AN OVERVIEW OF MULTIPLE IMPUTATION

Donald B. Rubin
2002-01-01
Abstract:Multiple imputation for nonresponse in public-use files replaces each missing value by two or more plausible values. The values can be chosen to represent both uncertainty about which values to impute assuming the reasons for nonresponse are known and uncertainty about the reasons for nonresponse. The theoretical underpinnings and several examples are given in Rubin (1987). Thispresentation illustrates the dramatic improvements possible when using multiple rather than single imputation and provides a brief overview of current technology and lacunae that, hopefully, will be addressed and filled by current research efforts. The two important applications of multiple imputation that this overview introduces, demonstrate the substantial improvements that can accrue from the straightforward use of multiple imputation in practice.
What problem does this paper attempt to address?