Bookmark and Share

Notice: On March 31, it was announced that Statalist is moving from an email list to a forum. The old list will shut down on April 23, and its replacement, is already up and running.

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: st: Bootstrapping question

From   Austin Nichols <>
To   "" <>
Subject   Re: st: Bootstrapping question
Date   Fri, 8 Feb 2013 10:02:45 -0500

Henry Ilian <> :
See also section 7.4.2 in
for why inference for a multinomial probability minimizes distance on
the surface of a hypersphere.

On Fri, Feb 8, 2013 at 9:49 AM, Austin Nichols <> wrote:
> Bootstrapping is designed to improve CIs by making their coverage
> better, not by making them smaller! A better approach for your case
> would be Bayesian, but again a better CI does not necessarily mean a
> smaller CI. Some priors might result in smaller credible intervals,
> but others in larger, and you need to describe the dependence of your
> results on your assumptions. If you estimate 15 ways and only report
> the one that accords with the conclusion you want, you are committing
> scientific fraud.
> For the multinomial case, a flat prior is not the face of a simplex:
> see also section 5.2 in
> Depending on how serious you are about getting the smallest possible
> CI, you may need to do a lot of reading about maximum entropy methods.
> On Thursday, February 7, 2013, Ilian, Henry (ACS) wrote:
> Nick, you're right. Some of the potential (and actual) outcomes have
> observed zeros. I looked everywhere I could think of for the formula
> to compute sample sizes for multiple categories but couldn't find it.
> In the process, I read somewhere that the problem of multiple
> categories reduces to a two-category problem. I asked one statistician
> about this, and he said not true and suggested I take his on-line
> advanced sampling class. I certainly am considering that, but for the
> meanwhile, I still have a sample size that results in very wide
> confidence intervals. Again, my understanding from reading about
> bootstrapping is that one of the things bootstrapping was designed to
> do was to improve estimates of confidence intervals in small samples.
> My question is, can I use it in this situation, and if I do, what do I
> report?
*   For searches and help try:

© Copyright 1996–2016 StataCorp LP   |   Terms of use   |   Privacy   |   Contact us   |   Site index