This article provides an adequate statistic for testing the rationality of point predictions of categorical outcomes under a subjective median or mode assumption. The test improves upon previous approaches by accounting for all the required moment inequalities and their respective covariance matrix in a single statistic. Monte Carlo simulations show that the test has an improved Type I error, especially in cases with multiple modes. Applying these tests to a dataset with parents' and students' predictions of school scores, I find that respondents are biased towards expecting higher grades and that students' expectations are correlated with the time of study.
