Ged Ridgway schrieb:
> Hi Cédric,
>
>> The question of how to deal with subjects not activating the ROI 
>> remains...
>>
>> -set the threshold to 1 and extract the VOI time-series ?
>>
>> -choose a liberal threshold (say 0.1) and exclude subjects without 
>> activation ?
>
> I'm no expert on this, but I think that either should be okay...
>
> I think the former would be more in keeping with standard statistical 
> practice. For example, in a summary statistic approach where straight 
> lines where fitted to within-subject data, and then slopes compared 
> across e.g. groups of subjects, I *think* it would be very 
> non-standard to exclude any subjects from the group analysis if their 
> regression slopes where insignificant according to any criterion.
>
> However, I think the key idea behind spm_regions is that the contrast 
> you subsequently investigate over the VOI is orthogonal to the 
> contrast used to select the VOI (e.g. define VOI based on main effect 
> of slope, then test the orthogonal slope-group interaction). So I 
> think you can therefore exclude subjects without biasing the 
> subsequent analysis.
>
> So I think my answer to this might have to be the rather woolly 
> "you'll have to decide based on what makes more sense to you in your 
> experimental context". Though I do hope that more statistically 
> knowledgeable people post more clear-cut answers to the list!
>
> Ged
Hi Cedric and Ged!
The more I think about it, the more interesting it gets. :-)  
Nonetheless, I would rather protest against the former suggestion of 
Cedric, that is,
to set the threshold to 1. If you extracted from a sphere of 0mm in the 
voxel taken from the group maximum activation, the threshold would make 
no difference.
But as soon as you extended your sphere size and set the threshold too 
liberal, you would include voxels that only showed a very very weak effect.
That will lead to unevitable problems in the following sense:
As far as I understood it, with the PPI you would try to investigate if
a) the contribution of the VOI region to another region X was higher (or 
lower) in the context of a psychological manipulation (namely, in the 
context of task A) or with another reading if
b) region X showed a greater sensitivity (or lower) to main effect A 
when activitity in the VOI region was high
As a prerequisite, your VOI region has to demonstrate a significant main 
effect in task B, since the effect of task B in region X is substituted 
(explained) by activity
in the VOI region and the pyscho-physiological interaction in region X 
is the product of activity in VOI (main effect task B) times task A.  If 
there were no activity (main effect)
in the VOI region related to task B,  there'd be no interaction since 
the product would be zero. That would mean including too many subjects 
or voxels demonstrating no
strong effect for task B would make it rather difficult for the PPI to 
be detected. That's why i would suggest to choose not a too liberal 
threshold and exclude subjects without
activation. What do you think?
Best
Markus

-- 
Dipl.-Psych. Markus Staudinger
Social Cognitive Affective Neuroscience Group
Department of Psychiatry
Division of Medical Psychology
University of Bonn
Sigmund-Freud-Strasse 25
D- 53105 Bonn
Germany

Tel: +49 (0)228 287 19709
Fax: +49 (0)228 287 16097
E-mail: [log in to unmask]