Cyril,
If I understand what you are saying, you are suggesting a fixed-effect
analysis where the A-B contrast is created for each of the conditions,
right?
I've been wondering about multisubject f.e. analyses, so maybe you can
clarify something for me...
I am also working with pilot data, so I set the experiment up to
eventually do a mixed effects analysis by creating a subject separable
design, but for now I am only looking at the overall group effects.
I've noticed that the subject-separable design has fewer degrees
of freedom than the non separable one....makes sense...So if I look at the
group
effects, doesn't this affect my t-thresholds because the d.o.f. used to
get the t-values are fewer? Should I choose a more liberal threshold when
looking at group effects within a subject separable design, or should I
just go ahead and created the non-separable design and trust those t-values?
Kerrie Tainter
-----Original Message-----
From: SPM (Statistical Parametric Mapping) [mailto:[log in to unmask]] On
Behalf Of cyril pernet
Sent: Thursday, September 21, 2006 11:32 AM
To: [log in to unmask]
Subject: Re: [SPM] paired t-test
=?iso-8859-1?Q?Rachel_Kozink?= a écrit :
>SPM experts,
>
>I am a first time user of SPM and wanted to double-check my design to
>ensure that I am performing my analysis correctly, especially since I am
>not getting the results that I had anticipated. I am using SPM2.
>
>Our study has 7 subjects so far (we are in the middle of the study and
>trying to analyze what we have so far to use as pilot data). Each subject
>is scanned twice and has condition 1 (C1) on one day and condition 2 (C2)
>on the other day. On each day, the subject performs two runs of the same
>task, which has two block conditions (A and B).
>
>I performed a first level analysis for each subject on each visit (e.g.
>Sub1, C1). I created contrast images that contrasted block A versus block
>B. I then took these contrast images and entered them into a paired t-
>test, where a pair consisted of 1) A-B contrast image for C1 and 2) A-B
>contrast image for C2.
>
>Just to clarify, it is my understanding that the p value and voxel extent
>thresholds you put in for the t-test after the first level analysis has no
>bearing on the con_* image that is created and used in the second level
>analysis. Is this correct?
>
>
Yes it is, the con images are the linare combinations of you beta images
- spmT images are the resulting statistical maps
>I am performing no masking and using very liberal p-values and cluster
>values and am still not seeing much activation. Does this mean that my
>data just doesn't have any significance or am I doing something wrong?
>
>
Everything looks ok to me - did you try with a single matrix per subject
par the 2 sessions?
Note also that 7 subjects may be not large enough to 'see' smthg - as it
is pilot data you can try to make one matrix (1st) with all subjects and
sessions ..
Best
Cyril
|