Safety performance functions (SPFs) form the analytical backbone of various roadway safety analyses. SPF are either developed as jurisdiction-specific models or adopted from the Highway Safety Manual (HSM) and calibrated to reflect local conditions. This paper first outlines data needs and availability, data processing methods, and approaches to gathering the required data for the calibration and development of SPFs. Rural two-lane two-way segments and intersections in New Jersey are used as a case study. The paper then presents a maximum likelihood–based calibration factor estimation method, an alternative to the one given in the HSM, allowing for the estimation of its standard deviation and the evaluation of sample size adequacy. The results reveal notable differences in calibration factors derived from these two methods. The paper also develops New Jersey–specific SPFs and presents a statistically rigorous evaluation of both the calibrated and developed SPF. A comparative analysis of the calibrated and developed SPF is performed using log-likelihood ratio tests, Rootogram plots, and chi-square tests. The findings caution against relying on mean absolute deviation alone for model comparison, as it fails to account for the intrinsic variability of crash data. The study also highlights that data preparation is often the most resource-intensive aspect of SPF development and calibration, requiring substantial programming and manual effort. Accordingly, the paper emphasizes the need for automated data extraction techniques to reduce costs and improve accuracy, and it underscores the critical role of crash location accuracy in model outcomes.
Bartin et al. (Tue,) studied this question.