Hi Kambiz -
>I am carrying out a 2x2x2 factorial analysis of MEG data in sensor space.
>The design includes the factors stim (prime,target), cond (id,nonid), and
>freq (low,high) in a repetition priming paradigm.
>My first question is whether it is advisable to go straight to the 2nd-level
>analysis of my data without ever specifying the 1st-level specifications in
>SPM? Essentially my scans in the design matrix (see attachment) are the
>trial-averaged (.img) files for each of my subjects.
Yes, that is fine.
>My second question concerns the contrasts I wish to carry out. Specifically,
>I am interested in finding out when & approx. where (in sensor space)
>repetition priming caused significant response attenuation following the
>target stimulus. Meaning that I am interested in an interaction between
>factors stim and condition. I am also interested in a main effect of
>frequency. Would someone, kindly, confirm that the contrasts I have defined
>(below) are indeed correct, given the specific question I have. Thank you in
>advance.
>
>DESIGN
> stim cond freq
>cell
>1 1 1 1
>2 1 1 2
>3 1 2 1
>4 1 2 2
>5 2 1 1
>6 2 1 2
>7 2 2 1
>8 2 2 2
>
>(F) CONTRASTS
>main effect(s)
>stim [1,1,1,1,-1,-1,-1,-1]
>cond [1,1,-1,-1,1,1,-1,-1]
>freq [1,-1,1,-1,1,-1,1,-1]
>2-way interactions
>stim x cond [1,1,0,0,0,0,-1,-1]
>freq x cond [1,0,0,-1,1,0,0,-1]
>3-way interaction
>stim x cond x freq [3,2,2,-2,2,-2,-2,-3]
Assuming that the columns of your 2nd-level design matrix are ordered as
your DESIGN above (ie stim rotating slowest; freq rotating fastest), then
your main effect contrasts are correct.
However, your interaction contrasts are not correct. Think of an interaction
as a difference of differences. Then a basic stim x cond interaction in a
2x2 design would be [1 -1] - [1 -1] = [1 -1 -1 1]. Then because you are
collapsing across a third factor (freq) in your design (for the two-way
interaction), you need to "average" (sum) over the levels of that factor, so
that the correct interaction is:
stim x cond [1,1,-1,-1,-1,-1,1,1]
Likewise:
freq x cond [1,-1,-1,1,1,-1,-1,1]
(and one you forgot?:)
stim x freq [1 -1 1 -1 -1 1 -1 1]
Then generalising to the three-way interaction, [[1 -1] - [1 -1]] - [[1 -1]
- [1 -1]], ie
stim x cond x freq [1 -1 -1 1 -1 1 1 -1]
(Note that the sign of these contrast weights does not matter for an
F-contrast, but will matter for a T-contrast (ie you can reverse all signs
of each of above to test opposite directional effects with a T-contrast).
More generally (eg for factors with more than 2 levels), it is easier to
think of the standard ANOVA effects in terms of the Kronecker product
(Matlab's "kron") of two component contrasts for each factor: a differential
effect (eg [1 -1]) and a common effect (eg [1 1]). For more details, see my
Technical Note on ANOVAs in SPM (available on my webpage).
Finally, for an example of such a 2nd-level analysis of sensor-time images
in a factorial repetition priming paradigm, see:
Henson, R.N., Mouchlianitis, E., Matthews, W.J., & Kouider, S. (2008).
Electrophysiological correlates of masked face priming. Neuroimage, 40,
884-895.
Best wishes
Rik
|