I want to estimate differences in income earned from wages after controlling for selection into full time employment using pooled 2014-2018 CPS data. The Heckman command states that I can’t use weights if twostep is specified. If I run the analysis without EARNWT, will my estimates of wage differences still be accurate?
The use of sampling weights does not necessarily make your estimates more or less accurate. It really depends on what you are aiming to estimate. If you want to calculate a statistic that is representative of some region (e.g., the entire US population) then you will want to incorporate sampling weights. If you are simply trying to calculate statistics for the specific sample of individuals sampling weights are likely not necessary. Although we generally do not provide support for use of statistical software, note that there are multiple ways to specify sampling weights in Stata. You can specify weights within given commands or by using the SVY command.