<abstract xmlns="http://www.w3.org/1999/xhtml"><p>Nonresponse is very common in epidemiologic surveys and clinical trials. Common methods for dealing with missing data (e.g., complete-case analysis, ignorable-likelihood methods, and nonignorable modeling methods) rely on untestable assumptions. Nonresponse two-phase sampling (NTS), which takes a random sample of initial nonrespondents for follow-up data collection, provides a means to reduce nonresponse bias. However, traditional weighting methods to analyze data from NTS do not make full use of auxiliary variables. This article proposes a method called nonrespondent subsample multiple imputation (NSMI), where multiple imputation (<xref ref-type="bibr" rid="j_jos-2016-0039_ref_005_w2aab2b8c15b1b7b1ab1ab5Aa">Rubin 1987</xref>) is performed within the subsample of nonrespondents in Phase I using additional data collected in Phase II. The properties of the proposed methods by simulation are illustrated and the methods applied to a quality of life study. The simulation study shows that the gains from using the NTS scheme can be substantial, even if NTS sampling only collects data from a small proportion of the initial nonrespondents.</p></abstract>

Nonresponse is very common in epidemiologic surveys and clinical trials. Common methods for dealing with missing data (e.g., complete-case analysis, ignorable-likelihood methods, and nonignorable modeling methods) rely on untestable assumptions. Nonresponse two-phase sampling (NTS), which takes a random sample of initial nonrespondents for follow-up data collection, provides a means to reduce nonresponse bias. However, traditional weighting methods to analyze data from NTS do not make full use of auxiliary variables. This article proposes a method called nonrespondent subsample multiple imputation (NSMI), where multiple imputation (Rubin 1987) is performed within the subsample of nonrespondents in Phase I using additional data collected in Phase II. The properties of the proposed methods by simulation are illustrated and the methods applied to a quality of life study. The simulation study shows that the gains from using the NTS scheme can be substantial, even if NTS sampling only collects data from a small proportion of the initial nonrespondents.

Nonrespondent Subsample Multiple Imputation in Two-Phase Sampling for Nonresponse

Division of Biostatistics & Epidemiology, Cincinnati Children’s Hospital Medical Center, OH 45229,

Department of Epidemiology & Biostatistics, College of Public Health, University of South Florida, Tampa, FL 33612-3085,

Department of Biostatistics, School of Public Health, University of Michigan and Survey Research Center, Institute for Social Research, University of Michigan, Ann Arbor, MI 48019,

Journal of Official Statistics

This work is licensed under the Creative Commons Attribution-NonCommercial-NoDerivatives 3.0 License.

Nonresponse is very common in epidemiologic surveys and clinical trials. Common methods for dealing with missing data (e.g., complete-case analysis,...

Nonrespondent Subsample Multiple Imputation in Two-Phase Sampling for Nonresponse

Published Online: Sep 23, 2016

Page range: 769 - 785

Received: Feb 01, 2015

Accepted: Aug 01, 2015

DOI: https://doi.org/10.1515/jos-2016-0039

Keywords
Double sampling, maximum likelihood, missing data, nonignorable missing-data mechanism, quality of life, weighting

© 2016 Nanhua Zhang et al., published by De Gruyter Open

This work is licensed under the Creative Commons Attribution-NonCommercial-NoDerivatives 3.0 License.

Nonrespondent Subsample Multiple Imputation in Two-Phase Sampling for Nonresponse

Published Online: Sep 23, 2016

Page range: 769 - 785

Received: Feb 01, 2015

Accepted: Aug 01, 2015

DOI: https://doi.org/10.1515/jos-2016-0039

KeywordsDouble sampling, maximum likelihood, missing data, nonignorable missing-data mechanism, quality of life, weighting

© 2016 Nanhua Zhang et al., published by De Gruyter Open

This work is licensed under the Creative Commons Attribution-NonCommercial-NoDerivatives 3.0 License.

Keywords
Double sampling, maximum likelihood, missing data, nonignorable missing-data mechanism, quality of life, weighting