|
Activity Number:
|
409
|
|
Type:
|
Topic Contributed
|
|
Date/Time:
|
Wednesday, August 5, 2009 : 8:30 AM to 10:20 AM
|
|
Sponsor:
|
Section on Survey Research Methods
|
| Abstract - #304085 |
|
Title:
|
A semi-parametric approach to fractional imputation for nonignorable missing data
|
|
Author(s):
|
Jae-kwang Kim*+ and Cindy L. Yu
|
|
Companies:
|
Iowa State University and Iowa State University
|
|
Address:
|
Snedecor Hall, Ames, IA, 50011,
|
|
Keywords:
|
EM algorithm ; Score equation ; Observed information ; Follow-up survey ; Multiple imputation
|
|
Abstract:
|
Parameter estimation with nonignorable missing data is a challenging problem in statistics. Fully parametric approach for joint modeling of the response model and the population model can produce results that are very sensitive against the failure of the assumed model. We consider a more robust approach of modeling by describing the model for the nonresponding part as a exponential tilting of the model for the responding part, which can be justified under the assumption that the response mechanism can be expressed as a logistic regression model. The model for the responding part can be estimated using a nonparametric method. Thus, the overall model can be called semi-parametric. In this paper, based on the exponential tilting model, we propose a fractional imputation method that can be a useful computational tool for missing data analysis with non-ignorable missing data.
|
