Labor economists have long used occupation indicators as a proxy for unobserved skills that a worker possesses. In this paper, we consider whether inter-occupational wage differentials that are unexplained by measured human capital are indeed due to differences in often-unmeasured skill. Using the National Compensation Survey, a large, nationally- representative dataset on jobs and ten different components of requisite skill, we compare the effects on residual wage variation of including occupation indicators and including additional skills measures. We find that although skills do vary across 3-digit occupations, occupation indicators decrease wage residuals by far more than can be explained by skill differentials. This indicates that "controlling for occupation" does not equate to controlling for skill alone, but also for some other factors to a great extent.
Additionally, we find that there is considerable within occupation variation in skills, and that the amount of variation is not constant across skill levels. As a result, including occupation indicators in a wage model introduces heteroskedasticity that must be accounted for. We suggest that greater caution be applied when using and interpreting occupation indicators as controls in wage regressions.