1

I am using the bootstrap to calculate 95% confidence intervals for the median of a population. The data can only take positive integer values. My sample indicates a fair bit of variability, though the data are definitely clustered around a single value close (but not equal to) zero.

Under these circumstances, can bootstrapping the median go wrong? I have seen warnings against bootstrapping statistics when the population is either discrete or asymmetric, but I have yet to see any good explanation why these properties would cause the bootstrap to fail/malfunction.

Thanks!

holomorphic32
  • 11
  • 1
  • There are some situations where the bootstrap doesn't work well but not because a distribution may be discrete or skewed. – Michael R. Chernick Jul 14 '19 at 21:39
  • Without knowing more about the distribution of your data it is hard to judge. Note that the [breakdown point](https://en.wikipedia.org/wiki/Robust_statistics#Breakdown_point) of the median is 0.5, which could in some circumstances lead to practical problems with bootstrapping even the median. Please look at [this answer](https://stats.stackexchange.com/a/220028/28500) to a similar question, and its links, and edit this question with more information on your data while specifying particular issues that are still unresolved. – EdM Jul 14 '19 at 21:45

0 Answers0