Our aim is to compare traffic safety among several counties in France, and explore whether observed differences can be explained by differences in road types distribution and by differences in socio-economic characteristics between counties. Traffic safety is measured by incidence and severity, where incidence is defined by the ratio of counts of injury accidents and exposure, measured by the amount of kilometres driven. Severity is measured by the ratio between fatal and injury accidents. These indexes are analysed in the framework of Generalised Linear Models: counts of injury accidents are analysed with a Negative Binomial regression, which accounts for over-dispersion. Severity being the proportion of fatal accidents among injury accidents corresponds to the probability of a Binomial setting and this is modelled by a logistic regression. This modelling provides an easy way to adjust for covariates such as road type, environment (urban/rural) and evolution over time, and to test their possible interactions. We find that the time trend of each indice (incidence and severity) is the same across counties and across road types. There is a significant interaction between county and road type, meaning that, first, differences in traffic safety between counties are not fully explained by different road type distributions, and second, that the "ranking" of counties in term of incidence or severity varies according to the road type considered, and vice-versa. It was planned to explore global characteristics of the counties (driving and socio-economic data) as possible explanatory factors of differences between counties, but the existence of an interaction of county with road types shows the necessity of collecting and exploring characteristics of the sub-levels of road type within county. (Author/publisher).
Abstract