Subject: | | Re: regressor order in the GLM |
From: | | Roberto Viviani <[log in to unmask]> |
Reply-To: | | [log in to unmask][log in to unmask]] >> > Sent: Friday, January 01, 2010 7:53 AM >> > To: [log in to unmask]; Vaidya, Jatin G >> > Subject: Re: 2 x 2 ANOVA >> > >> > On Thu, 31 Dec 2009 22:38:19 +0000, Jatin Vaidya <jatin- >> > [log in to unmask]> wrote: >> > >> >> Hi, >> >> >> >> I'm trying to run a 2 X 2 design with PET data. The first factor is >> > group >> > (user >> >> vs. control). The second factor is task (tapping vs. control); thus, >> > this is a >> >> repeated measures variable. >> >> >> >> I can't seem to determine how to run the analysis in SPM with this type >> > of >> >> dataset. Is there a good resource available for this? >> > >> > http://www.fil.ion.ucl.ac.uk/~wpenny/publications/rik_anova.pdf<http://www.f il.ion.ucl.ac.uk/%7Ewpenny/publications/rik_anova.pdf> >> > >> > In this particular case, if you're only interested in comparing tapping >> > with >> > control (i.e., not interested in tapping alone or control alone or their >> > average, >> > but only their difference), you could also just subtract control from >> > tapping at >> > the subject level (or using ImCalc), and bring the difference to the >> > group >> > level, resulting in a 1-way, between-subjects ANOVA with no repeated >> > measures. >> > >> >> >> >> Thanks, >> >> >> >> Jatin >> > > >-- >Swann Pichon, PhD >Laboratory for Behavioral Neurology and Imaging of Cognition >Department of Neuroscience, University Medical Center >1 rue Michel-Servet, 1211 GENEVA 4, Switzerland >Tel: +41 (0)22 379 5979 >Fax: +41 (0)22 379 5402 >Gsm: +33 (0)6 26 43 83 61 >http://labnic.unige.ch/ >40_6Jan201015:23:[log in to unmask] |
Date: | | Thu, 21 Jan 2010 11:20:32 +0100 |
Content-Type: | | text/plain |
Parts/Attachments: |
|
|
|
|
Hi Raphael
> But why should I worry about co-linearity at all, since ist does not
> alter my results? Though this way we allow the regressors to show
> activation that is not independent, which is rather not the proper
> analysis.
You do not show the regressors, or the beta estimates, you show the
parametric maps. T values are computed from the explained variance of
the relevant contrast, after all other regressors have been taken into
account. It is like entering the relevant regressor last. You do not
need orthogonal regressors to compute a correct t value, each t value
is computed as if the regressor were the last.
<...>
>
> I need my two almost inverted boxcar functions for separate parametric
> modulation. Though I dont feel good with this correlation in the GLM.
If you want orthogonal regressors with current techniques, you have to
design your task accordingly. Remember that for orthogonality the
matrix of the AR(1) autocorrelation must be taken into account.
Best,
Roberto Viviani
>
> Best regards, Rafael
>
>
> On Wed, 20 Jan 2010 16:47:38 +0000
> Steve Fleming <[log in to unmask]> wrote:
> Hi Rafael
> spm_orth is only called to orthogonalise regressors within trial-type
> ("condition" in Specify 1st level). Have a look within spm_fMRI_design.m
> at
> line 283 to see this.
> So SPM will only orthogonalise regressors attached to the same onset
> function (i.e. separate basis functions or parametric modulators).
> This is why your stats have not changed when you change the order, as you
> are looking at two separate trial types/conditions.
> Best
> Steve
> 2010/1/20 Rafael Lüchinger <[log in to unmask]>
> > Dear SPM user
> >
> > The order of regressors within the designmatrix of spm (spm5) does matter
> > because all regressors get orthogonalized to the first one (spm_orth.m).
> > I'm
> > interested in the severity of this effect and tested this having two
> >simple
> > boxcar function with high colinearity of r=.7 within the design matrix and
> > run two separate analysis with switched regressor order. Surprisingly the
> > two regressor had the identical activation independet of their order in
> >the
> > designmatrix. I', pretty sure not done any mistakes. How is this to
> > explain?
> >
> > Thanks for any comments
> > RL
> >
>
> ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
> Rafael Lüchinger
> Brain Mapping Research, UZH Department of Child and Adolescent Psychiatry
> Neumuensterallee 9, 8032 Zuerich, Switzerland. Phone: (+41) 43 499 26 50
> ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
|
|
|
|