The magnitude of the actual difference between two models; smaller effects require more samples to detect reliably.