Efficient estimation in a partially specified nonignorable propensity score model

Li, Mengyan; Zhao, Jiwei

doi:10.1016/j.csda.2021.107322

Cited by 3 publications

(2 citation statements)

References 27 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…To guarantee identifiability, several studies often assume the existence of a covariate, referred to as a non-response instrumental variable or shadow variable (Wang et al, 2014;Zhao and Shao, 2015;Miao and Tchetgen Tchetgen, 2016;Zhao and Ma, 2018;Li et al, 2021;Shetty et al, 2021). Although the instrumental variable is beneficial in several cases, the determination of its existence is not straightforward, and even in this case, its detection from the observed data is an elusive task.…”

Section: Introductionmentioning

confidence: 99%

“…Instead of modelling the complete data, numerous recently developed methods have modelled the distribution of the observed data, referred to as the respondents' outcome model (Kim and Yu, 2011;Riddles et al, 2016;Morikawa and Kim, 2021;Li et al, 2021;Shetty et al, 2021). This modelling is advantageous because the observed data are available, and consequently, we can select a better model for the candidates using information criteria based on the observed data, such as the Akaike information criterion (AIC) and the Bayesian information criterion (BIC), or other variable selection methods such as the adaptive lasso (Zou, 2006).…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Imputation with verifiable identification condition for nonignorable missing outcomes

Beppu¹,

Morikawa²,

Im³

2022

Preprint

View full text Add to dashboard Cite

Missing data often results in undesirable bias and loss of efficiency. These results become substantial problems when the response mechanism is nonignorable, such that the response model depends on the unobserved variable. It is often necessary to estimate the joint distribution of the unobserved variables and response indicators to further manage nonignorable nonresponse. However, model misspecification and identification issues prevent robust estimates, despite carefully estimating the target joint distribution. In this study we model the distribution of the observed parts and derived sufficient conditions for model identifiability, assuming a logistic distribution of 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 often been assumed to guarantee model identifiability but cannot be practically determined beforehand. To analyse missing data, we propose a new fractional imputation method which incorporates verifiable identifiability using only the observed data. Furthermore, we present the performance of the proposed estimators in numerical studies and apply the proposed method to two sets of real data, namely, Opinion Poll for the 2022 South Korean Presidential Election, and public data collected from the US National Supported Work Evaluation Study.

show abstract