02-10-2017 07:51 PM
I am trying to determine confidence intervals for a correlation, but without using Fisher's Z as the Ho=0 and Ha(do not) equal 0. So no Fisher Z required.
02-12-2017 06:55 PM - edited 02-12-2017 06:57 PM
Why you wouldn't accept the probability under H0: rho=0 produced by PROC CORR, I have no idea. Even if it uses the z transform, wouldn't it produce the correct probability?
However, if you must. Using the "testing rho=0" section of Estimation and Confidence Intervals in notes for Introduction to Statistics at Andrews.edu, it looks like you can calculate t=r•sqrt((n-2)/(1-r2)), where r is your sample correlation. You can compare your value ot t against the critical values for n-2 degrees of freedom and alpha level of your choice.
Assuming you know your N and r in, say, dataset mycorrs, you can do this:
put 'Sample stats: ' r= n= sample_t= ' Prob >|r| under H0: ' prob_abs_r_eq_0;
if lower_critval <= sample_t <= upper_critval then put "H0: rho=0 is not rejected at " alpha=;
else put "H0: rho=0 is rejected at " alpha=;
Lots of extra stuff in this program, but it is not clear what precisely you want.
Added two minutes later: you really might want to put this question in the Analytics section, sub-section IML.