De gustibus non est taxandum

Preferences over consumption and leisure play no role in the standard optimal tax model, which attributes all variation in earnings to differences in income-earning ability. We show how to incorporate these preferences, which like ability are publicly unobservable, into the standard model in a tractable way. In this more general model, the policy designer must guess at the relative importance of ability and preferences in explaining variation in earnings. We show that such preferences could, in principle, increase or decrease optimal redistribution. In the most plausible specifications of the model, however, the result is clear: greater variation in preferences lowers the optimal extent of redistribution. To generate more redistribution than in standard results, one must assume that the desire for income is inversely related to income earned. This result holds even when the conventional model accurately describes the average individual, and it suggests one potential resolution to the puzzle of why observed redistribution is in some cases weaker than conventional theory would suggest. We then establish a new empirical finding that confirms this model's central policy prediction across developed countries and U.S. states. In countries and states with more heterogeneous tastes for consumption relative to leisure, redistribution is statistically significantly lower.