Hypotheses Testing of Functional Principal Components

Abstract

: We propose a test for the hypothesis that the standardized functional principal components (FPCs) of functional data are equal to a given set of orthonormal bases (e.g., the Fourier basis). Using estimates of individual trajectories that satisfy certain approximation conditions, we construct a chi-square-type statistic, and show that it is oracally e(cid:14)cient under the null hypothesis, in the sense that its limiting distribution is the same as that of an infeasible statistic using all trajectories, known as the \oracle." The null limiting distribution is an in(cid:12)nite Gaussian quadratic form, and we obtain a consistent estimator of its quantile. A test statistic based on the chi-squared-type statistic and the approximate quantile of the Gaussian quadratic form is shown to be both of the nominal asymptotic signi(cid:12)cance level and asymptotically correct. It is further shown that B-spline trajectory estimates meet the required approximation conditions. Simulation studies demonstrate the superior (cid:12)nite-sample performance of the proposed testing procedure. Using electroencephalogram (EEG) data, the proposed procedure con(cid:12)rms an interesting discovery that the centered EEG data are generated from a small