Estimation of variance components from the results of interlaboratory comparisons

Interlaboratory comparisons are frequently carried out in analytical laboratories either as a part of their quality assurance procedures or as an important part of analytical method development. The total analytical error is usually composed of three components: the random measurement error, laboratory bias and sample—laboratory interaction. The significance of these error components and the estimates of their variances can in principle be obtained from a properly designed experiment by carrying out the analysis of variance. If the samples used in the interlaboratory comparison cover a wide concentration range it is usual that the errors are not independent of concentration and, consequently, the estimation of all three error components is difficult. In this paper a scaling procedure for the initial data is proposed that makes it possible to estimate all three error components.