A1385
Title: Confidence regions when the parameter is near the boundary
Authors: Karl Oskar Ekvall - University of Florida (United States) [presenting]
Abstract: The focus is on recent advances in the theory and methods for constructing reliable confidence regions when the parameter may be near the boundary of the parameter set. For reasons to be discussed, constructing such confidence regions is often substantially more difficult than the testing of boundary points. Recent work shows a connection between boundary problems and a singular Fisher information that can sometimes be leveraged to provide reliable conference regions. However, it is unclear how far that approach generalizes, and other arguments appear more useful in some settings. Examples using variance components and mixed models illustrate the general theory and also suggest several directions for future research. For example, the need for reliable confidence intervals for variance components suggests studying settings where both a parameter of interest and nuisance parameters can be near the boundary.
