Thanks @martinmodrak and @avehtari for the responses.
My model formulation is actually a bit more complex than the example I gave here, and relates to this question Missing data imputation for outcome missing for all rows for certain timepoints - #5 by jsocolar
I am imputing missing outcomes so that I can keep those rows of data which will help me better impute missings for those predictors.
Just to be clear - you used that same argument as
newdatawhen computing theloofor both models?
Yes, that is correct.
If so, then I think it might be mostly OK
Hmm…that’s not super reassuring haha. So is “integrating out the parameter representing the imputed value” the only solution for comparing models via ‘information criterion’ when I impute missing data? If I don’t do that (honestly not sure how), then is using the complete case dataset in the newdata argument the best I can do?