Assessing the safety effectiveness of citywide speed limit reduction: A causal inference approach integrating propensity score matching and spatial difference-in-differences

The New York City (NYC) initiated a new default speed limit law on November 7th, 2014, where speed limits on all road segments without a posted speed limit were reduced from 30 mph to 25 mph. The safety effectiveness of citywide speed limit reduction in an urban setting like NYC has been understudied in the literature. The high-density road network of NYC could lead to a significant spatial spillover effect of speed limit reduction on its neighboring sites. In addition, citywide speed limit reduction exerts much more treatment sites than control sites, which makes it challenging to identify sufficient control sites with similar covariates as the treated ones and thus may lead to confounding bias. Furthermore, there could also exist a time trend in crash observations caused by unobserved factors (e.g., enforcement, driving behaviors). To jointly account for spatial spillover effect, confounding bias, and time trend, this study proposes a novel causal inference approach integrating propensity score matching (PSM) and spatial difference-in-differences (SDID) to estimate the safety effectiveness of citywide speed limit reduction in NYC. The PSM utilizes a logistic generalized additive model (GAM) to capture the nonlinear relationship between covariates and the treatment indicator to reduce bias due to confounding variables. Moreover, the matched data are used to develop the SDID model that simultaneously captures spatial spillover effect and time trend via the extended difference-in-differences (DID) structure. The proposed causal approach suggests that the speed limit reduction would result in a 62.09% decrease in fatal crashes, with the spatial spillover effect found to be statistically significant. However, it does not indicate a significant change in injury and property-damage-only crashes as a result of the speed limit reduction. This study adds to the literature a robust causal inference approach for safety evaluation and provides researchers, practitioners, policy-makers insights into the safety effectiveness of the speed limit reduction in an urban setting.