Latent Class Models Approach to the Multi-rater Agreement

摘要

Kappa-like statistics measures the agreement among two raters for nominal scale. For the multi-rater studies the extended versions of kappa-like statistics are suggested. Latent class models also investigate inter-rater agreement and provide a powerful and flexible tool and allow us to analyze agreement among more than two raters simultaneously. The latent class models are the finite mixture models. This model treats a contingency table as a mixture of unobserved qualitative latent variable. When the latent class model fits to the data, the expected frequencies can be used to assess the rater agreement. In this study, it is purposed to describe the latent class approach in the agreement context and, to compare this approach to the classical agreement indices on various numerical examples.