Modelling time-series of British road accident data

Area of Science:

• Traffic safety analysis
• Statistical modeling
• Time series analysis

Background:

• Road accident data analysis is crucial for public safety.
• Understanding trends in vehicle accidents informs policy and prevention strategies.
• Previous studies have utilized various statistical methods to model accident data.

Purpose of the Study:

• To analyze monthly road accident data in Britain from 1970 to 1978.
• To compare the effectiveness of regression analysis versus Box-Jenkins time series methods for modeling accident data.
• To provide predictions for accident trends from 1979 to 1981 using both methods.

Main Methods:

• Regression analysis was applied to model two-vehicle accidents.
• Box-Jenkins time series methodology was explored for single-vehicle accidents.
• Model performance was evaluated based on data trends, residual autocorrelation, and predictive accuracy.

Main Results:

• Regression models were found adequate for two-vehicle accidents due to consistent time trends.
• Box-Jenkins models were investigated for single-vehicle accidents where trends were inconsistent.
• Both methods provided predictions for 1979-1981, with regression showing comparable results to Box-Jenkins for noisy accident data.

Conclusions:

• Box-Jenkins models are unlikely to significantly outperform standard regression for noisy accident time series data.
• Regression analysis, assuming uncorrelated residuals, remains a practical and often sufficient method for road accident modeling.
• The study highlights the importance of considering data characteristics when selecting time series analysis techniques.