How do you account for multilevel structure when imputing missing values?

[ejamro]

I have a multilevel dataset (individuals within counties) that has missing values. I am ultimately creating a multinomial, multilevel logistic regression model with these data.

I am currently using proc mi to ascribe the missing values, then proc glimmix to produce effect estimates, and finally proc mianalyze to pool the estimates from the n=8 imputed datasets. However, I recognize that this fails to account for the clustering in the imputation model. As per Ludtke 2017, "Multiple imputation is a widely recommended means of addressing the problem of missing data in psychological research. An often-neglected requirement of this approach is that the imputation model used to generate the imputed values must be at least as general as the analysis model. For multilevel designs in which lower level units (e.g., students) are nested within higher level units (e.g., classrooms), this means that the multilevel structure must be taken into account in the imputation model."

Is there any way to incorporate the multilevel structure into the multiple imputation procedure in SAS? 

[sbxkoenk]

Hello @ejamro ,

I haven't used PROC MI for 6 or 7 years now, but at the time

, there was no specific adjustment in Proc MI for an HLM type model (multi-level model).  The problem is that there is no widely accepted approach to deal with an HLM model and multiple imputation (afaik).  

You may open a track / ticket with SAS Technical Support.
Maybe they can tell you more.
Maybe something has been foreseen meanwhile to do this.

Thanks,

Koen