Citation

Kim D, Kim DK, Lee C. Transp. Res. Rec. 2013; 2398: 67-74.

Copyright

(Copyright © 2013, Transportation Research Board, National Research Council, National Academy of Sciences USA, Publisher SAGE Publishing)

DOI

10.3141/2398-08

PMID

unavailable

Abstract

The aim of this study is to develop safety performance functions reflecting the categorical impact of exposure variables that may vary with freeway segments. A four-step procedure is constructed. It includes clustering analysis, distribution selection, model specification, and model integration. First, clustering analysis is employed to classify freeway segments into three similar groups. A goodness-of-fit test is performed to select suitable distributions for the safety performance functions of each of the three groups. Three forms of relationships are compared between crash frequency and exposure variables, and the best one is selected by using statistical indexes. Two final models, based on the evaluations performed by a test for taste variations and a paired asymptotic t -test, are proposed: one for fatal injury crashes and one for total crashes. All coefficients and constants in the proposed models are statistically significant. In addition, both models show higher statistical significance than the models reflecting constant relationships between crash frequency and exposure variables. The proposed four-step procedure for safety performance functions makes it possible to predict crash frequency more accurately, and the procedure can be easily applied to predict the number of crashes without additional data or complex simulation procedures. If additional explanatory variables are available for criteria on the difference and similarity, the ability to explain the data may be enhanced. It would be necessary to select an appropriate tool for characteristics of targeted roads; this process is a subject for further research.