Can We Predict Road Safety by Simulation?

Abstract

The use of simulation programs to assess the impact of measures on traffic safety is possible, but needs a double validation: the simulation should be valid for the properties that influence traffic safety and there should be a criterion and observable measure that correlates well with traffic safety. Both requirements are difficult to satisfy for most simulation programs. Accidents happen too frequent, but still the causes of accidents are not always easy to identify and the accident process is often very sensitive to small influence factors and cooperative affects. Therefore, the use of simulation for the determination of effects of measures and technologies on traffic safety is often based on believes and expert knowledge: if in a driving simulator the number of errors made by a driver is counted and when this is considered as a measure of the unsafety, the simulation has at least face validity a logical consistency with the expected relationship with traffic unsafety. In macroscopic simulations, like e.g. traffic assignment programs and traffic flow models, the empirical relationship between macroscopic characteristics of the traffic situation and safety can be used to predict changes in accident after changes in the traffic conditions. Still this is at best an estimate of the expectation value and not accurate prediction of the change in number of accidents. For microscopic simulations no validated measure for safety exists, as far as the author knows. The time-to-collision is often used, sometimes modified to the TIT or TET, but the validity as safety predictors has still not been proven. A most difficult to assess is the rebound effect, the effect that drivers change their behavior after new technologies become available or certain measures have been taken. This adaptation leads to more risky behavior which may annihilate the effect of the improvement.