I'm dealing with stratified cluster sampling data. I would like to do multiple imputation first and then use weights in R. I looked at the survey-package for weighting and the mi- and mice-package for imputation but I don't know how to combine them.


Let me make a few points.

  1. No solid imputation method exist in the presence of sampling weights - a little more - it will NEVER exist - to construct such imputation you need to estimate a model for the weights - which is silly - because you can use ML without having to estimate such a model or make any model assumptions for the weights.

  2. My own research (unpublished) indicates you can ignore the weights during the imputation - no published example exist to show this otherwise.

  3. You want to at least take into account "stratified" and "cluster" designs for the imputation - that means you should do multiple group two-level imputation. Don't settle for single level imputation that assumes observations are independent across clusters.

  4. Once you have the imputations done then combining the results is trivial - the weights have no affect on the combining rules

  5. This is too hard - instead of imputation - use model estimation that can take into account the missing data


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.