Variance Estimation of Imputed Estimator In Complex Design

We have studied imputation and variance estimation method under the regression imputation model. We proposed the best linear unbiased predicted value as an imputed value under the model, and obtained unbiased variance estimator of the imputed sample mean. Basically, our method is an extension of Sarndal s(1992) model based approach to general regression model. Through a simulation study, we confirmed that the sample mean after the prediction imputation is nearly unbiased under nonignorable as well as ignorable response mechanism and the proposed variance estimator is more efficient than other known adjusted variance estimators. Furthermore, the proposed variance estimator is applicable regardless of ignorable or nonignorable response mechanism.