Higher Order Learning Gains