Thanks for your help. Again the original message was
"Correct Perm has max t 0 & rank 200 out of 200 completed permutations".
Thomas Nichols wrote:
> It's hard to tell what could be causing this; my first guess would
> be image data problem. Did an equivalent parametric analysis run OK?
Actually I don't know how to do a parametric analysis with one scan per
subject.
> You might try
>
> load SnPM_cfg
> load SnPM
> if (~bhPerms); MaxT=MaxT(:,1); end
> hist(MaxT(:),40)
>
> This will give you a plot of the permutation distribution. If it's just a
> bar at 0, then it would seem that you have all-zero data. Let me know
> what it looks like otherwise.
1) The histogram shows some permutations with value 0 but the distribution is
spread out across a range of MaxT values (0 to 0.74), not clustered at 0.
2) I can imagine that (since we're not doing a paired comparison) this group
comparison may have mostly negatives and no signif. positives, which may
explain part of these results.
3) When I run spm_snpm_pp with positives, even at a p=0.99 threshold, I get no
signif. pixels. But
when I run it with negatives, I get a big list of pixels with the max t value
being 3.09. Shouldn't this be counted in determining where in the list of
scans this comparison falls (i.e., shouldn't max(|t|) be the omnibus statistic
used rather than max(t))? How do I tell where t=-3.09 falls in the list of 200
scans? Can I generate a histogram for max(|t|) or at least max(-t)?
Thanks a bunch,
Kevin Black
|