Hot Deck Imputation of Nonignorable Missing Data with Sensitivity Analysis

• Activity Number: 499
• Type: Contributed
• Date/Time: Wednesday, August 7, 2013 : 8:30 AM to 10:20 AM
• Sponsor: Survey Research Methods Section
• Author(s): Danielle Sullivan*+ and Rebecca Roberts Andridge
• Companies: The Ohio State University and The Ohio State University College of Public Health
• Keywords: hot deck ; nonignorable nonresponse ; sensitivity analysis

Abstract

Hot deck (HD) imputation is a common method for handling item nonresponse in surveys, but most implementations assume data are missing at random. We combine the Approximate Bayesian Bootstrap (ABB) distance-based donor selection method of Siddique and Belin (2008) with the Proxy Pattern-Mixture (PPM) model (Andridge and Little 2011). The PPM model defines distances between donors and donees under different missingness assumptions, creating a proxy HD with distance-based donor selection to perform imputation, with an intuitive sensitivity analysis (SA). Missingness in the outcome is assumed to be a linear function of the outcome and the proxy variable, estimated from a regression analysis of respondent data. The SA allows for simple comparisons between ignorable and varying levels of nonignorable missingness. The PPM HD provides a more concise SA than using the more than 6 various `shaped' ABBs of Siddique and Belin. Compared to the parametric PPM model, the PPM HD is potentially less sensitive to model misspecification. We compare the bias and coverage of estimates from the PPM HD with the ABB HD through simulations and apply the method to data from the Ohio Family Health Survey.