Weighting in unbalanced panel

[Francesco Filippucci]

Dear all,
I've a 10 years unbalanced panel from GSOEP (German Socio-Economic Household Panel), with some respondants in the first year that drop out through the period and some respondants that start being observed during the period. I also have the inverse selection probabilities and the inverse staying probabilities for each individualin each year, based on some individual characteristics. I want to run a FE panel regression on this 10 yy period, but I need to weight it to make it representative of German population.
I can select only individuals who stayed from beginning to the end of the 10yy period, obtaining a balanced panel, and weight individuals by the product of inverse selection probabilities and the inverse staying probabilities from first to 10th year. This however makes me lose a lot of observations and variation.
Bottom line:
Do you have any references or suggestions on how to weight an unbalanced panel in stata?

[Phil Bromiley]

You didn't get a quick answer. You'll increase your chances of a helpful answer by following the FAQ on asking questions - provide Stata code in code delimiters, readable Stata output, and sample data using dataex.

A lot depends on whether the errors in the selection equation are correlated with the errors in the outcome equation. Throwing away everyone who doesn't stay the entire 10 years is problematic. It magnifies sample selection problems. As long as you're thinking fixed effects, I suspect you can go with Heckman in earlier versions or the eregress routines in Stata 15. Weighting is explained for almost every Stata procedure, but the question is the correct weights. I would also look at the survey documentation in Stata.