I'm doing some imputation with the MICE package. The outcome variables I am using are zero-inflated, and in the absence of imputation, I would analyze them with a zero-inflated negative binomial regression (ZINB). I would like to do the same with the imputed data; however, I have to manually combine parameters since MICE or its add-ons (as far as I know), do not generate estimates for ZINB.

My question is whether Rubin's Rule for combining regression parameters (which, to my knowledge, is simply the mean) would apply to regression parameters for ZINB. I believe that regular linear multiple regression parameters can be combined in such a way because they are approximately normally distributed. Do regression parameters for ZINB (and for that matter, Poisson and logistic regression) approximate a normal distribution in the same way? In other words, is the Rubin's Rule for combining ZINB and other generalized linear regression parameters the same as combining regression parameters from a linear multiple regression?

Thank you!

  • $\begingroup$ Are these parameters estimated using MLE? If so they should in general have an asymptotic normal distribution and Rubin's rules should be valid. $\endgroup$ – David Luke Thiessen 11 hours ago

Your Answer

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

Browse other questions tagged or ask your own question.