Right-angle collisions – a lesson learned

Abstract Right-angle post-encroachment data collected at 12 signalized interactions in 2004 were revisited. The available data did not allow checking the presence of evasive maneuvers. A Lomax-based analysis was conducted. The proper separation thresholds were typically 4 s or slightly longer. As anticipated, the expected crash estimates were insensitive to short thresholds. Nevertheless, the estimates were on average over 20 times higher than the reported crashes. The bias was caused most likely by not checking for the presence of evasion and the consequential inclusion of many non-conflicts. The expected pattern of the results indicates the Lomax distribution of non-conflict events. This points to the existence of alternative Lomax data-generating mechanisms, such as random arrivals, that can be avoided by strictly following the failure-based definition of traffic conflict.