Modified Tests of Linear Hypotheses Under Heteroscedasticity for Multivariate Functional Data with Finite Sample Sizes (2504.03161v1)

Abstract: As big data continues to grow, statistical inference for multivariate functional data (MFD) has become crucial. Although recent advancements have been made in testing the equality of mean functions, research on testing linear hypotheses for mean functions remains limited. Current methods primarily consist of permutation-based tests or asymptotic tests. However, permutation-based tests are known to be time-consuming, while asymptotic tests typically require larger sample sizes to maintain an accurate Type I error rate. This paper introduces three finite-sample tests that modify traditional MANOVA methods to tackle the general linear hypothesis testing problem for MFD. The test statistics rely on two symmetric, nonnegative-definite matrices, approximated by Wishart distributions, with degrees of freedom estimated via a U-statistics-based method. The proposed tests are affine-invariant, computationally more efficient than permutation-based tests, and better at controlling significance levels in small samples compared to asymptotic tests. A real-data example further showcases their practical utility.