Estimating the regression-to-mean effect associated with road accident black spot treatment: towards a more realistic approach

Abstract

Until recently, the evaluation of engineering remedial measures at road accident Black Spots was seen as a routine exercise. However, the regression-to mean effect has been found to distort comparisons between before-and-after accident data to a significant extent, and for this and other reasons the problem can no longer be considered as straightforward. Various methods for applying corrections for regression-to-mean have been proposed. They rely on various assumptions about the data which have not been properly investigated or justified. These assumptions relate to the nature of the distribution of mean accident rates between sites, the nature of the population from which the sites are drawn, and the validity of using total accident frequencies as evaluation criteria. The aim of this paper is to investigate the validity of the assumptions, to assess the robustness of the results which are based on them, and to put forward some suggestions for improving the quality of the results.