Analytic choices in road safety evaluation: Exploring second-best approaches

► Eight designs for before-and-after studies are compared. ► Estimates of effect are found to differ between the study designs. ► The empirical Bayes approach is methodologically stronger than traditional approaches for before-and-after studies. ► Two simple methods for controlling for regression-to-the-mean are tested. ► The methods were not found to be reliable and their use is discouraged.