DATA IMPUTATION: AN OPTIMIZATION APPROACH.

Due to the frequent occurrence of missing data in sample surveys, the major techniques for dealing with missing data (primarily imputation methods) have been formulated by survey statisticians in consultation with scientists performing survey research. One approach to the missing value problem is to formulate it as an optimization problem. This formulation allows assumptions regarding the performance of the method to be developed in an explicit manner, which can enable 'optimal' methods with respect to user-defined criteria. The aim of this paper is to develop a general imputation formulation as an optimization problem and to demonstrate the characteristics of this formulation in a few simple examples.