Multiple imputation for missing data

Abstract

Missing data occur frequently in survey and longitudinal research. Incomplete data are problematic, particularly in the presence of substantial absent information or systematic nonresponse patterns. Listwise deletion and mean imputation are the most common techniques to reconcile missing data. However, more recent techniques may improve parameter estimates, standard errors, and test statistics. The purpose of this article is to review the problems associated with missing data, options for handling missing data, and recent multiple imputation methods. It informs researchers' decisions about whether to delete or impute missing responses and the method best suited to doing so. An empirical investigation of AIDS care data outcomes illustrates the process of multiple imputation.