Some bias in road safety evaluation in Before/After Studies

New traffic management schemes - variable speed control, managed lane operations as Hard Shoulder Running, tidal flow, dedicated bus or taxis lane are nowadays developing. Their impact on traffic safety can be computed by means of Before/After studies. In these studies, the counts of accidents which occurred in the period after and in the period before the implementation of the new scheme are examined. As some contributing factors to accidents might have changed between before and after periods - due to the general safety policy, the betterment of vehicles or of the road, it is important to identify the impact of these changes on the accident number, in order not to confound this impact (called the "time effect") with the one of the new scheme; this time effect is identified on a "reference" site as the difference between the accident risks between the before and the after period on this reference site. There is an implicit assumption that there are no, or no important structural differences between the reference and the assessed sites (in terms of infrastructure, traffic, external conditions). If it is not the case, and if the time effect varies according to the elements of the infrastructure or of the traffic patterns, the "average" time effect computed on the reference site cannot be directly used on the assessed site, due to a bias. This bias is mitigated by splitting the reference site and its traffic into traffic scenarios or into infrastructure scenarios, and by computing the time effect on each element; then those time effects are used, with the appropriate weights, on the assessed site. The impact of this breakdown on the variance of the estimation is provided. A numerical application, for a French managed lane operation, is given to illustrate the interest of the method in road safety assessment