Dear Statalisters,

I currently analyze an unbalanced panel with about 200 different firms and aprximately 4000 different observations (for some firms I have only 2 observations; for others I have 50). I think of this dataset as a panel and include random effects for each firm. I am also (but not exclusively) interested in the sigma_u of the firms.
In principle, I see three different ways of fitting this model:
Code:
Option 1: xtreg dependent variable covariates, re
Code:
Option 2: xtreg dependent variable covariates, re mle
Code:
Option 3: xtreg dependent variable covariates, re sa
While all three options give me comparably similar estimates for the covariates, they drastically differ in their estimate of sigma_u. In fact, Option 3 > Option 2 > Option 1.
I am wondering what this actually means and whether this could be potentially relevant for me.

Best regards Peter