How Many Bootstraps?
Abstract
The bootstrap is a non-parametric method for assessing statistical accuracy. In approximating bootstrap quantities by monte carlo simulation, one must decide how many bootstrap samples to generate. This document proposes an adaptive sequential method that estimates the accuracy based on the current bootstrap samples. Bootstrap sampling is continued until the estimated accuracy is high enough. In the examples given, 100 to 300 bootstraps are sufficient for standard error and bias estimation, while 1000 bootstraps may be necessary for estimating a percentile. Additional keywords: Tables(data).
Document Details
- Document Type
- Technical Report
- Publication Date
- Aug 22, 1985
- Accession Number
- ADA160043
Entities
People
- R. Tibshirani
Organizations
- Stanford University