FIML and mixed

[Tracy Lam]

I would like to estimate a multilevel model but have about 5-10% missing data on my covariates. I would like to use FIML to handle missing data. Is this possible with the mixed command? Or, is it possible to estimate a multilevel model using the sem command to implement mlmv?

[Andrew Musau]

If multilevel model means a random effects model, then it is possible, but it will take some work. Full information maximum likelihood (FIML) is available in sem. gsem only allows maximum likelihood. It is as easy to estimate a random effects model in gsem as it is in mixed, but not so straightforward in sem. Among other things, you will need to have your data in wide format. This blog post by Vince Wiggins explains how to do it. If you are successful in setting up the data, use -method(mlmv)- in sem, and you have FIML.