Safety Performance Function Calibration and Development for the State of Alabama: Two-Lane Two-Way Rural Roads and Four-Lane Divided Highways

摘要

The Highway Safety Manual (HSM) published by the American Association of State HighwayTransportation Officials provides procedures and statistical tools for estimating the expectednumber of crashes for various roadway facilities. One critical component of the HSM method isthe Safety Performance Function (SPF). SPFs are essentially regression models that correlatequantitatively the expected number of crashes with traffic exposure and geometric characteristicsof the road. Since these models are developed using data from selected states, theirtransferability is not gauranteed. As part of a project peformed by researchers from theUniversity of Alabama to facilitate the implementation of the new HSM procedures in the Stateof Alabama, this study aims to evaluate the applicability of HSM predictive methods to Alabamadata, and to develop state-specific statistical models for two facility types, namely two-lane two-way rural roads and four-lane divided highways. This study first calibrates the HSM base SPFsusing two approaches. Besides the method recommended by HSM, this study also proposes anew approach that treats the estimation of calibration factors as a special case of a negativebinomail regression. In addition, new forms for state-specific SPFs are further investigated toidentify the best model using Poisson-Gamma regression techniques. Four new functional formsare studied in this project. The prediction capabilities of the two calibrated models and the fournewly developed state-specific SPFs are evaluated using a validation data set. Five performancemeasures are considered for model evaluation. They are the mean absolute deviance, the meansquared prediction error, the mean prediction bias, the log likelihood value, and the Akailke'sinformation criterion. The study is able to identify a particular state-specific SPF that fits theAlabama data well and outperforms other models, including the calibrated SPFs. The best modeldescribes the mean crash frequency as a function of annual average daily traffic, segment length,lane width, year, and speed limt. The study finds that the HSM-recommended method forcalibration factor estimation also performs well. Although it is not as good as the best state-specific SPF, it is still a good alternative considering the approach is very straightforward andcan be easily applied.