Abstract
Alternative approaches are developed for analyzing errors in evaluating sampling and analytical methods. The aim is to decide with given confidence (e.g., 90%) that an accuracy criterion is met. The accuracy criterion specifies the fraction (e.g., 95%) of an acceptable method's concentration estimates that must fall within a given range (e.g., 25%) of a true or reference value. Inaccuracy in the form of both bias and imprecision must be considered. Three approaches to analyzing experimental error in method evaluations are presented. Integral control establishes a single-sided rejection confidence band for bias and imprecision estimates. This band is defined by requiring a high probability of rejecting methods that are unacceptable according to the accuracy criterion. If the reference method imprecision is small enough and if the number of degrees of freedom of the evaluation experiment is large enough, this probability can be specified. Tolerance interval control, contrary to integral control, is intended to specify an acceptance range of bias and imprecision corresponding to given confidence that the accuracy criterion is met. Useful algorithms for both integral and tolerance control are presented for either determining compliance with an accuracy criterion or, alternatively, for determining the accuracy. Individual control requires that upper bias and imprecision bounds, computed individually at a given confidence (e.g., 95%) meet the accuracy criterion. The individual control approach can be overly restrictive, particularly when the reference method is imprecise. Integral and tolerance controls, on the other hand, more precisely define the overall confidence in the evaluation experiment and incorrectly eliminate fewer adequate methods. Bartley, D.L.; Fischbach, T.J.: Alternative Approaches for Analyzing Sampling and Analytical Methods. Appl. Occup. Environ. Hyg. 8(4):381-385; 1993.