CFE 2019: Start Registration
View Submission - CMStatistics
B1922
Title: Accurate likelihood inference on boundaries Authors:  Anthony Davison - EPFL (Switzerland)
Soumaya Elkantassi - Ecole Polytechnique Federale de Lausanne (Switzerland) [presenting]
Abstract: Statistics used to test hypotheses concerning parameters on the boundary of their domain often have non-standard limiting distributions, and these may be poor approximations to finite-sample distributions of the test statistics even when the sample size is very large. A canonical example of such a situation is testing for a zero variance component, which is equivalent to testing whether a spline expansion is needed in semi-parametric regression. An approach is described to small-sample approximation in such settings, based on higher-order approximations and in particular the tangent exponential model. Numerical results show that the approach can give much improved approximations, even in small samples.