This paper is based on a review of often-cited studies on the safety impact of traffic signal installation. Most of these studies are found wanting with respect to methods of analysis or inferences from the results. Two common pitfalls - regression to the mean and incorrect inferences from cross-section studies - are illustrated. The first might lead to erroneous conclusions about the circumstances under which signal installation is likely to improve or degrade safety. The second might lead to underestimation of the safety effectiveness of installing signals at relatively unsafe intersections. Most of the studies reviewed were conducted before these pitfalls came to light. Nevertheless, the upshot of these revelations is that there is very little substantial knowledge about the safety impact of traffic signal installation. The question of how to improve this somewhat embarrassing state of affairs using the latest methods of analysis is addressed, and recommendations are made on how to incorporate useful knowledge into the allocation of resources for signal installation.
Abstract