Missing data on predictors in a mixed model

[GiaLee]

I just noticed that "When it comes to missing data, the big limitation of PROC MIXED and PROC GLIMMIX is that they do nothing about missing data on predictor variables. In particular, cases with valid observations on the dependent variable but missing values on one or predictor variables contain potentially valuable information that will be completely lost."

(https://statisticalhorizons.com/wp-content/uploads/MissingDataByML.pdf)

I was wondering, could I first use multiple imputation to impute the missing predictors and covariates, and then allow the mixed linear model to handle the missing outcome using maximum likelihood estimation?

[PaigeMiller · Accepted solution]

Yes, you can do that.

If appropriate, you could use PROC PLS with option MISSING=EM, which uses the Expectation Maximization algorithm to fit a model with imputed values (which means you don't have to program the missing value imputation).

https://documentation.sas.com/doc/en/pgmsascdc/9.4_3.4/statug/statug_pls_syntax01.htm#statug.pls.pls...

[GiaLee]

Thank you!!