Developing context-specific safety performance functions for Florida intersections to more accurately predict intersection crashes

Abstract

Safety performance functions (SPFs) are vital tools used to predict and reduce intersection crashes. Because SPFs developed in the Highway Safety Manual (HSM) only use data from certain states, several states have developed region-specific SPFs. However, these SPFs typically only utilize the three roadway categories in the HSM. This research developed SPFs based on a new context classification system used by the Florida Department of Transportation (FDOT) which categorizes intersections into eight different categories. Zero-inflated negative binomial (ZINB), zero-inflated Poisson, and hurdle models were developed and compared to the commonly used negative binomial (NB) and Poisson models for four context classification groups. To develop these context-specific SPFs, data for 29 variables were collected based on the Model Inventory of Roadway Elements 2.0, allowing for standard data collection across agencies. A statistically significant linear regression model (adjusted R 2 = 0.684) was built to predict missing minor AADT volumes. ZINB models outperformed the other models for the two unsignalized intersection groups, whereas NB models performed the best for the two signalized intersection groups. The influential variables differed for each group, showing how FDOT’s context classification system can identify specific crash-influencing factors for different classifications, helping agencies better reduce intersection crashes.