This paper tests the usefulness of a multilevel model (MLM) for explaining the spatial occurrence of road accidents; it also shows and confirms how far the characteristics of the geographical environment influence the location and concentration of road accidents at two levels of spatial aggregation. The results are compared to those obtained from a more classical logistic regression. The analysis is performed on the southern periphery of Brussels (Belgium). The main conclusions are: (1) that MLM is a potentially useful technique for modelling road accidents, but that hierarchical levels are not easy to define for spatial data and so MLM are less useful than other regression techniques for modelling spatial occurrences of road accidents; (2) that the characteristics of the environment and the road itself significantly influence the occurrence of road accidents, and changes in these characteristics are quite important elements in the explanation, leading to the suggestion that road users do not adapt their behaviour sufficiently to changes in road conditions. Hence, concentrations of road accidents often correspond to places where improvements could be made in terms of road design, signalling and land-use planning. (Author/publisher).
Abstract