|
Abstract:
|
We present simple methods for testing the adequacy of linear mixed effects models against a general non-parametric alternative, for dense or sparse longitudinal data. Using spline representations and functional principal components, the testing procedure can be simplified to tests of single variance components or covariance structures, allowing us to build off existing methods and software. We highlight the limitations and strengths of each method through simulation studies, and compare performance under specific situations where existing methods can be used. We also demonstrate the utility and ease of use of these methods by applying them to an infant growth dataset.
|
