Serge,
> I included EVs for subject means (EV3 - EV12) since this is analogous to the
> webpage example of the paired t-test (if I am not mistaken).
> In my design cope1 and cope2 are the paired observations. But perhaps it
> does not make much of a difference if the pairedness of the data is not
> modeled.
That's right - you model out the subject means (and treat them as
confounds) when you are interested in the difference between pairs of
observations in a paired t-test. From what I understood, the question you
want to ask is on the group means calculated _separately_ for cope1 and
cope2. Hope that makes sense.
Cheers, Mark.
Mark Woolrich.
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 Thu, 2 Dec 2004, Rombouts, S.A.R.B. wrote:
> Dear Mark,
>
> Thanks very much.
> I included EVs for subject means (EV3 - EV12) since this is analogous to the
> webpage example of the paired t-test (if I am not mistaken).
> In my design cope1 and cope2 are the paired observations. But perhaps it
> does not make much of a difference if the pairedness of the data is not
> modeled.
>
> Cheers,
> Serge.
>
>
> -----Original Message-----
> From: Mark Woolrich [mailto:[log in to unmask]]
> Sent: Thursday, December 02, 2004 10:30 AM
> To: [log in to unmask]
> Subject: Re: [FSL] second level analysis: repeated measures with covariates
>
>
> Dear Serge,
>
> > I am trying to do a second level analysis with repeated measures per
> > subject. The simplified version is: a group of 10 subjects, consisting
> > of two subgroups of 5 subjects each, and 2 copes per subject (in
> > reality I have 3 groups, 87 subjects and 7 copes).
> >
> > Copes were generated using a model with orthogonal regressors at first
> > level. I want to apply an analysis with 2 repetitions per subjects
> > (the two copes) and test whether cope1 distinguishes the two subgroups
> > of 5 subjects better than cope 2.
> > My design then is as follows:
> > group EV1 EV2 EV3
> > EV4 ... EV12
> > subject 1, cope 1 1 1 0 1
> > 0 0
> > subject 1, cope 2 1 0 1 1
> > 0 0
> > subject 2, cope 1 1 1 0 0
> > 1 0
> > subject 2, cope 2 1 0 1 0
> > 1 0
> >
> > ......
> >
> > subject 9, cope 1 1 -1 0 0
> > 0 0
> > subject 9, cope 2 1 0 -1 0
> > 0 0
> > subject 10, cope 1 1 -1 0 0
> > 0 1
> > subjects 10, cope 2 1 0 -1 0
> > 0 1
>
> > Contrasts:
> > 1 0 0 0 0 0 0 0 0 0 0 0 : first 5 subjects > last 5 subjects for cope
> > 1
> >
> > 0 1 0 0 0 0 0 0 0 0 0 0 : the same for cope 2
> >
> > 1 -1 0 0 0 0 0 0 0 0 0 0: is the difference between the two subgroups
> > greater for cope 1 than for cope 2?
> >
> > etcetera
>
> I cannot see why you have included EVs for subject means (EVs 3-12). You
> should replace them with EVs for the cope means:
>
> group EV1 EV2 EV3
> EV4
> subject 1, cope 1 1 1 0 1
> 0
> subject 1, cope 2 1 0 1 0
> 1
> subject 2, cope 1 1 1 0 1
> 0
> subject 2, cope 2 1 0 1 0
> 1
>
> ......
>
> subject 9, cope 1 1 -1 0 1
> 0
> subject 9, cope 2 1 0 -1 0
> 1
> subject 10, cope 1 1 -1 0 1
> 0
> subjects 10, cope 2 1 0 -1 0
> 1
>
> then a [1 -1 0 0] contrast should give you what I think you are after.
>
> > So far I am happy with my design. But I need to include one covariate
> > for age and one for gender as well. The demeaned age EV (EV13) would
> > be something like: 12
> > 12
> > -7
> > -7
> > ...
> > -11
> > -11
> > 6
> > 6
>
> Yes, you can simply put this in as an extra EV.
>
> > To avoid rank deficiency, I can remove EV12 (the EV of subject 10),
> > but I am not sure whether this is allowed.
>
> with EVs3-12 removed this shouldn't be a problem.
>
> Best Regards,
>
> Mark Woolrich.
>
|