February 2007
Beginner to intermediate
464 pages
16h
English
In this section, we give an overview the most commonly used methods, which are valid under the MAR assumption. See Dmitrienko et al. (2005, Chapter 5) for details.
As stated earlier, likelihood-based inference is valid whenever the mechanism is MAR, and provided the technical condition holds that the parameters describing the nonresponse mechanism are distinct from the measurement model parameters (Little and Rubin, 1987). The log-likelihood then partitions into two functionally independent components, one describing the measurement model, the other one the missingness model. This implies that likelihood-based software with facilities to handle incomplete records provides valid inferences. ...
Read now
Unlock full access