Hello; I am trying to make sure the results of an analysis I ran are robust to the missing data properties. The missingness is either MAR or NMAR.
There is 1 exogenous predictor variable and 4 DVs.
The 4 DVs are also count data (negative binomial) which the AUXILIARY = (M) command says cannot be implemented with count data.
Is there an acceptable way to do this? I have implemented the MLR estimator to account for the data distribution with the AUXILLARY command with the variables coded for missingness but am unsure if there is a better method.
The missingness is most likely related to the response variables and is also a function of the predictor variable (more missingness in 1 group than the other)
It sounds like you want Aux=(M) done manually for counts. I don't know of an easy way to do that because we don't have WITH for counts. You say that the DV values might predict missingness on them. But I don't know how to do Diggle-Kenward type missing data modeling in this situation. The best you can do is probably stay with MLR, that is, act as if MAR holds.
Thank you. I noticed the WITH statement wasn't allowed as well with counts. So to confirm, stick with the MLR and use the AUXILLARY = (M) code or were you suggesting dropping the auxillary part as well?