Data are MCAR when there are no shared effects linking missingness on one variable and on another and when missingness on all variables is random. The wikipedia page about missing data, says:
Values in a data set are missing completely at random (MCAR) if the events that lead to any particular data-item being missing are independent both of observable variables and of unobservable parameters of interest, and occur entirely at random.[5] When data are MCAR, the analysis performed on the data is unbiased; however, data are rarely MCAR.
https://en.wikipedia.org/wiki/Missing_data#Missing_completely_at_random
That is making the crucial point that to be truly MCAR the missingness must be random not only with respect to measured variables but to “unobservable” ones too. That’s logical but essentially untestable making, to my mind, tests of whether data fit the MCAR state of rather limited use. If one thinks about factors likely to cause data to be missing across routine change/outcome data then MCAR is incredibly unlikely to apply!
Try also … #
Missing at random (MAR)
Missing values
MICE (Multiple Imputation by Chained Equations)
Chapters #
Not mentioned in the OMbook.
Online resources #
I’m unlikely to create any I think, the issues are too complex and too general to make it easy.
Dates #
Created on or before 13/6/21, last updated 12/9/24.
