Evaluating the Effectiveness of Roundabout Collision Prediction Models using Estimated Total Daily Conflicting Volumes

Author(s): Taha Saleem, Robert J. Henderson

Slidedeck Presentation not available


One of the primary goals of transportation agencies around the world is to reduce crashes and potential crashes attributable to transportation facilities as well as minimize the potential for human error and provide a forgiving intersection environment. Roundabouts are constructed mostly because of their safety and capacity benefits; they provide a solution that can potentially reduce crashes at intersections. According to Transportation Association of Canada’s “Synthesis of North American Roundabout Practice”, roundabouts are shown to reduce injury collisions by approximately 75% as compared to stop control or traffic signals. The main objective of this study is to evaluate the effectiveness of roundabout collision prediction models based on estimated total daily conflicting volumes. For the purpose of this paper, samples of roundabouts from the Region of Waterloo were used. State-of-the-art, generalized linear modeling (GLM), with the specification of a negative binomial (NB) error structure, was used to develop the crash prediction models. Models were developed linking the estimated total daily conflicting volume (TDCV). These models were also compared to the conventional models linking traffic volumes to crashes. The reliability of the model estimates were enhanced further by accounting for the variation/trends in crash counts due to the influence of factors that change from year-to-year. This variation was captured by treating the counts for each year as a separate observation using general estimating equations (GEE) to develop crash prediction models accommodating for time trend and/or the temporal correlation in crash data. The results are promising in that the proposed methodology yields results that closely match results from the conventional models.