Confidence and prediction intervals for generalised linear accident models.

Area of Science:

• Statistics
• Transportation Engineering
• Data Analysis

Background:

• Generalized linear models (GLMs) with log links and Poisson or negative binomial errors are standard for analyzing accident rates.
• Existing methods for calculating confidence and prediction intervals within these models can be complex.

Purpose of the Study:

• To develop and present accessible methods for generating confidence and prediction intervals for GLMs used in accident rate analysis.
• To extend the practical application of GLMs in traffic safety by leveraging common spreadsheet software.

Main Methods:

• The study details the use of spreadsheet software to implement calculations for confidence intervals.
• It also describes spreadsheet-based approaches for deriving prediction intervals for future accident numbers.
• The methods are demonstrated using examples relevant to traffic flow and accident data.

Main Results:

• The paper provides a practical, spreadsheet-based toolkit for producing confidence intervals for true accident rates.
• It also offers a method for generating prediction intervals for accident counts at new sites with specified traffic flows.
• These results facilitate more robust statistical inference in road safety studies.

Conclusions:

• Spreadsheet technology offers a viable and user-friendly approach to calculating essential intervals for GLMs in accident analysis.
• The developed methods enhance the practical utility of statistical modeling for traffic safety professionals.
• This work contributes to improved data-driven decision-making in transportation safety.