The issue, in my opinion, isn't with the inflated df, but with a reduced error term. The reduced error term increases the statistical results artificially. If you look at how the statistics are computed by hand for various effects according to the formulas from a statistics book, you will note that there are two-error terms: a within-subject error that is used to compute the statistics for within-subject effects (e.g. condition and condition*group) and a between subject error term. SPM only computes a single error term -- within-subject for a properly set up repeated measures design. This leads to the statement that if you are interested in the group effects, one should use two-sample t-tests -- which use the the between subject error term since that is all that is in the model.
On Fri, Jul 10, 2009 at 11:08 AM, Bernhard Haslinger
<[log in to unmask]> wrote:
Dear SPM experts,
I have a question raised by a reviewer's comments on our fMRI study:
We have conducted an event-related fMRI study on two groups (patients and controls) with 3 experimental conditions (3 fMRI runs/subject, each presenting the 3 conditions in ranomized order).
We were interested in activation differences between patients and controls for each of the three experimental conditions and for the differential contrast con1-con2).
We therefore calculated con-images for con1, con2, con3 on the first level and entered them into a second level full factorial model with two factors ("condition" with 3 levels; "group" with 2 levels) resulting in a design matrix containig 6 columns (condition1@controls, condition1@patients, condition2@controls ...), factor condition: independence="no", factor group: independence="Yes".
We tested for activation differences by applying t-contrasts (e.g. -1 1 0 0 0 0 for patients>controls during con1, ...-1 1 1 -1 0 0 for the differential contrast patients>controls during con1 vs. con2).
One of the reviewers is concerned that "Second level random-effects analysis should be based on simple t-tests which directly compare the contrast images of interest (rather than using a factorial design that inflates the degrees of freedom)."
We are now asking if our design is correct/valid or if indeed the second level analysis has to be performed by calculating separate two sample t-tests contrasting the corresponding first level con.images (con1, con2, con1-2).
Thank you very much for your help!
Best regards
Bernhard Haslinger
--
-----------------------------------------------------------------------------------------
PD Dr. Bernhard Haslinger
Oberarzt
Neurologische Klinik und Poliklinik
Klinikum Rechts der Isar
TU-Muenchen
Ismaninger Strasse 22
D-81675 München
E-mail: [log in to unmask]
WWW: http://www.neurokopfzentrum.med.tum.de/neurologie/
-----------------------------------------------------------------------------------------
--
Best Regards, Donald McLaren
=====================
D.G. McLaren
University of Wisconsin - Madison
Neuroscience Training Program
Office: (608) 265-9672
Lab: (608) 256-1901 ext 12914
=====================
This e-mail contains CONFIDENTIAL INFORMATION which may contain PROTECTED HEALTHCARE INFORMATION and may also be LEGALLY PRIVILEGED and which is intended only for the use of the individual or entity named above. If the reader of the e-mail is not the intended recipient or the employee or agent responsible for delivering it to the intended recipient, you are hereby notified that you are in possession of confidential and privileged information. Any unauthorized use, disclosure, copying or the taking of any action in reliance on the contents of this information is strictly prohibited and may be unlawful. If you have received this e-mail unintentionally, please immediately notify the sender via telephone at (608) 265-9672 or email.