Test for high-dimensional mean vector under missing observations (Q2237813)

Under a high-dimensional model for stochastically independent and identically distributed random vectors \(\mathbf{Y}_1, \ldots, \mathbf{Y}_n\) taking their values in \(\mathbb{R}^p\), the author is concerned with testing a point null hypothesis about the mean vector of \(\mathbf{Y}_1 = (Y_{1,1}, \ldots, Y_{p, 1})^\top\). In this, it is assumed that some of the \(Y_{i, j}\)'s cannot be observed due to a missing data mechanism. This missing data mechanism is formalized by indicator variables \((\xi_{i, j})_{1 \leq i \leq p, \, 1 \leq j \leq n}\) which are assumed to be jointly stochastically independent and stochastically independent of \((Y_{i, j})_{1 \leq i \leq p, \, 1 \leq j \leq n}\). A test statistic \(T_n\) is proposed and its asymptotic standard normality under the null hypothesis is proved under certain assumptions. Moreover, the asymptotic power of the proposed test based on \(T_n\) is discussed. Besides these theoretical contributions, the author also presents some numerical results from various simulation studies.