 |
 |
Dear Tom, I think there are two things that need to be shown to statisfy all curiosity. A map of %signal change and a logmap of T/Z statistic (or a map of variability). I am generally in favor of a %signal change map but the problem lies in the fact that a small change but consistent change would not necessarily be detected in a %signal change map. Consistency is what the statistical map shows. That is why Matthew showed one with the acitivity map on which the threshold was superimposed. We can extend that to show bands for -log(pvalues) on it, similar to a terrain map with contour lines on it. From a functional standpoint we get two different pieces of information from the two maps. One tells us "how much" and the other tells us "how reliably". Satra On Tue, 15 Mar 2005 17:07:17 -0600, IAIN T JOHNSTONE > One aspect of the discussion on thresholded versus unthresholded maps > that has been more or less lost in the to-and-fro is whether images > should use a colour scale to depict a statistical quanitity (e.g. Z or > T), or whether a unit more related to magnitude of signal change (e.g. % > signal change) should be reported (this applies equally to graphs of > extracted cluster activation). I would argue that the latter is a better > technique, because it gives us some indication of whether the failure of > a region to achieve statistical significance has more to do with > variability, or with lack of signal change.