It might be hoped that engineering schemes designed to improve the highway network would also save accidents. While before and after comparisons of observed accident frequencies may indeed imply a safety benefit, such comparisons tend to exaggerate the size of the benefit, due to bias arising from the regression-to-mean effect. In addition, it has been suggested that the apparent benefits of treatment may be eroded by an accident migration effect in which the reduction in accidents at the treated sites is partially offset by increases elsewhere. This paper describes an investigation into the safety effects of all types of engineering schemes implemented on rural and suburban highway networks in twelve local authority areas: some 2300 schemes in all. The effects of treatment are estimated in two ways: using control sites and using a simple graphical time series analysis. With a view to forming the basis of a method for allowing for regression-to-mean effects without the use of controls, the accuracy of the default accident frequencies recommended by the Department of Transport for use in its computer cost-benefit analysis program COBA is investigated. The results indicate that overall, the highway engineering schemes resulted in rather modest accident reductions although, since the majority of sites were not high risk sites, and the majority of schemes were not accident remedial schemes, the results were not conclusive regarding the reductions achievable at such sites. No migration effects were detected. The quality of the COBA estimates varied with link length, with very large errors for short links.
