 |
 |
I don't know if there are any standardized procedures for SVC. I thought it was pretty standard to first use p < .001 (unc), and then when you do an SVC on your *a priori* region, you use FWE p < 0.05. For fMRI reporting in general, you can see Poldrack et al. (2008), NeuroImage, and Carp (2012), NeuroImage. For an alternative to FDR and FWE, you can look into TFCE (Smith & Nichols). You can also look into permutation tests. Regarding your question of a more liberal threshold: 0.05 is pretty much arbitrary. Fisher wrote: "The value for which P=0.05, or 1 in 20, is 1.96 or nearly 2; it is convenient to take this point as a limit in judging whether a deviation ought to be considered significant or not." I suppose any other choice for a cut-off would be just as arbitrary. You could argue that p < 0.10 suggests a trend, and warrants further study. But I think that, from the standpoint of a reviewer or reader of a Results section, "accepting as truth" a result with e.g. p=0.06 will depend on many aspects, including * data quality (spatial & temporal resolution, lack/presence of artifacts and motion) * number of subjects (maybe this is a pilot study) * homogeneity of subject group (with a more heterogeneous group, maybe it is unreasonable to be extremely strict) * fMRI task (if you're doing a bulk motor task with block design of ~15-30s length and don't see anything with FWE 0.05, there is probably something seriously wrong with your data) * using a cluster extent threshold (e.g. if you have a cluster of say 25 voxels, then maybe seeing p=0.06 is worth another look) and so on. And then there's always the possibility that your hypothesis is wrong, and unless you find a journal that will publish "negative results", you'd be out of luck. On 02/24/2014 06:57 PM, Andy Yeung wrote: > Dear Donald, Helmut, Chris, Marko and all, > > The discussion is really inspiring and interesting. Apart from cluster > size setting concern, shall we also talk about if there's a widely > accepted series of procedures to carry out SVC by SPM? We came across > with initial p thresholding, and voxel vs cluster SVC. Are there any > literature out there giving some answers to such field? I think it's > crucial to come up with a standardized procedures to perform SVC (as > well as standardized way to report) so as to reduce misunderstanding. > SPM8 manual only got a paragraph with three lines or something which > surely looks rather thin and abstract for beginners like me. > > And a major point raised here which worths further discussion is: > biological reality. We've been talking FWE 0.05 as a statistical > cut-off point for distinguishing significance. I strongly agree that > the surviving voxels can be noise or just tip of an iceberg. In our > current days, how is the acceptance for the ideas of more liberal FWE > threshold/ FDR threshold/ uncorrected p threshold but with cluster > size threshold? Matthew Brett suggested the ideas of more liberal > threshold or unthresholded SPM reporting to explore and display the > possibility of a bigger, if not whole, picture of brain activation. > (https://www.jiscmail.ac.uk/cgi-bin/webadmin?A2=spm;664eed4.05) But > that was 9 years ago already. Are there any changes in the mainstream > thoughts? > > Best, > Andy