The Language of Missing Data Mechanisms

Missing completely random is actually what the name implies. It's an omnibus test of whether there are patterns of missing this on what you observed as a function of whether you were in the complete case of data or the partially missing case of data groups. I'm less enamored with it than modern approaches to handling missing data that establish whether their MCAR isn't relevant. And indeed, if somebody is doing really good due diligence in a paper and presents little test of MCAR, I'll often recommend they drop it because I don't find it useful.