More Efficient Bootstrap Computations
More Efficient Bootstrap Computations
复制标题
更高效的引导计算
DOI：
10.1080/01621459.1990.10475309
复制
发表时间：
1990
影响因子：
3.7
通讯作者：
B. Efron
中科院分区：
文献类型：
--
作者：
B. Efron
摘要本文讨论了自举法的计算方法，这种方法比通常使用的直接蒙特卡罗方法更有效。Bootstrap被认为是其最简单的形式：在单样本非参数问题中，目标是通过Bootstrap抽样估计某些统计量的偏差或方差，或者根据Bootstrap分布的各种分布为感兴趣的参数设置近似置信区间。在有利的情况下，本文的方法可以将引导复制的必要数量减少许多倍。此外，简单的诊断可以用来判断这些方法是否可以访问任何特定情况。
Abstract This article concerns computational methods for the bootstrap that are more efficient than the straightforward Monte Carlo methods usually used. The bootstrap is considered in its simplest form: in a one-sample nonparametric problem, where the goal is to estimate the bias or variance of some statistic by bootstrap sampling, or to set approximate confidence intervals for a parameter of interest in terms of various percentiles of the bootstrap distribution. The methods of this article can, in favorable situations, reduce the necessary number of bootstrap replications manyfold. Moreover, simple diagnostics are available to see whether or not any particular case is accessible to these methods.