Multiple membership multilevel model to estimate intersection crashes.

Area of Science:

• Transportation Engineering
• Traffic Safety
• Statistical Modeling

Background:

• Intersection crash prediction models traditionally focus on micro-level factors.
• Including macro-level factors, like zonal population, presents boundary challenges.
• Existing models may suffer from heterogeneity and misspecification issues.

Purpose of the Study:

• Introduce and evaluate the multiple membership multilevel model (MMMM) for intersection crash analysis.
• Address boundary problems and reduce heterogeneity in crash prediction.
• Compare MMMM performance against traditional single and conventional multilevel models.

Main Methods:

• Utilized five years of intersection crash data (2009-2013) from Regina, Saskatchewan.
• Developed and applied the multiple membership multilevel model (MMMM).
• Compared MMMM with single model (SM) and conventional multilevel model (CMM) fitting capabilities.

Main Results:

• The MMMM demonstrated superior fitting performance compared to SM and CMM.
• MMMM effectively avoided underestimation of macro-level variance and Type I statistical errors.
• Identified significant micro-level factors (e.g., AADT, signals, speed) and macro-level factors (e.g., land use, young drivers).

Conclusions:

• The MMMM is a more robust and accurate approach for intersection crash prediction.
• This model successfully integrates micro- and macro-level factors, overcoming limitations of previous methods.
• Findings provide valuable insights for enhancing traffic safety strategies at intersections.