Tell me more ×
Cross Validated is a question and answer site for statisticians, data analysts, data miners and data visualization experts. It's 100% free, no registration required.
up vote 4 down vote favorite
1

I am interested in non-parametric methods for building confidence intervals for an estimator (e.g. the mean) using few samples (e.g. 10). I think I have read somewhere that smoothing the bootstrapped estimator values can improve the quality of the derived percentiles interval. However I could not find any online reference that explains how to tune the bandwidth of the smoothing step.

share|improve this question
1  
I am not an expert, but I do not think that smoothed bootstrap always performs better. Anyway, in some cases it might be and a standard smooth estimate of a continuous pdf is the kernel density estimation for which there is a huge amount of literature on how to choose the bandwidth h including rules which make reference to the normal distribution, or cross-validation methods. – ocram May 2 '12 at 5:11
1  
For example, you can find details here: google.be/… – ocram May 2 '12 at 5:11
1  
In line with @ocram's comment, you might be interested on comparing different types of bootstrap intervals. Take a look at this question. – user10525 May 2 '12 at 7:39

2 Answers

up vote 1 down vote

Sorry to be answering so late. This question came just when I joined CV and i only found it by looking back. For your specific question about using the bootstrap in kernel density estimation I think you will find material in Bernard Silverman's book. I think he covers the use of bootstrap for bandwidth selection.

Efron and Tibshirani discuss the bootstrap for finding modes of a density via kernel methods.

Oddly there is not really much on it in the general text on bootstrap including mine. Maybe in the next edition I will add something.

share|improve this answer
3  
I think the OP is asking a slightly different question which is about using kernel density estimation for producing improved confidence intervals when the bootstrap sample is small: "I am interested in non-parametric methods for building confidence intervals for an estimator (e.g. the mean) using few samples (e.g. 10)". – user10525 Jul 24 '12 at 16:15
1  
That was my impression as well @Procrastinator (+1). Michael, note that there's no need to apologize for answering late. In fact, I believe that attending to old questions is actually encouraged, as evidenced by the existence of the Necromancer, Revival and Archaeologist badges. – Macro Jul 24 '12 at 16:50
Thanks guys. I know I was not directly answering the question but it was my impression that the OP might have been a little misinformed or misinterpreted what he read. With respect to bootstrap confidence intervals there are many variations that adjust the endpoints. Those I am familiar with but not anything applying a smoothed bootstrap. Of course there is a smoothed bootstrap which involves sampling from a kernel density fit to the original sample. – Michael Chernick Jul 24 '12 at 17:52
up vote 1 down vote

Bootstrap confidence intervals

Thomas J. DiCiccio and Bradley Efron Source: Statist. Sci. Volume 11, Number 3 (1996), 189-228.

http://projecteuclid.org/DPubS?service=UI&version=1.0&verb=Display&handle=euclid.ss/1032280214

Keywords: Bootstrap-t; BCa and ABC methods; calibration; second-order accuracy

share|improve this answer

Your Answer

 
discard

By posting your answer, you agree to the privacy policy and terms of service.

Not the answer you're looking for? Browse other questions tagged or ask your own question.