Missouri-Specific Crash Prediction Model for Signalized Intersections

Abstract

The Highway Safety Manual (HSM) provides guidance and tools to conduct quantitative safety analysis. Crash prediction models are used to estimate the expected number of crashes per year, by facility type, severity, and crash type. There are two approaches for applying the HSM crash prediction methodology to local conditions: (1) calibration of models provided in the HSM; or (2) development of jurisdiction-specific models. There are some instances in which model calibration may not be appropriate. To illustrate this case, 601 urban signalized four-leg intersections (U4SG) in Missouri were used to obtain the calibration factor, assess the quality of the calibration factor, and develop jurisdiction-specific models. For U4SG total crashes, the calibration factor for Missouri conditions was 3.98 (standard deviation, 0.13). The assessment of the calibration factor showed a disproportional difference between the observed data in Missouri and the HSM model. Thus, the calibration was deemed inappropriate and the development of Missouri-specific models was supported. The models were developed for severities Fatal and Injury (FI) and Property Damage Only (PDO) crashes. The predictor variables considered were intersection AADT, posted speed limit, signal control type, exclusive left turn lanes, exclusive right turn lanes, right turn on red prohibited, and facilities of interest within 1,000 ft from the intersection (bus stops, schools, and alcohol sale establishments). Functional forms for all predictor variables were optimized. The log-likelihood, inverse overdispersion, and Cumulative Residuals (CURE) plots showed satisfactory measures of model accuracy.