Prevalence Dependent AccuraciesMisstate the Diagnostic Power ofthe Standardized Field SobrietyTest.
In the 1990s as the legal blood alcohol limit for driving changed, validation studies reported the Standardized Field Sobriety Test (SFST) to be accurate at discriminating between Blood Alcohol Concentrations (BAC) above or below several legal limits: 0.10%, 0.08%, 0.05% and 0.04%. This effectively made the implications of the test depend on the legal jurisdiction involved and on whether the driverÃ¢â‚¬â„¢s license was passenger or commercial class. We investigated the contribution of the validation studiesÃ¢â‚¬â„¢ choice of accuracy statistic to the SFSTÃ¢â‚¬â„¢s reported accuracy. Methods: Using the data set from a commonly cited SFST validation study, we calculated the arrest accuracy and overall accuracy of the SFST at identifying BACs above or below 31 different target BACs from 0.00% to 0.30%. We organized the results in tables; we observe and explain trends. Results: At target BAC 0.30% the arrest accuracy of the SFST is 1%; at BAC 0.15%, 34%; at BAC 0.00%, 100%. The statistics arrest accuracy and overall accuracy describe the SFST, a test designed to identify changes caused by alcohol, as less accurate when the changes are severe, more accurate when changes are mild, and as 100% (arrest) and 93% (overall) accurate when there are no changes at all. Conclusion: The statistic arrest accuracy identifies the SFST as currently used by US law enforcement as 78% accurate. This number is an artifact of the prevalence dependence of arrest accuracy. Calculations independent of prevalence show that the SFST actually has no meaningful power to discriminate between drivers with high and low BACs. The statistics overall accuracy and arrest accuracy to not quantify the probability that impaired driving defendants who failed the SFST had an elevated BAC or were impaired.