Inference about misclassification probabilities from repeated binary responses.

Repeated binary responses provide efficient information for two purposes: (1) estimating two misclassification (false-positive and false-negative error) probabilities and (2) testing the hypothesis that either is zero in a reliability study. We focus on the assessment of reliability of a diagnostic test when there is no gold standard. This paper uses a latent class model and illustrates some of its properties. In addition, application to data containing variation among individuals is considered. We apply this model to the serological data on the MNSs blood group of atomic bomb survivors and their children. The results provide valuable information for examining measurement reliability.