As part of an emphasis on improving road safety, the Minnesota Department of Transportation seeks to identify the locations where older drivers were over-represented in accident records. This research project reports on the use of three methods to help improve the accuracy of identifying locations where older drivers were at increased risk: a basic statistical model, the Empirical Bayes statistical method and a clustering method.
Overall, the basic statistical model preformed the best. The clustering method and the Empirical Bayes method could both be usefully applied to the traditional task of high-hazard identification--that of automatically screening a large number of accident sites to identify potential candidates for improvement. This information can point the way to areas that may require a more detailed engineering analysis.
