Comparison of Shrinkage–Based Estimators in the Presence of Missing Data: A Multiple Imputation Analysis

摘要

In this paper we examined the performance of the mean square error of the Ordinary Least Square (OLS) estimator, Minimum Mean Square Error (MMSE) estimator, N/N shrinkage Estimator (N/NSE) and a proposed Adjusted Minimum Mean Square Error (PAMMSE) estimator in a multiple imputation analysis when data points are missing in different data sets. The program for the proposed adjusted minimum mean square error was written and implemented in R. It is shown by numerical computations that the PAMMSE Estimator seem to be the best choice among OLS, MMSE, N/NSE and PAMMSE estimators in terms of their mean square errors when applied in multiple imputation analysis.