www.cmm.bristol.ac.uk/forum

You should read the manual that is on the website. It does not used chained equations and you can spcify missing data anywhere and imputed values are available.
Harvey Goldstein

We haven't really explored this. There are some bounds that are placed around draws that may deal with this, but you would do best to avoid perfect predictions if you can. Interested to know if it works for you however.
Harvey Goldstein

When you specify a 2 level imputation model any level 1 variables that are declared responses at level 1 implicitly contribute through their level 2 random effects. However you cannot have a level 1 predictor for a level 2 response in any multilevel model so do not include level 1 variables as covar...

There is no feature in realcom to do this I'm afraid.
Harvey Goldstein

The IGLS estimates are indeed maximum likelihood ones. Re missing data the mlwin default is listwise deletion of all level 1 records where any model variable has a missing value. If you wish to do an efficient multiple imputation on 2-level data you have two possibilities now. Either REALCOM which h...

Should be original 2 level dataset
Harvey

Yes - that should be fine. Let me know if there are any problems.
Harvey Goldstein

Yes - imputation can be slow. We are switching now to STATJR (see CMM web site) and a 2-level imputation module has just been put up there - use of this is free.
You should ideally put all variables in as responses.
Harvey Goldstein

All variables in the MOI, whether responses or predictors do need to be in the imputation model.

Strictly speaking you should set up a 2-level imputation model where any level 1 vbles in the MOI are responses (at level 1 in the imputation model).
Harvey Goldstein