Highway Safety Manual, New Model Parameters vs. Calibration of Crash Prediction Models

Part C of the newly published Highway Safety Manual (HSM) presents crash prediction models for different types of facilities. Each of these models comprises of a Safety Performance Function (SPF) and a series of Crash Modification Factors (CMFs). These models are developed by using data from a few States therefore; the predictions made by these models for other regions could only be valid if they are calibrated for those regions. HSM also recommends the users to develop their own region’s SPFs to use in the crash prediction process. This paper presents a heuristic procedure for developing SPFs and CMFs for rural two-lane highway segments of Washington State comparable to the HSM model and compares the predictions of new models to the predictions of calibrated HSM model. Data from Washington State for over 5,000 miles are used. Half of the data is used for the development of new SPFs and four of the CMFs. The same half is used to calibrate the HSM model for Washington State conditions. The other half of the data is used for comparing the new models predictions to the prediction of the calibrated HSM model. The results show that the new models performances are as good as the performance of the calibrated HSM model. There is also a sensitivity analysis conducted for the minimum data size recommended in the HSM for model calibration. The final recommendations for new model developments vs. calibration of the HSM as well as calibration data size are offered.