I don't understand yet,
1. can i use inverse probability weighting for non-ignorable missing data?
2. Is MI a likelihood based model or imputation based?
3. what is the difference between MI and (selection models or pattern-mixture models or shared-parameter models)