A statistical method for evaluating how well a model predicts the outcome of additional experiments.