Professor Rauf Ahmad, Statistiska institutionen, Uppsala universitet
General linear models on high-dimensional spheres
Abstract
The classical F statistic to test significance of parameter vector in general linear model is well-known to be robust in the spherical class. The statistic however is inapplicable for high-dimensional models, even under normality, and there have been recent extensions of classical theory to cope with this problem. This project attempts to answer an extended Q: Can F be modified so that it not only works for high-dimensional data but also for spheres in high-dimensional spaces? It is becoming increasingly in demand, as e.g., in directional statistics, statistical shape analysis etc. A modification is shown to work in both cases, and even beyond, covering e.g., sub-Gaussian case, or generally any distribution with finite moments. A further extension to linear mixed models is also discussed.
Senast uppdaterad: 2026-05-06
Sidansvarig: Statistiska institutionen