Multiple imputation has become viewed as a general solution tomissing data problems in statistics. However, in order to lead toconsistent asymptotically normal estimators, correct varianceestimators and valid tests, the imputations must be proper.So far it seems that only Bayesian multiple imputation, i.e.\using a Bayesian predictive distribution to generate theimputations, or approximately Bayesian multiple imputations hasbeen shown to lead to proper imputations in some settings. In thispaper, we shall see that Bayesian multiple imputation does notgenerally lead to proper multiple imputations. Furthermore, itwill be argued that for general statistical use, Bayesian multipleimputation is inefficient even when it is proper.
"Proper and Improper Multiple Imputation." Internat. Statist. Rev. 71 (3) 593 - 607, December 2003.