific crash data in eight potential study areas around the U.S.: Arlington, VA (city/county); Bellingham, WA (city); Boulder, CO (city); Denver, CO (city/county); Minneapolis and St. Paul, MN (cities); Philadelphia, PA (city/county); Portland, OR (city); and San Diego, CA (county). The available online data from each were compared. The study city for future analysis (Boulder) was selected based on the availability of not just crash data but also the availability of continuous and short-duration bicycle and pedestrian traffic count data. In this analysis, a negative binomial model with log link was used to predict annual, non-fatal, motorist-bicyclist crashes on road segments per mile. This report adopts methods from the HSM used for motor vehicle safety performance functions (SPFs) in order to develop bicyclespecific SPFs for roadway segments in Boulder (9). The analysis shows that motor vehicle volume is a leading factor associated with more crashes between motor vehicles and bicyclists. Bicyclist exposure, population density, and percent retail land use are also predictive. While both vehicle volume and bicycle volume data are used in the model in order to account for the “safety in numbers” effect, the model did not demonstrate this effect that is seen so commonly in other research, including the bicycle SPF developed previously for intersections in Boulder (11). This effort at developing a bicycle-specific SPF for segments in the U.S. that utilize bicycle volumes is an important first step towards further understanding bicyclist safety and may inform future versions of the HSM. To that end, the report includes a table of motorist-cyclist crashes predicted future efforts to by the model for various values. The authors hope this table may serve as a potential format template and starting point for generalize the results of models for possible use in HSM updates
