 |
 |
oops! by hyperbola I meant Parabola :D Cheers On Mon, Apr 26, 2010 at 9:24 AM, Reza Salimi <[log in to unmask]> wrote: > Hi Gabor, > The Z-value in parametric cluster-based inference is not the Z-stat at > which you threshold for inference; it is your cluster-forming threshold. > In order to use random field theory (RFT), you need to define a threshold > above which you will have a group of surviving clusters. Given the curvature > of the field (smoothness/roughness/FWHM), cluster-forming threshold (i.e., > the Z=1.8) and the size of a surviving cluster, under the null hypothesis of > "your landscape is nothing but a standard Gaussian noise", you will get a > P-value, which of course is different from the P-value corresponding to your > Z threshold (i.e., the one you get from ztest(Z,0,1)) > > So, should we care about the Z? YES we should, as the lower this Z, closer > we are to violate the "clusters' geometry will have an approximate > hyperbola shape" assumption, which is an essential underlying assumption > of the RFT. > > So, it is safe to stay around FEAT's default value, i.e., ~2.5 > > Best, Reza > > > On Mon, Apr 26, 2010 at 9:12 AM, Gabor Perlaki <[log in to unmask]>wrote: > >> Hi all, >> >> What is the lowest Z for a group analysis if I use cluster-like >> thresholding? I've some results for a group with Z=1.8, P=0.05. The >> probability for Z=1.8 is P=0.0359, it is less than 5%, but I don't know I >> can believe it or not. >> >> Thanks a lot, >> Gabor >> > > > > -- > Reza Salimi-Khorshidi, > DPhil Candidate, Dept. of Clinical Neurology, University of Oxford (Linacre > College). > [log in to unmask] > FMRIB Centre, John Radcliffe Hospital, Oxford OX3 9DU > Tel: +44 (0) 1865 222704 Fax: +44 (0)1865 222717 > -- Reza Salimi-Khorshidi, DPhil Candidate, Dept. of Clinical Neurology, University of Oxford (Linacre College). [log in to unmask] FMRIB Centre, John Radcliffe Hospital, Oxford OX3 9DU Tel: +44 (0) 1865 222704 Fax: +44 (0)1865 222717