This report describes the collection, analysis, and modelling of crash and roadway data for intersections on rural roads in California and Michigan for the years 1993-1995. Three classes of intersections are considered: (1) three-legged intersections with major road four-lane and minor leg two-lane stop-controlled, (2) four-legged intersections with major road four-lane and minor legs two-lane stop-controlled, and (3) signalized intersections with both major and minor roads two-lane. Data were acquired from the Highway Safety Information System-4 State and Federal photologs, and field work at all intersections. The field work included morning and evening traffic counts by movement and vehicle " as well as alignment measurements out to 800 feet (244 metres) along the major road. The final data sets consist of 84 three-legged intersections, 72 four-legged intersections, and 49 signalized intersections. Other variables collected include Roadside Hazard Rating, number of driveways, channelization, intersection angles, and speed limits. Negative binomial models - variants of Poisson models that allow for overdispersion - are developed for each of the three data sets. Significant variables include major and minor road traffic, peak major and minor road left-turning percentage, number of driveways, channelization, median widths, vertical alignment, and, in the case of the signalized intersections, the presence or absence of protected left-turn phases and peak truck percentage. Models are developed for all crashes within 250 feet (76 metres) of the intersection centre, for intersection-related crashes within 250 feet (76 metres), and for injury crashes. For injury crashes, intersection angle and minor road posted speed are significant. Models of crashes at signalized intersections by approach flows are also investigated, and other model forms are proposed for future consideration. (A)
Abstract