Caitlin F posted on Friday, February 24, 2012 - 12:39 pm
We are working with nationally representative data that uses a complex survey design. We are specifying TYPE=COMPLEX with MLR, using a cluster variable, stratification, and a weight. Everyone has a weight variable, that is not an issue. Our issue is that we have two levels of missing data.
First, there are participants with legitimate skips that we do not want to use in our sample. We do not want to use the EM Algorithm for them. We are just including those participants in the dataset because they have a valid weight.
Second, there are participants that have missing data that ARE part of our sample. We do want to use the EM Algorithm for their answers. For example, if a participant does not report on an item, but could have, we still want to use their data.
So can we ask MPlus to run EM Algorithm with just a subsample of people who have weights - without deleting them?