Practical Predictive Analytics

Not Missing at Random (NMAR)

A variable that is not MCAR and not MAR can be considered Not Missing at Random (NMAR). That basically means that the missing value is related to the values of the missing value itself, or is related to a variable that's not in the model. NMAR values often show up in survey research and clinical research where subjects are measured over time. For example, a survey respondent who reported severe depression at the beginning of a study may drop out at the end of the study, so if they scored high on a depression scale, that would be related to the fact they have missing scores at a later date, since they might be more inclined to drop out.

NMAR is the most difficult situation to detect, since a variable may seem to ...