Our model is a logistic regression. We have been adviced that in our case we should do multiple imputation (mice for exemple). However, in this particular model we have some interaction terms (like 4 or 5) and I've heard that we cannot do classical MICE when there are interactions.

Can someone tell me why interactions are a problem for imputation? I'm not sure to understand where is the problem. What is the specific procedure that can be used then ?

Thanks for your answers.


You can impute interaction terms using MICE. You must first impute the components, and then use passive imputation to derive the interaction term directly from its components. This preserves the relationship between the interaction and its components for each observation.

For more detail on how to do this, see this source: https://www.jstatsoft.org/article/view/v045i03


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.