Application of Bayesian ordinal logistic model for identification of factors to traffic barrier crashes: considering roadway classification

One of the main objectives of policymakers is to reduce crash severity due to high social impacts and economic loss associated with severe crashes. One of the most efficient ways to achieve this objective is through identification of the contributory factors to severe crashes. Highway traffic barriers have been installed with the objective of protecting motorists who have drifted off the roadway. Although these traffic barriers save many lives, the crash severity for these crashes are disproportionally high. Only traffic barriers crashes were considered in this study to identify the factors for these types of crashes. Moreover, due to the importance of low volume crashes, especially in rural areas like Wyoming, this study investigated the effects of road classification on crash severity as well as how these effects impact the role of the contributory factors. Low volume roads often receive less attention in terms of road safety due to their low crash frequencies. A Bayesian approach was used to fit the models since this approach does not require large sample assumptions, it does not rely on approximations for estimating nonlinear functions of the parameters, and also it provides simpler interpretations for model unknowns. The factors identified by this study included the main effects of day of the week, seasonality, improper restraints as well as the interaction effects of low volume roads with shoulder width, road surface conditions, and lighting conditions. These interaction terms indicated that the effects of these contributory factors change with the traffic volume. Possible causes of the significant main and interaction terms are discussed in the manuscript.

Automated summary

One efficient way to reduce crash severity is to identify contributory factors to severe crashes. Low volume roads, despite their low crash frequencies, also need attention for road safety. A Bayesian approach was used in this study for fitting models as it does not require large sample assumptions.