|
Activity Number:
|
450
|
|
Type:
|
Contributed
|
|
Date/Time:
|
Thursday, August 7, 2003 : 8:30 AM to 10:20 AM
|
|
Sponsor:
|
Biometrics Section
|
| Abstract - #300997 |
|
Title:
|
Doubly Robust Estimation in Monotone Missing Data and Causal Inference Models
|
|
Author(s):
|
Heejung Bang*+ and James M. Robins
|
|
Companies:
|
University of North Carolina, Chapel Hill and Harvard School of Public Health
|
|
Address:
|
2701 Homestead Rd., Chapel Hill, NC, 27516,
|
|
Keywords:
|
causal inference ; doubly robust estimation ; longitudinal data ; marginal structural model ; missing data ; semiparametrics
|
|
Abstract:
|
The goal of this paper is to construct doubly robust augmented inverse probability of treatment weighted estimators in ignorable longitudinal missing data and causal inference models. In missing data models, an estimator is doubly robust if it remains consistent and asymptotically normal as long as either the model for missingness mechanism or the model for the distribution of the complete data is correctly specified. In a causal inference model, an estimator is doubly robust if it remains consistent when either the model for the treatment assignment mechanism or the model for the remainder of the distribution of the observed data is correctly specified. The semiparametric variance bound is achieved when both models are correct and the distribution of the complete data or the marginal structural model is unrestricted. It is shown that the estimators that achieve this goal can be represented as sequential regression estimators. Various statistical models can be studied in a unified framework. Practical algorithms to obtain such estimators are provided. Simulation studies show that our estimators perform well with finite samples.
|
