When we calculate sample - sample correlations, we actually have a bunch of correlations for each sample. Right now, we take the median, and then subject that to an outlier analysis within groups.
But, really, we could be testing if the distribution of some samples correlations are different than others. Maybe? Worth looking into. We have all these values, and we are limiting them to the medians for each one.