Comparing Statistical Software Packages for Multilevel Modeling

• Activity Number: 472
• Type: Contributed
• Date/Time: Wednesday, August 5, 2009 : 10:30 AM to 12:20 PM
• Sponsor: Section for Statistical Programmers and Analysts
• Author(s): Marie Gantz*+ and G. Gordon Brown and Randall Bender and Jill Fromewick and Scott Novak and Dhuly Chowdhury and Malavika Subramanyam and Theresa L. Osypuk
• Companies: RTI International and RTI International and RTI International and Summit Research Associates and RTI International and RTI International and Harvard School of Public Health and Northeastern University
• Keywords: software comparison ; mulitlevel modeling ; correlated data

Abstract

Analysis of multilevel data requires software that can account for complex correlation structures; however, identifying the most appropriate software can be difficult. In addition, most packages do not accommodate both model-based and design-based approaches to multilevel modeling. For example, while many surveys include weights, strata and clustering, packages that account for such design-based components do not often accommodate model-based features such as random effects. We compare the capabilities of SAS, SUDAAN, Stata, Mplus, HLM and MLwiN for analyzing multilevel data from the model- and design-based points of view. Comparisons are based on a review of software documentation and relevant literature, interviews with users, and implementation of a range of models using each of the software packages.