Creating non-parametric bootstrap samples using Poisson frequencies
Creating non-parametric bootstrap samples using Poisson frequencies
复制标题
DOI：
10.1016/j.cmpb.2006.04.006
复制
发表时间：
2006-07-01
影响因子：
6.1
通讯作者：
MacGibbon, Brenda
中科院分区：
文献类型：
--
作者：
Hanley, James A.;MacGibbon, Brenda
本文介绍了如何在非统计学家使用的高级软件包中创建和分析近似的非参数Bootstrap样本，而无需物理创建新的数据集，也无需进行复杂的编程。这种捷径的方法，它使用泊松，而不是多项式频率的副本的数量，每个观察，从理论上证明了通过评估自助方差的例子中，经典的估计的抽样方差的统计量的兴趣有一个已知的封闭形式。对于50个或更多的样本量，通过这种快捷方法获得的自助标准误差超过标准版本获得的标准误差不到1%。所提出的方法也评估了两个工作的例子，涉及统计的抽样分布是比较复杂的。其中第二个也用于说明何时可以和不能使用非参数自助样本。(c)2006爱思唯尔爱尔兰有限公司保留所有权利。
This article describes how, in the high-level software packages used by non-statisticians, approximate non-parametric bootstrap samples can be created and analyzed without physically creating new data sets, or resorting to complex programming. The comparable performance of this shortcut method, which uses Poisson rather than multinomial frequencies for the numbers of copies of each observation, is demonstrated theoretically by evaluating the bootstrap variance in an example where the classic estimator of the sampling variance of the statistic of interest has a known closed form. For sample sizes of 50 or more, bootstrap standard errors obtained by this shortcut method exceeded those obtained by the standard version by less than 1%. The proposed method is also evaluated in two worked examples, involving statistics whose sampling distribution is more complex. The second of these is also used to illustrate when one can and cannot use non-parametric bootstrap samples. (c) 2006 Elsevier Ireland Ltd. All rights reserved.