Models to estimate the safety of a signalised intersection on the basis of information about its traffic flow and accident history are provided. They are based on data from 145 intersections in metropolitan toronto. Several insights were obtained during the development of the models. First, logically sound models require that the frequency of collisions be related to the traffic flows to which the colliding vehicles belong and not to the sum of the entering flows. Second, it is therefore necessary to categorise collisions by the movement of the vehicles before the collision and not by the initial impact type, as is customary. Third, the relationship between collision frequency and the related traffic flows is at times unexpected in form. For each of the 15 accident patterns, an equation is given to estimate the expected number of accidents and the variance using the relevant traffic flows. When data about past accidents are available, estimates based on traffic flow are revised with a simple equation.Several practical questions can now be answered. Given the traffic flow for a signalised intersection, one can predict how many and what kinds of accidents should be expected to occur on it; one can also show the probability density function (pdf) of the estimate. Knowledge of the pdf allows the determination of what an unusually high number of accidents would be on such an intersection. If the traffic flow of the intersection changes from year to year, one can estimate what changes in safety should be attributed to changes in flow. Also, one can correctly compare the safety of several intersections that have different flow patterns. Most important, one can estimate safety when both flows and accident history are given and, on this basis, judge whether an intersection is unusually hazardous. This method of estimation is recommended for accident warrants in the manual on uniform traffic control devices.
Abstract