Comparison of the effectiveness of various procedures for the rejection of outlying results and assigning consensus values in interlaboratory programs involving determination of trace elements or radionuclides

Rejection of outlying results is usually necessary in the evaluation of interlaboratory comparisons, especially those involving trace elements and low-level activity measurements. The performance of several criteria (Dixon's test, Grubbs' test, coefficient of skewness test, and coefficient of kurtosis test) is described; authentic data sets from intercomparison runs on trace element or radionuclide determinations in spiked samples are used. The existence of “masking effects” in certain cases is demonstrated. It is shown that concurrent use of all four criteria at small (0.01) or moderate (0.05) significance levels gives better results than the use of any single test. The use of all four criteria at α = 0.05 followed by calculation of overall mean and pertinent confidence limits is suggested as a general method of data handling in interlaboratory comparison. This method normally ensures good selectivity of outlier rejection while avoiding excessive rejections, and results in generally good agreement between the “true” value and the final overall mean. The use of this approach for the establishment of recommended values in candidate reference materials is demonstrated and its advantages and limitations are discussed. The results are further compared with those obtained by other methods of data evaluation (Student's test, dominant cluster mode and gamma transformation).