Abstract
One impediment to wider use of bootstrap methods is the large amount of computer time often required to compute bootstrap estimates. This paper shows how the saddlepoint approximation method can be used in certain situations to give very accurate parametric bootstrap statistics with a much smaller amount of calculation. Essentially, the problem of computing an n-dimentional integral (where n is the sample size) by Monte Carlo is reduced to an integral whose dimension is that of the parameter space, which is usually much smaller. This allows accurate numerical integration techniques to be used in place of simulation. The method should be especially effective for maximum likelihood in linear exponential family problems; it is illustrated by application to a problem in reliability.
| Original language | English (US) |
|---|---|
| Pages (from-to) | 451-460 |
| Number of pages | 10 |
| Journal | Computational Statistics and Data Analysis |
| Volume | 15 |
| Issue number | 4 |
| DOIs | |
| State | Published - 1993 |
Fingerprint
Keywords
- Gaussian quadrature
- Maximum likelihood
- Numerical integration
- Reliability
- Stress-strength problems
ASJC Scopus subject areas
- Computational Theory and Mathematics
- Statistics, Probability and Uncertainty
- Electrical and Electronic Engineering
- Computational Mathematics
- Numerical Analysis
- Statistics and Probability
Cite this
Almost-exact parametric bootstrap calculation via the saddlepoint approximation. / Rocke, David M.
In: Computational Statistics and Data Analysis, Vol. 15, No. 4, 1993, p. 451-460.Research output: Contribution to journal › Article
}
TY - JOUR
T1 - Almost-exact parametric bootstrap calculation via the saddlepoint approximation
AU - Rocke, David M
PY - 1993
Y1 - 1993
N2 - One impediment to wider use of bootstrap methods is the large amount of computer time often required to compute bootstrap estimates. This paper shows how the saddlepoint approximation method can be used in certain situations to give very accurate parametric bootstrap statistics with a much smaller amount of calculation. Essentially, the problem of computing an n-dimentional integral (where n is the sample size) by Monte Carlo is reduced to an integral whose dimension is that of the parameter space, which is usually much smaller. This allows accurate numerical integration techniques to be used in place of simulation. The method should be especially effective for maximum likelihood in linear exponential family problems; it is illustrated by application to a problem in reliability.
AB - One impediment to wider use of bootstrap methods is the large amount of computer time often required to compute bootstrap estimates. This paper shows how the saddlepoint approximation method can be used in certain situations to give very accurate parametric bootstrap statistics with a much smaller amount of calculation. Essentially, the problem of computing an n-dimentional integral (where n is the sample size) by Monte Carlo is reduced to an integral whose dimension is that of the parameter space, which is usually much smaller. This allows accurate numerical integration techniques to be used in place of simulation. The method should be especially effective for maximum likelihood in linear exponential family problems; it is illustrated by application to a problem in reliability.
KW - Gaussian quadrature
KW - Maximum likelihood
KW - Numerical integration
KW - Reliability
KW - Stress-strength problems
UR - http://www.scopus.com/inward/record.url?scp=43949169312&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=43949169312&partnerID=8YFLogxK
U2 - 10.1016/0167-9473(93)90176-T
DO - 10.1016/0167-9473(93)90176-T
M3 - Article
AN - SCOPUS:43949169312
VL - 15
SP - 451
EP - 460
JO - Computational Statistics and Data Analysis
JF - Computational Statistics and Data Analysis
SN - 0167-9473
IS - 4
ER -