Dear Esa,
Esa Wallius wrote:
> Dear SPM'ers
>
> I have a few more questions concerning the multisubject correlation
> analysis.
>
> Does "Covariates only: interaction with subject" model take into account
> the dependence within subject?
> If it does, how does it do that?
> This is important because we have 8 scans per subject.
>
> We specified the following contrasts (11 subjects) to test
> 1) positive correlation 1 1 1 1 1 1 1 1 1 1 1
> 2) negative correlation -1 -1 -1 -1 -1 -1 -1 -1 -1 -1 -1.
> Are these t-contrasts correct for testing the average positive and
> negative correlation?
>
> I wonder whether these results are fixed effects analyses, that cannot
> be generalized into population level.
>
You are correct that this is a fixed effects analysis, and cannot strictly
speaking be generalised into a population level inference.
The way to perform a random effect analysis on this study is to generate a
contrast parameter estimate map for each subject (by simply looking at
contrasts [1 0 ... 0], [0 1 0 ... 0] ... [0 .. 0 1] in the results
section). You then perform a one-sample t-test (i.e. compare the average of
your parameter estimates with zero) on these (con_00*.img) maps. You do the
same thing with the [-1 0 ... 0], [0 -1 0 ... 0] ... [0 .. 0 -1] contrasts
to check the negative correlations.
The RFX model will tell you if all subject "activate" in the same location
and with roughly the same magnitude. If you want to answer the slightly
less stringent question "do they all activate in the same location?" you
can use a conjunction across subjects instead. Its real easy. Once you have
entered all the individual contrasts above you simply select them all
(positive and negative separately) in the results section using the control
button.
>
> Are there theorethical references concerning the multisubject
> correlation analysis with correlated scans?
> We have been searching such articles, but didn't find any.
>
Any stats textbook will contain sections on multi-level statistical models.
In neuroimaging I think we are still waiting for the full paper on RFX.
>
> Thanks in advance
> Esa Wallius, Turku PET centre, Finland
Good luck Jesper
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|