0

As I understand the Bootstrap Method: you have a set of data which is of size $N_{all}$, and then from this set you select sub-samples of size $N_{sub}$.

Is there a rule of thumb for the size of the subset you should choose? For now I simply take the $N_{all} = N_{sub}$ as for the sub samples repeated values from the original main set may be selected.

Is there a smarter way to choose $N_{sub}$?

Q.P.
  • 248
  • 1
  • 13
  • 1
    @einar the thread that you linked is about how many times to do the resampling, when each resample is the same size as the original sample. This question has to do with the size of each resample. – EdM Jun 12 '19 at 13:38
  • @EdM I believe you get that impression because the asker of the linked question uses "samples" where they should have said "observations." I answered a more precisely worded variant of that question (https://stats.stackexchange.com/questions/263710/why-should-boostrap-sample-size-equal-the-original-sample-size/275746#275746) which also got closed as duplicate. – einar Jun 13 '19 at 19:37
  • @EdM ... actuallly I'm no longer certain exactly what the asker of that linked question is asking because it seems to contradict itself (they can make the bootstrap distribution arbitrarily narrow but they are also doing a "proper bootstrap sample.") The Q I linked with my own answer is definitely a duplicate of this one. – einar Jun 13 '19 at 19:43

0 Answers0