I first conducted a fixed effect model using xtreg y x, fe and I found that all the variables are significant and R-squared is .51.
So I thought that maybe I should use two step system GMM to account for endogeneity. But, since I only have 3 years, when i include the lagged variable as a predictor using xtabond2 y l.y x y*, gmm ( l.y) iv (x y*), equation (level)) nodiffsargan twostep robust orthogonal small, the number of observations per group shrinks to two and I can't even run an AR(1) test or Sargan test. And Also the output shows insignificant lagged variable.
I am still new to dynamic panel data models. Do I need GMM in such small sample size and small number of observations? Should i use something else? If i only report fixed effects results would that be sufficient to be considered for publication?
I would love to hear your recommendation. Thank you very much,
Three periods is enough, what matters is to have a reasonably high N. The motivation for using the GMM lies on the inclusion of the dynamics (the lag of the dependent variable). You do not need it you think that the dynamics is not important.
Jose, thank you for your input. Building on what you said is it to correct to do it this way:
xtabond2 y x y*, gmm (l.y) iv (x y*), equation (level)) nodiffsargan twostep robust orthogonal small
Why don't you try a longitudinal models? the best intro is here:
https://www4.stat.ncsu.edu/~davidian/st732_sp2007/notes/toc.pdf
Prof. Davidian developed these models. Feel free to ask questions.
Best, D. Booth
- Badges
- Science topic
- Similar topics
- Mathematics
- Statistics
- Sampling Design
- Sample Size