Uncertainty matters: Bayesian modeling of bicycle crashes with incomplete exposure data

Area of Science:

• Urban planning
• Transportation engineering
• Public health

Background:

• Neighborhood-level bicycle safety analysis is hindered by incomplete and unreliable exposure data.
• Existing methods like travel-diary surveys and fitness apps often suffer from underreporting of bicycle activities.
• Accurate exposure data is crucial for understanding and mitigating bicycle crashes.

Purpose of the Study:

• To introduce a Bayesian simultaneous-equation model to address uncertainty from incomplete exposure data in bicycle crash modeling.
• To correct for underreporting in bicycle activity data.
• To investigate the relationship between built environment, bicycle activity, and bicycle-motor vehicle (BMV) crashes.

Main Methods:

• Developed and applied a Bayesian simultaneous-equation model.
• Utilized a crowdsourced dataset of 792 BMV crashes across 209 Hong Kong neighborhoods over 3 years.
• Simultaneously modeled bicycle activities and BMV crash frequency.

Main Results:

• Empirically demonstrated bias from omitting activity-based exposure or using uncorrected cycling distance.
• Found that expanding bicycle infrastructure correlated with increased cycling levels.
• Observed a substantial reduction in BMV crash risk despite a slight increase in absolute crashes.

Conclusions:

• The proposed Bayesian approach effectively adjusts for raw exposure data uncertainty and extrapolates missing values.
• The framework untangles the links between the built environment, bicycle activity, and BMV crash frequency.
• Recommends designated facilities to separate cyclists and motor vehicles for enhanced safety.