TY - JOUR
T1 - Robust estimation for moment condition models with data missing not at random
AU - Li, Wei
AU - Yang, Shu
AU - Han, Peisong
N1 - Funding Information:
Dr. Yang is partially supported by NSFDMS 1811245 and NCIP01 CA142538.
Funding Information:
Dr. Yang is partially supported by NSF DMS 1811245 and NCI P01 CA142538 .
Publisher Copyright:
© 2020 Elsevier B.V.
PY - 2020/7
Y1 - 2020/7
N2 - We consider estimation for parameters defined through moment conditions when data are missing not at random. The missingness mechanism cannot be determined from the data alone, and inference under missingness not at random may be sensitive to unverifiable assumptions about the missingness mechanism. To add protection against model misspecification, we posit multiple models for the response probability and propose a weighting estimator with calibrated weights. Assuming the conditional distribution of the outcome given covariates is correctly modeled, we show that if any one of the multiple models for the response probability is correctly specified, the proposed estimator is consistent for the true value. A simulation study confirms that our estimator has multiple robustness when the outcome data is missing not at random. The method is also applied to an application.
AB - We consider estimation for parameters defined through moment conditions when data are missing not at random. The missingness mechanism cannot be determined from the data alone, and inference under missingness not at random may be sensitive to unverifiable assumptions about the missingness mechanism. To add protection against model misspecification, we posit multiple models for the response probability and propose a weighting estimator with calibrated weights. Assuming the conditional distribution of the outcome given covariates is correctly modeled, we show that if any one of the multiple models for the response probability is correctly specified, the proposed estimator is consistent for the true value. A simulation study confirms that our estimator has multiple robustness when the outcome data is missing not at random. The method is also applied to an application.
KW - Empirical likelihood
KW - Identification
KW - Missing not at random
KW - Multiple robustness
KW - Semiparametric maximum likelihood estimator
UR - http://www.scopus.com/inward/record.url?scp=85078014307&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=85078014307&partnerID=8YFLogxK
U2 - 10.1016/j.jspi.2020.01.001
DO - 10.1016/j.jspi.2020.01.001
M3 - Article
AN - SCOPUS:85078014307
SN - 0378-3758
VL - 207
SP - 246
EP - 254
JO - Journal of Statistical Planning and Inference
JF - Journal of Statistical Planning and Inference
ER -