Hi Carlos,
>
> 1. Is it possible to do the group contrasts as I had originally
> performed
> them? If so, are there any disadvantages to this?
>
By this I assume you mean "Originally I thought the correct way to
perform a group comparison was to first estiamte the activation for
each of the groups and then contrast the 2 resulting .gfeat
directories".
I am not overly familiar with thow SPM sets up its group stats, but
the FEAT philosophy is to model each level in the group hierarchy
using a GLM and then have contrasts doing statistical tests on the
regression parameter estimates. With this approach, as you say, the 2
sample unparied t-test is the straightforward way to get what you
need: the GLM models the groups, and a (1 -1) contrast will perform
the group comparison. Doing separate 2nd level analysis on each group
and passing up group means to a 3rd level, where you compare the group
means, will be equivalent.
> 2. I don't quite understand why the setup for the unpaired has 2 EVs
> which
> seem to specify which group the subjects belong to, when there is
> already a
> Group column which specifies just that.
> 3. Concerning the above, how do the Group and EV variables interact?
>
The Group column on the setup GUI corresponds to variance groupings,
e.g. which subjects will be assumed to have the same between-subject
variance. This is distinct from modelling group effects, such as group
means, which are modelled using the EVs in the design matrix. This
gives the user freedom to specify if, for example, different groups
are assumed to have the same between-subject variance or not.
> I apologize if these questions seem rudimentary, but I guess my
> previous
> experience with SPM is preventing me from understanding the FSL
> method.
No worries - I hope the above helps to clarify some of the issues.
Cheers, Mark.
----
Dr Mark Woolrich
EPSRC Advanced Research Fellow University Research Lecturer
Oxford University Centre for Functional MRI of the Brain (FMRIB),
John Radcliffe Hospital, Headington, Oxford OX3 9DU, UK.
Tel: (+44)1865-222782 Homepage: http://www.fmrib.ox.ac.uk/~woolrich
On 2 Feb 2009, at 18:31, Carlos Faraco wrote:
> Dear FSL Users,
>
> I have a decent amount of experience with FEAT, but have never used
> it to
> perform group comparisons and would like some clarification on
> setting up
> the EVs.
>
> We have a data set with 20 subjects: 10 control and 10 concussed
> individuals. Each subject completed a number of different tasks, but
> with
> only 1 run per task. Originally I thought the correct way to perform
> a group
> comparison was to first estiamte the activation for each of the
> groups and
> then contrast the 2 resulting .gfeat directories (similar to what is
> done is
> SPM). Apparently this either doesn't work in FSL or I am setting it up
> incorrectly, as I do not end up with any cope directories.
>
> Going through the FEAT manual, it seems the setup provided for the Two
> Sample unpaired T-test (which is a 2nd level analysis, not a 3rd as I
> originally thought I would need) is the most similar to what I need.
> But I
> have a few questions about this.
>
> 1. Is it possible to do the group contrasts as I had originally
> performed
> them? If so, are there any disadvantages to this?
>
> 2. I don't quite understand why the setup for the unpaired has 2 EVs
> which
> seem to specify which group the subjects belong to, when there is
> already a
> Group column which specifies just that.
>
> 3. Concerning the above, how do the Group and EV variables interact?
>
> I apologize if these questions seem rudimentary, but I guess my
> previous
> experience with SPM is preventing me from understanding the FSL
> method.
>
> Thanks,
>
> Carlos
>
|