The use of a comparison area relies on the assumption that two areas which display similar patterns of accident distribution in the before period would have continued to do so in the after period if the scheme had not been introduced in one of them. This concept is examined by identifying a hypothetical treatment area and several candidate comparison areas. The accident histories of each of these areas are compared with that of the hypothetical treatment area and are ranked according to goodness-of-fit. These rankings are compared against a second ranking based on the ability of each of the possible comparison areas to predict accident frequencies in the hypothetical treatment area during an assumed after period, no treatment having taken place. The results from five TRRL urban safety project schemes are re-evaluated against several comparison areas, and using no comparison. Several models are fitted to determine the dependence of the prediction of reductions in accident numbers on the comparison area chosen. The results highlight the problems associated with the use of comparison areas. Not only do the predictions of reductions in accidents due to the scheme depend on the area selected as the comparison, and to a lesser extent on the model fitted, but also a comparison area which is selected due to historic similarities in accident pattern with the treatment area will not necessarily produce the best predictions of what would have happened, in the treatment area, if the scheme had not been introduced.
Abstract