 |
 |
Dear Takanori, > I am interested in the GRF and statisitcs in the spm. I carried out > some simulations to understand thresholding processes in spm and had > questions about spatial extent test. > > And I plot the theoretical probability of the spatial extent test when > simulated images is thresholded at different Z values and different > areas n. > > ...... > for u=1:500;for k=1:100;pnumax(u,k)= spm_P(1,[8 8],u*0.01,k,128*128);end;end > ...... > > From the graph,the uncorrected p-value is greater than the corrected p-value > by the combination of a certain u and k. > Why is this? I think one explanation for this anomaly is that you have used very low thresholds. This violates the assumptions of GFT. In our simulations we generally use thresholds of about Z = 2.4 or more. Your thresholds go down to 0.01. GFT results are only assymtopically correct at high thresholds. > In this simulation, the probability of getting one or more clusters > of size k or more, above high threshold u, almost corresponds to > the corrected p-value. This probability is the corrected p value. With very best wishes - Karl %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%