Questions and Comments:
1. It was stated that data are MNAR when the likelihood of being missing is associated with the value of the variable. It seems to be incorrect. According to Rubinn (1987), it should be nonignorable missingness. MNAR should be the missing data that is related to the unobserved data. For example, suppose that the black people are less likely to respond to a question than the white people, but the racial information was not recorded. Then, the missing data is not missing at random.
2. The parameters would be biased when the MAR assumption does not hold. Therefore, I’m wondering wether the parameters are recovered in the study when the MNAR was simulated. If the parameters are not correctly estimated, the DIF detection may not be proper.