On Mon, 6 Apr 2009 10:10:48 +0100, Thomas Stephan
<[log in to unmask]> wrote:
>Dear Laura, Lennart, Stephen, Darren,
>
>let me try to come back to my original question posted last week:
>https://www.jiscmail.ac.uk/cgi-bin/webadmin?A2=ind0904&L=SPM&P=R12110
>
<snip>
>
>But even if so, how can we explain the complex values of the
>standarddeviation originiating from the square root of a negative number
>that appears in a situation in SPM where numbers are not planned to be
>negative? Is this a result from using the wrong model, or is it a bug in SPM?
<snip>
Just commenting about this part for now.
I'm not sure what to do. It sounds like you're somewhat comfortable with
examining the SPM code. When I get into these situations, I use the matlab
debugger to track down the source of the error or problem.
In your original post, you wrote:
"It looks like somewhere around line 298 in spm_graph.m there is a negative
value appearing in the diagonal..."
I would think that the Bcov matrix should be nonnegative definite, so that if
it's in diagonal form, all elements should be nonnegative. So that's weird.
Only suggestions I have at the moment:
(1) Use the debugger to go through the various SPM *.m files to find where
this might be coming from.
(2) Maybe there's something wrong or awkward about the covariance and/or
ReML calculations. In your original post, you told SPM that side and condition
should both have "no" independence and "equal" variance. Glascher/Gitelman
instruct in at least one place to use "no/equal," but in my opinion if "no" is
specified for independence, then in most situations "unequal" should be
specified for variance. (While this is arguable, I think some other people on
the list agree with this.) So, you could try that and see what happens.
Best,
S
|