Gabriel Go.Es. wrote:
> Dear Cyril
>
> Thank you so much for all your answers,
> now I'm perfoming the analysis at once, but should I set the group
> just as [1 -1] or [1 0] and [0 -1]?,
doesn't matter either [1s and -1s] and a contrast 1 will show gpA > B
whereas -1 will show gpB > A or enter 2 columns with [1s 0s] and [0s 1s]
to model each group and then run the contrasts [1 -1] and [-1 1]
> I'm putting all regresors with No centering, is it right?
better is to have your data normalized ie centered and of unit variance
- if you have your regressor loaded in the workspace simply type
zscore(regressor_name)
Cyril
> Dear Cyril
>
> Thank you so much for all your answers,
> now I'm perfoming the analysis at once, but should I set the
> group just as [1 -1] or [1 0] and [0 -1]?, I'm putting all
> regresors with No centering, is it right?
>
> Best
> Gabriel
>
>
>
>
>
>
>
> > Date: Tue, 15 Dec 2009 08:20:25 +0000
> > From: [log in to unmask]
> > To: [log in to unmask]; [log in to unmask]
> > Subject: Re: [SPM] SPM contrasts
> >
> > Gabriel
> >
> > your matrix should look a lot like mine in the referenced paper - also
> > no matter if there is a significant effect of age or not you should not
> > remove it from the analysis as it will change how data Y are projected
> > onto X (see SPM book for explanation on the GLM) - linear independence
> > simply means that no regressor is a combination of the others ; this is
> > somehow different from correlation and orthogonality ; indeed many of
> > those combinations are possible (like e.g. correlated and independent
> > variables), see 'Linearly Independent, Orthogonal, and Uncorrelated
> > Variables' by Rodgers et al. The American Statistician
> > <http://www.jstor.org/action/showPublication?journalCode=amerstat>,
> Vol.
> > 38, No. 2, pp. 133-134.
> >
> > one design matrix should be enough for what you want to do, having gp
> > (1s and -1s), performance gp A (values and 0s), performances gp B, (0s
> > and values), age, total GM volume, and scanner as regressors - T
> > contrasts [1] and [-1] will test for gp differences, [1 0] and [0 1]
> for
> > performance correlations in each group, [0 1 1] and [0 -1 -1] for
> > correlations across groups, [0 1 -1] and [0 -1 1] for differences
> > between groups ; all those effects being controlled for age, total
> > volume and scanner differences (which exist no matter if significant or
> > not).
> >
> > good luck
> > Cyril
> >
> >
> > > Dear Cyril,
> > >
> > > Thank you very much for your quick reply and for the paper reference.
> > > I have learned a lot of the described statistical approach. I would
> > > like to confirm that what I'm going to do is correct and at the end I
> > > will ask you one more question.
> > >
> > > I will perform within and across-subjects correlations between
> > > gray/white matter and performance at the same time that I remove the
> > > confounding effects of age. In both designs my regressors are
> > > performance and age and the contrast should be 0 for scanners, 1 for
> > > performance (because I expect to find positive correlations), and 0
> > > for age. I will also compute partial correlations between performance
> > > and age to ensure lineal independence. If finally they are correlated
> > > I will remove age from the analysis because I don't have group
> > > differences for age.
> > >
> > > I want also to perform between-subjects correlations to determine
> > > whether one group shows higher correlation between GM/WM and
> > > performance than the other. Which model and which F-contrast do I
> have
> > > to use in this case?
> > >
> > > I’m very sorry to disturb you again. Thank you very much in advance
> > > for your help.
> > >
> > > Best regards,
> > > Gabriel
> > >
> > >
> > >
> > > > Date: Wed, 9 Dec 2009 11:23:52 +0000
> > > > From: [log in to unmask]
> > > > Subject: Re: [SPM] SPM contrasts
> > > > To: [log in to unmask]
> > > >
> > > > Gabriel
> > > > >
> > > > > Hello
> > > > >
> > > > >
> > > > > I want to make a correlation between my covariate1 and my
> volume loss
> > > > > in the scans, without confounding effects of covariate 2(age),
> what
> > > > > contrast should I use: a b or c?
> > > > >
> > > > >
> > > > > age covariate1 scan data
> > > > >
> > > > > a) 0 -1 0
> > > > >
> > > > > b) -1 -1 0
> > > > >
> > > > > c) 1 -1 0
> > > > >
> > > > >
> > > > >
> > > > >
> > > > >
> > > > > and would this be applied for many covariate effects? I mean
> to make
> > > > > the statistics without confundign the effects of one, 2 or three
> > > > > covariates, should i set them as you will say to me for
> cavariate 2?
> > > > >
> > > >
> > > > the short answer is c = [0 1 0]
> > > > the long answer is, assuming one group of subjects, to use a
> multiple
> > > > regression ion which you enter your covariates with also age,
> possibly
> > > > age^2, gender, global volume - these last 3 or 4 being the 'usual'
> > > > confounding factors - using a contrast with 0s everywhere except 1
> > > or -1
> > > > for the covariate of interest will show the voxel correlating or
> > > > anti-correlating with you covariate, independently of, or
> controlling
> > > > for the effect of the other covariates. For more complex design with
> > > > different groups it's pretty much the same except that you can also
> > > > contrast covariate - you can find an example of such a matrix in the
> > > > supplementary material of one of my papers published in HBM 30
> > > p2278-2292.
> > > >
> > > > good luck
> > > > cyril
> > > >
> > > >
> > > > --
> > > > The University of Edinburgh is a charitable body, registered in
> > > > Scotland, with registration number SC005336.
> > >
> > >
> ------------------------------------------------------------------------
> > > ¡Windows Phone ya está aquí! Mucho más que un teléfono. ¡Hazte con
> > > uno! <http://www.windowsphone.com/spain>
> >
> >
> > --
> > Dr Cyril Pernet,
> > fMRI Lead Researcher SINAPSE
> > SFC Brain Imaging Research Center
> > Division of Clinical Neurosciences
> > University of Edinburgh
> > Western General Hospital
> > Crewe Road
> > Edinburgh
> > EH4 2XU
> > Scotland, UK
> >
> > [log in to unmask]
> > tel: +44(0)1315373661
> > http://www.sbirc.ed.ac.uk/cyril
> > http://www.sinapse.ac.uk/
> >
> >
> > The University of Edinburgh is a charitable body, registered in
> > Scotland, with registration number SC005336.
> >
>
> ------------------------------------------------------------------------
> ¡Windows Phone ya está aquí! Mucho más que un teléfono. ¡Hazte con
> uno! <http://www.windowsphone.com/spain>
--
Dr Cyril Pernet,
fMRI Lead Researcher SINAPSE
SFC Brain Imaging Research Center
Division of Clinical Neurosciences
University of Edinburgh
Western General Hospital
Crewe Road
Edinburgh
EH4 2XU
Scotland, UK
[log in to unmask]
tel: +44(0)1315373661
http://www.sbirc.ed.ac.uk/cyril
http://www.sinapse.ac.uk/
The University of Edinburgh is a charitable body, registered in
Scotland, with registration number SC005336.
|