Do we really need to account for run bias when producing analytical results with stated uncertainty? Comment on ‘Treatment of bias in estimating measurement uncertainty’ by G. E. O'Donnell and D. B. Hibbert

Abstract

Treatment of bias is an important issue relating to analytical quality. Recently, G. E. O'Donnell and D. B. Hibbert (Analyst, 2005, 130, 721) recommended to always correct analytical results for ‘run bias’ determined by a single analysis of a certified reference material (CRM) in each analytical run. In the authors' opinion, this is necessary for the results obtained to be comparable from run to run. It is argued here that such a recommendation is logically inconsistent and stems from misinterpretation of measurement uncertainty as being estimated under repeatability conditions. The fundamental principle underlying the measurement uncertainty methodology is that all relevant sources of error should be taken into account, which results in overall uncertainty assessment and thus provides a means for a global comparability of measurement and test results. The local, i.e. run-to-run, comparability is not a factor if analytical results are interpreted on the basis of their associated uncertainty.