With randomise, null-hypothesis exchangeability is essential, so if you
have dependencies among repeated measurements, then no, your results
would not be strictly valid if you just threw all measurements into
randomise "as is". There is an example on the randomise page for using
"exchangeability-blocks" to model repeated measures for something like a
plain vanilla repeated measures ANOVA. However, the model that you are
proposing to maintain is more complicated than that because it has two
different factors of potentially correlated measurements (repeated
measures in the same subjects AND measures from twins in the same
family).
cheers,
-MH
On Wed, 2011-07-27 at 12:44 +0200, Diederick Stoffers wrote:
> Hi guys,
>
>
> Thanks for the advice. Averaging over twin pairs is not really an
> option, the twins were specifically selected because they were
> discordant on a certain psy scale. I could average over sessions, but
> am hesitant to do so as the behavioural measure I am interested in can
> be quite different over sessions; i would lose a lot of power in my
> group comparison.
>
>
> If I were to go about this a different way; I would just do a
> regression of fMRI activation over all sessions versus the behavioural
> measure without including any information about family ties or session
> and not averaging. I will lose some power because I am not moddelling
> family ties and sessions, but will gain by having more observations.
>
>
> A little extra background, this is actually resting-state data on
> which I performed concatenated ICA and now want to do dual regression
> and compare the resulting 160 spatial maps (80 subjects * 2 sessions)
> using randomise. Will the results be valid if I don't include
> information on the dependencies?
>
>
> Cheers,
>
>
> Diederick
>
>
>
>
> On 26 jul. 2011, at 20:28, Michael Harms wrote:
>
> > Hi Diederick,
> > To my knowledge (someone please correct me if I'm wrong), there
> > aren't any
> > packages currently that can easily construct voxel-based maps in the
> > context of the complicated variance relationships that one might
> > want to
> > model for sibling and/or twin studies. However, if you have ROI-
> > based
> > data, you can import it into a package that allows one to specify
> > covariance structures between subjects in a sibling pair, such as
> > SAS's
> > PROC MIXED. I'm sure that R and SPSS have something equivalent as
> > well.
> >
> > cheers,
> > -MH
> >
> >
> > > You need to eliminate the repeated measurements.
> > >
> > > The issue is that if you have between-subject effects and within-
> > > subject
> > > effects in the same model, then you only investigate the within-
> > > subject
> > > effects because the error term is for the within-subject effects.
> > > Once
> > > software becomes available to have multiple error terms for the
> > > between
> > > and
> > > within-subject effects, then you'll not need to collapse them.
> > >
> > > Additionally, if you have more then one within-subject effect,
> > > then you
> > > can
> > > only look at the interaction for the same reason.
> > >
> > > Given these points, it seems that it would be best to collapse
> > > (e.g.
> > > average
> > > your two conditions) to eliminate the repeated measurement.
> > >
> > > I'm not sure how you should deal with the twins not being
> > > independent. I
> > > would say average each twin pair, but that seems like it would
> > > ruin your
> > > research question.
> > >
> > >
> > > Best Regards, Donald McLaren
> > > =================
> > > D.G. McLaren, Ph.D.
> > > Postdoctoral Research Fellow, GRECC, Bedford VA
> > > Research Fellow, Department of Neurology, Massachusetts General
> > > Hospital
> > > and
> > >
> > > Harvard Medical School
> > > Office: (773) 406-2464
> > > =====================
> > > This e-mail contains CONFIDENTIAL INFORMATION which may contain
> > > PROTECTED
> > > HEALTHCARE INFORMATION and may also be LEGALLY PRIVILEGED and
> > > which is
> > > intended only for the use of the individual or entity named above.
> > > If the
> > > reader of the e-mail is not the intended recipient or the employee
> > > or
> > > agent
> > > responsible for delivering it to the intended recipient, you are
> > > hereby
> > > notified that you are in possession of confidential and privileged
> > > information. Any unauthorized use, disclosure, copying or the
> > > taking of
> > > any
> > > action in reliance on the contents of this information is strictly
> > > prohibited and may be unlawful. If you have received this e-mail
> > > unintentionally, please immediately notify the sender via
> > > telephone at
> > > (773)
> > >
> > > 406-2464 or email.
> > >
> > >
> > >
> > > On Tue, Jul 26, 2011 at 3:31 PM, Diederick Stoffers
> > > <[log in to unmask]>wrote:
> > >
> > > > Hi all,
> > > >
> > > > I have a group of 80 subjects (all dizygotic twins) from whom I
> > > > have two
> > > > fMRI measurements per subject. For all these subjects I also
> > > > have a
> > > > behavioural score per scan. I would like to compare scans
> > > > associated
> > > > with a
> > > > high score with those with a low score, while correcting for the
> > > > fact
> > > > that
> > > > measurements in dizygotic twins are not independent and
> > > > measurements in
> > > > the
> > > > same subject are not independent. For now, I am not interested
> > > > in
> > > > within-subject effects over scans.
> > > >
> > > > I have been thinking how to best model this, but quite frankly I
> > > > can't
> > > > wrap
> > > > my head around it and I wasn't able to deduce this from the
> > > > mailing list
> > > > or
> > > > FSL site. Could anyone shed some light on how to set up my
> > > > design
> > > > matrix?
> > > >
> > > > Thanks,
> > > >
> > > > Diederick
> > > >
> > >
>
|