Comparison of Class Mean Imputation and Weighted Class Mean Imputation under Complex Survey Design

Unit non-response and item non-response are the two types of non-response in surveys. Imputation, a process in which values are assigned for missing responses, is typically employed to compensate for item missing data. Under complex survey design, weighted class mean is usually used as the imputed value. This paper compares three estimators related to class mean imputation and weighted class mean imputation under two models: equal expectation, equal and unequal variance within each class respectively. The conclusion is: For the equal variance model, the simple class mean estimator is the best, the class mean imputation estimator is the second and the weighted class mean imputation estimator is the worst. Under the specified unequal variance model, and the specified survey design, for large sample, the weighted class mean imputation estimator is better than class mean imputation estimator, which is better than the simple class mean estimator.