Analysing the role of traffic volume as mediator in transport policy evaluation with causal mediation analysis and targeted learning

Traffic volume has often been analysed as the outcome of interest in transport policy evaluation. However, its role as mediator lying in the causal pathways between policies and other outcomes has rarely been explored. We investigate this issue by using a targeted learning-based causal mediation analysis approach. Compared to the traditional approach that has been used in transport research, namely the Baron-Kenny multiple-regression approach, the causal one incorporating potential outcomes has clearer causal definitions and interpretations. Also, targeted learning has higher functional flexibility by enabling the use of supervised learning algorithms. Simulations indicate that targeted learning outperforms the traditional approach in complex settings with nonlinearities and interactions. We present an empirical example, quantifying the direct effect of the London Cycle Superhighways (LCS) on traffic speed, and the indirect effect via traffic volume as mediator. Our results indicate that the installation of LCS has reduced motor traffic along the routes. The average causal effect on annual average daily traffic (AADT) relative to the AADT in the pre-intervention period is − 9.2 %. Regarding the direct and indirect effects, we find that LCS has a negative direct effect on traffic speed, which might be due to less space available for motor vehicles, while LCS can increase traffic speed via reducing the amount of motor traffic. The direct effect on traffic speed relative to the speed in the pre-intervention period is − 2.0 %, whereas the indirect effect is + 1.3 %. As a result, the total causal effect on speed is small.