Road safety research is a wonderful combination of counting fatal accidents and using a toolkit containing prior, posterior, overdispersed Poisson, negative binomial and Gamma distributions, together with positive and negative regression effects, shrinkage estimators and fiercy debates concerning the phenomenon of accident migration. Accidents are counted at the level of, e.g., roundabouts of some specific architecture and also over all roundabouts. For any individual roundabouts a Poisson distribution is used to asses its `unsafety'; for the mean of the accidents over roundabouts of the same architecture, a Gamma distribution is used. The combination of the two leads to a mixed distribution, `a negative binomial distribution' for the accidents for all roundabouts together. Methodological problems are the regression effect, the apparent reduction of fatal accidents after remedial treatment, the selection of, e.g., roundabouts, for remedial treatment, and assessing the influence of the remedial treatment. For the selection of systems for remedial treatment, decision theoretical concepts like `false' and `correct' positives and negatives are used. (A)
Abstract