Performance and Robustness of Covariance Matrix Equality Tests under High-Dimensional Covariance Structures

When the dimension is close to or above the number of observations, classical tests for equal covariance matrices cannot be used, and the behaviour of their high-dimensional alternatives may change with the covariance structure. We examine how well three such procedures, the Srivastava–Yanagihara (SY), Ahmad (AHM), and Qayed–Han (QH) tests, hold the nominal level and how often they detect unequal covariance matrices. Monte Carlo simulations with 5,000 replications were conducted for three multivariate normal populations with sample sizes from 10 to 160 and dimensions equal to or larger than the sample size, under first-order autoregressive (AR(1)), banded Toeplitz (BTOEP), and simple diagonal (SIM) covariance structures. QH kept its empirical Type I error rate between 0.041 and 0.059 in all settings, and SY stayed inside Bradley's interval in all but one setting, whereas AHM was liberal for sample sizes of 20 or less, with rates up to 0.104. SY was the most powerful test in most AR(1) and SIM settings, while QH was the most powerful under BTOEP, with power of at least 0.91 even for samples of size 10. Because QH depends on the data only through the sum of the variables and is equivalent to Bartlett's test applied to that sum, it has high power when the variance of this sum differs between populations but only trivial power, equal to its size, against alternatives that leave it unchanged. SY is therefore recommended as the primary test when the covariance structure is unknown, with QH as a complementary test.

Authors

Institutions

Publication Details

Journal
International Journal of Analysis and Applications
Published
2026-11-16
DOI
https://doi.org/10.28924/2291-8639-24-2026-382
Primary Topic
Statistical Methods and Inference
Type
article
Field-Weighted Citation Impact
0.00
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

Performance and Robustness of Covariance Matrix Equality Tests under High-Dimensional Covariance Structures

Boonyarit Choopradit, Sirithip Wasinrat
1 links
International Journal of Analysis and Applications
Statistical Methods and Inference
article

Performance and Robustness of Covariance Matrix Equality Tests under High-Dimensional Covariance Structures

Boonyarit Choopradit, Sirithip Wasinrat
article en

Abstract

When the dimension is close to or above the number of observations, classical tests for equal covariance matrices cannot be used, and the behaviour of their high-dimensional alternatives may change with the covariance structure. We examine how well three such procedures, the Srivastava–Yanagihara (SY), Ahmad (AHM), and Qayed–Han (QH) tests, hold the nominal level and how often they detect unequal covariance matrices. Monte Carlo simulations with 5,000 replications were conducted for three multivariate normal populations with sample sizes from 10 to 160 and dimensions equal to or larger than the sample size, under first-order autoregressive (AR(1)), banded Toeplitz (BTOEP), and simple diagonal (SIM) covariance structures. QH kept its empirical Type I error rate between 0.041 and 0.059 in all settings, and SY stayed inside Bradley's interval in all but one setting, whereas AHM was liberal for sample sizes of 20 or less, with rates up to 0.104. SY was the most powerful test in most AR(1) and SIM settings, while QH was the most powerful under BTOEP, with power of at least 0.91 even for samples of size 10. Because QH depends on the data only through the sum of the variables and is equivalent to Bartlett's test applied to that sum, it has high power when the variance of this sum differs between populations but only trivial power, equal to its size, against alternatives that leave it unchanged. SY is therefore recommended as the primary test when the covariance structure is unknown, with QH as a complementary test.

International Journal of Analysis and ApplicationsVol. 24
Thammasat University (TH), Chandrakasem Rajabhat University (TH)
Statistical Methods and Inference
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon