[math] Wilcoxon Signed Rank - Why return the maximum W?

[Nathanael Costa <nc...@gmail.com>]

Hi,

I downloaded the version 3.0 and I noted that the Wilcoxon Signed Rank
returns the maximum W. But anywhere I read about this test says that
the minimum W should be returned. Am I missing something?

Another question is why do not ignore the rank when differences are zero?

Thanks in advance,

-- 
Nathanael de Castro Costa

Brasil

"No meio da dificuldade encontra-se a oportunidade." A. Einstein

---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@commons.apache.org
For additional commands, e-mail: dev-help@commons.apache.org

Re: [math] Wilcoxon Signed Rank - Why return the maximum W?

[Phil Steitz <ph...@gmail.com>]

On 5/30/11 2:54 PM, Nathanael Costa wrote:
> Hi,
>
> I downloaded the version 3.0 and I noted that the Wilcoxon Signed Rank
> returns the maximum W. But anywhere I read about this test says that
> the minimum W should be returned. Am I missing something?
>
> Another question is why do not ignore the rank when differences are zero?
>
> Thanks in advance,
>
Thanks for the feedback.  I am not sure I understand exactly what
you mean, though.  By "the minimum W" do you mean the minumum of W-
and W+ (per javadoc of the wilcoxonSignedRank method)?  I am sorry,
but I don't follow the second question either. 

The implementation follows the algorithm referenced in the
calculateExactPValue method.  This is new code, so it is quite
possible that we have missed something or are using non-standard
definitions.  The test cases (look in src/test) include validation
of the exact p value computation in one case against R.  It would be
great if you could provide a test case or external reference showing
us what is incorrect / nonstandard about the implementation.

Thanks!

Phil



---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@commons.apache.org
For additional commands, e-mail: dev-help@commons.apache.org