Comparative Analysis of Techniques for Identifying Locations of Hazardous Roads

This paper compares five techniques for identifying hazardous road locations. These techniques embody different degrees of control for randomness in accident counts. They are tested by means of data for Norwegian roads. As a basis for the comparison, a hazardous road location is defined as one that has a higher expected number of accidents than similar locations due to local risk factors. The five techniques were compared: recording the number of accidents during a specific period, observing the accident rate (accidents per million vehicle kilometers) during a specific period, combining a critical count of accidents and an accident rate above normal during a specific period, using the empirical Bayes estimate of the expected number of accidents at each location, and determining the size of the contribution of presumably local risk factors to the empirical Bayes estimate of the expected number of accidents at each location. Each of the techniques was applied to the upper 1%, upper 2.5%, and upper 5% of the distribution of sites and evaluated against various criteria (including accident count, accident rate, etc). The diagnostic performance of the five techniques was assessed in relation to epidemiological criteria (sensitivity and specificity). The empirical Bayes technique was found to perform best according to the epidemiological criteria. It is concluded that hazardous road locations are most reliably identified by applying the empirical Bayes technique.