Dependence of the treatment of systematic error in interlaboratory comparisons on different classes of standards

Abstract

A few years ago the question arose whether different types of measurands can be the consequence of the different intrinsic nature of different types of standards, which prompted an analysis resulting in the proposal of two distinct ‘classes’ of standards be considered. They correspond to different situations depending upon whether the aim is to compare laboratory artefact standards (Class 1), or, instead, to compare standards whose values are a measure of the value of the same unequivocally defined physical or chemical condition (Class 2). This distinction is relevant to the statistical treatment of the comparison data, in particular, with respect to the issue of the correct modelling of the data and of the correct choice of the summary statistics for populations of non-repeated data. This paper, after a summary of the rational of making a distinction of standard in “classes”, and discussing the general issue of bias, is discussing some implications of the above distinction, concentrating on cases where the effect of the systematic error is dominating the experimental results, as it commonly happens in several metrology fields. A preferred way to tackle the problem for standards of Class 2 is presented.