Estimation with missing data: beyond double robustness

Han, Peisong; Wang, Lu

doi:10.1093/biomet/ass087

Cited by 141 publications

(175 citation statements)

References 28 publications

Supporting

Mentioning

171

Contrasting

Order By: Relevance

“…To provide additional protection, Han and Wang (2013) introduced the concept of multiple robustness; see also Han (2014aHan ( , 2014bHan ( , 2016aHan ( , 2016b, Chan and Yam (2014), Chen and Haziza (2017) and Duan and Yin (2017).…”

Section: Introductionmentioning

confidence: 99%

Multiply robust nonparametric multiple imputation for the treatment of missing data

Chen¹,

Haziza²

2019

STAT SINICA

View full text Add to dashboard Cite

Imputation is an effective way for handling missing values. We propose a nonparametric multiple imputation procedure that makes use of multiple outcome regression models and/or multiple propensity score models. Our procedure leads to a multiply robust point estimator in the sense that it remains consistent if all the models but one are misspecified. A variance estimate is readily obtained by applying the customary rule advocated by Rubin (1987). The asymptotic properties of the proposed method are established. Results from a simulation study, assessing the proposed method in terms of bias, efficiency and coverage probability, support our findings.

show abstract

Section: Introductionmentioning

confidence: 99%

Multiply robust nonparametric multiple imputation for the treatment of missing data

Chen¹,

Haziza²

2019

STAT SINICA

View full text Add to dashboard Cite

show abstract

“…Recently, many estimators of the form n i=1 R i w i Y i have been proposed where the w i are derived by optimizing an objective function, such as the empirical likelihood (Tan 2006(Tan , 2010Qin and Zhang 2007;Chen, Leung and Qin 2008;Kim 2009;Han and Wang 2013;Chan and Yam 2014), the exponential tilting (Kim 2010) or some generalizations of them (Tan and Wu 2015), subject to certain constraints on w i . Different objective functions and/or sets of constraints lead to different w i , for which the two most common forms are Because of this difficulty, we propose to circumvent the optimization and directly derive the calibration on π(α; X) by solving certain empirical equations.…”

Section: Notation and Some Existing Methodsmentioning

confidence: 99%

“…Estimators in Rubin and van der Laan (2008), Tan (2008;, Cao, Tstiatis and Davidian (2009) and Rotnitzky et al (2012) have improved efficiency: with a correctly specified propensity score model, each of these estimators is asymptotically equivalent to the most efficient AIPW estimator among a class of AIPW estimators for which the data distribution parameters in the augmentation terms are fixed but arbitrary. Estimators in Han and Wang (2013), Chan and Yam (2014), Han (2014aHan ( , 2014bHan ( , 2016 and Chen and Haziza (2017) are multiply robust:…”

Section: Introductionmentioning

confidence: 99%

“…Estimators in Tan (2006;, Qin and Zhang (2007), Kim (2009;, Chan (2012), Han and Wang (2013), Chan and Yam (2014) and Tan and Wu (2015) are convex combinations of the observed outcomes, and thus always fall within the range of observed values, known as sample boundedness property (Robins et al 2007). …”

Section: Introductionmentioning

confidence: 99%

“…Nice properties of these estimators mainly rely on delicate construction of augmentation terms and/or careful estimation of data distribution parameters. In recent literature, many researchers proposed to use weight derived by modifying the raw propensity score (Tan 2006(Tan , 2010Qin and Zhang 2007;Chen, Leung and Qin 2008;Qin, Shao and Zhang 2008;Kim 2009Kim , 2010Chan 2012;Han and Wang 2013;Chan and Yam 2014;Han 2014aHan , 2014bTan and Wu 2015;Han 2016).…”

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

A Further Study of Propensity Score Calibration in Missing Data Analysis

Han¹

2018

STAT SINICA

View full text Add to dashboard Cite

Methods for propensity score (PS) calibration are commonly used in missing data analysis. Most of them are derived based on constrained optimizations where the form of calibration is dictated by the objective function being optimized and the calibration variables used in the constraints. Considerable efforts on pairing an appropriate objective function with the calibration constraints are usually needed to achieve certain efficiency and robustness properties for the final estimators. We consider an alternative approach where the calibration is carried out by solving the empirical version of certain moment equalities. This approach frees us from constructing a particular objective function. Based on this approach, under the setting of estimating the mean of a response, we establish intrinsic, improved and local efficiency and multiple robustness in the presence of multiple data distribution models. A revisit to the generalized pseudo exponential tilting estimator and generalized pseudo empirical likelihood estimator of Tan and Wu (2015) is also provided.

show abstract