|
Activity Number:
|
152
|
|
Type:
|
Contributed
|
|
Date/Time:
|
Monday, August 7, 2006 : 10:30 AM to 12:20 PM
|
|
Sponsor:
|
Biometrics Section
|
| Abstract - #306440 |
|
Title:
|
Sieve Maximum Likelihood Estimation for Missing Covariates in Regression Models
|
|
Author(s):
|
Qingxia Chen*+ and Donglin Zeng and Joseph G. Ibrahim
|
|
Companies:
|
Vanderbilt University and The University of North Carolina at Chapel Hill and The University of North Carolina at Chapel Hill
|
|
Address:
|
1161 21st Ave., S., Nashville, TN, 37232,
|
|
Keywords:
|
missing covariates ; generalized linear model ; model misspecification ; semiparametric efficient ; b-spline ; profile likelihood
|
|
Abstract:
|
Misspecification of the covariate distribution in missing data problems is studied for regression models with several missing covariates. We propose a new semiparametric method which specifies a fully nonparametric model for the conditional distribution of the missing covariates given the completely observed covariates, assuming the missing covariates are missing at random (MAR). For ease of exposition, we first deal with the problem of one missing continuous covariate. To obtain the estimates, the log of the fully unspecified covariate joint density is approximated by B-spline functions and the estimates of the regression coefficients are obtained by maximizing a pseudo-likelihood function over a sieve space. Such estimators are shown to be consistent and asymptotically normal with their asymptotic covariance matrix achieving the semiparametric efficiency bound. Profile likelihood metho
|
