ENBIS: European Network for Business and Industrial Statistics

Measurement system analysis for binary classifications: a taxonomy of methods

21 September 2009, 15:45 – 16:05

Abstract

Submitted by

Tashi Erdmann

Authors

J. de Mast T.P. Erdmann W.N. van Wieringen

Affiliation

IBIS UvA

Abstract

Several methods are available to assess a binary measurement system. Two factors are highly relevant in deciding which method to use: 1) whether a reference value is available and 2) whether the underlying measurand is continuous or a true dichotomy. Treating a continuous measurand as an artificial dichotomy can result in misleading reproducibility metrics, because the assumption of independence of the measurements conditional on the measurand does not continue to hold. This paper provides a taxonomy of methods for studying and quantifying reproducibility of binary classifications, indicating in which situation each of them can be used, and discussing experimental design, statistical modelling, and reproducibility metrics.

Return to programme