Konferens: Road safety on four continents: 16th international conference, 2013, Beijing, China

Safety performance models are essential tools for traffic safety work, and are used to estimate the expected number of accidents in an intersection, on a road segment or other entity of the road network. The models are based on accident history for several years and for many intersections/segments, thus trying to counter the large random variation in accident counts between individual years and places. Much research has been done within the area of safety performance models, but the vast majority of it has been done for motorized traffic. With an increased focus on sustainable traffic and increased demand for walking and biking there is also need for good models for these transportation modes, but very limited research has been carried out so far in this area. This paper reports mainly on two recent Swedish studies with a special focus on safety performance for pedestrian and bicyclist. The first Swedish study was based on police reported accidents on 400 urban street segments and the second on both hospital and police reported accidents on 360 segments and 63 intersections. Short counts of pedestrians and bicyclists were carried out to account for exposure in the models. The statistical approach used for the model development was quasi-Poisson in the first study and Negative Binomial in the second one, both approaches accounting for over/underdispersion in the accident count. The developed models exhibit a ‘safety in numbers’, i.e. the risk for an individual pedestrian/bicyclist to be involved in an accident decreases with an increase in the number of pedestrians/bicyclists. This effect is stronger for bicyclists than for pedestrians. The short counts were shown to significantly improve the predictive capability of the models compared to only using proxies for their number, such as street type or type of built environment along the street.