Browse Teach Curate

Heavily Right Combination Tests for Convex Confidence Regions under Dependence

A 'heavily right' strategy for combining dependent studies: truncating the Cauchy combination test to a Half-Cauchy test (and recognizing the harmonic-mean test as its Pareto-tail analogue) restores convexity of the resulting confidence region, so that every linear projection is a valid confidence interval, provably when studies are summarized by Hotelling T-squared or chi-squared statistics, with exact algorithms and application to network meta-analysis.

Data Text

A Heavily Right Strategy for Statistical Inference with Dependent Studies in Any Dimension Tianle

Studies 'dependence-resilient' global hypothesis tests that combine many possibly-dependent studies without modeling their dependence, of which the Cauchy combination test is the influential example.…