The practical application of statistical methods of data processing always requires the certain assumptions concerning the initial data. When there is not enough a priory information for promotion or check of these assumptions, naturally, some doubts in the received results and their interpretation arise. As to the evaluation of key comparisons data, the measurement results obtained from different laboratories are, as a rule, correlated, but it's quite difficult to express quantitatively a degree of this correlation. So in practice often there is no valid information about the correlation matrix, that is necessary for uncertainty evaluation of reference value, as an estimate of the measurand, and uncertainty of deviations of measurement results from the reference value. It is necessary to emphasize one more aspect of key comparisons. There are two interconnected, but different tasks of key comparison: the assessment of equivalence of the national standards and the check of measurement result uncertainty assigned by each laboratory. The first task meets directly practical needs. According to MRA "The degree of equivalence is taken to mean the degree to which these standards are consistent with reference values determined from the key comparisons and hence are consistent with one another". The quantitative expression of equivalence, as a matter of fact, is a new way of expression of closeness (reproducibility) of the measurement results obtained from participating laboratories. The estimation of reproducibility is just important for the mutual recognition of measurement results from different laboratories. As to the second task, it is directly connected with the evaluation of the accuracy of the measurement results, finding - out of the reasons of systematic deviations between results of different laboratories. The knowledge of the correlation matrix of measurement results is necessary for the decision of the second task. For the decision of the first task - assessment of the equivalence of the national standards - the approach that does not require the knowledge of the correlation matrix and uses the key comparisons data model based on the mixture of probability distributions of measurement results in each laboratory is offered in the paper.
展开▼