This paper presents a test of the predictability of the effects of road lighting and traffic separation on road safety. It is only intended as a first attempt to test the performance of evaluation studies in predicting the safety effects of various measures. It begins with a general discussion of some general arguments for and against attributing predictive value to the results of evaluation studies of safety effects. These arguments show that the predictability depends on many factors, and cannot be assessed on theoretical grounds. The main part of the paper presents simple tests for two safety measures for which meta-analyses of evaluation studies have been performed: road lighting and traffic separation. The tests divided the evidence from evaluation studies into quintiles, and used the first 20%, 40%, 60% and 80% of evidence for the respective prediction of the second, third, fourth, and fifth 20%. When the safety effect was stable over time, the predictions were found to be reasonably accurate. When it showed a trend, the predictions were not found to become more accurate. The ability to predict the results of future studies is not necessarily improved by attempting to explain the variability of past evaluation studies; the data required for such explanation are often not available.
Abstract