Tutorial and spreadsheets for Bayesian evaluation of risks of false decisions on conformity of a multicomponent material or object due to measurement uncertainty

被引：20

作者：

da Silva, Ricardo J. N. B. ^{[1
]}

Pennecchi, Francesca R. ^{[2
]}

Hibbert, D. Brynn ^{[3
]}

Kuselman, Ilya ^{[4
]}

机构：

[1] Univ Lisbon, Ctr Quim Estrutural, Fac Ciencias, Edificio C8, P-1749016 Lisbon, Portugal

[2] Ist Nazl Ric Metrol INRIM, Str Cacce 91, I-10135 Turin, Italy

[3] UNSW Sydney, Sch Chem, Sydney, NSW 2052, Australia

[4] Metrol, 4-6 Yareh St, IL-7176419 Modiin, Israel

来源：

CHEMOMETRICS AND INTELLIGENT LABORATORY SYSTEMS | 2018年 / 182卷

关键词：

Conformity assessment; Risk of false decision; Measurement uncertainty; Multicomponent material; Markov chain Monte Carlo simulations; Spreadsheet;

D O I：

10.1016/j.chemolab.2018.09.004

中图分类号：

TP [自动化技术、计算机技术];

学科分类号：

0812 ;

摘要：

A tutorial and a user-friendly program for evaluating risks of false decisions in conformity assessment of a multicomponent material or object due to measurement uncertainty, based on a Bayesian approach, are presented. The developed program consists of two separate MS-Excel spreadsheets. It allows calculation of the consumer's and producer's risks concerning each component of the material whose concentration was tested ('particular risks') as well as concerning the material as a whole ('total risks'). According to the Bayesian framework, probability density functions of the actual/'true' component concentrations (prior pdfs) and likelihood functions (likelihoods) of the corresponding test results are used to model the knowledge about the material or object. Both cases of independent and correlated variables (the actual concentrations and the test results) are treated in the present work. Spreadsheets provide an estimate of the joint posterior pdf for the actual component concentrations as the normalized product of the multivariate prior pdf and the likelihood, starting from normal or log-normal prior pdfs and normal likelihoods, using Markov chain Monte Carlo (MCMC) simulations by the Metropolis-Hastings algorithm. The principles of Bayesian inference and MCMC are described for users with basic knowledge in statistics, necessary for correct formulation of a task and interpretation of the calculation results. The spreadsheet program was validated by comparison of the obtained results with analytical results calculated in the R programming environment. The developed program allows estimation of risks greater than 0.003% with standard deviations of such estimates spreading from 0.001% to 1.5%, depending on the risk value. Such estimation characteristics are satisfactory, taking into account known variability in measurement uncertainty associated with the test results of multicomponent materials.

引用

页码：109 / 116

页数：8

共 19 条

[1]

Anderson T.W., 1986, STAT ANAL DATA, V2nd, DOI DOI 10.1007/978-94-009-4109-0

[2]

[Anonymous], 20160071500 IUPAC

[3]

[Anonymous], 2008, GUID EXPR UNC MEAS

[4]

BIPM I. F. C. IFCC I. L. A. C, 2012, 1062012 BIPM IEC IFC

[5]

Ellison, 2007, EURACHEM CITAC GUIDE

[6]

Ellison SLR., 2012, Quantifying uncertainty in analytical measurement

[7]

Gelman A., 2013, BAYESIAN DATA ANAL, DOI DOI 10.1201/B16018

[8]

Geyer C.J., 2017, mcmc: Markov Chain Monte Carlo. R package version 0.9-5

[9] Markov chain Monte Carlo methods: an introductory example [J].

Klauenberg, Katy ;

Elster, Clemens .

METROLOGIA, 2016, 53 (01) :S32-S39

[10] Total risk of a false decision on conformity of an alloy due to measurement uncertainty and correlation of test results [J].

Kuselman, Ilya ;

Pennecchi, Francesca R. ;

da Silva, Ricardo J. N. B. ;

Hibbert, D. Brynn ;

Anchutina, Elena .

TALANTA, 2018, 189 :666-674

← 1 2 →