13: Towards a Unified Approach to Causal Analysis in Traffic Safety using Structural Causal Models

Effective engineering arguably requires causal models of the systems to be engineered. By this we mean models which not only describe how systems behave under existing conditions, but also how they would behave if subjected to different conditions. For some problems it is possible to start with well-established physical models, but for others the choice of an appropriate model is less clear, and problems involving road traffic operations and safety tend to fall in this latter group. For example, in a recent review of research on the relationship between speed and accident risk, Shinar (1998) identified several different modeling approaches, such as those seeking to relate individual vehicle speeds to accident risk, those seeking to relate aggregate measures of speed to accident risk, and those looking at speed as a factor in individual accidents. This variety in methodology appears in turn to have produced a variety of findings. Increases in average speed might, or might not, be associated with increased accident rate. Some studies have found both faster and slower-moving vehicles to be over-involved in accidents, while others report over-involvement only for faster vehicles. There appears to be a reliable association between accident rate and speed variance, but the correct interpretation of this association is subject to debate (Davis 2001a).