Mandatory speed limits and crash frequency on motorways — A causal machine learning approach
利用因果森林分析德国高速公路强制限速对受伤事故频率的影响，发现限速显著减少严重和致命事故，尤其在交通量小或有出入口匝道的路段效果更大。
This study analyzes the effects of binding segment-level speed limits on injury crash frequencies on German motorways. Various geo-spatial data sources are merged to a rich novel data set, providing detailed information on 500-meter segments of large parts of the network. A causal forest is applied to estimate effects under fairly weak assumptions about the underlying data generating process and to offer insights into effect heterogeneity. Furthermore, the study explores potential biases through a phenomenon called spatial overfitting and examines potential solutions. Substantial negative effects of three levels of speed limits on crash frequencies are found, particularly for crashes involving severe or fatal injuries, while effects on crashes involving light injuries are comparably small. The heterogeneity analysis suggests larger crash rate reductions on roads with less traffic, as well as on roads with entrance and exit ramps, while heterogeneity regarding shares of heavy vehicle traffic is inconclusive. • A new comprehensive data set is constructed from multiple geo-spatial sources. • Causal forests reveal nuanced speed limit effects under minimal assumptions. • Spatial overfitting and potential remedies are introduced to causal machine learning. • Clear negative effects of mandatory speed limits on severe and fatal crashes. • Indications of larger reductions for roads with less traffic or entrance/exit ramps.