14

Bootstrapping works well to access the uncertainty in the mean estimate, however I remember reading somewhere the bootstrap does not do a good job in assessing the uncertainty in quantile estimates (particularly the median).

I don't remember where I read this, and I couldn't find much with a quick Google search. Thoughts on this and any references would be greatly appreciated.

whuber
  • 281,159
  • 54
  • 637
  • 1,101
Glen
  • 6,320
  • 4
  • 37
  • 59
  • It sounds strange to me, as bootstrapping is how the `sqreg` (simultaneous-quantile regression) command in Stata estimates the standard errors. But this does not prove anything, I know. – boscovich Aug 28 '12 at 15:42
  • See also: Rogers, W. H. 1992. sg11: Quantile regression standard errors. Stata Technical Bulletin 9: 16–19. Reprinted in Stata Technical Bulletin Reprints, vol. 2, pp. 133–137. College Station, TX: Stata Press. --- Rogers, W. H. 1993. sg11.2: Calculation of quantile regression standard errors. Stata Technical Bulletin 13: 18–19. Reprinted in Stata Technical Bulletin Reprints, vol. 3, pp. 77–78. College Station, TX: Stata Press. – boscovich Aug 28 '12 at 15:46
  • 1
    The reference you mention might be related to (1) [A Note on Bootstrapping the Sample Median](http://projecteuclid.org/DPubS?service=UI&version=1.0&verb=Display&handle=euclid.aos/1176346731), (2) [Exact convergence rate of bootstrap quantile variance estimator](http://www.springerlink.com/content/v1kr16751768m216/) –  Aug 28 '12 at 16:07
  • 3
    I wonder if there was a miscommunication. It is well understood that the bootstrap works better in the middle of a distribution than at the tails. Thus, eg, bootstrapping the median would be the *most* robust quantile, whereas bootstrapping the min or max necessarily fails. You may find @cardinal's answer [here](http://stats.stackexchange.com/questions/9664/what-are-examples-where-a-naive-bootstrap-fails/9722#9722) to be of interest. – gung - Reinstate Monica Aug 28 '12 at 16:20
  • 1
    @Procrastinator Thank you for the two very relevant references that you cite. My book that I cite in my answer is loaded with references to bootstrap articles and both the references that you cite are listed in the book. – Michael R. Chernick Aug 28 '12 at 16:23

1 Answers1

13

The median can be bootstrapped and estimation of the median is a good application of the bootstrap. Staudte and Sheather (1990, pp.83-850 described here derive the exact calculation of the bootstrap estimate of the standard error of the estimate of the median that was originally derived in a paper by Maritz and Jarrett in 1978. Details of this can be found on pages 48-50 of my book on the bootstrap here on amazon.com.

Michael R. Chernick
  • 39,640
  • 28
  • 74
  • 143
  • 3
    (+1) Although a bit of attention has to be paid to the convergence of the bootstrap variance estimator as mentioned in references I posted. From (1) "The natural conjecture that the bootstrap variance estimator converges almost surely to the asymptotic variance is shown by an example to be false unless a tail condition is imposed on $F$". –  Aug 28 '12 at 16:13
  • 1
    @Procrastinator Yes it is good that you pointed out the mild restriction for consistency. Basically, it requires a moment alpha>0 to exist. – Michael R. Chernick Aug 28 '12 at 17:55
  • (+1) @Michael, I was expecting to see an answer from you in this question. –  Aug 28 '12 at 17:58
  • @Procrastinator Yes my eyes light up when I see the term bootstrap in the question. – Michael R. Chernick Aug 28 '12 at 18:05