The RATS Software Forum

Hello!
I have a question.
Using RATS I have estimated the cross-spectra coherences between a number of variables. But now I need some critical values to determine which coherences are statistically significant. I have found a number of suggestions in the literature, most of which are somewhat unclear, Rory Thompson (1979) Journal of Atmospheric Science, and Fishman (1969) Spectral Methods in Econometrics, among others.
One formula (Thompson 1979) for the coherence critical value is: C_crit= sqrt(1- α^(1/(n-1)) ) , where alpha is the size of a type 1 error. But it’s not clear what n is. From the text of the article, n is clearly not sample size (my sample size is 518), because the author suggests that the ideal n is between 5 and 10.

So n could be the size of the spectral window. But from the RATS program crosspec.scr, it is not clear what the size of the spectral window is. So my question is, first, what should the actual cross-spectral coherence critical value be? Two, what is the value of n? And three, what is size of the spectral window?

Thank you very much for your help.
- David

The n is, indeed, the window width or (more generally if the window isn't flat) the equivalent band width. You can retrieve that as %EBW.

Hi David. I'm a RATS user like you. I'm going to use cross-spectral estimation for a research project. I see you've already run one and there are no examples on the Board. I'd appreciate it If you're willing to share your program or post it, since I normally have to contact Tom a few times to get the syntax right! It may speed things up for me. Please let me know either way. My email is adrangi@up.edu. Thanks much. Baram