The key problems in missing data models involve the identifiability of two distributions: the target law and the full law. The target law refers to the joint distribution of the data variables, while the full law refers to the joint distribution of both the data variables and the response indicators. It has not been clearly stated how identifiability of the target law and the full law relate to multiple imputation. We show that imputations can be drawn from the correct conditional distributions if only if the full law is identifiable. This result means that direct application of multiple imputation may not be the method of choice in cases where the target law is identifiable but the full law is not. In such cases, alternative imputation approaches sometimes enable estimation of the target law. For this purpose, we introduce decomposable multiple imputation.

翻译：暂无翻译