Thanks so much Mark, I really appreciate your advice!
Clare
On 1/6/07, Mark Jenkinson <[log in to unmask]> wrote:
> Hi,
>
> Yes, I replied hastily. Sorry.
> The sigmasquareds volume is the unbiased estimate of the noise
> standard deviation in the GLM.
>
> The avwmaths++ command will calculate the sum of squares divided
> by the number of timepoints, which isn't quite the same thing.
>
> However, looking at the original email, I see that the objective is not
> to get the residuals and calculate variances so much as to compare
> models. This should really be done either with Bayesian model
> comparison techniques, or (within the GLM) using F-tests. For the
> F-test, the "additional" EVs (e.g. nuisance regressors) EVs have
> individual t-contrasts and then are combined into a set for a single
> F-test. The result of this will show when the "additional" EVs explain
> a statistically significant amount of signal over and above that
> explained by the remaining EVs. This is equivalent to testing the
> difference between models with and without these "additional" EVs.
> You can do this for different sets, where each time those included
> in the F-test are the "additional" EVs and the rest are the remaining
> ones.
>
> We would definitely recommend either of the above approaches rather
> than just looking at the residuals.
>
> All the best,
> Mark
>
>
> On 5 Jan 2007, at 22:57, Doug Greve wrote:
>
> > Mark,
> >
> > isn't this what the stats/sigmasquareds volume is? btw, the
> > avwmaths cmd below will under estimate the variance slightly as it
> > will take into accnt the proper DOF.
> >
> > doug
> >
> > Mark Jenkinson wrote:
> >
> >> Hi,
> >>
> >> You can calculate the voxel-wise variance from the 4D residuals
> >> by using
> >> avwmaths++ res4d -Tstd -sqr variance_map
> >>
> >> That should be what you want.
> >>
> >> All the best,
> >> Mark
> >>
> >>
> >>
> >> On 5 Jan 2007, at 19:55, Clare Kelly wrote:
> >>
> >>> Thanks Michael, I had seen that but what I was thinking of was a 2D
> >>> file that gives one value for the error variance for each voxel
> >>> - in
> >>> the same way that the varcope images provide a measure of how much
> >>> signal is not accounted for by one particular EV - how do I
> >>> calculate
> >>> this for the whole model... or does that not make sense?!?
> >>>
> >>> Thanks!
> >>> Clare
> >>>
> >>> On 1/5/07, Michael Hanke <[log in to unmask]> wrote:
> >>>
> >>>> Hi,
> >>>>
> >>>> On Fri, Jan 05, 2007 at 06:30:44PM +0000, Clare Kelly wrote:
> >>>> > Hi all,
> >>>> >
> >>>> > I apologise if this is a really basic question but I can't
> >>>> seem to find the answer I'm looking for in the
> >>>> > archive.
> >>>> >
> >>>> > I want to calculate the residuals for a particular model that
> >>>> included 4 task regressors plus their
> >>>> > temporal derivatives, 8 nuisance regressors, and 6 motion
> >>>> parameters, as I want to compare the
> >>>> > residuals between different models with different numbers of
> >>>> regressors.
> >>>> >
> >>>> > I understand that each varcope gives me the error related to
> >>>> that particular EV, but I'm not clear on
> >>>> > how I combine them - I have looked at section 14.5 of the
> >>>> Jezzard et al. book but I still don't know
> >>>> > what to do.
> >>>> FEAT can output the residuals for you (no need to calculate
> >>>> anything).
> >>>> In fact it always stores them to disk, but deletes them at the
> >>>> end of
> >>>> the analysis. Just look at the end of you design.fsf file and
> >>>> you'll
> >>>> see:
> >>>>
> >>>> # Delete GLM residuals?
> >>>> set fmri(cleanup_residuals_yn) 1
> >>>>
> >>>> Change this default into:
> >>>>
> >>>> # Delete GLM residuals?
> >>>> set fmri(cleanup_residuals_yn) 0
> >>>>
> >>>> and you'll find a res4d.nii.gz (or whatever suffix) in the stats
> >>>> directory when Feat is finished.
> >>>>
> >>>> HTH.
> >>>>
> >>>>
> >>>> Cheers,
> >>>>
> >>>> Michael
> >>>>
> >>>>
> >>>>
> >>>> --
> >>>> GPG key: 1024D/3144BE0F Michael Hanke
> >>>> http://apsy.gse.uni-magdeburg.de/hanke
> >>>> ICQ: 48230050
> >>>>
> >>
> >>
> >
> > --
> > Douglas N. Greve, Ph.D.
> > MGH-NMR Center
> > [log in to unmask]
> > Phone Number: 617-724-2358 Fax: 617-726-7422
> >
> > In order to help us help you, please follow the steps in:
> > surfer.nmr.mgh.harvard.edu/fswiki/BugReporting
>
|