Estimation of crash type frequency accounting for misclassification in crash data

Area of Science:

• Transportation Safety
• Statistical Modeling

Background:

• Crash misclassification (MC) is a common issue in transportation safety data.
• MC errors can bias crash frequency models, leading to ineffective safety countermeasures.

Purpose of the Study:

• To develop a novel methodological formulation to directly account for MC error in crash frequency prediction.
• To incorporate this formulation into Poisson and Negative Binomial (NB) regression models.
• To evaluate the proposed models' ability to estimate true parameters in the presence of MC error.

Main Methods:

• Developed a framework introducing probabilistic MC rates among crash types.
• Modified the likelihood function of Poisson and NB count models.
• Integrated the approach into reformulated discrete choice models.
• Conducted simulation analysis to examine parameter estimation accuracy.
• Applied models to empirical transportation safety data.

Main Results:

• Proposed models demonstrated capability to estimate true parameters despite MC error.
• Simulation analysis confirmed the effectiveness of the novel formulation.
• Empirical data analysis showed improved model fit compared to models ignoring MC error.
• MC rates in empirical data were found to be very low.

Conclusions:

• The proposed methodology effectively addresses crash misclassification in transportation safety.
• Models incorporating MC error provide more reliable parameter estimates.
• This approach enhances the accuracy of crash frequency prediction and countermeasure planning.