|
Abstract:
|
Traffic safety studies rely on a combination of highway-related crash data, roadway characteristics, traffic data, and statistical modeling. Crash data are reported as count data, typically yearly, and vary significantly due to factors unrelated to the roadway (e.g., environmental, human, vehicle). Often, safety assessments are based on data available for only a select number of sites, for a few years, or on roadways with low traffic volumes. Over the past decades, predictive modeling of crash frequencies based on roadway geometrics and traffic volumes has evolved from OLS regression models to models assuming a negative binomial error structure with an overdispersion modeled as a function of those variables. We discuss this evolution, the various modeling issues, present directions in the field, and challenges in the assessment and modeling of roadway safety.
|
