My data consists of measurements on patients with cancer and the variables are some indicators regarding the cancer as well as the stage and the grade of the cancer and some personal info about the patients. According to the doctors the missing values occured either because the part of the tissue they took from the patients was destroyed or because the doctor who did the biopsy didn't take a tissue sample big enough to estimate the stage of the cancer. Taking all these into account I have no reason to believe that the variables are MNAR and so two options remain, MAR or MCAR.

To my question now, let's say the MCAR is true (I read in another question that I can check this by doing a t-test, if someone knows another way please let me know) an easy way to deal with the missing values is to follow the complete-case method since the results will be unbiased, but my problem is that my data isn't big enough and if I delete cases I am afraid that the results probably won't be efficient. That's why I plan to do the MI but I'd like to know others opinion on that.


There is no way to test the MCAR or MAR assumptions fully, because the data you need to do the test are missing. You can, at best, test the differences between the group where some variables are missing and that with complete data on variables where you have complete data.

That said, if MAR or MCAR is reasonable, then MI is probably a good option. If the amount of data missing is very small, then it probably won't make much difference, and might not be worth the added complexity of MI.

  • 3
    $\begingroup$ Unfortunately the amount of missing data can't be considered as small, so i guess I will proceed to the MI. Thanks $\endgroup$ – Nick Mar 11 '12 at 10:19

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

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