A new approach to determining the key comparison reference value

摘要

National Measurement Institutes are becoming much more involved in international activities as world trade expands. One important activity is to organize and participate in comparisons aimed at establishing their calibration measurement capabilities. A typical comparison circulates a number of artefacts to between 10 and 20 institutes, which measure them following a defined technical protocol. A report is then written, reporting the results and drawing some conclusions about each laboratory's performance relative to a reference value, taken in the context of their declared uncertainties. The determination of the reference value is a very important first step and often results in a lot of discussion. In general laboratories have different capabilities and the reference value needs to be a weighted mean of some kind. This paper evaluates an approach which determines a participant's weighting factor from the reported results, without using the participant's uncertainty estimates. It is applied to a recent key comparison which required expert judgment from the pilot to exclude some results that contained measurement errors. This method avoids the need to exclude participants and is relatively insensitive to artificial noise, or an offset, added to one of the data sets.