The "Road Infrastructure Safety Management Evaluation Tools (RISMET)" project targets objective A (Development of evaluation tools) of the Joint Call for Proposals for Safety at the Heart of Road Design ("The Call"). This project aims at developing suitable road safety engineering evaluation tools that will support the aims of the Call as described in the Guide for Applicants (GfA) and furthermore those of the Directive for Road Infrastructure Safety Management (2008). These evaluation tools allow the easy identification of both unsafe (from accidents or related indicators) and potentially unsafe (from design and other criteria) locations in a road network. With such evaluation tools estimates of potential benefits at the local and the network level can be calculated and potential effects on aspects such as driver behaviour can be estimated. Such tools empower road authorities to improve their decision making and to implement (ameliorative) measures to improve the road safety situation on the roads. Since evaluation tools rely on good quality data, RISMET aims at reviewing available data sources for effective road infrastructure safety management in EU-countries, linked to a quick scan and assessment of current practices. Furthermore, RISMET aims at exploiting results related to the development and use of Accident Prediction Models (APMs) in road safety management. The present deliverable provides APMs for data collected at junctions from the rural road networks of Austria, Norway, Portugal and Holland. For the first three countries it was possible to obtain accident prediction models for each country individually. For Holland, however, and due to restrictions on the dimension of the data set, it was only possible to analyse these data together with the other countries data, i.e. analysing aggregated data sets. The data consists, per junction, of injury accident counts, type of junction, traffic control, speed limit and annual average daily traffics entering from the major and the minor road. The regression models had the injury accident frequencies as the dependent variable and the remaining variables as explanatory and were fitted using Bayesian statistical techniques with vague or non-informative prior and hyper-prior distributions. These models consisted on the Poisson regression model, hierarchical Poisson-Gamma and Poisson Log-Normal hierarchical regression model. The Poisson regression model was found to be not appropriate to model the junction data in any of the data sets due to not being able to capture variations and attributes of the data, namely the over-dispersion. The Poisson-Gamma and the Poisson Log-Normal models obtained similar results and in general performed equally well. It was found that accidents occurring at junctions in all countries depend on the junction’s entering traffic volume as well as the other explanatory variables considered. This report provides descriptions of the several data sets, equations for the expected injury accident frequencies, per year, on rural road network junctions for Austria, Norway and Portugal and for the conjoint set of the combined data (including Dutch data) as well as posterior means of the expected number of accidents for minimum, mean, median and maximum profiles obtained by the explanatory variables and measurements of model fit together with the major results obtained. (Author/publisher)
Abstract