It may be believed that the great efforts of the European Commission and Eurachem have consolidated quality assurance (QA) for applications in science. Useful and novel practices and procedures have definitely been published in the form of recommendations and guidelines 1-3, but there are important aspects yet to be resolved. Simplicity is crucial, and it should be recognized that straightforward, manageable methods are required to promote dissemination and implementation in research and industry. Alas, most guidelines are impractical and create confusion in analytical chemistry. The BIPM Guide to Expression of Uncertainty in Measurement (GUM) 1 and the Eurachem/CITAC guide to Quantifying Uncertainty in AnalyticalMeasurement (QUAM) 2 focus on the concept of uncertainty, but despite progress in some areas of QA there are still important shortcomings that need to be addressed. Uncertainty is by far the most important variable of science because it is used as a tool to decide the course of future experiments. Recent results indicate that scientists operate with two types of uncertainty, founded on the basis of short-term precision and long-term precision 4. In statistics is it understood that a sufficiently high number of repetitions automatically generates the true value, but experiments have disproved this; a high number of repetitions consolidate the laboratory's experimental result, which may be very different from the expected value 5. This is the reality of analytical chemistry, and there is a demand for means to address this problem. Determination of contents in certified reference materials (CRMs)must be independent of the apparatus and must also be independent of the calibration procedure, for example linear regression 6, bracketing 7, standard addition 8, or internal standards 9. If calibration methods yield different results, this is most probably caused by an incorrect estimate of uncertainty. It is most important to recognize that statistics cannot help ensure accuracy. The old concept of accuracy has been replaced with a new concept, termed "trueness", which has essentially the same meaning as the old concept 10. Trueness is determined by use of a consensus value that is an average of contributions from many independent laboratories who deliver results with appropriate levels of uncertainty. The consensus value may differ from the certified value, and the statistical test used to prove the deviation must be able to assume a reliable uncertainty. Experience has shown that professional laboratories provide results that may differ by orders of magnitude, with levels of uncertainty that also differ by orders of magnitude 5. However, there would be no need to worry about reliability of analytical results if consensus science were fully adopted. Manufacturers of CRMs may argue that amounts determined by weight provide samples whose contents are known with a high degree of trueness (~accuracy, old concept), and a true value is thus known. Scientists are prone to report a reference value and uncertainty very close to the certified values. However, because the number of repetitions is different and different types of apparatus are used, it is expected that the result of a laboratory must be significantly different from the certified value. So, can the accuracy only be estimated by combining the results of several independent laboratories to obtain a consensus value with a combined (elevated) uncertainty estimated from the pooled data of all laboratories? This would give an overall high, yet true, value of the uncertainty. Would all contributing laboratories arrive at the certified value, within the limits of uncertainty, if the uncertainty were estimated correctly? This has been revealed to be possible under the presumption that outliers were not removed from the original data set.
展开▼
