Modeling temporal correlation and heterogeneity in real-time conflict rates using Bayesian Tobit models for...

Area of Science:

• Traffic Engineering
• Transportation Safety
• Statistical Modeling

Background:

• Real-time traffic conflicts offer valuable insights into intersection safety dynamics.
• Observed traffic conflict data often exhibits temporal correlation and unobserved site heterogeneity, complicating analysis.
• Existing models may not adequately capture these complex data characteristics.

Purpose of the Study:

• To develop advanced statistical models for real-time traffic conflict rates.
• To simultaneously address temporal correlations and unobserved heterogeneity in traffic conflict data.
• To enhance the accuracy and reliability of intersection safety assessments.

Main Methods:

• Utilized signal cycle-level traffic data from six signalized intersections.
• Developed and compared three Tobit models: conventional, temporal Tobit (T-Tobit), and temporal grouped random parameters Tobit (TGRP-Tobit) within a Bayesian framework.
• Employed Deviance Information Criteria (DIC) to evaluate model fit.

Main Results:

• Significant temporal correlations were identified in T-Tobit and TGRP-Tobit models, improving model goodness-of-fit.
• TGRP-Tobit models demonstrated the best performance, indicated by the lowest DIC, highlighting the benefit of accounting for unobserved heterogeneity.
• Key factors significantly associated with real-time traffic conflict rates include traffic volume, shock wave area and speed, queue length, and platoon ratio.

Conclusions:

• Advanced modeling techniques are crucial for accurately analyzing traffic conflict data.
• Accounting for temporal correlation and unobserved heterogeneity substantially improves traffic safety model performance.
• Traffic volume, shock wave dynamics, queue length, and platoon characteristics are critical determinants of intersection safety.