|
Activity Number:
|
428
|
|
Type:
|
Contributed
|
|
Date/Time:
|
Wednesday, August 9, 2006 : 10:30 AM to 12:20 PM
|
|
Sponsor:
|
Section on Statistics in Epidemiology
|
| Abstract - #307567 |
|
Title:
|
A Comparison of Three Categorical Data Analysis Methods Applied to Survey Data
|
|
Author(s):
|
Barbara Neas*+ and Hani Dimassi and David M. Thompson and Betty J. Pfefferbaum
|
|
Companies:
|
The University of Oklahoma Health Sciences Center and The University of Oklahoma Health Sciences Center and The University of Oklahoma and The University of Oklahoma Health Sciences Center
|
|
Address:
|
801 NE 13th Street, Oklahoma City, OK, 73104,
|
|
Keywords:
|
categorical survey data ; GEE ; log-linear analysis ; latent class analysis
|
|
Abstract:
|
This work was motivated from a secondary analysis of Oklahoma City bombing survey data of middle school students. We applied generalized estimating equations (GEE), log-linear analysis (LLA) and latent class analysis (LCA) to a subset of data and compared the results. These results provide the basis for a series of simulated data sets with different patterns of associations. The GEE analysis included school as a cluster variable and resulted in separate models for males and females with significance between the risk factor and outcome for males only. LLA showed the dependencies among the variables. LCA produced underlying constructs for the outcome. The construct representing presence of the outcome included the risk factor and a pseudo-age variable, while the construct representing absence of the outcome included the risk factor and gender. Other examples are discussed.
|
