In case you want to perform SVC on peak level you should go this way:
(1) Select your contrast, enter any threshold, be it corrected or uncorrected, this is irrelevant
(2) Define your small volume
(3) Look at the updated FWEp at the bottom up the results table, this is the critical T (or F) value for .05 FWE on peak level for the small volume
(4) Re-load the contrast, choose "none" for "p value adjustment to control", enter the T (or F) value from (3)
(5) Define your small volume like you did in (2)
If you don't do so, then the SPM is not thresholded properly, as it would still reflect e.g. .001 uncorrected on peak level or any other value that you might have entered during (1). As a consequence, depending on whether the corresponding T (or F) value is smaller or larger than the one reflecting .05 FWE for the small volume, you either miss voxels that would actually be sig., or you include voxels that are in fact non-sig. Accordingly, the k in the table would be incorrect, and the table might be incomplete, as peaks being non-sig. based on the threshold from (1) but sig. based on the threshold in (4) would not show up.
> btw, I found many articles did not describe the initial threshold they used in svc
I wouldn't be surprised if many of the SVC were conducted incorrectly = with an incorrect threshold. In addition, if people really had a-priori regions they should in fact start with the SVC and only then look at the whole-brain results for the rest of the brain. However, in all of the papers I remember it's the opposite, that is after failing to detect sig. effects on whole-brain level (or finding some in a subset of "a-priori" regions only) they turn to SVC. This doesn't make sense; if I had several a-priori regions and assuming I want to rely on SVC then I would want to combine those into a single mask and then look at all of them based on the SVC threshold.
Hope this helps
Helmut
|