 |
 |
Dear Victoria, As far as I understood, you're looking at *cluster-level* significance. Cluster-level significance usually varied with the predefined threshold due to greater / smaller number of voxels that survive the control level of significance.However, peak-level significance is independent of initial threshold. Regarding reporting results, p < 0.05 FWE-corrected on the *peak-level*after SVC is usually used. "An ideal or widely-accepted initial threshold" prior to SVC depends on the ROI size that is used for SVC. Hope this helps, Ella On Thu, Jan 16, 2014 at 7:15 AM, Victoria Klimaj <[log in to unmask]>wrote: > Hi SPM experts, > > I have a question regarding SVC analyses and initial thresholding that > I've seen asked in the SPM archives before, but also have some follow-up > questions to: > > I notice that when I do SVC analyses, I often get different results > depending on what the initial whole-brain threshold is set to. Sometimes I > find P-FWE results in an SVC analyses when the brain is initially set to a > .001 uncorrected threshold, but such results do not hold when the brain is > initially set to .05 FDR or FWE instead. This has confused me, as I thought > the the small-volume correction was independent of the whole-brain > thresholding. Am I understanding small-volume corrections incorrectly? What > would be the statistical explanation for these threshold-dependent > differences in SVC results? > > Regarding reporting results, is there an ideal or widely-accepted initial > threshold that should be used before performing small-volume corrections? > And is a P-FWE cluster result obtained for a small-volume analysis within a > .05 FDR thresholded whole brain more valid than a similar result obtained > from a .001 uncorrected brain? Can such cluster results be reported in the > same manner? > > Thanks for your help! > > Best, > > -Victoria > -- Ella Gabitov Ph.D. student The E.J. Safra Brain Research Center for Learning Disabilities The Laboratory of Human Brain and Learning University of Haifa Israel Phone: +972-52-6722700 E-mail: [log in to unmask]