Statistics Meeting Book (June 20, 2018)

Problem • Student t C.I. fails if all d i

= 0. In this case

SE(dPOD) = 0, so get zero width C.I. • For modern qualitative methods POD > 0.95, so for small studies it is not uncommon to find all measured d i = 0. • Need a simple (Excelable) C.I. estimator that does not have this degeneracy

9

copyright 2016 by Least Cost Formulations, Ltd., all rights reserved.

Solution • Use ‘continuity-corrected’ standard error of dPOD • b = # of (M1,M2) = (+, -) in n test portions • c = # of (M1, M2) = (-, +) in n test portions • 6( G32' ¥^ > E F Q – ((b – c) / (n + 2)) 2 ] / n } • 95% C.I. = dPOD + z SE(dPOD), where z = 1.960 for 95% confidence

10

copyright 2016 by Least Cost Formulations, Ltd., all rights reserved.