Dear Eduardo,
Usually one would stick to one way of defining significance but if
some effect that you'd really like to report doesn't quite cross the
threshold but is significant according to a different criterion I
think you can mention that the same way that people report trends and
if your reviewers don't like it they will ask you to remove it. There
can definitely be cases when something is significant at the peak
level but not at the cluster level and vice versa. These two methods
are just sensitive to different kind of effects. As long as you decide
about your criterion in advance there is no problem. What is wrong is
to choose your criterion after you've already looked at the results
and know what will give you p<0.05. This kind of procedure will have
an inflated false positive rate.
See also this thread
https://www.jiscmail.ac.uk/cgi-bin/webadmin?A2=SPM;48daaf99.1601
regarding the current recommended practices.
The tool from Liege as far as I understand its description just helps
you to do something that you can also do without it, find an extent
threshold such that only significant clusters will be displayed. You
can do the same thing by looking at the results table and taking some
number in between the size of the smallest significant and largest
insignificant cluster.
Best,
Vladimir
On Fri, Apr 8, 2016 at 3:11 PM, Eduardo Ramirez <[log in to unmask]> wrote:
> Dear SPM experts,
>
> I've seen a couple postings about the reporting of peak vs cluster-level results, but only for imaging studies. If I'm only interested in analyzing time-frequency images, can I report both, or should I stick to one level? Also, what happens with cases where I obtain a significant cluster but no significant peak, is that meaningful/reportable? And lastly, does obtaining an extent threshold (e.g. as in https://github.com/CyclotronResearchCentre/SPM_ClusterSizeThreshold) make sense for this kind of data?
>
> Thank you very much,
> Eduardo
|