help - where can I post stat questions?

Art Eschenlauer eschen at biosci.cbs.umn.edu
Fri Jul 28 10:02:37 EST 1995


Hello:

Does anyone know of a newsgroup where I can post statistics questions?

Here's my problem:

I want to calculate a p-value for a hypothesis test of r, the correlation
coefficient. Specifically, if I observe r for n samplings, and I want to test
the hypothesis that rho (the  real value that r estimates) = 1, that is,
that the actual correlation is perfect, do I proceed as follows:

Ho: rho = 1.   Ha: rho < 1 (one tailed, because rho cannot exceed 1)
n observations of two variables X and Y. r is calculated.

Fisher showed that f(r) = ln( (1+r)/(1-r) )/2 is approximately normally
distributed with a mean of zero and a standard deviation of 1/sqrt(n-3).

     f(r)~N( 0, 1/sqrt(n-3) )

Therefore, If I understand the stats book correctly, my test statistic 

    Z = f(r) * sqrt(n-3)

and my p-value is the probability that the standard normally distrubuted
random variable z [ z~N(0,1) ] is greater than Z.

Is this correct?

Thanks for your assistance. Please email me if possible.

- Art (eschen at biosci.cbs.umn.edu)

--
eschen at biosci.cbs.umn.edu (Art Eschenlauer, U of M Agronomy & Plant Genetics)
Opinions expressed represent all living and inanimate beings in the universe.



More information about the Ag-forst mailing list