Accounting for Data Collection Mode in Hot Deck Imputation

• Activity Number: 181 - Contributed Poster Presentations: Government Statistics Section
• Type: Contributed
• Date/Time: Monday, July 31, 2017 : 10:30 AM to 12:20 PM
• Sponsor: Government Statistics Section
• Author(s): Jiantong Wang* and Peter Frechtel and Amang Sukasih and David Kinyon
• Companies: RTI International and RTI International and RTI International and U.S. Energy Information Administration
• Keywords: item nonresponse ; missing data ; cyclical tree based hot deck ; multimode survey ; Residential Energy Consumption Survey

Abstract

Item missing values are common in survey data. Hot deck imputation is one popular method to address item missing values. Research Triangle Institute (RTI) developed the Cyclical Tree-Based Hot Deck (CTBHD) imputation technique that implements a regression tree to construct imputation cells and uses weighted sequential hot deck imputation to select the donor. It is possible to account for multi-mode used in data collection during the imputation. RTI implemented CTBHD in the 2015 Residential Energy Consumption Survey (RECS) data, where the data collection was conducted using different modes: Computer Assisted Personal Interviewing (CAPI), internet (Computer-Assisted Web interviewing, or CAWI), and mail (Paper and Pencil Interviewing, or PAPI). There were more than half of the 2015 RECS cases responded in CAWI/PAPI. The response patterns might be different across these modes. In this paper, we investigated how the response patterns are different, and whether it is necessary to impute nonresponses by modes to account for potential mode effects; that is, whether to use mode as an additional class variable in the CTBHD.