I read an article today that stressed the need when data is missing more than two percent of cases to test whether it is MCAR MAR, MNAR.

Does anyone know how to do such a test. I understand MAR is tied to being associated with the predictors and the MNAR with Y. But not how to do a formal test of this.


Less is more. Stay pure. Stay poor.
My knowledge says there is no test for any of them. MAR can be examine with modelling, but the presence of a latent variable associated with informative missingness could always be possible.