Timeline for Is it true that the percentile bootstrap should never be used?
Current License: CC BY-SA 4.0
7 events
| when toggle format | what | by | license | comment | |
|---|---|---|---|---|---|
| Dec 5, 2023 at 12:21 | comment | added | cdalitz | That's interesting, but if I understand Hall's remark correctly, the "right table" is for a transformed estimator with known convergence rate (fixed to $\sqrt{n}$ by Hall), and a variance estimated by a different way than bootstrapping. Although he considers the basic bootstrap to make "one error less", he does not discuss the problem that it can lead to impossible values for bounded paramters of interest, e.g. confidence limits greater than one for a correlation coefficient (the same problem affects his "correctly loooked up" values, too, though). | |
| Dec 4, 2023 at 18:45 | comment | added | Closed Limelike Curves | @cdalitz See Peter Hall's paper on why a frequentist would think of the percentile bootstrap as being like "looking up the wrong tables backwards": jstor.org/stable/2241604?seq=1 | |
| Dec 2, 2023 at 8:59 | comment | added | cdalitz | Is the frequentist justification not simply that the bootstrap distribution is representative for the distribution of the estimator and thus can be used to construct confidence intervals for it? This obviously assumes that the estimator is asymptoticlally unbiased, but this assumption typically holds for reasonable estimators. | |
| Nov 26, 2023 at 19:13 | history | edited | Closed Limelike Curves | CC BY-SA 4.0 |
Add link to more rigorous explanation
|
| Feb 7, 2023 at 18:45 | history | edited | Closed Limelike Curves | CC BY-SA 4.0 |
deleted 73 characters in body
|
| Nov 13, 2021 at 7:20 | review | Late answers | |||
| Nov 13, 2021 at 7:33 | |||||
| Nov 13, 2021 at 7:02 | history | answered | Closed Limelike Curves | CC BY-SA 4.0 |