This report describes the collection, analysis, and modelling of accident and roadway data pertaining to segments and intersections on rural roads in the States of Minnesota (1985-1989) and Washington (1993-1995). The segments are on two-lane roads, and the intersections are three-legged and four-legged intersections of such roads, stop-controlled on the minor legs. Data were acquired from the Highway Safety Information System, photologs, construction plans, and State data bases. More than 1,300 segments and more than 700 intersections are included in the final samples on which the modelling is based. Variables collected include accident counts, traffic exposure, surface and shoulder width, Roadside Hazard Rating, number of driveways, channelisation, horizontal and vertical alignments, intersection angles, speed limits, and commercial traffic percentage. Models of Poisson type, negative binomial type, and extended negative binomial type (the latter due to Shaw-Pin Miaou) are developed, and advanced statistical techniques are applied to assess the explanatory value of the models in the presence of Poisson randomness and overdispersion. The models derived from these data indicate that exposure and traffic counts are the chief highway variables contributing to accidents, but that surface and shoulder width, roadside conditions, and alignments are also significant, especially in the segment models. Unexpected behaviour of intersection angle, Roadside Hazard Rating, number of driveways, and channelisation in the intersection models is worthy of note. Despite the incompleteness of the data and uncertainties in the values of some variables, the quantity, quality, and variety of the data give the models both descriptive and predictive value. (A)
Samenvatting