Darren,
That is a problem, and the lack of "+ eps" also gives rise to
occasional division by zero error for cluster statistics.
The fix is to replace "/P(D)" with "/(eps+P(D)". The latest version,
2.10, reflects this and is now in spm99_updates...
ftp://ftp.fil.ion.ucl.ac.uk/spm/spm99_updates/spm_P_RF.m
-Tom
On Wed, 19 Sep 2001, D. Gitelman wrote:
> Dear SPM:
>
> Changes seem to have crept into SPM_P (2.5) => SPM_P_RF (2.9) that produce
> different cluster threshold results. Since the results are more
> conservative, activations we thought were present are now
> non-significant. I'd appreciate if someone could tell me which code is
> correct.
>
> The change is centered on lines 61-64 of both the old and new versions.
>
> Version 2.5 of spm_p had
> Em = (R./G)*P;
> EN = P(1)*R(D);
> Em = Em + eps;
> En = EN/Em;
>
> Version 2.9 of spm_p_rf now has
> Em = (R./G)*P;
> EN = P(1)*R(D);
> En = G(D)*P(1)/P(D); % i.e. En = EN/Em;
>
>
> The problem is that in the new version using the calculated EN and Em does
> not produce the same result as the En equation. Thus if one uses debug and
> steps through the spm_p_rf code then
>
> EN/Em ~= G(D)*P(1)/P(D);
>
>
> Thanks for any help with this.
> Darren
>
>
> -------------------------------------------------------------------------
> Darren R. Gitelman, M.D.
> Cognitive Neurology and Alzheimer¹s Disease Center
> E-mail: [log in to unmask]
> WWW: http://www.brain.northwestern.edu
> Voice: (312) 908-9023
> Fax: (312) 908-8789
> Northwestern Univ., 320 E. Superior St., Searle 11-470, Chicago, IL 60611
> -------------------------------------------------------------------------
>
|