Rank-based Maxsum test for high dimensional regression coefficient

TL;DR

This paper introduces adaptive rank-based maxsum tests for high-dimensional linear models that are robust to heavy-tailed errors and effective across various sparsity levels, supported by theoretical and simulation results.

Contribution

It establishes joint asymptotic independence of rank-based sum and max statistics, enabling principled p-value aggregation for robust inference in high-dimensional regression.

Findings

Accurate size control across diverse error distributions.

Strong power for both dense and sparse alternatives.

Robustness to heavy-tailed errors demonstrated through simulations.

Abstract

We study global inference for regression coefficients in high-dimensional linear models under potentially heavy-tailed errors. While sum-type tests are powerful for dense alternatives and max-type tests excel for sparse alternatives, practical applications rarely reveal the sparsity level, and many existing procedures rely on light-tail assumptions. Motivated by the Wilcoxon-score sum test of Feng et al. (2013) and the two Wilcoxon-score maximum tests of Xu and Zhou (2021), we establish under $H_0$ the asymptotic independence between the rank-based sum statistic and each max statistic. These joint limit results justify principled $p$-value aggregation, and we propose two adaptive rank-based maxsum tests via the Cauchy combination method (Liu and Xie, 2020). The proposed procedures inherit robustness from rank-based construction and adaptivity from combining dense- and sparse-sensitive components. Simulation studies confirm accurate size control and strong power across a wide range of error distributions and sparsity regimes.