Two-phase sampling approach to fractional hot deck imputation

Abstract Hot deck imputation is popular for handling item nonresponse in survey sampling. In hot deck imputation, imputed values are taken from the respondents in the same imputation cell, where imputation cells are used to approximate the imputation model. We extend the fractional hot deck imputation of Kim and Fuller (2004) to the case where the imputation cells are not defined in advance. The proposed method of fractional hot deck imputation is performed in two steps and has a structure similar to that of two-phase systematic sampling. The proposed hot deck imputation method is applicable to multivariate missing data. A replication method is used for variance estimation. Results from two simulation studies are presented.