Full Bayes Before-and-After Evaluation of Traffic Safety Improvements in City of Edmonton, Canada

Abstract

The objective of this study was to evaluate the safety performance of a sample of intersections that had been improved with the implementation of certain safety countermeasures targeting right-turn collisions in the city of Edmonton, Canada. A full Bayes approach was used to determine the effectiveness of the improvements by employing a before-and-after design with matched (yoked) comparison groups. Three linear intervention models were considered: a multivariate model that modeled treatment effects as a gradual change, a similar model with the addition of a jump treatment effect, and a univariate model that specifically analyzed right-turn collisions. The results indicated that the safety improvement program was effective; up to 40% of right-turn collisions were reduced. Despite the small sample size, these reductions were statistically significant. The results show the usefulness of the full Bayes technique in performing before-and-after evaluations of traffic treatment programs and in eliminating the need for a reference population and also in allowing for additional types of analysis, including multivariate analysis (modeling collisions of different types and severities at the same time), temporal effects (for both treatment and long-term trends), and greater freedom in the selection of error structure.