Measurement uncertainty evaluation from analytical validation data: Asymmetric coverage interval due to uncorrected significant bias

Reliability in analytical measurements depends on appropriate measurement uncertainty estimation. In top-down approaches, precision and recovery data obtained during analytical method validation are commonly used for this purpose. Although significant systematic effects should preferably be corrected, correction is not always feasible, and uncorrected bias may instead be incorporated into the uncertainty estimate. However, treating directional bias as a symmetric uncertainty component may result in unnecessarily wide intervals. This study proposes an asymmetric approach (ASYM) in which bias shifts the uncertainty interval without increasing the random uncertainty component.

A Python-based Monte Carlo algorithm was developed to evaluate coverage probability and interval width over recovery values from 40% to 120% and precision up to 30%. ASYM was compared with a conventional root-sum-of-squares symmetric approach (RSSu or SYM) and alternative uncertainty models (SUMU, SUMUmax, and Ue). Moreover, the models were additionally evaluated using real analytical validation datasets from pesticide residues by LC-MS/MS. When bias was small relative to precision, the approaches generally provided similar coverage. As bias increased, ASYM and SUMU generally maintained coverage closer to the nominal 95% level, whereas other models produced wider intervals and misstated coverage levels. Application to real analytical data supported these trends. A practical decision framework based on the bias-to-precision ratio was suggested.

This study systematically evaluates asymmetric uncertainty estimation for uncorrected bias, distinguishing directional bias from random uncertainty. ASYM avoids artificially inflated intervals, provides coverage closer to the nominal level, and supports uncertainty model selection using the bias-to-precision ratio.