Recent advances in semiparametric estimation under nonignorable nonresponse

 Research Interests:

Survey Sampling, Statistical Analysis with missing data, Measurement error models, Multi-level models, Causal Inference, Statistical Computing, Bayesian inference, Prediction with big data, etc.

Work Experience

1.    Academic Positions: Hankook University of Foreign Studies (2002-2003), Yonsei University (2004-2008), Iowa State University (2008-current), KAIST (2016-current).

2.    Other job experience: Mathematical Statistician at US Census Bureau (1999-2000), Senior Statistician at Westat (2000-2001) 

Education

§  1994-2000: Department of Statistics, Iowa State University (PhD)

§  1991-1993: Department of Statistics, Seoul National University (MS)

§  1987-1990: Department of Computer Science and Statistics, Seoul National University (BS)

Semiparametric estimation with nonignorable nonresponse data is an important but notoriously difficult problem in statistics. We first consider a parametric nonresponse model and discuss a new identifiability condition without relying on an instrumental variable assumption. After that, a very general result for semiparametric optimal estimation under nonignorable nonresponse is presented and then two adaptive estimators for semiparametric optimal estimation are proposed. Because the proposed estimators achieve the semiparametric optimality, their statistical performances are better than  the existing methods, which was demonstrated in the simulation study and in the real data analysis.  Furthermore, some extensions to semiparametric nonresponse models are also discussed. This is a summary of two papers, one with Kosuke Morikawa at Osaka University and the other is with Masatoshi Uehara at Harvard University