Citation

Gullberg RG. J. Can. Soc. Forensic Sci. 2006; 39(1): 15-24.

Copyright

(Copyright © 2006, Canadian Society of Forensic Science, Publisher Informa - Taylor and Francis Group)

DOI

10.1080/00085030.2006.10757133

PMID

unavailable

Abstract

Jurisdictions measuring duplicate breath samples in drunk-driving cases must decide on a decision rule to employ in order to identify unacceptable differences. Selecting the appropriate decision rule always involves balancing the probability of a false-positive error with that of error detection. The most common decision rules currently used, 0.020 g/210 L (20 mg/100 mL reported blood alcohol in Canada) or some specified percent of the mean, do not adequately accommodate these probabilities nor account for the non-constant variance of breath alcohol measurement. A decision rule is proposed that accommodates the relevant probabilities and measurement variability by employing normal distribution theory. The proposed decision rule will have a probability of error detection of at least 80% and a probability of false-positive error of less than 5% for concentrations up to approximately 0.23 g/210 L (230 mg/100 mL reported blood alcohol). Compared to other commonly used decision rules, the proposed rule maintains the maximum power for error detection between the relevant forensic concentration of 0.05 and 0.28 g/210 L (50 and 280 mg/100 mL reported blood alcohol). The proposed decision rule provides a sound statistical basis for establishing fitness-for-pur- pose in duplicate breath alcohol measurement.