Talk:Missing completely at random
The contents of the Missing completely at random page were merged into Missing data on 12 May 2014 and it now redirects there. For the contribution history and old versions of the merged article please see its history. |
This redirect does not require a rating on Wikipedia's content assessment scale. It is of interest to the following WikiProjects: | ||||||||||||||||||||||||||||
|
I think this needs a rewrite
editThis article is in need of a more thorough treatment. It is missing formal mathematical definitions of the terms as well as standard implications of the definitions.
I'm going to remove the last sentence of the article for now. It is absolutely NOT advisable to be throwing out the items with incomplete responses in MNAR data prior to running an analysis. This produces asymptotically biased answers (see e.g. this book [1])! In general, you can only throw out missing data when it is MCAR, although this is a bad idea if you are a fan of being efficient in using your data. — Preceding unsigned comment added by 24.136.36.175 (talk) 19:01, 5 January 2012 (UTC)
I am not happy with the examples
editThe one for MAR (accidentally omitting an answer on a questionnaire) looks more like MCAR than MAR to me. The one for NMAR looks more like MAR! Shouldn't NMAR be an answer omitted because of the answer itself (like very depressed people skipping the depression item)? If it's missing due to some other (measured?) characteristics, then it seems like MAR.
I'm not a missing data expert, so I'm just throwing this out to the readership.