Dear Claus, Stephen and Dorian,
Thank you for help!
To Stephen:
"Not sure I understand---why wouldn't you get the effect of that contrast across subjects?"
I meant, if I use contrast (A1-A2)>(B1-B2) in the 1st level, maybe I couldn't got (A1-A2) VS (B1-B2) effect in 2nd level. But now I think I'm wrong. Because if I use F-contrast [1] in the 2nd level with one-sample t-test model, I can acquire the (A1-A2) VS (B1-B2) effect. Am I right?
PS: I use '(A1-A2) VS (B1-B2) effect' to describe '(A1-A2) not equal (B1-B2)'.
Feng Lu
-----Original Message-----
From: Stephen J. Fromm [mailto:[log in to unmask]]
Sent: 2010年3月24日 19:42
To: [log in to unmask]; Lu Feng
Cc: Stephen J. Fromm
Subject: Re: Questions about random- and fixed-effect in factorial design^^
"You mean that, my procedure(taking 4 contrast images per subject from 1st-level to a 2nd-level analysis) would be 'textbook rfx', so it wouldn't be 'SPM rfx'. Am I right?"
No, I think by "textbook rfx" he meant "typical SPM rfx".
Actually, if you read textbooks, what people in neuroimaging call RFX is actually _mixed_ effects, because there's both fixed and random effects. Furthermore, as far as I can tell, while the method of doing fixed effects at the subject level and then taking the con images to the group level to do the random effects isn't unique to neuroimaging, it's more obscure---the textbooks I've seen consider the model with a single level combining both.
Note: if you bring four (i.e., more than one) contrasts per subject to the group level, it's repeated measures. There's nothing wrong with that, but there might be additional issues to consider. You can read up on it on this listserve, and there's a manuscript, at
http://www.fil.ion.ucl.ac.uk/~wpenny/publications/rik_anova.pdf
"Maybe I have to get one contrast image per subject using (A1-A2)>(B1-B2), feed these con-images to 2nd-level analysis and do a one-sample t-test. Maybe it's a standard rfx. But in this way I won't get the effect of (A1-A2) vs (B1-B2) across subjects."
Not sure I understand---why wouldn't you get the effect of that contrast across subjects?
"In spm8 manual, an informed basis set are described. I think it's similar to my procedure in a sense."
Yes, in the sense that they're both repeated measures ANOVAs.
__________ Information from ESET NOD32 Antivirus, version of virus signature database 4971 (20100324) __________
The message was checked by ESET NOD32 Antivirus.
http://www.eset.com
__________ Information from ESET NOD32 Antivirus, version of virus signature database 4972 (20100324) __________
The message was checked by ESET NOD32 Antivirus.
http://www.eset.com
|