No Detail Found!

Bias reduction: The delata method versus the jackknife and the bootstrap.

Author(s)
Christopher S. Withers Applied Mathematics Group. Industrial Research Limited Lower Hutt, New Zealand
Saralees Nadarajah School of Mathematics, University of Manchester Manchester M13 9PL, UK
Abstract
Calculation of Quenouille (1956)’s jackknife for an estimator based on a sample of size n takes ~n2 calculations in general. The bootstrap takes over twice as many. The dalta estimator (based on Taylor’s series) takes only ~n calculations. So, the jackknife has computational efficiency (CE) ~1/n and asymptotic CE (ACE) zero. These three estimators are all second-order: that is they reduce bias from ~1/n to ~1/n2. For higher order bias reduction the CE of the jackknife to the delta estimator is even worse. For estimators which can be written as a function of one or more sample means, the second-order jackknife only requires ~n computations, though still more than required by the delta method-more than twice as many for a univariate sample; the ACE of higher order jackknives remains zero. In short (a) the bootstrap is only second order and has bad computational efficiency; (b) the jackknife has bad CE except for a second order jackknife of a function of a mean, in which case its ACE (relative to the delta estimator) is less than one, and less than 0.5 for a univariate mean.
Publication Details
Page(s) 143-151
DOI DOI not available
Published Journal: Pakistan Journal of Statistics, Volume: 30, Issue: 1, Year: 2014

Request the full text from PASTIC

0

Citations

0

Downloads

37

Views