Data Analysis with R, Second Edition - Second Edition

So which one is it?

You may have noticed that the place of a particular dataset in the missing data mechanism taxonomy is dependent on the variables that it includes. For example, we know that the mechanism behind qsec is MAR, but if the dataset did not include am, it would be MNAR. Since we are the ones that created the data, we know the procedure that resulted in qsec's missing values. If we weren't the ones creating the data—as happens in the real world—and the dataset did not contain the am column, we would just see a bunch of arbitrarily missing qsec values. This might lead us to believe that the data is MCAR. It isn't, though; just because the variable to which another variable's missingness is systematically related is non-observed ...