Stephan,
Thanks for handling my question. In response to the following:
>> If I am interpreting this 2nd model correctly, am I also correct in the
>> following? In this AnCOVA model, and given 2 groups and one nuisance
>> (covariate) variable, the contrast 1 -1 0 0 tells me where group 1 is
higher
>> than group 2 in activation (adjusted for the covariate).
>> The contrast 0 0 0
>> 1 tells me where the effect of the nuisance (covariate) is found in the
>> activations, collapsing across groups.
>What does the 3rd basis function in your model stand for?
The design matrix is: [group1 group2 u covariate], so the third term is
'mu'. I actually don't know
what this represents. The average of group1 and group2?
-----Original Message-----
From: Stefan Kiebel [mailto:[log in to unmask]]
Sent: Thursday, January 18, 2001 2:33 PM
To: Kareken, David A.; SPM
Subject: Re: Group Comparison w/ Co-variate
Dear David,
> I have a couple of basic design questions: First, I am doing a 2-group
> PET analysis, and my groups differ on one variable that might be
> contributing to the difference.
>
> Having individual subject *con.imgs already from a MultiSub PET model
> without a covariate, can I test for group differences and include the
> covariate in the 2nd level/Basic Models Section? For example, I noted
that
> the AnCOVA model appears to compare two groups and also requests a
nuisance
> variable.
If you have one measured scalar per subject, modelling this covariate at
the 2nd level is correct.
>
> If I am interpreting this 2nd model correctly, am I also correct in the
> following? In this AnCOVA model, and given 2 groups and one nuisance
> (covariate) variable, the contrast 1 -1 0 0 tells me where group 1 is
higher
> than group 2 in activation (adjusted for the covariate).
> The contrast 0 0 0
> 1 tells me where the effect of the nuisance (covariate) is found in the
> activations, collapsing across groups.
What does the 3rd basis function in your model stand for?
>
> Is it prefereable to instead include the covariate in the PET model that
> generates the *con.imgs (i.e., MultiSub: Conditions x Subject Interaction
&
> Covariates). Including the covariate (nuisance) variable at the 2nd level
> seems to permit me to see where in the brain the covariate is having its
> effect, whereas I don't see how to do that in the first level analysis
> (i.e., I don't see an estimable contrast for the covariate).
Given that you have one scalar value for each subject, I would take the
contrast images of your choice to the 2nd level and choose a model, with
which you can model the interaction between conditions and covariate. I
don't know whether there are better solutions, but multi-subject cond x
subj interaction & covariates seems to do it. Choose all your scans
(from both groups) and 1 subject, model group by condition and enter
your covariate. Choose interaction by condition. Make sure that you
don't choose global normalization, because you did this already at the
1st level. You can test then differences in the group main effect by [-1
1 0 0] and the interaction group x covariate with [0 0 -1 1].
Stefan
--
Stefan Kiebel
Functional Imaging Laboratory
Wellcome Dept. of Cognitive Neurology
12 Queen Square
WC1N 3BG London, UK
Tel.: +44-(0)20-7833-7478
FAX : -7813-1420
email: [log in to unmask]
|