Over a 30-day period, morning and afternoon peak period crashes along a 17.7 km section of Interstate 5 north of Seattle, Washington, were analysed with respect to the real time traffic conditions present in the spatial and temporal vicinity of the crash. Twenty-second volume and occupancy data were collected over an Internet connection from loop detectors present in each travel lane every 0.5 to 0.8 km. Earlier work had suggested that large scale changes in the variability in traffic volumes across individual lanes in the immediate vicinity of the crash might be related to an increase in the probability of a crash. The present study extended the notion of variability to include variation in vehicle speeds in the immediate vicinity of the crash site. At a 'macroscopic' level, crashes were not limited to major points in time and space defined by systematic transitions from non-congested to congested operations (that is, along the edge of the wave). At a more microscopic level however, there was some evidence for an increased likelihood of crashes under conditions characterised by large amplitude changes in the 'slope' of average vehicle speeds in the immediate vicinity of the crash. The limited sample of crash data available for this exploratory effort was not sufficiently large to confirm the presence of a reliable relationship between variation in traffic speeds and the likelihood of a crash. The interpretation is however consistent, at least in concept, with results from the UK showing that the dynamic use of variable speed limits could be effective in reducing crashes. (a).
Abstract