Estimating heterogeneous treatment effects in road safety analysis using generalized random forests.

Area of Science:

• Road safety analysis
• Causal inference
• Statistical modeling

Background:

• Road safety measures are widely evaluated, but treatment effect heterogeneity remains understudied.
• Existing causal methods may not fully capture complex variations in treatment effects.

Purpose of the Study:

• Introduce generalized random forests (GRF) for estimating heterogeneous treatment effects (HTEs) in road safety.
• Compare GRF's performance against traditional causal methods via simulations.
• Apply GRF to analyze the UK's speed camera program and identify factors influencing effectiveness.

Main Methods:

• Generalized Random Forests (GRF) for flexible HTE estimation.
• Simulation experiments comparing GRF with outcome regression, propensity score, and doubly robust methods.
• Case study analysis of UK speed camera data, incorporating socio-economic and traffic variables.

Main Results:

• GRF demonstrates superiority in model specification, particularly with non-linear and non-additive data.
• Speed cameras significantly reduce road accidents, with statistically significant heterogeneity in effects.
• Camera effectiveness is greater at sites with higher baseline accidents, traffic volume, and in deprived areas.

Conclusions:

• GRF is a powerful tool for uncovering treatment effect heterogeneity in road safety.
• Findings provide actionable insights for optimizing speed camera placement and policy.
• Evaluating HTEs enhances road safety program performance and informs policy decisions.