Simple and efficient methods for constructing bootstrap confidence intervals (Q1965936)

An iterated bootstrap procedure for confidence intervals construction is considered. In this procedure the confidence interval obtained by bootstrap is improved by some new bootstrap-based procedure. To minimize the time of computation balanced importance resampling is proposed. The resamples are taken with probabilities depending on the level of the confidence interval and on the empirical influence function of the estimated statistic. A simulation study is made to compare the mean squared error and the computation time for the efficient balanced bootstrap and the naive bootstrap procedures. A program written in C is presented which can be used to implement the balanced importance resampling.