On the number of bootstrap simulations required to construct a confidence interval (Q1087272): Difference between revisions

Suppose we obtain one-sided confidence intervals for a parameter \(\theta\) by ''smooth'' Studentized statistic using B bootstrap simulations. It is shown that the reduction of error of coverage probability from \(O(n^{- 1/2})\) to \(O(n^{-1})\) is available uniformly in B provided nominal coverage probability is a multiple of \((B+1)^{-1}\). This improvement is possible even when B is fixed and n is allowed to increase. Further it is shown that for large n, the simulated statistic values behave like random observations from a continuous distribution unless B increases faster than any power of n. The effect of discrete nature of the bootstrap statistic is felt only if B increases exponentially with n.