I see - thanks very much for clarifying this!
Hi - right, that is the correction for degrees of freedom taking into account autocorrelation - ie the "standard error" part you can see a few equations into https://en.wikipedia.org/wiki/Fisher_transformation
The adjustment for autocorrelation is done empirically via the data.
If you want to turn this scaling off then just change the "1" to "-1" - see help usage.
If you don't have that option then just upgrade your version of FSLNets.
On 22 Aug 2017, at 21:20, SUBSCRIBE FSL Anonymous <[log in to unmask]> wrote:
Thanks for the reply, I realize that the z score can be potentially infinite, however when I look up the z-score equivalent of r= .5155, it should be z = .5702, not z = 5.1373 - to get such a high z-score, my r value would have to be closer to .99995 (assuming z = (1/2)[log(1+r) - log(1-r)] ). Does this mean something has gone wrong?
Hi - yes that's the point of the (nonlinear) transformation - instead of being bounded as r (-1 : 1) Z is (-inf : inf). The closer r gets to one, Z really takes off...into the tail of the Gaussian.
On 22 Aug 2017, at 04:39, SUBSCRIBE FSL Anonymous <[log in to unmask]> wrote:
I'm wondering if I'm interpreting the results of nets_netmats (for FSLNets) incorrectly. I understand that if I use, e.g. netmats=nets_netmats(ts,1,'corr'), then the "1" will facilitate a Fisher r to z transform. However when I compare the untransformed matrix of r values to the transformed matrix, I see that, for example, r=.5155 gets converted to z=5.1373, and r=.9290 gets converted to 14.8730! How can the z-scores be so high? Am I interpreting these results incorrectly somehow?
Thanks very much!