Semiparametric Bayesian models for evaluating time-variant driving risk factors using naturalistic driving data and...

Area of Science:

• Traffic Safety Research
• Statistical Modeling
• Behavioral Science

Background:

• Driver behavior is a primary cause of traffic accidents in the U.S.
• Naturalistic driving studies (NDS) offer advanced methods for analyzing driver behavior using in-vehicle data.
• Traditional statistical models have limitations in analyzing complex, clustered driving data.

Purpose of the Study:

• To develop and validate a novel statistical model for evaluating driver behavior risk using NDS data.
• To address limitations of standard models in handling unbalanced and clustered outcomes in crash risk analysis.
• To quantify the increased risk associated with specific driver behaviors like cell phone use and drowsiness.

Main Methods:

• A case-crossover approach was employed to assess driver-behavior risk.
• A semiparametric hierarchical mixed-effect model was proposed to manage within-stratum and among-strata variations in driving data.
• Two methods for calculating marginal conditional probability were developed and their consistency analyzed.

Main Results:

• The proposed model demonstrated superior efficiency and robustness compared to alternative methods in simulation studies.
• Analysis of the 100-Car NDS data revealed that cell phone dialing increased crash/near-crash risk by 2.37 times (OR: 2.37, 95% CI: 1.30-4.30).
• Drowsiness was found to increase crash/near-crash risk by 33.56 times (OR: 33.56, 95% CI: 21.82-52.19).

Conclusions:

• The study introduces a robust statistical framework for analyzing driver behavior risk in NDS.
• Specific driver behaviors, notably drowsiness and cell phone use, are significantly associated with increased crash risk.
• The findings provide valuable insights for developing targeted interventions to improve road safety.