A New Methodology for Before-After Safety Assessment Using Survival Analysis and Longitudinal Data

Area of Science:

• Traffic Safety Engineering
• Statistical Modeling
• Transportation Research

Background:

• Traditional empirical Bayes (EB) and full Bayes (FB) methods for before-after safety assessments require extensive data from reference sites, posing limitations.
• The selection of appropriate and sufficient reference sites can be challenging, potentially affecting the reliability of safety evaluations.
• Existing methods do not adequately address censored data or incorporate longitudinal covariates in before-after safety analyses.

Purpose of the Study:

• To propose a novel before-after safety evaluation methodology using survival analysis and longitudinal data as an alternative to EB/FB methods.
• To develop a Bayesian survival analysis (SARE) model with a random effect to account for unobserved site heterogeneity.
• To validate the proposed SARE model through simulation and apply it in a case study evaluating red-light-running photo enforcement.

Main Methods:

• Development of a Bayesian survival analysis (SARE) model incorporating a random effect for site heterogeneity.
• Validation of the SARE model via a simulation study.
• Application of the SARE model in a case study analyzing the safety effectiveness of red-light-running photo enforcement in New Jersey, using individual crashes as the unit of analysis.

Main Results:

• The survival analysis method provides valid safety estimates using only data from treated sites, eliminating the need for reference sites.
• The SARE model effectively handles censored data arising from the before-to-after period transition.
• The model successfully incorporates longitudinal covariates like traffic volume and weather, accounting for temporal heterogeneity.

Conclusions:

• The proposed survival analysis approach offers a robust alternative to EB/FB methods for before-after safety evaluations, particularly when reference site data is limited or problematic.
• The SARE model's ability to handle censored data and longitudinal covariates enhances its applicability and accuracy in traffic safety research.
• This methodology provides a more comprehensive and flexible framework for assessing the safety effectiveness of traffic interventions.