Dear Alfonso,

Warm greetings.


I am currently using the conn toolbox for a functional connectivity, and the fisher-Z bivariate correlation matrix between the brain regions has some weights which are above 1. e.g 1.635.


I'm a bit concerned about the high weights especially since its a correlation between the brain regions.

Please what do you think or suggest? Is this expected? or is it possible i might be doing something wrong?


I would really appreciate getting a feedback. Thanks very much.

Kind regards,
Joan

Dear Joan,

Fisher transformed values range from -inf to inf. The transformation between Pearson correlation coefficients r and Fisher transformed scores z is r = tanh(z), so a z=1.635 value corresponds approximately to a r=0.93 correlation. That is a high value, it could be just fine if those values happen between two regions that are either overlapping or small and contiguous. If that is not the case (if such high values appear between many different regions) this likely indicates some residual confounder effects in the data which may have not been properly removed during denoising.

Hope this helps
Alfonso
Originally posted by Joan Amos:

Dear Alfonso,

Thanks very much for your prompt response and detailed explanation. It really does help.

Best,
Joan
Originally posted by Alfonso Nieto-Castanon: