Measurement system analysis for binary data

We describe a methodology for the assessment of the repeatability and reproducibility (R&R) of measurement systems that measure on a binary scale, such as pass-fail inspections. We focus on the situation where no reference values can be obtained for the objects in the experiment and consequently model the results of the R&R experiment as a latent class model. We provide estimators based on the maximum likelihood approach and the method of moments, and compare their properties. We also give guidelines for model checking and recommendations for sample sizes. The methodology is illustrated by an example.