To further explore whether they are some relationship between missing values of observed predictors, a new dataset has created by replacing missing values with 1 while non-missing with 0.


               AST        ALT        CD4    marriage_f
AST        1.00000000 0.77589893 0.22670360 0.07317342
ALT        0.77589893 1.00000000 0.29791660 0.09978341
CD4        0.22670360 0.29791660 1.00000000 0.07826039
marriage_f 0.07317342 0.09978341 0.07826039 1.00000000

Output show the missing values in AST variable are strongly correlated with missing values in ALT variable.

My question are the follows

  1. Is that belong to MAR or MNAR, of course I know it is not MCAR?
  2. How to handle such issue? Is that fine to use multiple imputation by mice() function?

1 Answer 1


If the fact of missingness can be predicted by another variable in your substantive model then you have either MAR or MNAR but unfortunately you cannot tell which. It may be that from your scientific or clinical knowledge of the liver you can tell whether the measurements were in fact related to the value you would have obtained but that is external to the problem.

Multiple imputation by chained equations should be fine.

  • $\begingroup$ To be specific, my case is the missingness can be predicted (or dependent) by the "missngness" rather than available values of another variable. A little subtle difference. But I believe it's difficult to tell MAR and MNAR apart as you mentioned. $\endgroup$
    – juanli
    Mar 21, 2017 at 20:13

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.