Pharmaceutical Statistics Using SAS

12.7. MNAR Modeling

Even though the assumption of likelihood ignorability encompasses the MAR and not only the more stringent and often implausible MCAR mechanisms, it is difficult to exclude the option of a more general MNAR mechanism. One solution for continuous outcomes is to fit an MNAR model as proposed by Diggle and Kenward (1994). In the discrete case, Molenberghs, Kenward, and Lesaffre (1997) considered a global odds ratio (Dale) model.

However, as pointed out in the introduction and by several authors (discussion to Diggle and Kenward, 1994; Verbeke and Molenberghs 2000, Chapter 18), we must be extremely careful with interpreting evidence for or against MNAR in a selection model context, especially in large studies with a lot of power. ...