Freeway Accident Likelihood Prediction Using a Panel Data Analysis Approach

Abstract

The ability to predict freeway accident likelihood promises significant benefits to freeway operations. However, the development of such prediction models has proven to be very challenging because of the random nature of accidents, as well as the impact of site-specific factors. In addition, accident data has a pronounced nature of discrete response—a preponderant portion of nonaccident cases. To address these challenges, this research investigates the use of a discrete response model designed for panel data—the random effects ordered probit model, in predicting freeway accident likelihood. Panel data refers to data sets that combine time series and cross section (i.e., from different individuals, groups, etc.) observations. The empirical results of this research illustrate that the random effects ordered probit model performs well in identifying factors associated with traffic accidents. In addition, when applied in a predictive setting, the model provides benefits in forecasting the likelihood of accidents based on both time-varying and site-specific parameters.