Evaluation of CRM homogeneity in cases of insufficient method repeatability: Comparison of Bayesian analysis with substitutes for ANOVA based estimates

Insufficient method repeatability is a problem for the evaluation of certified reference materials (CRMs). Investigating the homogeneity studies of 216 certified parameters over the of 36 CRMs released over the last four years, it was found that in 1/3 of the cases the method repeatability (swb) was too high to calculate the standard deviation between units (sbb). It was also found that the application of the repeatability requirement stated in ISO Guide 35:2017 is not feasible, as it would require unrealistically low repeatabilities (swb) or impossible high number of replicates per unit. Evaluation of the uncertainty of homogeneity (ubb) as evaluated by ANOVA using both the maximum of sbb and 0, the maximum of sbb and u*bb, the uncertainty hidden by method repeatability, the maximum of sbb and sbb/n and Bayesian analysis using both informative and uninformative priors and using the standard deviation of the unit means were compared using simulated homogeneity studies with repeatabilities of 1-8 % and sbb between 0.2 and 2.8 %. It was found that using the maximum of sbb and sbb/n as estimate of ubb guards against severe underestimation, but usually results in severe overestimation. Using the maximum of (sbb, 0) shows the least average bias, but results in severe underestimation of ubb in a high fraction of cases. Using the maximum of (sbb, u*bb) limits, but does not eliminate cases of severe underestimation and also leads to average results biased towards high values. It was found that Bayesian analysis performs worse than max (sbb, u*bb) in limiting severe underestimation of ubb, but limits the average bias towards high results. A risk-based approach to cases of insufficient method repeatability is proposed where, after evaluating the other contributions to the uncertainty of certified values, the effect of severe over- and underestimation of ubb is evaluated and an appropriate approach is chosen based on this analysis.

LINSINGER Thomas; 

2020-08-11

Elsevier

JRC116160

2590-1346 (online),   

https://www.sciencedirect.com/science/article/pii/S2590134620300116?via%3Dihub#appsec1,    https://publications.jrc.ec.europa.eu/repository/handle/JRC116160,   

10.1016/j.acax.2020.100049 (online),