The Effects of Mandatory Speed Limits on Crash Frequency - A Causal Machine Learning Approach
This study analyzes the effects of binding local speed limits on crash frequency on German motorways. Various geo-spatial data sources are merged to a new data set providing rich information on characteristics for 500-meter segments of large parts of the German motorway network. The empirical analysis uses a causal forest, which allows to estimate the effects of speed limits on crash frequency under fairly weak assumptions about the underlying data generating process and provides insights into treatment effect heterogeneity. The paper is the first to discuss the issue of spatial over-fitting and potential solutions in the context of causal machine learning. Substantial negative effects of three levels of speed limits on accident rates are found, being largest for severe, and especially fatal crash rates, while effects on light crash rates are rather moderate. The heterogeneity analysis suggests that the effects are larger for less congested roads, as well as for roads with entrance and exit ramps, while heterogeneity regarding shares of heavy traffic is inconclusive.