Why use a complicated accident prediction model when a simple one is just as good

This paper discusses accident prediction models, which express the expected number of injury accidents, on a road section or at a junction, as a function of the traffic and road characteristics there. The authors developed such models for two main reasons: (1) estimating the expected number of accidents more accurately than by counting only accidents reported by the police; and (2) gaining more insight into the relationships between traffic and road conditions and the number of accidents. The models were applied to Finnish data, and can be set up quite easily for rural roads, using data on accidents, roads and traffic. These data were not sufficient to allow accident models to be developed for urban streets. All the models were developed for homogeneous road sections, using generalised linear modelling techniques and the GLIM software. Some comparisons were made between different models, to test their strengths and weaknesses, and testing how well they fitted with accident data in the same year or in different years. Data are presented from the comparison of four different accident models. It was found that quite simple models adequately estimated expected numbers of accidents on a specific road section.