A1485
Title: Inference for nearly directional data
Authors: Thomas Verdebout - Universite Libre de Bruxelles (Belgium) [presenting]
Abstract: Models for noisy directional data are considered, in which a radial noise makes the observations deviate from their theoretical hyperspherical sample space, namely a hypersphere centered at and with radius. Inference hypothesis testing, point estimation, and confidence zone estimation are considered on the location parameter. Several asymptotic scenarios are introduced, in which the radius of the hypersphere and, most importantly, the noise magnitude may depend on the sample size in an essentially arbitrary way. This allows for considering very diverse cases in which the prior information that the data belongs to a hypersphere is more and more, or on the contrary, less and less relevant. The investigation is based on Le Cams asymptotic theory of statistical experiments and aims at a full understanding of the resulting limiting experiments.
