Background: So I have a dataset with 32 Variables. There are several missing values so I decided to impute them with the mice package in R. However, I plan to combine most of those variables through factor analysis, so that I am only left with 12 variables in my final analysis (with which I will conduct a multilevel analysis).

Question: should I combine the variables before imputation and then impute with my smaller 12 Variables or with the "raw" dataset of 32 variables?

I know you can do passive imputation, where you can include combined items, but I can hardly just use the factor scores from the unimputed dataset to be used in the passive imputation process like this (meth["Index1"] <- "~I(x1*0.789+x2*0.678+x3*0.786)", or can I?

Because here is another problem I have: I also read that you should include interaction terms, but my interaction terms will be between those indices. So I would need to have my indices already created in order to passively impute the interaction terms, or am I overthinking this?

I have read some similar questions here, but nothing that would really answer my specific question (1) (2) (3).

If I can help you out with further information, please let me know!



Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.