Modeling safety effects of geometric design consistency on two-lane rural roadways using mixed effects negative binomial regression

Open Access
Butsick, Andrew Joseph
Graduate Program:
Civil Engineering
Master of Science
Document Type:
Master Thesis
Date of Defense:
March 19, 2014
Committee Members:
  • Paul Peter Jovanis, Thesis Advisor
  • geometric design consistency
  • SPFs
  • mixed effects negative binomial regression
  • Sites with Potential (SWiPs)
Previous research has examined the relationship between roadway safety and design consistency using measures such as the difference between design and operating speeds and the difference in operating speeds on successive elements. While such measures have proven effective in identifying inconsistencies in the roadway, they do not directly identify the conditions associated with safety performance. The purpose of this research was to directly quantify the effects of geometric design consistency on roadway safety using measures that can be linked to specific geometric elements. To do so, five years of crash data and roughly 5,000 miles of alignment data from the state of Washington were utilized to model crash experience on 2.5 mile segments. Using mixed effects negative binomial modeling, three safety performance functions (SPFs) were developed. The first contained typical roadway parameters that were suggested for use by several contemporary safety management tools, while the second contained various geometric design consistency measures developed from the dataset. The final SPF contained both typical and design consistency parameters. After Empirical Bayes adjustments were applied using the conditional overdispersion parameters from the mixed effects negative binomial models, sites with potential (SWiPs) for safety improvements were ranked for each model using the scaled differences in frequencies between the predicted and adjusted number of crashes. A comparison was then made based on differences in SWiP rankings between the typical parameter model and the final model containing additional design consistency parameters. Ultimately, 40 unique segments were identified by each SPF out of the top 220 segments ranked; this constitutes a 19 percent change in the top 10 percent of segments identified as SWiPs. Additionally, there was marked variation in the order in which SWiPs were ranked. This disparity may lend credence to the incorporation of geometric design consistency parameters in the development of predictive safety models. Ultimately, by directly modeling the inconsistencies in geometric roadway design, practitioners may be able to better identify and categorize unsafe roadways both in the design stage and post-construction. However, it is important to note that the use of design consistency parameters does not ameliorate the modeling process solely based on a difference in SWiP identification; rather, it should encourage further avenues of research into the use of such measures in predictive safety modeling. Although this investigation is only preliminary, the results may help to burgeon the ever-expanding body of literature regarding the relationship between geometric design consistency and roadway safety.