www.cmm.bristol.ac.uk/forum

Yes -weights should certainly be used in model of interest. MLwiN will allow this.
It is often not serious if you do not use weights in the imputation.
Harvey Goldstein

The equations displayed are for the complete imputation model that you have specified. See notes on realcom for an example.
Harvey Goldstein

Mlwin will by default always use casewise deletion for missing data - even when carrying out a Bayesian fit using MCMC. REALCOM was written to overcome this and it links with mlwin - but will handle only 2 levels at present. You could impute using 24 dummies for country. Go to realcom web site to se...

The weights facility is still not implemented. Recent work by Carpenter has addressed this issue - see new book by Carpenter and Kenward (Wiley, 2013)

This is often what you might get when you do imputations - suggests that the complete cases analysis produced biased results.
Harvey Goldstein

Mlwin will not automatically combine residuals. You will need to do this manually. We will look into doing this.
Harvey Goldstein

You should generally use all variables that will appear in the model of ineterst. In addition you can condition on any auxiliary variables that might be useful for satisfying the missing at random assumption.
Harvey Goldstein

An interesting question. At the moment the MOI should not contain interaction terms. There are different approaches in the literature to incorporating interactions (including polynomials) - see e.g. van Buuren S (2007). Multiple imputation of discrete and continuous data by fully conditional specifi...

I'm puzzled by this. I don't think that realcom will ever write 'NA' in these files. It is just possible that if the model fails to converge you could get 'NAN'.
Harvey Goldstein

Re your second question categorical expl variables should be entered with dummy variable coding.
Re first Q try running as a single level model. Can you also confirm which are responses and which explanatory.
Harvey Goldstein