A new method is proposed for estimating BAC distributions for persons with unknown BAC test results on the Fatal Accident Reporting System (FARS) files. The method utilizes discriminant analysis to form linear combinations of variables associated with alcohol involvement in drivers and non-occupants, and uses these linear functions to estimate posterior BAC distributions based on various person, vehicle and accident attributes. Accident-level BAC distributions can be computed directly from the person-level BACs as the joint probability distribution of all drivers and non-occupants involved in each accident. The FARS database of drivers and non-occupants is stratified by vehicle body type, known vs. unknown police-reported alcohol involvement, and driver age, resulting in twenty separate model strata for estimating BAC distributions. Variables found to be most useful in estimating BAC are: police-reported alcohol involvement, accident hour, person age, vehicle role, injury severity, weekday/weekend, use of occupant restraint, driver license status, number of entries on driver record, person sex, location of non-occupant in relation to roadway, and whether or not the driver could drink legally (minimum drinking age in accident state). Validation tests are conducted using cases with known BAC test results from the 1984 and 1985 FARS. Results of these tests are presented and discussed. (Author/publisher)
Abstract