E-Values for Exponential Families: the General Case
Abstract
We analyze common types of e-variables and e-processes for composite exponential family nulls: the optimal e-variable based on the reverse information projection (RIPr), the conditional (COND) e-variable, and the universal inference (UI) and sequen\-tialized RIPr e-processes. We characterize the RIPr prior for simple and Bayes-mixture based alternatives, either precisely (for Gaussian nulls and alternatives) or in an approximate sense (general exponential families). We provide conditions under which the RIPr e-variable is (again exactly vs. approximately) equal to the COND e-variable. Based on these and other interrelations which we establish, we determine the e-power of the four e-statistics as a function of sample size, exactly for Gaussian and up to in general. For -dimensional null and alternative, the e-power of UI tends to be smaller by a term of than that of the COND e-variable, which is the clear winner.
Cite
@article{arxiv.2409.11134,
title = {E-Values for Exponential Families: the General Case},
author = {Yunda Hao and Peter Grünwald},
journal= {arXiv preprint arXiv:2409.11134},
year = {2025}
}
Comments
50 pages, 11 figures