Logistic Models in Crash Severity Classification Based on Road Characteristics

Abstract

The influence on traffic safety of certain road geometric parameters and characteristics is identified. The research involves real data about multivehicle accidents without pedestrians on nonurban roads for a chosen region of Poland. The accident data were integrated with the road data for the accident location by means of the author's computer programs. Logistic regression in the form of the proportional odds ratio model and continuation ratio logits was investigated in the analysis. The threat on the road is represented by accident severity, which is the ordered response variable subject to the classification task. The set of explanatory variables consists of the road design elements (e.g., vertical and horizontal curves) as well as information on road access (e.g., bus stops, public and private access lanes) and land use (e.g., gas stations, parking places). The modeling results indicate that the following features have a statistically significant influence on accident severity: the type of shoulders on both sides of a roadway, area type, pedestrian sidewalks, and the presence or absence of intersections.