Agreement between raters on disease diagnostics is typically assessed using the kappa coefficient. There has been considerable work using logistic regression to provide summary estimates of inter-rater agreement. Shoukri and Mian (1996) used covariates to predict the marginal probability of classification by each rater using a likelihood based approach. Klar et al (2000) used covariates to predict kappa using an estimating equations approach. We propose a method that identifies covariates predictive of kappa using a likelihood based approach under Bayesian framework. To illustrate this procedure, we study how patient age and physician gender affect inter-rater agreement in the diagnosis of the disease of endometriosis.