Safety Performance Function for Undivided Rural Two-Lane Roadways Using Regression Tree Analysis

The objective of the research was to develop a safety performance function (SPF) for undivided rural two-lane roadways, using all crash types, specific to the state of Wisconsin with the use of regression tree analysis to enhance the development of the models. The initial steps in the study involved a regression tree analysis of the Wisconsin roadway segment database. The analysis split the segments into smaller subsets at statistically significant locations in the data, including number of lanes and divided/undivided segments. One of the splits separated rural two-lane segments from the rest of the data which led to the following research study. The model development methodology was designed to complement the practices of Wisconsin Department of Transportation (WisDOT). Regression tree analysis was used directly with chosen significant variables, instead of using peer groups or the Highway Safety Manual predictive method because it minimized bias and worked well with WisDOT practices. The significant variables included curve presence, lane width, right shoulder total width, right shoulder paved width, truck percentage, international roughness index (IRI), and posted speed. Regression tree results identified the annual average daily traffic (AADT) at 2,000 vpd to be the point where there was significant difference in crashes. Models were developed using a data set of 7,883 segments with more than nine variables. A Negative Binomial regression model was used along with backward stepwise selection to determine the final crash prediction models. Results show that using two models, split by AADT at 2,000 vpd, were more accurate models than using a single grouping of AADT.

  • Supplemental Notes:
    • This paper was sponsored by TRB committee ANB25 Standing Committee on Highway Safety Performance.
  • Corporate Authors:

    Transportation Research Board

    500 Fifth Street, NW
    Washington, DC  United States  20001
  • Authors:
    • Martz, Paige
    • Bill, Andrea R
    • Khan, Ghazan
    • Noyce, David A
  • Conference:
  • Date: 2017

Language

  • English

Media Info

  • Media Type: Digital/other
  • Features: Figures; References; Tables;
  • Pagination: 14p
  • Monograph Title: TRB 96th Annual Meeting Compendium of Papers

Subject/Index Terms

Filing Info

  • Accession Number: 01626395
  • Record Type: Publication
  • Report/Paper Numbers: 17-05652
  • Files: TRIS, TRB, ATRI
  • Created Date: Feb 23 2017 2:56PM