Imaging technologies for detecting beta-amyloid in patients with suspected Alzheimer’s disease have advanced. It is important to identify factors that contribute to diagnoses based on these images, especially qualities that contribute to reader concordance, as patients might seek a second opinion. There are statistical methods that quantify agreement, such as Cohen’s kappa statistic, but none have clear interpretations, differentiate positive and negative agreement, or account for within rater variation. We developed a method that addresses these challenges. First, a binomial mixed model with random effects for reader and image was fit. Second, random and fixed effect values were used in the trinomial distribution to obtain three probabilities of concordance: positive agreement, negative agreement and disagreement. Simulation results described concordance between readers at a range of fixed effects and demonstrated situations where disagreement may be high. Using this method, important factors, such as clinical presentation or reader training, could be identified and utilized to minimize disagreement of Alzheimer’s diagnoses based on imaging.