Finite sample properties of multiple imputation estimators

Finite sample properties of multiple imputation estimators under the linear regression model are studied. The exact bias of the multiple imputation variance estimator is presented. A method of reducing the bias is presented and simulation is used to make comparisons. We also show that the suggested method can be used for a general class of linear estimators. 1. Introduction. Multiple imputation, proposed by Rubin (1978), is a procedure for handling missing data that allows the data analyst to use standard techniques of analysis designed for complete data, while providing a method to estimate the uncertainty due to the missing data. Repeated im-putations are drawn from the posterior predictive distribution of the missing values under the specified model given a suitable prior distribution. Schenker and Welsh [(1988), hereafter SW] studied the asymptotic properties of multiple imputation in the linear-model framework, where the scalar outcome variable Y i is assumed to follow the model