|
Abstract:
|
Inherently, mixed modeling is computationally intensive and numerically tricky. Real-life data rarely works out as well as examples in text books and software manuals. The analysis of split-plot experimental designs and other mixed models often leaves the analyst dazed and confused. Default algorithms in software might fail to converge for some data sets and models, or might converge to fits that don't make statistical sense (e.g., negative variance components). In this roundtable, we will discuss experiences with these kinds of problems and more. We will leverage our own experiences and literature knowledge to provide recommendations for circumventing them. Or maybe we will pave the way for a new area of research in statistical practice. Both those with answers and questions are encouraged to attend.
|
