Permutation tests for reflected symmetry (Q1275409): Difference between revisions

A \(d\)-variate random variable \(x\) is said to be reflectedly symmetric about a center \(\mu\) if the imaginary part of the characteristic function of \(x-\mu\) equals zero, i.e., \[ E\biggl(\sin\bigl(t'(x-\mu)\bigl) \biggl) =0\quad \text{for }t \in\mathbb{R}^d, \tag{1} \] where \(t'\) stands for the transpose of \(t\). Let \(x_1,\dots, x_n\) be i.i.d. copies of \(x\) and define an empirical process \[ \Bigl\{V_n (X_n,t): =\sqrt nP_n\biggl(\sin \bigl(t'(X_n-\mu) \bigl)\biggl): t\in A\Bigr\} \tag{2} \] and an estimated empirical process \[ \Bigl\{V_{n1} (X_n,\widehat \mu,t): =\sqrt nP_n\biggl(\sin \bigl(t'(X_n-\widehat \mu)\bigr) \biggr): t\in A\Bigr\}, \tag{3} \] where \(X_n=(x_1, \dots, x_n)\), \(\widetilde\mu\) is an estimate of \(\mu\) when it is unknown, \(A\) is a so-called working region of \(\mathbb{R}^d\), and \(P_n(\cdot)\) stands for the empirical measure \(P_n(f(X_n)): =n^{-1} \sum^n_{j=1} f(x_j)\). When \(\mu\) is known, \textit{S. Ghosh} and \textit{F.H. Ruymgaart} [Can. J. Stat. 20, No. 4, 429-440 (1992; Zbl 0774.62054)] derived some asymptotic properties of the test based on \(\int_A(V_n(X_n,t))^2dw(t)\), where \(w(\cdot)\) is a distribution function on \(\mathbb{R}^d\), while \textit{C.R. Heathcote}, \textit{S.T. Rachev} and \textit{B. Cheng} [J. Multivariate Anal. 54, No. 1, 91-112 (1995; Zbl 0863.62059)] studied the limiting behaviour of the test based on \(\sup_A\|V_{n1}(X_n, \widehat \mu,t) |\) for the case where the center \(\mu\) is unknown. For further preliminary details we refer to Section 2 of the paper under review. The present contribution defines an empirical permutation process, a version of (2), for the known center case, and an estimated empirical permutation process, a version of (3), for the case when \(\mu\) is unknown. The resulting permutation tests that are based on \(L_2\) and sup-functionals of the just mentioned two permutation processes are shown to be strictly distribution-free under the null hypothesis that the underlying variables are symmetrically distributed around a center. Furthermore, the permutation tests are strictly valid if the center of symmetry is known, and are asymptotically valid if the center is an unknown point and hence it has to be estimated. It is shown as well that tests based on respective \(L_2\) and sup-functionals of (2) and (3) are asymptotically equivalent to those that are based on their permutation counterparts. The power behaviour of the just mentioned tests is investigated for local alternatives. In order to demonstrate the performance of the permutation tests, some small-sample simulation experiments have been performed and reported in the paper.