Estimating the Safety Effects of Horizontal Curves on Pennsylvania Two-lane Rural Roads

Open Access
Gooch, Jeffrey Paul
Graduate Program:
Civil Engineering
Master of Science
Document Type:
Master Thesis
Date of Defense:
March 25, 2015
Committee Members:
  • Eric Todd Donnell, Thesis Advisor
  • Vikash Varun Gayah, Thesis Advisor
  • Martin T Pietrucha, Thesis Advisor
  • Highway safety
  • traffic safety
  • horizontal curves
  • propensity scores
  • potential outcomes
  • negative binomial
  • mixed effects
Roadway departure crashes are three times more likely on horizontal curves than on tangent sections of two-lane rural highways. Research is needed to better quantify the safety performance (e.g., crash frequencies) of these crash-prone locations. Existing safety performance functions (SPFs) for two-way rural highways in the Highway Safety Manual (HSM) rely on base conditions that assume all roadway segments are tangent sections. The resulting predictions of crash frequency are then modified using a crash modification factor (CMF) that includes as independent variables the curve length, curve radius and presence of a spiral transition. Unfortunately, the CMF in the HSM does not have a standard error associated with it, which greatly limits its practical application and provides no indication of the level of uncertainty associated with the CMF. Other existing CMFs for horizontal curves in the FHWA CMF Clearinghouse also suffer from a lack of standard error and dated statistical methods, resulting in low to average star quality ratings (three or less). Furthermore, these existing CMFs treat individual curves as isolated geometric elements, even though recent research has shown that the safety performance of a horizontal curve is significantly influenced by its proximity to neighboring curves. Curves in close proximity to each other are expected to have lower crash frequency than those that are isolated because they are less likely to violate driver expectation. Failure to account for this distance may lead to erroneous predictions of crash frequency on horizontal curves. The objective of this study is to develop a high-quality CMF for horizontal curves on two-lane rural roads that takes into consideration the proximity of neighboring geometric elements using the most statistically rigorous modeling technique available, which reduces the potential for bias in the estimate of the CMF and provides a standard error that can be used to estimate the CMF uncertainty. This CMF is estimated using eight years of crash data (2005-2012) obtained from over 10,000 miles of state-owned two-lane rural roads in Pennsylvania. This comprehensive data includes information on roadway geometry (e.g., horizontal curvature, lane width, and shoulder width), traffic volume, access density, roadside hazard rating, and the presence of various low-cost safety countermeasures (e.g., centerline and shoulder rumble strips, curve and intersection warning pavement markings, and aggressive driving pavement dots). The propensity scores-potential outcomes method is applied, which matches each horizontal curve with tangent sections that are similar with respect to all other site characteristics (excluding crash frequency). The propensity scores are estimated using binary logistic regression, and curves and tangents are matched based on the propensity scores using the nearest neighbor matching technique with calipers and without replacement. Matching is performed across county lines to avoid individual horizontal curves being matched with its neighboring upstream or downstream tangent sections, which are likely to have very similar features and endogenous effects. Crash prediction is performed by means of random effects and mixed negative binomial regression using the explanatory variables mentioned above as well as distance to adjacent horizontal curves. The results indicate that degree of curvature, curve length, and traffic volume must be considered when predicting the frequency of total and fatal and injury crashes on horizontal curves. The presence of a horizontal curve and degree of curvature increase crash frequency, while the length of curve and traffic volumes decrease expected crash frequency. These results were consistent for both random and mixed effects models. The impact of the distance to adjacent curves was not found to be statistically significant. When predicting fixed object crashes, a proxy for roadway departure crashes, only degree of curvature and the presence of a horizontal curve were found to be statistically significant. All crash modification estimates for degree of curvature were consistent with the existing literature. The crash modification functions estimated are supplemented with formulas to estimate a conservative value for the standard error of the resulting CMF. The resulting crash modification functions in this thesis are recommended to evaluate safety at rural two-lane horizontal curves in Pennsylvania.