A Comparison of the Reporting of Problems Encountered in the Estimation of Covariance Parameters in Linear Mixed Models Using SAS, SPSS, R, Stata, and HLM

• Activity Number: 204
• Type: Contributed
• Date/Time: Monday, August 7, 2006 : 2:00 PM to 3:50 PM
• Sponsor: Section on Statistical Computing
• Author(s): Brady West*+ and Kathy Welch and Andrzej Galecki
• Companies: University of Michigan and University of Michigan and University of Michigan
• Keywords: linear mixed models ; statistical software ; covariance parameters ; estimation

Abstract

Computational problems with maximum likelihood estimation of covariance parameters in linear mixed models frequently arise in practice. In some cases, estimates of covariance parameters converge to a solution close to the boundary of the parameter space, or fall outside it. As a result, the variance-covariance matrix for the random effects in the model violates (or nearly violates) positive definiteness constraints. We fit a linear mixed model to a real-world longitudinal dataset in which this problem is encountered and compare the way it is reported when using the mixed modeling procedures in SAS, SPSS, R, Stata, and HLM. We compare the estimates derived using these packages and present practical alternatives for dealing with the problem. This work can be found in Linear Mixed Models: a Practical Guide Using Statistical Software.