Despite more than 50 years of research and advancements in road safety evaluations using traffic conflict methods, a major lacuna that hampered their widespread application was their inability to accurately predict the severity component of crashes in addition to their frequency. Ignoring crash severity in road safety evaluations could be counterproductive as the safety treatments being implemented could decrease the number of crashes occurring at a location, but, possibly, increase their severity. This paper is the first effort to fill this research gap and predict the frequency of severe and non-severe crashes using traffic conflict indicators alone. Using bivariate extreme value modelling with prevalent traffic conflict indicators like Time-to-Collision (TTC) and the expected change in velocity after a crash (Delta-V), this paper illustrates how the problem can be solved using crash and conflict data for signalised intersections in Brisbane.
The paper marks another major advance towards establishing traffic conflict methods as a viable alternative to crash-based safety evaluations. Crash-based evaluations are a long, slow, and unethical process – one has to wait for 3-5 years for enough crashes to occur BEFORE starting the analysis. Having credible severity information for those crashes is another potential area of bias in the estimates. Thus, this research paves the way towards providing a fast and reliable procedure to road safety engineers worldwide.
Ashutosh Arun has 7+ years in safety modelling, analysis & design engineering projects and is a Qualified Road Safety Auditor (2000+ km of roads and highways). He is a researcher in statistical and econometric modelling, the application of machine learning algorithms to problems in traffic engineering and crash modelling. Ashutosh has worked on a Supra Institutional Network Project, “Development of Indian Highway Capacity Manual” and leads the work package aimed at “Capacity Estimation of Multi-lane Interurban Highways”.