Spatial Modelling of Risk in Traffic Safety on the Road Network
E Moons, C Faes, T Brijs, Hasselt University - Transportation Research Inst., BE
One of the main policy aims within the domain of traffic safety is to produce safer roads and, as a consequence, to reduce the number of (fatal) accidents. In order to make the step from evidence to policy, one needs to make an inventory of the hazardous locations and try to understand the adhering risk on these roads. However, it is well-known that in order to properly analyse traffic safety data, one needs to overcome problems (such as low means or an excess of zeroes) that occur due to the stochastic nature of accidents. This paper represents a series of models with an increasing complexity for the purpose of estimating the relative accident risk. The most classic Poisson regression model will be extended to hierarchical Bayesian models that now have become standard in traffic safety literature. Although, one acknowledges that the geographical aspect is highly important to determine and to handle the most unsafe traffic sites in a scientifically sound and practical way, very often statistical ? non-spatial ? regression models are used to model the number or the risk of accidents. This paper goes beyond this classic Bayesian approach, because it also incorporates the spatial correlation structure. It seems only logical that there exists a geographical relationship between different locations, so therefore, in the spatial models represented here, the vicinity of the accident locations alongside the road network is taken into account to predict this accident risks. In trying to predict the variability, land use characteristics, infrastructural components and the proximity of attraction poles inducing traffic (such as e.g. schools, dancings, etc.) are taken into account. The results are presented on two case studies: the first deals with accidents that occurred on highways in Flanders (Belgium), while the second example concerns accidents within a city environment. The results of all models are compared to each other based on several goodness-of-fit measures. Based on these conclusions, recommendations for policy making will be formulated.
Association for European Transport