A hierarchical technique for estimating location parameter in the presence of missing data

Abstract

This paper proposes a hierarchical method for estimating the location parameters of a multivariate vector in the presence of missing data. At i th step of this procedure an estimate of the location parameters for non-missing components of the vector is based on combining the information in the subset of observations with the non-missing components with updated estimates of the location parameters from all subsets with even more missing components in an iterative fashion. If the variance-covariance matrix is known, then the resulting estimator is unbiased with the smallest variance provided missing data are ignorable. It is also shown that the resulting estimator based on consistent estimators of variance-covariance matrices obtains unbiasedness and the smallest variance asymptotically. This approach can also be extended to some cases of non-ignorable missing data. Applying the methodology to a data with random dropouts yields the well known Kaplan-Meier estimator.