Cyclical tree-based hot deck imputation

Abstract

Hot deck imputation is a method for filling in a missing value in a survey item (item nonrespondent) with a valid reported value from a donor (item respondent) within the survey. Our paper presents a multivariate hot deck imputation method called Cyclical Tree-Based Hot Deck (CTBHD). This method was developed to handle missing values in complex survey data with many different types of variables and allows the user to customize imputation classes, use sorting variables, impute vectors and compositional variables, and even edit or recode data “on-the-fly.” Additionally, CTBHD employs a cycling approach to get more stable imputed values with less bias and variance. Our paper evaluates the performance of CTBHD imputation through a simulation study using publicly available survey data from the 2020 Residential Energy Consumption Survey. Developed as a system for imputation, the CTBHD system is proprietary to RTI International.