This is a resend and append to an email sent last week. I have yet to receive any response so I want to try again...
Dear SPM experts,
We have the following design with exactly 3 visual stimuli:
A: event 1 (400ms in duration)
B: event 2 (400ms in duration)
c: cross-hair fixation (1.3s)
Event stream is something like:
AcAcAcAcAcAcBcAcAcAcBcBcA..... (Stated in words,we present lots of As with occasional Bs dropped in, and cross-hairs separating all stimuli from one another.)
We want to ask the question, what areas of the brain appear to be active in response to stimuli A and stimuli B (clarificantion: each separately, NOT the interaction).
Seems obvious to us that a multisubject (we have 24 subjects) conjunction analysis is appropriate, do you agree? If not, what analytic method is recommended?
If conjunction analysis is appropriate then could someone please offer us specific instructions as to how to implement. Below are 3 strategies, the first we think is correct for fixed-effects, the second we think is not a conjunction but may be an approximation, the third we think represents a "true" random effects conjunction.
Strategy 1(we think is correct, but fixed effects):
1. Create GLM with onset times for A as one vector and B as a second vector.
2. Determine fixed effects of A across all 24 subjects in the single matrix created in step 1.
3. Determined fixed effects of B across all 24 subjects in the single matrix.
4. Conjoin contrasts created in steps 2 and 3 above (with control key selection in the contrast manager)
Strategy 2(which we think is just averaging the effects and may be a variation of a conjunction):
1. Create GLM as stated in Strategy 1
2. Determine fixed effects for a [1,1] (x24) contrast corresponding to A AND B effects for all of the 24 subjects in one matrix.
3. Bring the 24 con. images from step 2 into a basic model and use a t-test to find where in the brain the group con.images have activation >0.
Strategy 3 (which we think is best and representes a "true" random effects conjunction)
1. Create GLM as noted in strategy 1
2. Determine fixed effects as noted in steps 2 and 3 of strategy 1
3. Carry con.imgs from step 2 to the second level and run and conjoin those effects with a multiple regression absent the constant term as described in :
http://www.jiscmail.ac.uk/cgi-bin/wa.exe?A2=ind0104&L=spm&P=R48885
Note that we have run strategies 2 and 3 and results are similar (but not exactly overlapping) so we posit that they test for similar effects. We would appreciate any insights as to which analysis is best and why. Note we did not yet run strategy 1 one because we prefer random effects and because it is more labor instensive to set up.
Thank you and best wishes,
Michael Abrams
Kennedy Krieger
Johns Hopkins
Balrimore,MD
|