ABSTRACT
As discussed in other chapters in this book, when the mechanism governing missingness is that of missing at random (MAR), three main approaches to inference on parameters describing the complete data using the observed data have been advocated. Broadly speaking, these may be characterized as likelihood methods, imputation methods, and methods involving the use of so-called inverse probability weighting. Likelihood, and imputation methods generally are predicated on the assumption of a parametric statistical model for the complete data; that is, the class of probability densities that is believed to contain the true density generating the data may be described using a finite number of parameters. One of the advantages of these approaches is that estimators for the model parameters may be derived without the need to model explicitly the missing data process.
