The purpose of this project was to create and validate a method of identifying homogeneous groups of high risk drivers. This approach was based on the premise that this strategy would be more effective in identifying high risk drivers prior to their accident involvement than would the point count system now in place. Also, if successful, it would permit the tailoring of driver improvement efforts to groups with similar driving problems. Starting with a 6-year driver record on a sample of over 250,000 licensed drivers, five risk groups were identified based on a 3-year driver record. For each of the five groups, a regression equation was derived to maximize the prediction of accident involvement in a future 3-year period. These equations were then cross validated on an independent sample which met the risk group definition. The drivers identified as being high risk using this approach were compared to the drivers identified as being high risk by three other methods: (1) A surrogate for California's negligent operator point count; (2) A regression equation developed on a similar data base using total accidents and total convictions as predictors applied to the total population; and (3) A regression equation identical to that described in method #2 which also includes age and sex as predictors. The drivers identified by the regression equations had the highest future accident expectancy (53 to 54 accidents per 100 drivers during a subsequent 3-year period) while the risk group approach identified drivers with an expectancy of 50 accidents per 100 drivers. The point count identified drivers with an expectancy of 45 accidents per 100 drivers. Recommendations included modifying the negligent operator point count to include total accidents and convictions, and weighting them in such a way as to maximize accuracy of prediction of subsequent accident involvement. They also included fully automating the new selection criteria and continuing to assess the selection procedure on an ongoing basis.
Abstract