ABSTRACT
Analyses in the MNAR framework try in some manner to model or otherwise take into account the missingness process and its impact on outcomes of interest. However, moving beyond MAR to MNAR poses fundamental problems. In MAR, it is assumed that the statistical behavior of the unobserved data is the same as it had been observed, such that the unobserved data can be predicted from the observed data. Of course, the characteristics and statistical behavior of the missing data are unknown (Mallinckrodt 2013). Therefore, it is impossible to know if MAR is valid and it is impossible to know if any specific MNAR model is correct.
