Treatment analyses based on average outcomes do not immediately generalize to the case of ordered responses because the expectation of an ordinally measured variable does not exist. The proposed remedy in this paper is a shift in focus to distributional effects. Assuming a threshold crossing...