> I've tried to do a little bit more research and have discovered actually quite a few articles
> about resting-state connectivity in which they plot t-statistic maps instead of mean
> z-transformed correlations...

What is prevalent in the field does not necessarily mean such a practice should be treated as something carved in stone. I would argue that the colored maps should even present the correlation values (converted back from z-values through the inverse Fisher transformation). To support the argument, let me quote a paragraph I wrote elsewhere:

"A statistic (e.g., t-statistic) depends on the effect amplitude, signal-to-noise ratio, and sample size. As it has no physical dimension, a statistic only serves the purpose of a binary inference between null and alternative hypotheses, and reporting significance alone could render a distorted representation of reality. For example, two voxels (or regions) with the same t-statistic value in the brain do not mean the same response amplitude (or correlation), and vice versa. The distorted impression from the colored (and thresholded) statistic values was strongly evidenced in recent surveys (e.g, Engel and Burton, 2013). It is the response amplitude or effective size, not the thresholded statistic value, that should be the primary product of scientific investigation. The omnipresence (and over-obsession) of focus on only statistic values (color-coded blobs of t-values), while ignoring the effect magnitudes, leads to a situation where one would be unable to gauge the false negative rate (or power) of the study, the probability of failure (or success) to detect the effect. Activation identification in FMRI data analysis heavily relies on the contrasting between conditions; however, the contrast between a significant effect and a nonsignificant one is not necessarily statistically significant. The lopsided focus on statistical significance (e.g., peak definition based on statistic value) may enforce two prevalent fallacies: 1) the probability of obtaining the current data given the null is the same as that of the null given the data; and 2) if the result is not statistically significant, it proves that no effect exists."

Gang



Edited 2 time(s). Last edit at 07/29/2014 02:00PM by Gang.