Analysis of Multilevel Complex Survey Data: Comparison of Procedures from Several Software

• Activity Number: 149 - Advances in Modeling Multilevel Observational Data from Complex Surveys
• Type: Topic Contributed
• Date/Time: Monday, July 31, 2017 : 10:30 AM to 12:20 PM
• Sponsor: Section on Statistics in Epidemiology
• Author(s): Lin Dai* and Mulugeta Gebregziabher
• Companies: Medical University of South Carolina and MUSC
• Keywords: Complex survey ; Multilevel modeling ; Pseudolikelihood ; Weights ; Statistical software

Abstract

Multilevel complex surveys are widely used in large-scale research. In a multilevel complex survey, clusters and units within the sampled clusters may be sampled with unequal probabilities. Multilevel models that take into account hierarchical structure are commonly used to analyze multilevel complex survey data. For example, a pseudolikelihood approach extends the multilevel model framework to incorporate weights at different levels. The increasing use of this approach has led software such as SAS, R and Stata to develop their own package or procedure to analyze complex survey data with sampling weight. In this study, we demonstrate how to estimate multilevel models for complex survey data using SAS, R and Stata. We clarify the weight options offered in each package in SAS, R and Stata, and provide the syntax for estimating multilevel model in each package. We then compare the results of these different software under different scaling weights.