Dear SPM experts/advanced users,
I would be grateful for advice on the following issue:
We recorded EEG simultaneously with fMRI during a motor paradigm (LH, RH, both hands tapping, no tapping [rest]).
At the first level, with 6 regressors we model the three tapping conditions (LH, RH, BH) and three measures derived from the EEG (EEG1, EEG2, EEG3).
At the single subject level, we construct simple T-contrasts, e.g. effect of LH, RH, BH, EEG1, EEG2, EEG3 as simple "ones" above the regressors (conditions): 1 0 0 0 0 0, 0 1 0 0 0 0, etc, contrasting each condition against baseline.
Also, we are interested in what variance the EEG (EEG1, EEG2, EEG3) explains beyond the paradigm (LH, RH, BH), e.g. construct an F-test [zeros(3,3) eye(3)].
Now, we would like to perform a 2nd level analysis for both the T- and the F-test described for the first level.
We are not sure which model we need to build:
One option (according to the SPM8 manual (Chapter 29, p. 245)) is a one-way ANOVA. In this case, the analysis can be specified in SPM as a ‘Full Factorial’ design with one factor (‘condition’) having 6 levels. So the single-level con images (one for each subject) are grouped into 6 cells within one factor).
The question is: do we take into account the inter-subject variability using this method? Do we need to?
An alternative way taking the inter-subject variability into account, would be a “Flexible Factorial’ design with the factors: ‘subject’ and ‘condition’ ; the con images are grouped into 42 cells (the number of subjects) by 6 images (the number of conditions) in each cell. Then, depending on whether only one (‘condition’) or both ‘condition’ and ‘subject’ factors are specified as ‘Main effects’ within this design, we get different results.
We are interested only in the ‘condition’ factor, but we would like to model ‘subject effect’ to be sure that it is taken into account (if this has to be).
So the final explicit questions are:
1. how should we model our data at the 2nd level?
2. will the ‘subject’ effect be taken into account in the described Flexible Factorial design if only one (‘condition ‘) factor is specified as a main factor?
Thank you for your help in advance.
Sergey
|