Series: Risks of conformity assessment of a multicomponent material or object in relation to measurement uncertainty of its test results
When components of a substance or material are linked by a mass balance constraint (sum of their mass fractions, molar fractions or any other positive quantity ratios is 100 % or 1), test results of the components’ contents are named “compositional data”. These data are correlated because of the constraint: the correlations are called “spurious”. Such correlations may influence measurement uncertainty of the test results, and therefore risks in conformity assessment of the substance or material. That is important in testing geological and environmental objects, products of metallurgical and food industries, etc. A special case is the evaluation of purity of substances and development of corresponding (pure) certified reference materials, based on a mass balance.
There is an extensive literature stressing how traditional statistical techniques may produce inadequate results if applied on raw compositional data without suitable transformation. However, the relevant techniques of the compositional data analysis are still not implemented in metrology and analytical chemistry, as well as in conformity assessment.
In the IUPAC project 2016-007-1-500 a general Bayesian approach was elaborated for evaluation of risks of false decisions in conformity assessment of multicomponent materials or objects, taking also into account possible correlations between the measured property values of an item’s components. This ‘conventional’ approach applies integration of the relevant posterior multivariate probability density function on the tolerance/specification multi-domain for evaluation of the item conformance probability and corresponding risks of false decisions in the conformity assessment.
In the case of a mass balance constraint, the components’ contents of a substance or material form a multi-dimensional simplex, to which, in general, Euclidean geometry cannot be straightforwardly applied. A problem in compositional data analysis is also influence of spurious correlations between test results on singular covariance matrices. Spurious correlations should be considered as well at evaluation of measurement uncertainty of the test results, e.g. when considering uncertainty of a substance purity equal to the difference between 100 % and the sum of the test results of impurities mass fractions.
In this project, we will study how to deal with compositional data in conformity assessment. The proposed approach consists in resorting to a Monte Carlo method taking into account the mass balance constraint. The influence of the mass balance constraint on measurement uncertainty and risks in conformity assessment will be highlighted.
Page last updated 25 July 2019