The BGLIMM Procedure
Models with Missing Values
Missing response values are treated as parameters by default and sampled in the MCMC simulation. This mechanism of modeling missing values is referred to as missing at random (MAR). You can delete all observations that contain missing values by using the MISSING=CC option in the PROC BGLIMM statement.
Suppose that
The response variable 
consists of 
observed values, 
, and 
missing values, 
. At each iteration, PROC BGLIMM samples every missing response value (by using the likelihood function as the sampling distribution). After these samples are drawn, the GLMM is reduced to a full data scenario with no missing data. PROC BGLIMM then proceeds to update 
, 
, 
, and 
sequentially, in the same way as described in the section How PROC BGLIMM Works.