I have read this question, this question which refer to the imputation of multilevel level data using the 2l.pan or one of the other 2l methods in mice. However, my question is about using method "pmm" and setting the cluster variable in the predictorMatrix, like so:
pred1 [,"school"] <- -2
The method is "pmm" (default) for all continuous variables. The code seems to run without an issue. My question is:
Without the use of a specific 2l method, what is setting the cluster variable doing? How is it being used in the multiple imputation?
Since my analysis model is a single-level model with cluster-robust SEs, is it acceptable to use the "pmm" method or other methods in the imputation model instead of a multilevel imputation model?
The reason to prefer a single-level instead of a multilevel imputation model is that the multilevel imputation methods (e.g 2l.pan) only work with continuous variables whereas my data is ordinal (questionnaire data) and hence the multilevel imputation produces implausible values.