Activity Number:
|
260
|
Type:
|
Contributed
|
Date/Time:
|
Tuesday, August 13, 2002 : 2:00 PM to 3:50 PM
|
Sponsor:
|
Biometrics Section*
|
Abstract - #301852 |
Title:
|
Alternative Strategies for Dimension Reduction with Multivariate Incomplete Data
|
Author(s):
|
Juwon Song*+ and Thomas Belin
|
Affiliation(s):
|
University of California, Los Angeles
|
Address:
|
10920 Wilshire Blvd., Suite 300, Los Angeles, California, 90024, United States
|
Keywords:
|
incomplete data ; factor analysis ; simulation
|
Abstract:
|
When data have large numbers of variables measured on a modest number of cases, even small rates of missingness on individual variables can result in a large number of incomplete cases. As a result, complete-case analysis can lead to bias or loss of efficiency. When we apply multiple imputation, it is recommended to include available information to the fullest extent possible to reduce systematic difference between completely and partially observed cases (Rubin 1996). However, when the sample size is modest, even a simple model can be overparameterized. To avoid this problem under multivariate normal data, Schafer (1997) introduced a method using a ridge prior and Song and Belin (1999) proposed a dimension reduction technique based on a common factor model. Here, we explore strategies to choose the appropriate number of factors in the latter approach. Simulations are conducted to compare commonly used methods with alternative model selection techniques.
|
- The address information is for the authors that have a + after their name.
- Authors who are presenting talks have a * after their name.
Back to the full JSM 2002 program |