Um? - I only posted this question once, are you confusing this with the other question I posted earlier today about influence of estimator in MI? I couldnt find an earlier post by anyone else on this exact topic.
Jonathon - the difference in the scale should not be a problem at all. As you say there is no assumption of time invariance in the imputation model. The variables are treated as 9 different variables that help impute 30 categorical variables.
I want to impute data with 14 categorical variables and 1 continuous variable, and i declared variables as categorical by specifying a (c) after the respective categorical variables. So I specified this like this:
My questions are (1) are the imputations now done with MLR estimation / or is bayes used by default? (2) does this specification generate H1 imputations? (3) according to the output, the 'covariance' setting was applied with the imputations, but how is this possible since the imputation involves categorical data?
The imputations are based on Bayesian estimation using the H1 model. The H1 model estimates an unrestricted polychoric/polyserial correlation matrix. You can find more information here http://statmodel.com/download/Imputations7.pdf