A sample of approximately 140,000 records of licenses California drivers containing information on age, gender, and driving record variables was examined. The goal of this paper was to assess the accuracy of predicting future accident risk using various combinations of demographic and prior driving record variables as predictors in 17 regression models. All of the models were consistent in demonstrating that increased probability of subsequent accident involvement is associated with increased prior citation and prior accident frequencies, being young, and being male. Results from the regression model indicated the following: (i) models that use prior total accidents as a predictor variable perform better than models that do not use prior total accidents as predictors; (ii) models that use prior culpable accidents as a predictors do not perform as well as models that use prior total accidents as predictors; (iii) a comparison of models in which 17 individual violation types are used as predictors to those in which only total citations is used as a predictor shows only a small advantage of using individual citation types; (iv) models that use as predictors the demographic variables of age, gender, and license class along with various combinations of citations and accidents perform better than California's current negligent-operator system, which uses a weighted combination of countable citations and responsible accidents. It was concluded that if the goal of driver record adjudication systems is to identify and apply sanctions to high-risk drivers in order to intervene before this risk is realised, then the results presented in this report support the current point-count strategy which attempts to optimise the identification of drivers having a high probability of subsequent accident involvement. (A)
Abstract