|
Activity Number:
|
64
|
|
Type:
|
Contributed
|
|
Date/Time:
|
Sunday, July 29, 2007 : 4:00 PM to 5:50 PM
|
|
Sponsor:
|
Biometrics Section
|
| Abstract - #308584 |
|
Title:
|
Comparison of Design-Based and Model-Based Methods to Estimate the Cluster Effect Using National Population Health Survey Data (1994--2003)
|
|
Author(s):
|
Sunita Ghosh*+ and Punam Pahwa
|
|
Companies:
|
University of Saskatchewan and University of Saskatchewan
|
|
Address:
|
1042711 112th St, Edmonton, AB, T6J 4M1, Canada
|
|
Keywords:
|
Multi-stage sampling ; Bootstrap ; Generalized Estimating Equation ; NPHS ; Asthma ; Logistic Regression
|
|
Abstract:
|
Analysis of data collected using multi-stage sampling design, should account for stratification, clustering and unequal probability of selection. This paper aims to account for clustering and unequal probability of selection. Design-based approaches Rao-Wu bootstrap methods were used. Generalized Estimating Equation (GEE) approach was used for model-based methods. Both these approaches were applied and compared using National Population Health Survey (NPHS) dataset. Two time points Cycle 1 (1994--95) and Cycle 5 (2002--03) were used for analytical purpose. Longitudinal weight provided by Statistics Canada was used. The variable of interest was self-reported physician diagnosed asthma. Its relationship with other variables was studied. Accounting for the two features of survey design, produced similar results for the design-based and model-based methods.
|
