No label defined (Q4944834)

Let \(U_n\) be a one-sample \(U\)-statistic with kernel \(h(x_1, x_2)\) of degree two, such that \[ Eh(X_1, X_2)= \theta \quad\text{and}\quad \sigma_g^2= \operatorname {Var} E[h(X_1, X_2)- \theta\mid X_1]>0. \] Let \(\sigma_{g^*}^{*^2}\) be the bootstrap estimator of \(\sigma_g^2\). \textit{X. Shi} [Chin. J. Appl. Probab. Stat. 2, 144-148 (1986; Zbl 0634.62013)] showed that \(\sigma_{g^*}^{*^2}\to \sigma_g^2\) a.s. under \(Eh (x_1,x_2)^2< \infty\), and \(E|h(x_1,x_2)|< \infty\). Hence \(W_n= \sqrt{n} (U_n-\theta)/ 2\sigma_{g^*}^*\) converges in distribution to \(N(0,1)\). This paper shows that when the 4th absolute moment of the kernel \(h\) exists, we have \(\sup_x|p(W_n\leq x)- \Phi(x)|\leq C/\sqrt{n}\) as \(n\to \infty\).