Statistics for industrial consumer studies12 June 2009
Measurement System Analysis for classifications in industry12 June 2009, 14:15 – 15:00
Jeroen de Mast IBIS UvA firstname.lastname@example.org The precision of measurement devices in the engineering and physical disciplines, as well as the reliability of medical diagnoses and psychometrical instruments and tests, and finally the legal implications of metrology, are topics with substantial impact. The standards for measurement system analysis in industry are laid down in the MSA Manual of the Automotive Industry Action Group (AIAG, 2002) and ISO’s Guide to the Expression of Uncertainty in Measurements (ISO, 1995). All three topics and both standards struggle with fundamental and practical hiatuses in insight, for which it is plausible that statistics can make contributions and possibly even breakthroughs. For this talk we will focus on the reproducibility and repeatability (R&R) of classifications, such as ratings and grading of products in quality inspection and consumer studies. We start with a brief review of fundamental concepts in measurement theory, such as the notion of measurement as a map of empirical ordering to a measurement scale, and the typical lay-out of standard R&R studies in industry. Then we turn to classifications, and we discuss various options for assessing their R&R, such as agreement metrics, association measures, and model-based approaches. Finally, we will demonstrate the use of models borrowed from item response theory (IRT). The theory is illustrated from a real-life example.