Certification of reference materials is far more than just characterisation of a selected homogeneous batch of material. From the perspective of the ISO Guide on the Expression of Uncertainty in Measurement (GUM) all uncertainty sources relevant to the user of an individual certified reference material (CRM) sample at a moment in time should be part of the CRM uncertainty. This not only includes the full uncertainty of the batch characterisation (rather than the statistical variation), but also all uncertainties related to possible between-bottle variation, instability upon long-term storage and instability during transport to the customer.

 A pragmatic method is proposed for the implementation of the Guide to the expression of uncertainty in measurement in the certification of reference materials by laboratory intercomparison. It is based on the establishment of a full uncertainty budget for each laboratory result and the estimation of the impact of various laboratory standard uncertainties and of between-units variability on the certified reference material (CRM) uncertainty.

The preparation and certification of reference materials is a rapidly developing area. Many innovative reference materials have limited homogeneity and stability, and, additionally, the uncertainty estimation of the property values must be brought in agreement with the principles of the “Guide to the expression of uncertainty in measurement” (GUM). The results of the homogeneity and stability studies must be included to a certain extent in the uncertainty of the property values of the reference material, in order to comply with these requirements. The basic theory needed to accomplish this is essentially the theory of analysis of variance (ANOVA). As GUM also allows alternative evaluations other than Type A evaluations, a reinterpretation of the theory of ANOVA is necessary to establish a model for the certification of reference materials that is widely applicable. For this, analysis of variance can be used as a statistical technique to derive standard uncertainties from homogeneity, stability and characterisation data.

A pragmatic method is proposed for the implementation of the Guide to the expression of uncertainty in measurement in the certification of reference materials by laboratory intercomparison. It is based on the establishment of a full uncertainty budget for each laboratory result and the estimation of the impact of various laboratory standard uncertainties and of between-units variability on the certified reference material (CRM) uncertainty.

Many reference materials undergo a batch certification, which implies that a small number of samples is taken from a batch, characterised, and these results are then assumed to be representative of all remaining samples. An important aspect in this design is the translation of the characterisation data to a single sample, as usually the laboratory will be using only one sample of the batch. This form of homogeneity is very important and can be influenced to a certain extent by well-designed sample preparation procedures. Another subsampling problem associated with many reference materials is that only a small test portion is drawn from the sample to carry out the measurement. Obviously, this test portion must be representative of the sample, otherwise the certified value is still not applicable. Both kinds of homogeneity tests are examined in the paper and evaluated using practical examples.

Abstract 

To serve as a measurement standard, a (certified) reference material must be stable. For this purpose, the material should undergo stability testing after it has been prepared. This paper looks at the statistical aspects of stability testing. Essentially, these studies can be described with analysis of variance statistics, including variant regression analysis. The latter is used in practice for both trend analysis and for the development of expressions for extrapolations. Extrapolation of stability data is briefly touched upon, as far as the combined standard uncertainty of the reference material is concerned. There are different options to validate the extrapolations made from initial stability studies, and some of them might influence the uncertainty of the reference material and/or the shelf-life. The latter is the more commonly observed consequence of what is called ’stability monitoring’.

To serve as a measurement standard, a (certified) reference material must be stable. For this purpose, the material should undergo stability testing after it has been prepared. This paper looks at the statistical aspects of stability testing. Essentially, these studies can be described with analysis of variance statistics, including variant regression analysis. The latter is used in practice for both trend analysis and for the development of expressions for extrapolations. Extrapolation of stability data is briefly touched upon, as far as the combined standard uncertainty of the reference material is concerned. There are different options to validate the extrapolations made from initial stability studies, and some of them might influence the uncertainty of the reference material and/or the shelf-life. The latter is the more commonly observed consequence of what is called ‘stability monitoring’.