Mark Schram Christensen wrote:
> Dear Will Penny,
> Last week I posted a question on the SPM-mailing list regarding ANOVA
> contrast estimates with non-sphericity correction. Unfortunately I have not
> recived any responses. You have previously been able to answer some of my
> related questions, I would therefore ask you the same question. I fully
> understand if you are too busy answering questions from students. Here is my
> post to the mailing list:
>
> I am running a 2nd level analysis using a one-way ANOVA. I have four
> groups. I have made non-spericity correction, since the groups are four
> different effects measured in the same 14 subjects.
>
> If I make the t-contrast [1 -1 0 0] I assume the estimate of the first
> group will we larger than for the second group in the significant voxels in
> my SPM{T} map. I now choose a voxel (in this case the global max of the
> contrast [1 -1 0 0]) and make a plot of the contrast estimates. I
> choose 'effects of interest' (i.e. the one generated by default after the
> estimation (I can see this is not a diagonal matrix)))
You should ignore the 'effects of interest' contrast that is created by default.
In patched versions of SPM2 this is no longer produced. Please ignore
these results.
in order to get a
> display of all the four estimates. To my surprise it seems that the
> estimate of the second group is LARGER than the estimate for the first
> group (even though the voxel is picked from a thresholded map of the [1 -1
> 0 0] contrast). This is displayed on page 1 of the attached document. I
> assume this has something to do with the non-sphericity correction.
>
> I now create an 'effects of interest' as a diagonal matrix (F-contrast) [1
> 0 0 0; 0 1 0 0; 0 0 1 0; 0 0 0 1].
> If I then make a plot of the contrast estimates and use the diagonal
> matrix, the contrast estimates now look more sensible (group 1 is larger
> than group 2 as indicated by the contrast I'am testing [1 -1 0 0]). See
> page 2 of the attached document
>
> My questions:
> If I want to report the contrast estimates for all four groups in a
> particular voxel, can I then use the default 'effects of interest' created
> by SPM or should I use the diagonal matrix I have created my self in order
> to report something that is physiological sensible, i.e. "In group 1 there
> is a larger response in voxel x than in group 2"
> If I can use default 'effects of interest', how should I then interpret the
> sizes of the estimates?
> If I use my 'own' diagonal 'effects of interest', how should I then
> interpret the sizes of the estimates?
>
These will be in units of percentage of whole brain mean. eg. a value of
0.5 indicates an activation of 0.5% of the whole brain mean.
Best,
Will.
> I hope some of the experts could clarify this point and maybe explain the
> default 'effects of interest' when non-spericity correction is used.
>
> Sincerely,
>
> Mark S. Christensen
>
> _________________________________________
> Mark Schram Christensen
> MSc, PhD-student
> Danish Research Centre for Magnetic Resonance
> Copenhagen University Hospital Hvidovre
> Kettegerd Alli 30
> DK-2650 Hvidovre
> &
> Institute of Physical Exercise and Sport Science
> University of Copenhagen
> Denmark
> Phone: +45 3632 3323/+45 2695 6275
> Fax: +45 3647 0302
> E-mail: [log in to unmask]
> Web: www.drcmr.dk
> www.ifi.ku.dk
--
William D. Penny
Wellcome Department of Imaging Neuroscience
University College London
12 Queen Square
London WC1N 3BG
Tel: 020 7833 7475
FAX: 020 7813 1420
Email: [log in to unmask]
URL: http://www.fil.ion.ucl.ac.uk/~wpenny/
