estimating wages with EARNWT after Heckman selection - follow up

I asked a question on this previously, but I believe I was unclear. I’d like to know how to include the earnwt when using heckman selection. I use STATA and the help documents for this command state that I cannot use weights if twostep is specified (which it needs to be for the analysis I’m doing). Since the IPUMS description of earnwt state that it needs to be used for any analyses using income variables, I’d like to know if there is a way around this when conducting heckman selection analyses that control for selection into employment before estimating wages.

Thanks for following up. Note that we (IPUMS) do not generally provide support for specific questions relating to statistical software packages. Better resources for Stata support are available (see, for example, StataList).

That being said, I can provide you with a suggestion for how I would go about this procedure. I would estimate the Heckman selection two-step regression manually. I would first estimate the selection equation. Next I would estimate the predicted value of the dependent variable and generate the inverse mills ratio. Finally, I’d estimate the second-stage regression including the inverse mills ratio as a right-hand-side variable. That is, I’d run something like this:

probit y x
predict y_hat, xb
gen IMR = normalden(y_hat)/normal(h_hat)
reg y x IMR

This procedure will provide “correct” point estimates but “incorrect” standard errors because the second-stage regression includes a generated variable. So, you’ll need to wrap this all in a boostrapping program.