Citation

Claros BR, Sun C, Edara P. J. Transp. Saf. Secur. 2020; 12(2): 309-328.

Copyright

(Copyright © 2020, Southeastern Transportation Center, and Beijing Jiaotong University, Publisher Informa - Taylor and Francis Group)

DOI

10.1080/19439962.2018.1477896

PMID

unavailable

Abstract

There are various ways in which a transportation agency can approach safety prediction. One could calibrate the Highway Safety Manual (HSM) using adjustment factors by ranges of exposure variable, calibrate it using a function, or develop jurisdiction-specific models. Various tradeoffs are involved in deciding on which approach to undertake, including minimum sample size, data required, data processing, modeling, statistical expertise, labor involved, and accuracy of estimate. A comprehensive case study involving urban freeway four-lane segments (FU4) in Missouri is presented along with a discussion of general tradeoffs. Highway Safety Manual calibration was performed over the entire range of predictor variables. For calibration functions, regression modeling was performed, and calibration function forms were explored using annual average daily traffic (AADT) and segment length as additional predictor variables. Missouri specific safety performance functions (SPF) were developed and cumulative residuals (CURE) plots were used for comparative analysis of all model approaches. For calibration functions and SPFs, the inverse overdispersion and log-likelihood were evaluated in addition to the CURE plots. The results showed that calibration by AADT ranges outperformed all other calibration factors and functions proposed. The jurisdiction-specific SPF had similar accuracy as fully loaded and calibrated HSM models while not requiring the extensive data collection and processing of freeway-related crash modifications factors.

Language: en

Keywords

calibration; crash prediction; freeway segments; Highway Safety Manual