Dear Marko, and all
several month ago, we had similar differences in results between SPM2 and SPM99 analysis. At that time, Russ Poldrack helped us to identify the problem - it was due to the global normalization ('Scaling')- we had 'scaling' with SPM99 but not with SPM2 analysis. After we corrected that in the batch code, the differences disappeared (or became negligible). Russ also pointed us to an interesting paper, Aguirre's "The Inferential Impact of Global Signal Covariates in Functional Neuroimaging Analyses" in Neuroimage 1998.
We did not use 'Scaling' at all this time in SPM2 or SPM5. It seems I will have to look for some other possible causes (maybe differences in the preprocessing) this time.
Thanks for your reply,
Ping
======================================
Ping Zou Stinnett, Ph.D.
Neuroimaging Scientist
Radiological Sciences (MS-210)
St. Jude Children's Research Hospital
(tel) 901-495-2892 (fax) 901-495-5706
[log in to unmask]
======================================
-----Original Message-----
From: SPM (Statistical Parametric Mapping) [mailto:[log in to unmask]] On Behalf Of Marko Wilke
Sent: Tuesday, January 31, 2006 5:31 AM
To: [log in to unmask]
Subject: Re: [SPM] Re-Post: Results are different between spm2 and spm5. .
Dear Ping, Dear All,
> I saw similar differences between SPM2 and SPM5 results. Did you ever
> get a good explanation?
well, for a time I had problems in re-creating the difference which is kind of embarrassing ;) I then doug through my batch and found that in spm2, I rather inadvertedly had Global Normalization ("Scaling") enabled (default setting in a script posted by Karl [1], while "None" is set in a script created by Rik [2]). While I was aware it may not be necessary I did not think it would do much harm. Now I know better since even within spm2, T-map-differences of a similar magnitude can be seen (with scaling enabled, values are higher).
Now, while the difference between the two versions herewith seems to resolve, I am still pretty confused why in a single session fMRI analysis of native-space data global scaling makes that much of a difference. I know that correlations between task and globals may play a role (Kalina's website is still very instructive on this phenomenon [3]). However, in the setting at hand, the correlation between task and globals is only r = 0.0712, not significant. Moreover, the difference remains (and increases) even if global drifts are removed (Macey et al, NI 2004) so that the globals are practically identical over the session... odd.
So, I seem to need to apologize for following a red herring there (and for likely confusing people in the process - sorry!), but a herring like this may not be all that uncommon and, whith perhaps some comments from others who actually know what they are doing, may even turn out to be an instructive one :) Best, Marko
[1] www.jiscmail.ac.uk/cgi-bin/webadmin?A2=ind02&L=SPM&P=R282882&I=-3
[2] www.fil.ion.ucl.ac.uk/~wpenny/datasets/face-rep/SPM2/batch.m
[3] http://www-psych.stanford.edu/~kalina/SPM99/Tools/rd_global.html
--
=====================================================================
Marko Wilke (Dr.med./M.D.)
[log in to unmask]
Universitäts-Kinderklinik University Children's Hospital
Abt. III (Neuropädiatrie) Dept. III (Pediatric neurology)
Hoppe-Seyler-Str. 1, D - 72076 Tübingen
Tel.: (+49) 07071 29-83416 Fax: (+49) 07071 29-5473
=====================================================================
|