Evaluating Local and Tribal Rural Road Design with the Interactive Highway Safety Design Model (IHSDM)

Establishing performance-based safety goals and objectives becomes more attainable with the Highway Safety Manual (HSM). However, the safety performance functions (SPFs) in the HSM may not be accurate as they are not calibrated to local conditions. In addition, each SPF and crash modification factor (CMF) assumes a set of base site conditions which may not be realistic for local roadways. Although calibration procedures are available in HSM Part C Appendix A, they should be refined or modified to accommodate local data availability and roadway, traffic, and crash characteristics. It is also necessary to determine a set of base conditions applicable to local highways. This document presents the application of the HSM for rural local two-lane two-way highway segments in South Dakota (SD). The calibration was based on three-year (2009-2011) crash data from 657 roadway segments constituting more than 750 miles of roadways. The calibration process includes establishing new base conditions, developing SPFs, converting CMFs to base conditions as well as substituting default values with state-specific values. Five models have been developed and compared based on statistical goodness-of-fit and calibration factors. The same procedures were also conducted for the tribal two-lane two-way highway segments in SD based on three-year (2009-2011) crash data from 56 roadway segments constituting 199.5 miles of roadway. Results show that the jurisdiction-specific crash type distribution for CMFs can be drastically different from what is presented in the HSM. For rural local two-lane two-way roadways, the HSM method without modification underestimates SD crashes by 35 percent. The method based on SPFs developed from a full model has the best performance. For tribal two-lane two-way roadways, the HSM method without modification overestimates SD crashes by 122 percent. The method using the exponential from of annual average daily traffic (AADT) performs the best. This documentation provides important guidance and empirical results regarding how to calibrate HSM models.