Negative Binomial Analysis of Intersection-Accident Frequencies

TLDR

Urban intersection accidents impose significant societal costs, yet the factors influencing their frequency remain poorly understood, hindering prediction of improvement effectiveness. The study applies a negative‑binomial regression to seven‑year accident data from 63 Bellevue intersections targeted for operational upgrades to model accident frequency at intersection approaches. Results reveal key interactions between geometric and traffic variables and accident rates, offering exploratory evidence that could inform estimates of accident‑reduction benefits for proposed intersection improvements.

Abstract

Traffic accidents at urban intersections result in a huge cost to society in terms of death, injury, lost productivity, and property damage. Unfortunately, the elements that effect the frequency of intersection accidents are not well understood and, as a result, it is difficult to predict the effectiveness of specific intersection improvements that are aimed at reducing accident frequency. Using seven-yr accident histories from 63 intersections in Bellevue, Washington (all of which were targeted for operational improvements), this paper estimates a negative binomial regression of the frequency of accidents at intersection approaches. The estimation results uncover important interactions between geometric and traffic-related elements and accident frequencies. The findings of this paper provide exploratory methodological and empirical evidence that could lead to an approach to estimate the accident reduction benefits of various proposed improvements on operationally deficient intersections.