permutation P value for very large BF

[Shizhong]

Hi Grant,

Using BIMBAM I got several large log10BF(around 20) in a candidate
region. In this situation, will it be possible to get the P value by
permuation? I worried the P value might be too small to get it by
permutation. Is this correct?

Thank you for your help,

Sincerely,


Shizhong

[Yongtao Guan]

Hi Shizhong,

At least you can get an upper bound for p-value. The p-value in bimbam
was calculated by (extreme_count + 1) / (total_permutation + 1) (in
version 0.95; the denominator will be changed to +2 in a new release).
If you use -pval 1000000 and get a p-value = 1/1000001=9.9999e-7, you
can report pval < 1e-6.

Grant

--
Yongtao Guan, PhD
Postdoctoral Fellow
Dept of Human Genetics
University of Chicago
http://home.uchicago.edu/~ytguan/

[Matthew Stephens]

basically you are correct that it will not be possible to get an
accurate p value.
You can proabbly feasibly do several million permutations, but you
will surely find that p is less
than any number of permutations you want to do.
You are lucky!

Matthew

On Mon, Oct 13, 2008 at 11:17 PM, Shizhong <hans...@gmail.com> wrote:
>