I am working on a project where we have just switched from cross-sectional data (characteristics of individual observations measured at a certain point in time) to a panel dataset looking at these observations over a period of about 30 years (unbalanced panel as some observations exit the sample early). We've been using a Heckman, two step model to control for selection bias in the outcome stage and would like to employ the same method for the quarterly panel dataset over time. I am wondering what the most methodologically robust option may be to run this model with the panel data. I have done substantial research around Statalist/Stack Exchange and will summarize the main suggestions that I have seen so far.
1. Run the original Heckman with year dummy variables in the selection as a proxy for fixed effects
2. Run the Heckman (ML) with clustered standard errors on the observation level
3. Utilize the GLLAMM user-written command to simulate a Heckman (http://www.stata.com/statalist/archi.../msg00290.html)
4. Run xtprobit, estimate the inverse mills ratio and add the inverse mills term to the xtreg (will not run for me, citing not enough observations)
Any input would be much appreciated. Thank you very much.
1. Run the original Heckman with year dummy variables in the selection as a proxy for fixed effects
2. Run the Heckman (ML) with clustered standard errors on the observation level
3. Utilize the GLLAMM user-written command to simulate a Heckman (http://www.stata.com/statalist/archi.../msg00290.html)
4. Run xtprobit, estimate the inverse mills ratio and add the inverse mills term to the xtreg (will not run for me, citing not enough observations)
Any input would be much appreciated. Thank you very much.
Comment