Empirical likelihood-based hot deck imputation methods

The non-response problem commonly exists in survey data and has been investigated by various methods. We propose an empirical likelihood-based hot deck imputation method, which resamples the observed data by using the weights from the empirical likelihood ratio for missing values. We demonstrate that the estimator of the mean is unbiased and the corresponding variance estimator of the mean is asymptotically unbiased under mild conditions. Next, we extend our method for U-statistic estimators and show that the estimator converges to the real U-statistic in probability. The proposed method can also incorporate multiple imputations and/or regression imputations easily. Simulations and a real example illustrate that our method outperforms some of the existing approaches, such as simple hot deck imputation and fractional hot deck imputation. We conclude with a discussion of the advantages of the empirical likelihood-based hot deck imputation method.