Fixed Effects in General Linear Mixed Effect Modelling in Longitudinal Studies: Bootstrap and Modified Classical Procedures

• Activity Number: 18
• Type: Contributed
• Date/Time: Sunday, August 11, 2002 : 2:00 PM to 3:50 PM
• Sponsor: General Methodology
• Author(s): Dawen Sui*+ and Rachel Fouladi+ and Carl deMoor
• Affiliation(s): U. T. M. D. Anderson Cancer Center and U. T. M. D. Anderson Cancer Center and U. T. M. D. Anderson Cancer Center
• Keywords: Linear mixed models ; Repeated measures design ; Bootstrap ; Type I error control

Abstract

Equivalent asymptotic procedures can yield conflicting inference when applied to the same dataset (Aubin & Cordeiro, 2000). Among the class of likelihood ratio tests, modified likelihood ratio tests (e.g., Bartlett, 1937) have been proposed, and have documented improved performance over classical tests. Other procedures including modified F tests have also been proposed. Bootstrapping is generally proposed as another means by which to obtained improved procedures under varied sampling distributions. In the current paper, the performance characteristics of bootstrap tests are contrasted with the performance of modified asymptotic tests of fixed parameters in mixed linear models of data from small sample repeated measures studies. In particular, the current study contrasts the relative performance characteristics of bootstrap statistics with modified asymptotic statistics involved in tests of fixed effects in mixed linear models with data simulated following small sample balanced and unbalanced repeated measures designs. In particular, the Type I error control of these procedures under varied small sample distributional conditions is documented.