297 – Design and Analysis of Therapeutic Medical Device Trials
A Score Test for Detecting Heterogeneous Within-Group Variances in Linear Mixed Models
Sponsor: Section on Medical Devices and Diagnostics
Keywords: Diagnostic test, model violation, heteroscadasticity, score statistic
Chul H. Ahn
FDA-CDRH
One of the common assumptions in linear mixed models is that the within-group variances are homogeneous. This paper investigates the impact of heterogeneous within-group variances on the estimate of a regression coefficient especially when the within-group variance depends on a group-level covariate. We also develop a diagnostic test based on score statistic to detect heterogeneous within-group variances in linear mixed linear models. The method is applied to the data from an orthopedic device clinical trial.