Investigating effect of collision aggregation on safety evaluations with models of multivariate linear intervention

Abstract

This study investigated the effect of collision aggregation on safety evaluation through a case study from the 2001 Signal Head Upgrade Program of the Insurance Corporation of British Columbia, Canada. Three types of evaluations were performed. Bivariate intervention models were used in the first two evaluations to assess the impacts of different collision severity levels [severe and property damage only (PDO)] and the impact of the time of the collision (daytime and nighttime) on safety. In the third evaluation, multivariate intervention models were used to determine the safety impacts of the program on each combination of collision severity and time of occurrence (i.e., severe-daytime, severe-nighttime, PDO-daytime, PDO-nighttime). Overall, the results indicated that the program was effective in improving the safety of the treated intersections. However, the results revealed that aggregate analyses could lead to misleading results. Aggregation of collisions over time of day indicated that the treatment resulted in significant reductions in PDO collisions but not in severe collisions. Alternatively, aggregation of collisions over severity levels indicated that the treatment resulted in significant reductions in both daytime and nighttime collisions. These results were different from the results of the disaggregate analysis, in which significant reductions were found for all collision types, except for severe collisions during the daytime.