Ooups, just realised a mistake in my previous message: instead of the
con-image, it has to be the spmT-image!
Hi Iris,
One thing to bear in mind is that when you create a one-way ANOVA with the
two contrasts, you end up with twice as many degrees of freedom, therefore
if you select the two contrasts at p<.001 in this models, this corresponds
to a lower threshold in the single contrasts compared to when they were
done as normal one-sample t-tests. This difference (in the direction of
being too liberal) becomes more and more pronounced the more contrasts one
would include in this masking precedure.
There is a way to do inclusive masking without creating a common model for
the two. Run one-sample t-test models for c1 and c2, and create your
t-contrasts. Then write the results of one of them (e.g, c2) out as an
image (using the save function) at the threshold that you want to use. Then
use imcalc, select the con-image for c1 and then the created mask-image for
c2, and evaluate a function like: i1.*i2./i2. If everything goes right, the
model for c1 now includes the new masked con-image.
Best,
Evelyn
> After I posted my first message I got some help and I decided to pass
>on the following information that might prove useful to others:
>"""""""""") does sth different… "". You can not do anythin clever as
>in doing any between subjects designs (e.g. rep meas t-tests, not even 2
>sample t-tests and not rep meas Anova) as this leads to unestimable
>contrasts. You are allowed to correct for non-sphericity, though, if that
>reliefs you.
>So, one-way ANOVA, two groups, select all c1s of all subj. in group 1 and
>all c2s of all subj. in group 2 etc.
>then proceed as I first intended to, i.e. do two second level t-contrasts
>i) 1 0
>ii) 0 1 and then look at one masked (incl.) with the other. Select same
>(uncorr.) threshold for both.
> I would very much appreciate if, in case any SPM geak saw a flaw in this
>procedure, he or she could hint to it. Many thanks Iris
>
-----------------------------------------------
Dr Evelyn Eger
Wellcome Department of Imaging Neuroscience and
Institute of Cognitive Neuroscience
Alexandra House
17 Queen Square
London WC1 N3AR
phone 020-7679-1123
fax 020-7813-2835
email [log in to unmask]
|