www.cmm.bristol.ac.uk/forum

I'm running Mlwin 2.35 under Win Server 2012 (64 bit) with 12 GB memory. I have a dataset that is about 150k individuals. I want to build a cross classified multilevel logistic model (all covariates at level 1) with random intercepts only for now. The documentation suggests MCMC as the method of choice for estimation, but I prefer ML since I have some missing data on my DV and IVs. My understanding is that IGLS constitutes an ML approach.

1) Am I correct?
2) Can I do this in Mlwin on a cross classified model?

Yes, IGLS is equivalent to maximum-likelihood estimation. When fitting models using (R)IGLS in MLwiN any rows of data containing missing values are ignored. For information on fitting cross-classified models using (R)IGLS see chapter 18 of the user guide (available at http://www.bristol.ac.uk/cmm/software/m ... nuals.html). Note however that this is fitted using constraints (as described in the chapter) so you may find that the estimation becomes difficult with larger amounts of data.

Yes, I understand the potential estimation problem. However, this sounds more like listwise deletion rather than true FIML, correct?

Yes, this is listwise deletion.