Identification and Handling of Discrepant Measurements in Key Comparisons

Abstract

In a key comparison one or several measurement objects are circulated among a number of laboratories, each of which measures the quantities represented by the objects. In order to compare the measurement results obtained by the participating laboratories, the values represented by the circulated objects have to be established. These values, known as the key comparison reference values, and their associated uncertainties can easily be calculated by the method of least squares from the measurement results provided by the participating laboratories. Since this method requires that the measurement results be mutually consistent, a hypothesis that sometimes has to be rejected at a reasonable level of significance, a procedure for identification and handling of discrepant measurements is needed. In this paper such a procedure is suggested. It is demonstrated that although a key comparison reference value is attracted to a discrepant measurement result that has an uncertainty much smaller than the remaining results, the suggested procedure is able to identify this discrepant result. It is also demonstrated that the exclusion of a discrepant measurement result from the calculation of the reference values does not amplify the discrepancy of that result. As the discrepant result is not excluded from the comparison itself, the exclusion of the result in the calculation of the reference values should therefore be uncontroversial.