Abstract:
A statistical model was estimated for the frequency of nonintersection accidents on urban four-lane undivided roads. The model was based on electronic data from the state of Washington as supplemented by information from videologs. Modeling emphasized the selection of appropriate functional form for each variable. Results point to three principal lessons. First, the fit depends mostly on variables such as annual average daily traffic, number of commercial driveways, and speed limit, and the fit depends only weakly on variables such as vertical alignment or lane and shoulder width. Second, although some results are in line with common beliefs, many are not. This raises the important question of whether or when results of multivariate models can be assumed to indicate cause and effect. Third, relationships that hold for two-lane rural roads may not hold for urban four-lane roads. In particular, on four-lane roads, horizontal curves of moderate degree may be safer than tangent road sections.
