Robust estimation in interlaboratory measurements with small number of measurements

In this paper two robust methods of assessing the value and the uncertainty of the measurand from the samples of small number of experimental data are presented and compared. Those methods can be used when some measurements results contain outliers, i.e. when the values of certain measurement results significantly differ from the others. They allow to set a credible statistical parameters of the measurements with the use of all experimental data. The following considerations are illustrated by the numerical examples of multi-laboratory measurement data key comparison. Compared are the results obtained by a classical method with rejection of outliers with two robust methods: a rescaled median absolute deviation MADS and an iterative two-criteria method. The paper also presents the advantages of the robust iterative statistical method in estimating the accuracy of the tested laboratory measurement results during its accreditation on the sample of four elements with outlier. A comparison with the estimates obtained by the standard procedure for evaluating performance accuracy is also provided.