Bootstrap Confidence Intervals and Bootstrap Approximations.
Abstract
This document studies the BC sub a bootstrap procedure for constructing parametric and non-parametric confidence intervals. The BC sub a interval relies on the existence of a transformation that maps the problem into a normal scaled transformation family. The authors show how to construct this transformation in general. Exploiting this, they derive an interval that equals the BC sub a interval to second order, computable without bootstrap sampling. As a further benefit, this construction provides a second order correct approximation to the bootstrap distribution of a statistic, computed without bootstrap sampling. Both the new interval and the approximation require only n+2 evaluations of the statistic, where n is the sample size. (Author)
Document Details
- Document Type
- Technical Report
- Publication Date
- Jun 04, 1986
- Accession Number
- ADA168480
Entities
People
- Robert Tibshirani
- Thomas Dicicco
Organizations
- Stanford University