ABSTRACT
If missing data arise from an MCAR or MAR mechanism, a likelihood-based analysis can yield unbiased estimates and valid inferences from the available data such that the missing data can be ignored. With GEE and least squares estimation, for the missingness to be ignorable, it must arise from an MCAR mechanism. Therefore, the plausibility of ignorable missing data is greater with likelihood-based analyses than with GEE or least squares (Verbeke and Molenberghs 2000).
