Causal decision-making for speed camera allocation: Methodology and an application

Abstract

Speed enforcement cameras are implemented worldwide to regulate driving behaviours and enhance road traffic safety. Proper allocation of speed cameras is quite important. In practice, we should first identify road sites likely to experience larger crash reductions with speed cameras, but this step is commonly simplified as ranking sites based on the historical crash frequency. This paper proposes the use of causal decision-making to refine speed camera allocation rules. Within this framework, the heterogeneous treatment effects (HTEs) of speed cameras on crash frequency across different sites are first modelled by applying causal machine learning methods. Subsequently, by exploiting the trained HTE model, sites with larger predicted road safety benefits (i.e., crash reductions) will be prioritised for allocation. A UK case study is presented to demonstrate the superiority of the proposed method. Different speed camera allocation rules, including the HTE-based, historical crash-based, and random allocation, are compared with respect to the number of prevented road traffic crashes. Our empirical results indicate that a larger number of past crashes in general implies a larger safety benefit of the speed camera. Therefore, the historical crash frequency could be regarded as a useful criterion for camera site selection in the absence of additional information. Nonetheless, the HTE-based rule has been found to further enhance the allocation performance. That is, more road traffic crashes could be prevented by adopting the HTE-based rule. In future transportation research and practice, the causal decision-making framework could be applied more generally to costly resource allocation tasks.