Evaluating e-bike safety at unsignalized roundabouts using a Bayesian mixed logit model

Abstract

Roundabouts are a unique intersection design for calming traffic and improving vehicle safety without traffic signal control. While a few past studies have examined the impact of the roundabout on bicyclist and pedestrian injury crashes, little is known about its effect on the safety of electric bike (e-bike) riders. This study uses a Bayesian mixed logit model to quantify the impact of roundabout geometry, traffic flow and conflict characteristics on the severity of e-bike-vehicle conflicts. Surrogate safety indicators were used to measure the severity of conflicts. Statistical results show that the main safety issue at unsignalized roundabouts is conflicts between entering e-bike riders and vehicles, with a high propensity for serious conflicts, followed by exiting conflict. Marginal effects of the combined best-fit model showed that the probability of slight and no conflicts increased as the diameter of the roundabout increased, while an increase in the number of lanes could lead to a higher probability of serious conflicts. Whether e-bikes or vehicles, an increase in speed increases the probability of serious and slight conflicts, while high traffic volumes show the opposite effect. In addition, conflict severity reduced with additional factors of conflict characteristics compared to no conflict. The combined best fit model performs well in the validation dataset with a prediction accuracy of 67.1 % and better performance for the exiting conflict and entering conflict models. These findings can be used to assess e-bike safety at unsignalized roundabouts without dedicated e-bike facilities and to enhance safety through targeted measures such as driver and e-bike rider education, the implementation of dedicated or shared lanes, and speed limits.