Hi Gang,
In talking with a statistician here, we've been advised to transform the z scores (obtained using Fisher's r to z transformation) to a standard gaussian distribution by dividing by the sqrt(n-3). In this way, we would also be adjusting for the dof of the original data. Since we know that the original BOLD time-series has alot of temporal autocorrelation, our dof (n) should not the number of timepoints in the BOLD data, but an estimate of the effective dof of the BOLD time-series.
Thanks,
Judith