On the Correct Approach to the Modeling and Analysis of Quantitative EMC Proficiency Testing Data for the Purpose of Evaluating Test Laboratory Claims of Competency

Traditional statistical procedures used for analyzing quantitative electromagnetic compatibility (EMC) proficiency test (PT) data are overly simplistic, given that they are implicitly based upon the following erroneous assumptions: first, PT data in a given round constitute a random sample drawn from an underlying population that is distributed normally, and therefore, the normality criteria can used to set the pass/fail threshold for PT participants based upon an arbitrary, predefined choice of <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">Z</i>-value; and second the maximum permissible error values are unimportant and can be ignored. Those two fundamental errors produce misclassified PT pass/fail results, which can have serious economic consequences for both EMC test laboratories and their customers. This article first reviews the published literature on quantitative EMC PT from the standpoint of assessing the statistical methodologies used and how the pass/fail criteria were applied. Next, this article discusses the effects of these statistical procedure errors. Finally, it details a practical, more accurate alternative Bayesian method that incorporates the maximum permissible error as the <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">a priori</i> information in its analysis model. This method treats the quantitative EMC PT data as a population with no specific presumed distribution.