Describir: Blending Bayesian and Classical Tools to Define Optimal Sample-Size-Dependent Significance Levels.