Book Description

The primary objective of practitioners working on traffic safety is to reduce the number and severity of crashes. The Highway Safety Manual (HSM) provides practitioners with analytical tools and techniques to estimate the expected crash frequency and severity with the aim to identify and evaluate safety countermeasures. Expected crash frequency can be estimated using Safety Performance Functions (SPFs) provided in Part C of the HSM. The HSM provides simple SPFs which are developed using the most frequently used crash counts model, the negative binomial regression model. The rural nature of Wyoming highways coupled with the mountainous terrain (i.e., challenging roadway geometry) make the HSM basic SPFs unsuitable to determine crash contributing factors for Wyoming conditions. In this regard, the objective of this study is to implement advanced statistical methods such as the different functional forms of Negative Binomial, and Bayesian approach, to develop crash prediction models, investigate crash contributing factors, and determine the impact of safety countermeasures. Bayesian statistics in combination with the power of Markov Chain Monte Carlo (MCMC) sampling techniques provide frameworks to model small sample datasets and complex models at the same time, where the traditional Maximum Likelihood Estimation (MLE) based methods tend to fail. As such, a novel No-U-Turn Sampler for Hamiltonian Monte Carlo (NUTS HMC) sampling technique in a Bayesian framework was utilized to investigate the crash frequency, injury severity of crashes on the interstate freeways and some rural highways in Wyoming. The Poisson and the Negative Binomial (NB) models are the most commonly used regression models in traffic safety analysis. The advantage of the NB model can be further enhanced by providing different functional forms of the variance and the dispersion structure. The NB-2 is the most common form of the NB model, typically used in developing safety performance functions (SPFs) largely due to the mean-variance quadratic relationship. However, studies in the literature have shown that the mean-variance relationship could be unrestrained. Another introduced formulation of the NB model is NB-1, which assumes that there is a constant ratio linking the mean and the variance of the crash frequencies. A more general type of the NB model is the NB-P model, which does not constrain the mean-variance relationship. Thus, leveraging the power of this unrestrained mean-variance relationship, more accurate safety models could be developed, and these would lead to more accurate estimation of crash risk and benefits of potential solutions. This study will help practitioners to implement advanced methodologies to solve traffic safety problems of rural highways that have plagued the researchers for a long time now. The methodologies proposed in this study will help practitioners to replace the outdated and inefficient traditional models and obtain more accurate traffic safety models to predict crashes and the resulting crash injury severity. Moreover, this research quantified the safety effectiveness of some unique countermeasures on rural highways.