Multilevel Modeling with Scaled Weights vs. Models Based on Generalized Estimating Equations (GEE) with Bootstrap Variance Estimation

• Activity Number: 217
• Type: Contributed
• Date/Time: Monday, August 3, 2009 : 2:00 PM to 3:50 PM
• Sponsor: Section on Statistics in Epidemiology
• Author(s): Alomgir Hossain*+ and Punham Pahwa
• Companies: University of Saskatchewan and University of Saskatchewan
• Keywords: Multilevel,Bootstrap ; Pseudo maximum likelihood and Quasi-likelihood ; scaled weights ; BMI (Body mass index) ; NPHS (national population health survey),

Abstract

Longitudinal complex survey data have a hierarchical structure where repeated measurements are nested within subjects. Multilevel models with scaled weights based on weighted pseudo maximum likelihood approach, and GEE approach based on quasi-likelihood with bootstrap variance estimation technique are commonly used methods to analyze the longitudinal survey data. Objective of this study was to investigate the similarities and differences between multilevel models based on two types of scaled weights (level 1) and models based on GEE with bootstrap variance estimation. The objective was accomplished by analyzing longitudinal data from Canadian NPHS. The NPHS began in 1994/95 with a sample size of 17,276 and data being collected every two years. The outcome variable of interest was self-reported diabetes (yes, no) and explanatory variables were BMI, household income, education, sex and age.