Isotropic randomization for one-sample testing in metric spaces

TL;DR

This paper develops a new hypothesis testing method for the Fréchet mean in general metric spaces, extending classical Euclidean tests by leveraging geometric structures and randomization techniques.

Contribution

It introduces a novel approach to mean testing in metric spaces, combining geometric insights with randomization to ensure valid inference beyond Euclidean contexts.

Findings

The method provides theoretical guarantees for validity.

Numerical experiments demonstrate effectiveness across various metric spaces.

Application to wind data illustrates practical utility.

Abstract

We address the problem of testing hypotheses about a specific value of the Fr\'echet mean in metric spaces, extending classical mean testing from Euclidean spaces to more general settings. We extend an Euclidean testing procedure progresively, starting with test construction in Riemannian manifolds, leveraging their natural geometric structure through exponential and logarithm maps, and then extend to general metric spaces through the introduction of admissible randomization techniques. This approach preserves essential geometric properties required for valid statistical inference while maintaining broad applicability. We establish theoretical guarantees for our testing procedure and demonstrate its effectiveness through numerical experiments across different metric spaces and distributional settings. The practical utility of our method is further illustrated through an application to wind data in western Denmark, showcasing its relevance for real-world statistical analysis.