Asymptotic results for multiple imputation (Q1118276): Difference between revisions

For analyzing incomplete data in surveys, \textit{D. B. Rubin} [ibid. 6, 34- 58 (1978; Zbl 0383.62021)] proposed multiple-imputation where the missing data are replaced by two or more values representing a distribution of the possible values. These studies assume that the outcome variable Y is a scalar, sample selection is by simple random sampling with no covariates and there is ignorable nonresponse. In this paper, the authors assume that the scalar outcome variable Y follows a linear model involving certain covariates X and that there is ignorable nonresponse. The asymptotic sampling distributions of the multiple-imputation estimators are obtained using an ordinary least- squares method for the complete data. Some applications of the results and extensions to alternative estimation procedures are also given.