This paper aims at analyzing variations in road mortality between and within 24 European countries and seeks to attribute underlying structural factors to them. We introduce a Bayesian spatial multi-level regression model using generalised linear mixed model (GLMM) framework, which allows the inference for the model to be carried. The population density and country affiliation were used as covariates and were fitted into the model at using four levels of spatial aggregation known as NUTS regions - official statistical units in the EU defined by Eurostat. Population density has been found to have a significant influence on road mortality at regional level. For all countries, the elasticity estimate is -0.33, meaning that a 10% increase in population density will lead to a 3.3% decrease in road fatalities. Multilevel model defined at NUTS-3 level, taking into account NUTS-2 aggregation enables to take into account infra-regional variances in road mortality and produce most reliable model parameter estimates. Variations in Bayes relative risk (mortality ratio standardized by population density and country affiliation) is highest at NUTS-3 level, while it decreases for country level and NUTS-2 level, what suggests the existence of other important underlying factors being responsible for the variations among regions. Mapping Bayes relative risk allows identifying those regions, which should be targeted by national and regional policies. Last, not least, the new ranking of countries according to their road mortality risk adjusted for population density is presented. (Author/publisher)
Samenvatting