 |
 |
Hi again Matthew, Just a quick note that using a z-stat = 3.1 threshold as the value of <thresh> appears to give me better agreement with the SPM implementation, compared to when I use a t-stat = 3.38 as input instead (where 3.38 I think corresponds to p = 0.001 for my group of 31 subjects in a one-sided scenario. I used the p-value calculator here: http://www.socscistatistics.com/pvalues/tdistribution.aspx). My hunch is that <thresh> should use a z-stat, but maybe a lot of the time (with much larger group sizes) z and t are very close such that it makes no difference. Cheers, Donal On 29 Mar 2016, at 16:28, Hill, Donal <[log in to unmask]<mailto:[log in to unmask]>> wrote: Hi Matthew, I’m trying to compare what randomise with -c <thresh> gives me compared to similar permutation implementations in SPM. In SPM, I explicitly provide an uncorrected p-value threshold of 0.001. If FSL requires a t-stat threshold instead, I need to know what the equivalent t-statistic for a p-value threshold of 0.001 is. This isn’t completely trivial, as it depends on the number of subjects. It would be more unambiguous if the <thresh> argument was, for instance, a z-statistic value. In that case, I could use <thresh> = 3.1 to match the p = 0.001 requirement. Is there a reference somewhere for the exact form of the -c <thresh> argument i.e. is it a t-statistic or a z-statistic? I’ve looked high and low and can’t find any mention of it! Thanks for your help, Donal On 29 Mar 2016, at 16:20, Matthew Webster <[log in to unmask]<mailto:[log in to unmask]>> wrote: Hi Donal, the threshold is the t-statistic value you want to threshold at. Thresholding at a given uncorrected-p would require modifying randomise to run twice, using the voxelwise t-statisic null-distributions from the first run to apply p-thresholds for the second run. Kind Regards Matthew Dear experts, I am trying to use the -c option within randomise to perform cluster-based thresholding. I am wondering what the input argument <thresh> should be in this instance? I would like to threshold my statistic images at an uncorrected p-value of 0.001. Can I supply 0.001 as the value of <thresh>, or do I need to supply a t-statistic value? Kind regards, Donal