Likelihood ratio tests under model misspecification in high dimensions

TL;DR

This paper demonstrates that likelihood ratio tests for high-dimensional covariance matrices are robust to distributional assumptions and model misspecification, providing asymptotic normality results and practical validation through simulations and real data analysis.

Contribution

It extends the asymptotic analysis of likelihood ratio tests to a broader class of distributions and establishes their robustness in high-dimensional settings.

Findings

Likelihood ratio tests are asymptotically invariant under distributional misspecification.

Asymptotic normality of the test statistic is established for multiple covariance matrices.

Simulation and real data confirm the practical robustness of the proposed methods.

Abstract

We investigate the likelihood ratio test for a large block-diagonal covariance matrix with an increasing number of blocks under the null hypothesis. While so far the likelihood ratio statistic has only been studied for normal populations, we establish that its asymptotic behavior is invariant under a much larger class of distributions. This implies robustness against model misspecification, which is common in high-dimensional regimes. Demonstrating the flexibility of our approach, we additionally establish asymptotic normality of the log-likelihood ratio test statistic for the equality of many large sample covariance matrices under model uncertainty. For this statistic, a subtle adjustment to the centering term is needed compared to normal case. A simulation study and an analysis of a data set from psychology emphasize the usefulness of our findings.