Subject: | | full flexible anova spm5: covariate |
From: | | "<Ellen> <Greimel>" <[log in to unmask]> |
Reply-To: | | [log in to unmask][log in to unmask]] On Behalf Of PET Newbie >Sent: Thursday, September 06, 2007 8:15 PM >To: [log in to unmask] >Subject: [SPM] Calculating DVR's in PET > > >Looking for some advice. >I have a mood induction experiment with many subjects, 1 group, and 2 >conditions. >In SPM I have created T-Contrasts to produce activation (condition A - >Condition B), and have identified desired ROI's. >Everything seems to have gone smoothly up until this point. >Then, I try to extract the DVR's from the respective Subject's >ROI's and >calculate BP with BP=DVR-1. This is where things look a bit >fishy...I >get some individual subject negative values and if I average per >condition, for some ROI's I get negative average BP's as well. I can't >make any sense of this. I would expect that I should only >have positive >BP's (at least positive averages) for regions that were found to be >significant on the SPM analyses. Anyone have any suggestions. > >Thanks so much, > >PET Newbie >38_7Sep200717:07:[log in to unmask] |
Date: | | Wed, 19 Sep 2007 13:04:54 +0200 |
Content-Type: | | multipart/mixed |
Parts/Attachments: |
|
|
|
|
Dear SPM experts,
I have already posted my request some time ago and still would very much
appreciate your advice:
We have some questions regarding the covariate option in the full flexible
anova model in spm5.
We have a
2(group) x 2(within-subject factor stimulus type) x 2(within-subject factor
task) factorial design plus a high-level baseline.
The groups differ in one demographic variable. We thus want to include this
variable as a covariate, in order to rule out the possibility that
differences between groups in any activation patterns can be traced back to
differences in this particular demographic variable.
In spm 5 we set up:
Design: Flexible factorial
-Factor 1: condition
Independence: no
Variance: unequal
-Factor 2: subj
Independence: yes
Variance: unequal
-Factor 3: group
Independence: yes
Variance: unequal
-Main effects & interactions:
Main effect: factor number: 2 (i.e. subject)
Interaction: factor number 3 1 (i.e. condition x group)
-Covariate: Vector: the corresponding value for each subject.
Interaction: None
Centering: overall mean
We then set up contrasts that allow for comparisons between conditions
within a group, or for condition x group interactions.
When we then look at the results and compare them with the results we
yielded with the regular Anova without the covariate, the results do not
differ at all, not in any voxel and any parameter.
We suspect that this may be due to the fact that the covariate can be
constructed as a linear combination of the subject regressors and therefore
the parameter estimates for the condition-related regressors may remain
unchanged. Could that be the reason that our results don't get affected at
all when including the covariate?
We are now unsure how to correctly set up the model, in order to control
for subject-related variance that can be explained by a demographic
variable.
Does anybody have ideas concerning this problem?
I attach the printouts of the two design matrices (anova and ancova).
Thanks a lot for your help,
Ellen
(See attached file: 5-Design_ancova_neu.zip)(See attached file:
6-Design_anova_neu.zip)
|
|
|
|