Cross-sectional accident models on flemish motorways based on infrastructural design. Paper presented at Ecomod 2006 conference on regional and urban modeling, Brussel, 1-3 June, 2006.

Predictive models for accidents have been researched intensively in the world. From studies that predict all accidents to studies that predict a specific kind of accident; based on road design, environment variables, social parameters, etc. are investigated. On the dense motorway network in Flanders this has not been done yet. This paper presents a study of predictive models for accidents on motorways in Flanders based on infrastructural design. Starting with motorway accidents from 1996 to 2001, road infrastructure measurements of 2003 and traffic intensities from 1996 to 2003, a crosssectional database is created for negative binomial regression modelling. The results found on the Flemish motorways are compared with previous studies done in other countries. Road characteristics like traffic intensity, lane width, number of lanes, outer shoulder width and maximum speed limit seem to be statistically significant associated with the number of accidents. The results are consistent with many recent studies. While a lot of research combines fatal and severe accidents, this paper gives also models that predict only fatal accidents. An important conclusion is the useful split up of motorways in three different zones: link, exit and entry zones. Keywords: Predictive models; Motorways; Road characteristics; Negative binomial regression. (Author/publisher) The report is available at: http://www.steunpuntverkeersveiligheid.be/nl/modules/press/store/82.pdf