Test modelling single accidents with the basic evaluation model

This report elaborates and illustrates the proposed methodology for the evaluation and exploration of developments in Dutch road safety in the SWOV projects ‘Road safety assessments’ (Verkeersveiligheidsbalansen ) and ‘Road safety outlooks’ (Verkeersveiligheidsverkenningen ). Previous work in this area resulted in what is called the basic evaluation model, which considers the development of traffic safety to be the product of the developments in two latent, unobserved variables: exposure and risk. The basic evaluation model is a bivariate local linear trend model, and well-suited for handling the dependencies in time series data, thus yielding residuals satisfying the model assumption of independence. In the present report the basic evaluation model is extended to the incorporation of explanatory variables. Two methods are presented for including explanatory variables in the basic evaluation model. The first method uses a standard regression set-up in which the explanatory variables are treated as fixed and known. In the second method, on the other hand, they are treated as being subject to stochastic variation. The two methods are applied to evaluate the effects of wet weather conditions and drink-driving on the annual numbers of single motor vehicle accidents involving people being killed or severely injured in the Netherlands. It is found that only the second method uncovers the expected relationships between the explanatory variables and the risk of getting killed or severely injured in single motor vehicle accidents.