Multiple testing with optimal individual tests in Gaussian graphical model selection

TL;DR

This paper develops optimal unbiased individual tests for Gaussian graphical model selection, demonstrating their equivalence to sample partial correlation tests and comparing their performance with standard methods.

Contribution

It introduces a new class of optimal unbiased individual tests for Gaussian graphical models, showing their equivalence to sample partial correlation tests and improving model selection procedures.

Findings

Optimal tests are equivalent to sample partial correlation tests.

The proposed multiple decision procedure outperforms standard methods.

The approach enhances accuracy in Gaussian graphical model selection.

Abstract

Gaussian Graphical Model selection problem is considered. Concentration graph is identified by multiple decision procedure based on individual tests. Optimal unbiased individual tests are constructed. It is shown that optimal tests are equivalent to sample partial correlation tests. Associated multiple decision procedure is compared with standard procedure.