Efficiency of Complete Case Statistics in Semiparametric Regression Models with Responses Missing at Random

• Activity Number: 590
• Type: Invited
• Date/Time: Thursday, August 2, 2012 : 8:30 AM to 10:20 AM
• Sponsor: Section on Nonparametric Statistics
• Author(s): Anton Schick*+
• Companies: Binghamton University
• Keywords: Transfer principle ; efficient estimation ; semiparametric regression ; missing at random

Abstract

It is well known that a statistical analysis which ignores cases that are only partially observed does not always perform well, and that an approach which imputes missing values often has better properties. There are, however, situations where a complete case analysis is appropriate, and sometimes even asymptotically optimal.

We give a general method, the transfer principle, for obtaining limiting distributions of complete case statistics for missing data models from those of the corresponding statistics for the full model. This provides a convenient tool to obtain the asymptotic behavior of complete case versions of established methods without (reproducing) lengthy proofs.

We then use the transfer principle to show that complete case versions of efficient estimators of the finite-dimensional parameter in semiparametric regression models remain efficient in the model when responses are missing at random.

This talk is based on joint work with Uschi Mueller and Hira Koul.