The rate of the normal approximation for jackknifing U-statistics (Q1098184)

Let \(U_ n\) be a U-statistic with symmetric kernel \(h(x,y)\) such that \[ E h(X_ 1,X_ 2)=\theta \quad and\quad Var E[h(X_ 1,X_ 2)-\theta | X_ 1]>0. \] Let f(x) be a function defined on r and f'' be bounded. If f(\(\theta)\) is the parameter of interest, a natural estimator is \(f(U_ n)\). It is known that the distribution function of \[ z_ n=\sqrt{n}\{Jf(U_ n)-f(\theta)\}/S^*_ n \] converges to the standard normal distribution \(\Phi\) (x) as \(n\to \infty\), where \(Jf(U_ n)\) is the jackknife estimator of \(f(U_ n)\), and \(S_ n^{*2}\) is the jackknife estimator of the asymptotic variance of \(n^{1/2}Jf(U_ n)\). It is of theoretical value to study the rate of the normal approximation of the statistic \(z_ n\). Assuming the existence of fourth moment of \(h(X_ 1,X_ 2)\), we show that \[ \sup_{x}| P\{z_ n\leq x\}-\Phi (x)| =O(n^{-} \log n). \]