OK - yep, that all makes sense then.
Cheers.
On Wed, 28 Jan 2004, Rombouts, S.A.R.B. wrote:
> Hi Steve,
>
> Thanks: so I should use:
>
> Group EV1 EV2 EV3 EV4 (age)
> 1 1 0 0 0.25
> 1 1 0 0 0.25
> 1 0 1 0 -0.75
> 1 0 1 0 0.5
> 1 0 0 1 -0.5
> 1 0 0 1 0.25
>
> Further, my model was simplified: my actual design contains 90 subjects
> divided over three groups. That is why I tried to model variances separately
> for each group.
>
> Regards, Serge.
>
>
>
>
> -----Original Message-----
> From: Stephen Smith [mailto:[log in to unmask]]
> Sent: Wednesday, January 28, 2004 1:29 PM
> To: [log in to unmask]
> Subject: Re: [FSL] covariates of no interest at higher level
>
> hi - that's nearly right, but you don't actually have the means in the
> final model! You are right that you will need to set the group memberships
> all equal (in fact if you only have 2 members in each group you shouldn't
> be modelling the variances separately in the different groups separately
> anyway). The model you should use is the original one with it's 3 EVs and
> then add on the demeaned age EV as a 4th EV.
>
> Cheers, Steve.
>
>
>
> On Mon, 26 Jan 2004, SARB Rombouts wrote:
>
> > Hi Steve,
> >
> > Thanks for the reply.
> > This is what I did first (without age as covariate):
> >
> > Group EV1 EV2 EV3
> > 1 1 0 0
> > 1 1 0 0
> > 2 0 1 0
> > 2 0 1 0
> > 3 0 0 1
> > 3 0 0 1
> >
> > Contrasts:
> > A-B = EV1-EV2 = 1 -1 0
> > A-C = EV1-EV3 = 1 -1 0
> > B-C = EV2-EV3 = 0 1 -1
> >
> > Unfortunately, this model does not allow me to include one extra covariate
> > for age, since I have 3 groups, and each EV should contain non-zero values
> > for one group only, the other values must be zero....
> >
> > So the best model with age as covariate I could think of, was to pretend
> as
> > if all subjects belong to 1 group:
> >
> > Group EV1 EV2 EV3 (age demeaned)
> > 1 1 1 0.25
> > 1 1 1 0.25
> > 1 -1 0 -0.75
> > 1 -1 0 0.5
> > 1 0 -1 -0.5
> > 1 0 -1 0.25
> >
> > Contrasts:
> > A-B = EV1 = 1 0 0
> > A-C = EV2 = 0 1 0
> > B-C = EV2-EV1 = -1 1 0
> >
> > Is this how I should do it, or is there a better way to include age as
> > covariate in one model? I would prefer modelling three different groups
> > instead of just one.
> >
> > Thanks again,
> > Serge.
> >
> >
> >
> >
> > -----Original Message-----
> > From: FSL - FMRIB's Software Library [mailto:[log in to unmask]] On Behalf
> > Of Stephen Smith
> > Sent: Monday, January 26, 2004 6:24 AM
> > To: [log in to unmask]
> > Subject: Re: [FSL] covariates of no interest at higher level
> >
> > Hi Serge.
> >
> > Correlating with age in a separate analysis isn't too useful - doesn't
> > really tell you anything about the main analysis. Just because the effect
> > is under threshold doesn't mean there isn't some correlation with age.
> >
> > So I would do the analysis all in one model. I'm not quite sure how you
> > modelled age in the all-in-one analysis. I am assuming that the main part
> > of the model has one EV each for A, B and C, with unpaired t-tests carried
> > out with contrasts between these EVs. I would have thought the first thing
> > to try with the age confound is to just add one extra EV with all the ages
> > in it (but it must then be demeaned before putting into the model). Is
> > this what you tried?
> >
> > Cheers, Steve.
> >
> >
> >
> >
> > On Fri, 23 Jan 2004, Rombouts, S.A.R.B. wrote:
> >
> > > Hi,
> > >
> > >
> > >
> > > In a higher level FEAT analysis, I have 3 groups (A, B, C) and used a
> > > 3-sample t-test to test for differences between groups (A-B, A-C, B-C).
> > >
> > > Since age may be a confounding factor in each comparison, I additionally
> > > correlated data across groups with age (that is A + B (one group) with
> age
> > > as EV, A+C with age, and B+C with age). This appeared not to be the
> case.
> > >
> > >
> > >
> > > Alternatively, when I analyse the data in a different way, with all EVs
> in
> > > one analysis (that is, EVs testing for group differences, and one EV
> > > representing all ages), many group differences I saw in the first
> > analysis,
> > > disappear.
> > >
> > >
> > >
> > > I am confused about the proper way to analyse my data: which of the two
> > > results is correct? Is it enough to check whether a covariate
> > significantly
> > > explains group differences in a separate analysis, or must I include all
> > EVs
> > > (both those of interest and those of no interest) in one big analysis?
> > >
> > >
> > >
> > > Thanks,
> > >
> > > Serge.
> > >
> > >
> > >
> > >
> >
> > Stephen M. Smith DPhil
> > Associate Director, FMRIB and Analysis Research Coordinator
> >
> > Oxford University Centre for Functional MRI of the Brain
> > John Radcliffe Hospital, Headington, Oxford OX3 9DU, UK
> > +44 (0) 1865 222726 (fax 222717)
> >
> > [log in to unmask] http://www.fmrib.ox.ac.uk/~steve
> >
>
> Stephen M. Smith DPhil
> Associate Director, FMRIB and Analysis Research Coordinator
>
> Oxford University Centre for Functional MRI of the Brain
> John Radcliffe Hospital, Headington, Oxford OX3 9DU, UK
> +44 (0) 1865 222726 (fax 222717)
>
> [log in to unmask] http://www.fmrib.ox.ac.uk/~steve
>
Stephen M. Smith DPhil
Associate Director, FMRIB and Analysis Research Coordinator
Oxford University Centre for Functional MRI of the Brain
John Radcliffe Hospital, Headington, Oxford OX3 9DU, UK
+44 (0) 1865 222726 (fax 222717)
[log in to unmask] http://www.fmrib.ox.ac.uk/~steve
|