National Institute of Standards and Technology, Chemical Science and Technology Laboratory, Gaithersburg, MD
(Received 6 June 1992; accepted 9 September 1992)
The widespread practice of rejecting outliers on the basis of statistical tests of significance is shown to accomplish nothing in terms of finding the causes for discrepancies. A graphical procedure, based on the h-statistic and the k-statistic, is shown to provide a vivid picture of the biases and the internal precisions of the participating laboratories. It is highly recommended that the analysis of the data be followed by an investigation of causes of outhers and thereby lead to improvements in repeatability and reproducibility.
Paper ID: JTE11757J