Is it really that much effort to re-run the lower-level analyses? It is
just processor time :)
My bias would be to say that no, what you want to do is unfortunately
not ok, unless you can establish equivalence by testing both approaches
in a subset of subjects.
cheers,
-MH
On Tue, 2011-09-20 at 12:47 -0400, Leslie Engineering wrote:
> Thanks. Yes it is only one group. There isn't necessarily a
> correlation between EV1 and EV2 so I was trying to bypass redoing the
> lower-level analysis. I guess I just wanted to know if you thought
> that was ok?
>
>
> On Tue, Sep 20, 2011 at 12:19 PM, Michael Harms
> <[log in to unmask]> wrote:
> I'm confused by your usage of the term "groups". Do you
> actually have
> two groups of subjects that you are trying to contrast on a
> given lower
> level contrast?
>
> Or are you simply trying to figure out a way to effect the
> equivalent of
> a (1,-1) contrast that should have been included in the 1st
> level, but
> wasn't? If this is the case, then it all depends on the
> degree of
> correlation of EV1 and EV2, as I noted when you inquired about
> this
> previously
> https://www.jiscmail.ac.uk/cgi-bin/webadmin?
> A2=ind1109&L=FSL&P=R16891&1=FSL&9=A&J=on&d=No+Match%3BMatch%
> 3BMatches&z=4
>
> The "proper" way to do this is to redo the lower-level
> modeling with
> (1,-1) included as a contrast. As a general matter, desired
> contrasts
> should be included in the first level modeling so that you
> properly
> account for the correlation between EVs.
>
> cheers,
> -MH
>
>
>
> On Tue, 2011-09-20 at 11:55 -0400, Leslie Engineering wrote:
> > I think you misunderstand. The last question was in regard
> to 1 group
> > that individual subjects were run with EV1 and EV2
> paradigms.
> > Unfortunately I ran contrasts (1,0) (0,1) (1,1). But, now I
> want to
> > see activation differences between EV1 and EV2. This would
> have been
> > straight forward had my contrasts been (1,-1) and (-1,1) (i
> believe).
> > But, since I don't have that, may I run each subject twice
> and
> > consider the EV1s and EV2s different groups so I don't have
> to rerun
> > the individuals with the proper contrasts?
> >
> >
> > Thanks
> >
> > On Tue, Sep 20, 2011 at 11:50 AM, Michael Harms
> > <[log in to unmask]> wrote:
> > To compare groups you need to run a "Higher-level
> analysis" in
> > FEAT
> > using the lower-level FEAT directories as inputs.
> This will
> > then create
> > a group contrast for each of your lower-level
> copes. Assuming
> > that you
> > have already created the desired lower-level copes
> at the
> > individual
> > level, then no, you don't have to rerun individuals.
> >
> > cheers,
> > -MH
> >
> >
> > On Tue, 2011-09-20 at 11:38 -0400, Leslie
> Engineering wrote:
> > > Or can I just treat them as two separate groups so
> i don't
> > have to
> > > rerun individuals?
> > >
> > > On Tue, Sep 20, 2011 at 11:36 AM, Leslie
> Engineering
> > > <[log in to unmask]> wrote:
> > > So I am a little confused, then. If i have
> > individual subjects
> > > that were run with two EVs contrasts (1,0)
> (0,1)
> > (1,1), do I
> > > have to go back and rerun each individual
> subject
> > with
> > > contrasts (1,-1)(-1,1) in order to get
> group
> > differences or
> > > can I manipulate these in some way?
> > >
> > > On Tue, Sep 20, 2011 at 11:14 AM, Michael
> Harms
> > > <[log in to unmask]> wrote:
> > > Group contrasts (and statistics)
> should be
> > calculated
> > > from the
> > > appropriate copes, not done "post-
> hoc" by
> > subtracting
> > > z-maps.
> > >
> > > See
> > >
> >
> http://www.fmrib.ox.ac.uk/fsl/feat5/detail.html#UnpairedTwoGroupDifference
> > > which provides a clear example of
> how to
> > proceed.
> > >
> > > cheers,
> > > -MH
> > >
> > >
> > > On Tue, 2011-09-20 at 11:07 -0400,
> Leslie
> > Engineering
> > > wrote:
> > > > does it make a difference if
> instead I
> > have one
> > > group that went
> > > > through a single paradigm, EV.
> If EV is
> > made up of
> > > two separate
> > > > stimuli blocks and I want to
> separate them
> > into EV1
> > > and EV2, can I
> > > > run a single group average using
> EV1 and a
> > single
> > > group average using
> > > > EV2 then subtract the z scores
> for an
> > activation
> > > difference between
> > > > the two stimuli?
> > > >
> > > >
> > > > thanks so much
> > > >
> > > > On Tue, Sep 20, 2011 at 10:08
> AM, Leslie
> > Engineering
> > > > <[log in to unmask]>
> wrote:
> > > > Hello-
> > > >
> > > >
> > > > I ran two separate
> simple group
> > average
> > > analysis. Both groups
> > > > underwent the same fmri
> scan
> > sequence.
> > > Instead of rerunning a
> > > > group analysis and
> included all
> > subjects
> > > from both groups and
> > > > setting up contrasts for
> group
> > differences
> > > (1,-1)(-1,1), is it
> > > > statistically sound to
> simply
> > subtract their
> > > zscores?
> > > >
> > > >
> > > >
> > > >
> > > >
> > > >
> > >
> > >
> > >
> > >
> >
> >
> >
>
>
>
|