Crash data analysis: Collective vs. individual crash level approach

Traffic safety literature has traditionally focused on identification of location profiles where “more crashes are likely to occur” over a period of time. The analysis involves estimation of crash frequency and/or rate (i.e., frequency normalized based on some measure of exposure) with geometric design features (e.g., number of lanes) and traffic characteristics (e.g., Average Annual Daily Traffic [AADT]) of the roadway location. In the recent past, a new category of traffic safety studies has emerged, which attempts to identify locations where a “crash is more likely to occur.” The distinction between the two groups of studies is that the latter group of locations would change based on the varying traffic patterns over the course of the day or even within the hour. Hence, instead of estimation of crash frequency over a period of time, the objective becomes real-time estimation of crash likelihood. The estimation of real-time crash likelihood has a traffic management component as well. It is a proactive extension to the traditional approach of incident detection, which involves analysis of traffic data recorded immediately after the incident. The units of analysis used in these studies are individual crashes rather than counts of crashes. In this paper, crash data analysis based on the two approaches, collective and at individual crash level, is discussed along with the advantages and shortcomings of the two approaches.