The Sad Truth About Happiness Scales

We show that, without strong auxiliary assumptions, it is impossible to rank groups by average happiness using survey data with a few potential responses. The categories represent intervals along some continuous distribution. The implied CDFs of these distributions will (almost) always cross when estimated using large samples. Therefore some monotonic transformation of the utility function will reverse the ranking. We provide several examples and a formal proof. Whether Moving-to-Opportunity increases happiness, men have become happier relative to women, and an Easterlin paradox exists depends on whether happiness is distributed normally or log-normally. We discuss restrictions that may permit such comparisons.