> Hi Cyril,
>
> By saying "if you stick to p=0.001 default and then check the corrected cluster values, you should be fine" you mean 0.001 uncorrected,
yes you set 0.001 uncorrected (k=0)
> Actually I have a puzzle. If an activation blob is identified via 0.001, k=10, uncorrected in my fMRI experiment,
not all those blobs are correct - now you can only keep the ones with
cluster size or height FWE p<=0.05 (or FDR, but that doesn't control
type 1 error rate per se)
> but (1) my experiment has several trial conditions and this blob is only seen in a specific condition (i.e., it's highly selective), and (2) activity of this blob is also correlated with my behavioral measures which reflect underlying mental process (e.g. the performance of memory task), should I still have to stick on the 0.05 FWE or even more stringent threshold?
well, alpha is just a threshold - my view (other have similar but I'd
not generalize) is that interpretation should be based on context and
effect size
-- the alpha value is the long term prob to make a type 1 error / the
p-value is the observed prob to be = or bigger than a threshold under
the null --> not the same thing
-- stats maps can be converted to percentage signal change for effect
size (see eg
http://journal.frontiersin.org/article/10.3389/fnins.2014.00001/full)
-- stats maps should be looked at without threshold, and then you can
see which parts are significant (to me there is a big difference between
a significant blob which is the tip of a big mountain or if that blob is
the mountain)
-- a blob can be seen in one condition and not the other, that doesn't
mean that these two conditions differ ; you need to test that explicitly
(that means that this selective blob might not be as selective as you
thing -- in all cases plot parameter estimates or percentage signal
change for all conditions)
-- no matter what test is done (one sample or regression with
behaviour), as long as you do full brain analyses, yes you must do a
correction for multiple comparisons
Cyril
--
The University of Edinburgh is a charitable body, registered in
Scotland, with registration number SC005336.
|