Using Bayes methods and mixture models in inter-laboratory studies with outliers

Inter-laboratory studies (especially so-called key comparisons) are conducted to evaluate both national and international equivalence of measurement. In these studies, a reference value of some measurand (the quantity intended to be measured) is developed and results for all laboratories are compared to this single value. How to determine the reference value is not completely obvious if there are observations and/or laboratories that could be considered outliers. Since ignoring results from one or more participating laboratories is untenable in practical terms, developing methods that are robust to the possibility that a small fraction of the laboratories produces observations unlike those from the others is critical. This paper outlines two Bayesian methods of analyzing inter-laboratory data that have been proposed in the literature and suggests three modifications of one that are more robust to outliers. A simulation study is conducted to compare the five methods.