Sep 6 – 10, 2026
Centro Didattico Morgagni
Europe/Rome timezone

A study of prior sensitivity in random effects models for decisions in interlaboratory comparisons: credible intervals versus Bayesian evidence

Not scheduled
20m
Centro Didattico Morgagni

Centro Didattico Morgagni

Viale Morgagni 40, Firenze
Measurement Uncertainty and Metrology

Speaker

Séverine Demeyer (Laboratoire National de Métrologie et d'Essais)

Description

Random effects models are widely used in interlaboratory comparisons to estimate between-laboratory variability τ and assess degrees of equivalence of laboratories [1]. In this context, decisions are often based on 95% credible intervals [2], while Bayesian hypothesis testing provides an alternative probabilistic framework based on Bayes factors for assessing laboratory effects [3].
This work compares these two decision paradigms, focusing on the sensitivity of conclusions to the prior specification of the between-laboratory standard deviation. We investigate how credible interval based decisions and Bayes factor based hypothesis testing behave under different weakly informative priors on τ , including Half-Cauchy prior, inverse Chi-square and data-informed scale choices.
We further assess the robustness of credible interval and Bayes factor conclusions using posterior predictive simulations [4]. This predictive additional step to standard analysis allows us to examine whether decisions remain stable under replicated data scenarios, and to identify cases where apparent evidence may not be supported by predictive behavior.
Results from simulation studies and real interlaboratory data from CCQM-K53 [5] show that credible interval decisions are relatively stable with respect to prior choices, whereas Bayes factors exhibit strong sensitivity to the prior on τ. Posterior predictive checks provide additional insight to the robustness of the decisions. These findings emphasize the importance of combining inferential and predictive perspectives when making decisions in metrological applications such as laboratory equivalence assessments.

[1] Toman B, Possolo, A, Laboratory effects models for interlaboratory comparisons, Accred Qual Assur (2009) 14:553–563
[2] CCQM-KCWG/01, Guidelines for the CCQM KCWG on the review of CCQM CMCs for inclusion in the key comparison database, 2020.
[3] Wübbeler G, Bodnar O, Elster C. Bayesian hypothesis testing for key comparisons. Metrologia. 2016:1131-8.
[4] Raghu N Kacker RK Alistair Forbes, Sommer KD. Bayesian posterior predictive p-value of statistical consistency in interlaboratory evaluations. Metrologia. 2008:512-23.
[5] Lee J, Lee JB, Moon DM, Kim JS, van der Veen AMH, Besley L, et al. Final report on international key comparison CCQM-K53: Oxygen in nitrogen. Metrologia. 2010;47

Classification Both methodology and application
Keywords Bayesian hypothesis testing, Bayes factors, posterior predictive distribution, degrees of equivalence, prior knowledge

Primary authors

Dr Gerd Wübbeler (PTB) Séverine Demeyer (Laboratoire National de Métrologie et d'Essais)

Presentation materials

There are no materials yet.