simulants of the world unite [Nonparametrics]

posted by d_labes  – Berlin, Germany, 2010-09-15 12:14 (5262 d 15:50 ago) – Posting: # 5909
Views: 21,745

Dear Martin!

❝ you may find the following simulation results interesting

❝ ...

print(cover1/nsim)

[1] 0.937    # coverage of wilcox_test (coin)

print(cover2/nsim)

[1] 0.9505   # coverage of wilcox.exact (exactRankTests)

(comments by me)
Wow! Really interesting. And unbelievable :ponder:.
I was told that the Wilcoxon rank sum test is due to the discrete nature of the permutation distribution function always to some degree conservative (achievable confidence with n1, n2=10 is 0.956743).
And here it comes out that the implementation in the package coin is distinct anti-conservative! And the implementation in wilcox.exact does not show signs of conservativeness.

I have one small doubt. I'm not quite sure but should we not include the CI borders in counting the coverage due to the definition of the CI?
If we do this one gets (only 5000 sims, see BTW below):
[1] 0.9608   # coverage of wilcox_test (coin)
[1] 0.9608   # coverage of wilcox.exact (exactRankTests)

Moreover there seems to be some interaction between both packages. If I simulate without coin I get coverage of wilcox.exact = 0.9688 (including the CI borders in counting coverage).

BTW: Warning to all! The code is really a stress-test for the CPU of your computer (at least for mine). Submit the code and go for one and another beers. Next morning the result is astonishing enough :-D.
Therefore I could answer not until today.

Regards,

Detlew

Complete thread:

UA Flag
Activity
 Admin contact
23,379 posts in 4,913 threads, 1,661 registered users;
166 visitors (0 registered, 166 guests [including 11 identified bots]).
Forum time: 03:05 CET (Europe/Vienna)

Statistics is the art of never having to say you’re wrong.
Variance is what any two statisticians are at.    C.J. Bradfield

The Bioequivalence and Bioavailability Forum is hosted by
BEBAC Ing. Helmut Schütz
HTML5