I'm hoping to receive feedback on how to set up a fully within-subjects factorial design for 27 subjects. I have two factors, Factor A with levels A1 and A2, and Factor B with levels B1, B2, B3, and B4. Each subject completed 3 runs (randomized event-related with all conditions represented in each run). I'm thinking that the best way to proceed is to model the two main effects and the interaction at the first-level for each run (where EVs 1 through 8 represent the 2*4 = 8 conditions):

                 A1B1 A1B2 A1B3 A1B4 A2B1 A2B2 A2B3 A2B4
                  =    =    =    =    =    =    =    =
                 EV1 EV2 EV3 EV4 EV5 EV6 EV7 EV8 F1 F2 F3
C1: A(1-2)         1    1    1    1   -1   -1   -1   -1   x
C2: B(1-4)         3   -1   -1   -1    3   -1   -1   -1      x
C3: B(2-4)        -1    3   -1   -1   -1    3   -1   -1      x
C4: B(3-4)        -1   -1    3   -1   -1   -1    3   -1      x
C5: A(1-2)*B(1-4)  3   -1   -1   -1   -3    1    1    1         x
C6: A(1-2)*B(2-4) -1    3   -1   -1    1   -3    1    1         x
C7: A(1-2)*B(3-4) -1   -1    3   -1    1    1   -3    1         x

Then, at the second-level, I would used a fixed effects analysis to calculate the mean for each cope within each subject (across the 3 runs).

        EV1                        EV1
Input 1 (run 1) 1   C1: subject 1 mean 1
Input 2 (run 2) 1
Input 3 (run 3) 1
        EV1                        EV1
Input 1 (run 1) 1   C1: subject 2 mean 1
Input 2 (run 2) 1
Input 3 (run 3) 1

and so on... for all 27 subjects.

Then, at the highest-level, I would use a mixed effects analysis across all 27 subjects to model the group mean.

                EV1 EV2 EV3..EV27     EV1 EV2 EV3..EV27          Input 1 (subj 1) 1  0  0.. 0     C1:   1 1  1..   1
Input 2 (subj 2) 0  1  0.. 0
Input 3 (subj 3) 0  0  1.. 0
.
.
.
Input 27 (subj 27)0  0  0.. 1

I've done my best to adapt the examples posted on the website for the Multi-Session & Multi-Subject (Repeated Measures - Three Level Analysis). Does this plan make sense?

As a follow-up, subjects provided a rating after each trial. If I want to enter that in the first-level analysis, I need to demean it, but should that be demeaned *within* each condition so as not to soak up mean differences between conditions? Second, should the demeaned values be scaled to a maximum of 1? Finally, if I'm using a 3 column format text file, would the demeaned rating value be entered as the "event", i.e., replacing the constant value of 1 that's currently entered in column 3 of my 3 column format EVs at the first level?

Thanks in advance for any comments you have!

Best,

Heather

--
Heather L. Urry
Department of Psychology
Tufts University
490 Boston Avenue
Medford, MA 02155

email: [log in to unmask]
phone: 617-627-3733
fax: 617-627-3181