An ordinal multivariate analysis of accident counts as functions of traffic approach volumes at intersections

This research is concerned with the statistical analysis of accident counts at nonsignalized intersections. The objective is to develop a method for determining general (nonlinear) relationships between approach volumes and accident counts. The method must accommodate the testing of whether intersections of differing physical designs have higher or lower rates of accidents than predicted by traffic levels. It is assumed that only aggregate data are available: (1) counts of total accidents by type (e.g., injury versus property damage) without details concerning the locational position(s) of the vehicle(s) involved; and (2) aggregate traffic intensity on each intersection entry without details concerning turning volumes. The method involves the application of nonlinear multivariate methods to variables treated as ordinal scales. A case study application involving four-leg and three-leg ("T") nonsignalized major arterial intersections in the Netherlands is described. The effect of bicycle traffic on accident rates is included in the case study analysis. The results indicate that there are three groups of each of the two types of intersections based on traffic flow patterns. For each group, a different functional form was found to relate accident rates and specific variables measuring traffic volumes. There were no significant differences among the physical design categories of the intersections in each group that were not accounted for by differences in traffic intensities.