|
Activity Number:
|
151
|
|
Type:
|
Contributed
|
|
Date/Time:
|
Monday, July 30, 2007 : 10:30 AM to 12:20 PM
|
|
Sponsor:
|
Section on Statistics in Epidemiology
|
| Abstract - #309075 |
|
Title:
|
Nonparametric Regression with Missing Data Using Kernel-Estimating Equations
|
|
Author(s):
|
Lu Wang*+ and Xihong Lin and Andrea Rotnitzky
|
|
Companies:
|
Harvard University and Harvard School of Public Health and Universidad Torcuato Di Tella
|
|
Address:
|
655 Huntington Ave, SPH2, 4th Floor, Boston, MA, 02115,
|
|
Keywords:
|
nonparametric mean model ; two-stage study ; inverse probability weighted kernel GEE ; augmented estimating equations ; double robustness ; asymptotic efficiency
|
|
Abstract:
|
In epidemiology studies when we do not know the functional pattern of how the mean outcome depends on a covariate, nonparametric regression becomes attractive. However, if nonresponse occurs nonrandomly (e.g., in stratified two-stage designs), naive approach using complete cases is usually biased. We propose inverse probability weighted kernel generalized estimating equations (GEE) and a class of augmented inverse probability weighted (AIPW) kernel GEE to correct for dependent censoring and nonrandom missing. Both approaches are asymptotically unbiased and do not require full likelihood specification. Further, the double robustness of AIPW kernel estimator provides us two chances to achieve valid inferences and it has potentials to gain efficiency. Simulations are done to evaluate practical performances, followed by an application on the AIDS Costs and Services Utilization Survey data.
|
