I'm currently working with the MICE algorithm to impute missing data.
After I did the imputation I wanted to do some kind of quality check of the imputed data set.
There are some suggestions here https://www.r-bloggers.com/imputing-missing-data-with-r-mice-package/ where I can plot the original data and the imputed data to compare them. For example compare their densities.
Now I'm wondering: Since MICE assumes MAR, meaning that the original data and the missing data have the same distribution, isn't it obvious that the density is almost the same? So why even performing that "test"?
What I was thinking instead is:
Remove some data points from the original data. Then perform MICE and them compare the imputed datas with the original one. That makes sense to me, but why isn't standard procedure? Sure if I do that, my imputation model will be not as good as if I don't remove the data, since less data -> less information.
However,I just can't find any information about that in the internet.
I would really appreciate it, if some of you can give me some thought or advices.