Economists and other social scientists frequently treat scales that are ordinal (or ordinal with respect to anything of social scientific interest) as if they were cardinal. I review the problems with treating such scales as cardinal both when the data represent points on an ordinal scale and when they represent intervals. In the former case, it is sometimes possible to compare means between groups but often difficult to make strong statements about differences-in-differences. In the latter case, it is generally impossible to compare means over time or between groups without strong and, perhaps, implausible parametric assumptions. I present examples from the literature on the black-white test score gap, the fadeout of early childhood interventions and the happiness literature. In the first case, I provide bounds on the evolution of the black-white test-score gap in the early school grades and show that it might increase or decrease. To create a meaningful cardinal scale, I tie test scores to predicted eventual completed education and show that using this scale, the test-score gap does not increase over the first seven years of school. In the second case, I show that the extent and even the existence of fadeout for one mathematics intervention depends on the choice of scale. Finally, I show that we can reverse every major result in the happiness literature using plausible alternative assumptions about the scale.