Dear list,
besides this old suggestion from me that you just mentioned, recently
someone posted another reason for this errror and a solution to this
issue in AR convergence, that I can second by now.
It seems that when you set the masking threshold in your defaults too
low (as compared to the default 0.8 of average intensity), tissue
outside the head with very low signal is included in the AR estimation,
which then might not converge due to a noisy signal without much
structure as assumed in AR(1) modeling. The masking threshold is used to
exclude voxels outside the brain (eg, lower than 0.8xaverage signal
intensity by default) from statistical analysis.
See:
http://www.jiscmail.ac.uk/cgi-bin/wa.exe?A2=ind0801&L=SPM&P=R11223&I=-3&
m=29912
Back then, I indeed lowered the threshold to 0 to see outside of the
brain signal for quality control reasons. Later, when I increased that
threshold again, the AR did converge, and I could leave AR(1) modeling
on, as Mike also noticed recently.
In short, you might try setting your masking threshold back to 0.8, and
perhaps you can use AR(1) modeling because the spdiags error disappears
as it dids for us. Some form of autocorrelation correction is necessary,
or you will greatly overestimate your degrees of freedom.
When your masking threshold already is 0.8, the problem might be
elsewhere and you might be forced to refrain from AR(1) modeling.
Good luck,
Bas
--------------------------------------------------
Dr. S.F.W. Neggers
Division of Brain Research
Rudolf Magnus Institute for Neuroscience
Utrecht University Medical Center
Visiting : Heidelberglaan 100, 3584 CX Utrecht
Room B.01.1.03
Mail : Huispost B.01.206, P.O. Box 85500
3508 GA Utrecht, the Netherlands
Tel : +31 (0)88 7559609
Fax : +31 (0)88 7555443
E-mail : [log in to unmask]
Web : http://www.fmri.nl/people/bas.html
--------------------------------------------------
-----Oorspronkelijk bericht-----
Van: SPM (Statistical Parametric Mapping) [mailto:[log in to unmask]]
Namens Carolyn L. Fort
Verzonden: woensdag 23 januari 2008 18:49
Aan: [log in to unmask]
Onderwerp: Re: [SPM] Error: Index exceeds matrix dimensions.
Many thanks to those of you who replied (both offlist and on). Several
of you suggested checking that the # of scans specified in the model
versus the # of scans selected. Though that would certainly cause
problems it was not the case here.
A search in the archives for "spdiags error" produces the following
(among others), which ultimately solved it:
http://www.jiscmail.ac.uk/cgi-bin/webadmin?A2=ind05&L=SPM&P=R514650&I=-3
&X=39211A7A01151E95BB&Y
It was actually a result of the SPM2 software not being able to perform
non-sphericity correction appropriately for our design as it was set up.
Not wanting to either increase the number of images or decrease
complexity of the design, we chose to estimate the model without serial
autocorrelations.
Carolyn
Carolyn L. Fort
Research Specialist, Senior
Department of Psychiatry
University of Arizona
PO Box 245002
Tucson, AZ 85724-5002
v: 520-626-8568
f: 520-626-6050
|