My research requires that I predict the expected labor market earnings of individuals who are not employed. To do this I have run an expected earnings regression, with incwage as the explained variable and demographic characteristics, level of education, region in which one live, etc. as the explanatory variables. I’ve choice to weigh the data in Stata using i weights (iw) and using WTSUPP.
I used WTSUPP, as opposed to EARNWT, because what I have read about weighting in the CPS March Supplement seems to suggest that I ought to use WTSUPP if I use non-earnings related variables, such as education and sex. I have also used WTSUPP because my research requires that I run a number of other regression that use the same explanatory variables but have different y variables that are not earnings related (such as labor force participation).
My question is whether I should continue to use WTSUPP or should I use EARNWT for those regressions where the explained variables are earnings related?
Additionally, I have been using WTSUPP when creating tables with the “mean” command in stata. The variables that I am finding the mean for are variables which concern transfer payments, such as incssi, incwelfr, and income variables like inctot and incwage. Should I be using EARNWT instead? Should I use the EARNWT command if I look at transfer income by sex or quintiles? Below I provide two examples of mean commands I have:
mean mthwelfr [iw=wtsupp], over(sex)
mean incssi [iw=wtsupp], over(quintile)