Classes of inter-comparisons and the case of temperature standards

摘要

The paper discussed a classification of the inter-comparisons that is relevant to identify the proper statistical method to combine the data provided for each participant to the inter-comparison. The proposed approach for Class 2 IC's constructs a single probabilistic model for the reference distribution probability function, based on the use of the mixture density model. This approach allows the estimate of the reference value simply as the expected value of the mixture density function. The method does not require strong assumptions -as N large- or limitations on the local probability distributions, such as the Normality density in each participant laboratory. It is particularly valuable for inter-comparisons of physical-state realisations, e.g., for temperature standards, where the population associated to an IC can be viewed as a super-population. However, Class 2 inter-comparisons probably include a wider range of IC's, such as the ones where a single standard is circulated and measured. The paper also compared the main features of the statistical treatments suitable for the IC outcomes in both cases of Class 1 and Class 2. Some aspects of the treated problems, especially the meaning of uncertainty in Class 2 IC's and the consequent approach, still deserve a deeper subsequent insight.