[Lisa Feliks]

Thank you Clyde! I decided to include it. Te imputation worked, so the problem is gone. Following your argumentation I don't really have a choice . So I agree.

[Nick Cox]

Here's a not too wild thought. If you keep track of how many observations were used in the imputation, you have a cautionary measure. At a minimum you could check extreme imputations against how much information was used.

egen land = group(S003)
su land, meanonly  

gen N = .  
forvalues i = 1/`r(max)' {      
    capture regress educ (c.X047 c.X003 c.female)##c.wave if land==`i'    
    if _rc == 0 {          
          predict educ_hat if land==`i', xb        
          replace educ = educ_hat if missing(educ) & land==`i'        
          replace N = e(N) if land == `i'          
          drop educ_hat    
    }
}

[Lisa Feliks]

Thank you Nick! I will do this check tomorrow.

[Oscar Cerquera]

Hello,
I need help with a problem.
I am estimating a university value added model through a multilevel or mixed (mixed command in stata), with two levels (student level and university level); the model estimates me the residuals and/or variances of level 1 (var(Residual)) and level 2 ( var(_cons)). I need to predict the level 2 variance, because it is the one that calculates the aggregate value, but I don't know which command to use (predict resid, residual estimates the level 1 residuals).