An introduction to the three key missing data assumptions: Missing Completely at Random (MCAR), Missing at Random (MAR) and Missing Not at Random (MNAR).
Please you should definitely work on more videos. You are creating an impact with your teaching. Your approach is awesome and I love your voice. Thank you
First of all, thank you for existing. i really amazed by how calm and beautiful your voice is. second, please keep going T_T i really love your explanation. Instant subscribe!
how to assume whether data is MCAR or MAR or MNAR? is there any method of using it or we simply assume in mind at our own will....is there any method where we select this assumption??? or do we hypothesis it???
Apologies for missing this comment/question earlier. You can use Little's MCAR test to conduct a hypothesis test for the MCAR assumption. I now have a video on this topic in case it's still useful: ru-vid.com/video/%D0%B2%D0%B8%D0%B4%D0%B5%D0%BE-h9CzBtLpt_8.html You cannot run any tests to determine whether data are MAR or MNAR; you can only make assumptions based on what you think is plausible. You can additionally sensitivity analyses for the assumptions.
Wow!!! thank you SO SO MUCH! SO clear! In case NAs are missing due to the fact that participants didn't show up (for no specific reason, they simply didn't show up) on the test day, it would be a MCar case, right?
Hi Larissa, sorry for missing this comment/question earlier. I now have a video for a test for MCAR, which might be helpful for you: ru-vid.com/video/%D0%B2%D0%B8%D0%B4%D0%B5%D0%BE-h9CzBtLpt_8.html
How do I deal with MNAR? Can I assume MCAR even if Little's MCAR test is significant? The reason Little's MCAR is significant, I believe, is because a lot of data is missing.
Apologies for missing this comment/question earlier. I now have a video on Little's MCAR test, in case it's still useful: ru-vid.com/video/%D0%B2%D0%B8%D0%B4%D0%B5%D0%BE-h9CzBtLpt_8.html
Hiya! What should I do if my data is MNAR? My t tests from Little's MCAR test are showing significant results and I'm not unsure as to what to do with this data set. Please help
Hi Karolina, sorry for my delayed reply! If your missing data mechanism is not MCAR, what you could do is an analysis with multiple imputation under the MAR assumption and do sensitivity analysis to see how robust the results are to departures from the MAR assumption.
Great video! I was wondering for the MAR case, how do we know for sure that the two groups can be separated into outside/inside pieces? If we didn't know beforehand Mr. Pickles removed more outside pieces than inside pieces, in theory there could have been another unknown property responsible for the differing probabilities of missingness correct? Would this still be considered MAR? E.g. Mr. Pickles only removed pieces that had dirt on them (and they just happened to be mostly outside pieces). Thanks!
Thank you for your question. You can't be sure that the MAR assumption holds, so it's important to explore potential departures from the MAR assumptions and see what impact it has on results through sensitivity analyses (something I didn't explore much in the video)
Hi Mia, I enjoyed your video, well done. I was wondering when you might release the video on the statistical/sensitivity tests for different assumptions.
Thanks Amir. I now have a video on approaches to handling missing data: ru-vid.com/video/%D0%B2%D0%B8%D0%B4%D0%B5%D0%BE-ACN29i_fqkk.html I hope to make a video on exploring the validity of different assumptions soon.
Cute, but not helpful. The notation is poorly explained and leads to great confusion. Without explaining and demonstrating Rubin's difficult notation through clear and simple cases, everything else just goes to waste. For instance 'Missingness depends only on observed data' seems improperly defined or even like an instance of circular reasoning: Missingness means that the data is not observed, so which data you 'depend only on' itself depends on the missingness. That's not a clear definition. If you really want to help us, please focus on clear and thorough explanation, not on acting cute.