A0653
Title: Imputation with verifiable identification condition for nonignorable missing outcomes
Authors: Kenji Beppu - Osaka University (Japan) [presenting]
Kosuke Morikawa - Osaka University and The University of Tokyo (Japan)
Jongho Im - Yonsei University (Korea, South)
Abstract: Missing data often cause undesirable results such as bias and loss of efficiency. These results become more substantial problems when the response mechanism is nonignorable such that the response model depends on the unobserved variable. It is often required to estimate the joint distribution of the unobserved variable and response indicator to handle nonignorable nonresponse. However, model misspecification and identification issues prevent obtaining robust estimates even if we carefully estimate the target joint distribution. We model the distribution for the observed parts and derive sufficient conditions for the model identifiability, assuming a logistic distribution on the response mechanism and a generalized linear model as the main outcome model of interest. More importantly, the derived sufficient conditions are testable with the observed data and do not require any instrumental variables, which have been often assumed to guarantee the model identifiability but cannot be practically determined in advance. To analyze missing data, we propose a new fractional imputation method which incorporates the verifiable identifiability using observed data only. Furthermore, we present the performance of the proposed estimators in numerical studies and apply the proposed method to two sets of real data, the Opinion Poll for 2022 South Korean Presidential Election and public data collected from the US National Supported Work Evaluation Study.
