Despite the fact that millions DKK each year are spent on improving roadsafety in Denmark, funds for traffic safety are limited. It is therefore vital to spend the resources as effectively as possible. This thesis is concerned with the area of traffic safety denoted "hot spot safety work", which is the task of improving road safety through alterations of the geometrical and environmental characteristics of the existing road network. The presently applied models and methods in hot spot safety work on the Danish road network were developed about two decades ago, when data was more limited and software and statistical methods less developed. The purpose of this thesis is to contribute to improving "State of the art" in Denmark. Basis for the systematic hot spot safety work are the models describing the variation in accident counts on the road network. In the thesis hierarchical models disaggregated on time are derived. The proposed models are shown to describe variation in accident counts better than the models currently at use in Denmark. The parameters of the models are estimated for the national and regional road network using data from the Road Sector Information system, VIS. No specific accident models are estimated for the local road network as road data on local roads are not collected in a systematic manner. Methods are developed for targeting intersections and road sections in the road network with an unusual high number of accidents, the so-called "hot spots". The methods are based on the proposed accident models and they are shown to outperform the methods used in Denmark today. Also, guidelines on how to improve the prioritizing of hot spots and safety improving measures are provided. A new model for estimating the effect of treating hot spots is proposed. The model takes into account the so-called "regression to the mean effect" and results in better estimates of the effect of treatment than the model currently at use on the Danish road network. The proposed method is also shown to outperform the methods as yet suggested in the international literature.
Poisson-gamma hierarchical linear models; Regression to the mean effect; Hot spots; Evidence of hotness; Level og hotness; Traffic safety; Before and after studies