Counterfactual safety benefits quantification method for en-route driving behavior interventions.

Area of Science:

• Traffic Safety
• Behavioral Science
• Data Science

Background:

• Driving behavior interventions are crucial for traffic safety but face challenges in quantifying effectiveness due to complex factors.
• Traditional methods using observational data often yield biased results by failing to control for confounding variables.

Purpose of the Study:

• To propose a counterfactual safety benefits quantification method for en-route driving behavior interventions.
• To establish a closed-loop evaluation and optimization framework for behavior interventions.

Main Methods:

• Utilized empirical data from online ride-hailing services.
• Employed structural causality model based on the Theory of Planned Behavior (TPB) to infer counterfactuals.
• Applied Extreme Value Theory (EVT) to link speed maintenance behavior changes with crash probabilities.

Main Results:

• Safety broadcasting reduced driving speed by approximately 6.30 km/h.
• Contributed to an approximate 40% reduction in speeding-related crashes.
• The framework reduced the fatality rate per 100 million km from 0.368 to 0.225.

Conclusions:

• The proposed counterfactual quantification method and optimization framework are effective for evaluating and improving driving behavior interventions.
• The findings demonstrate significant safety improvements in the ride-hailing sector through targeted interventions.