Dear Chikashi,
> Let me ask one simple question about fMRI Random Effects
Analysis(SPM99).
> I have 3 conditions, A, B, and Rest. I compared A and B for each
subject,
> specifying the t-contrast (1,-1), and obtained con_***.imgs. Then I
tried to
> perform a group analysis using random effects model. I selected "one
sample
> t test" from Basic Stats menu and selected the con_***.imgs. So far so
good.
Yup - that's all fine.
> Then the program requested to specify contrasts again. My question is
how to
> specify this second level contrast and what it means. I assume that
one
> sample t-test refers to a significant test against the null
hypothesis: the
> populaton mean=0, so I guess that the second level contrast should be
"1".
> Is this correct? and If so, why should we specify it? (because I
cannot
> think of any other numbers. As long as one sample t test is concerned,
it
> must be always 1, isn't it? What does it mean when it is , say, -1?)
You're right: for the second level analysis you specify, your H0 is that
the
population mean is not significantly different to 0 (i.e. no consistant
mean
effects). Doing an F at the second level would test for ANY mean effects
significantly different to 0 - WITHOUT any constraints on the direction
of these
effects.
However, using your one-sample t, you can test for the direction of the
difference if you have prior predictions that some areas will show
increases
while other may show decreases (I am sidestepping the issue of how one
wishes to
interpret 'deactivations' or mean decreases in fMRI). So a contrast of
(1) means
'test for a significant mean (+ve) effect' and a contrast of (-1) means
'test
for a significant mean (-ve) effect'.
Best
Dave.
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|