Print

---

Hi Helmut (and anybody else who may take an interest),

I am back ;-) I'm responding to this email, as my current question is most
relevant to this last email of yours, as well as your email prior to that.

I now have 9 regressors in total, which are:
#1) the task/rest design (which I believe is input automatically as soon as
I define a condition with onsets)
#2) a PM - which is the behavioral performance data
#3-8) the realignment motion parameters (from the rp.txt file)
#9) a constant

My question is:
I think that for sure I am interested in including the realignment motion
parameters in my model.
How can I set up a contrast that tests for these AND for the task/rest
design for instance?
Or a contrast that tests for these AND the behavioural performance?

If I set up a contrast like: 1 0 0 0 0 0 0 0 0, I am ONLY testing for the
task/rest design I think, right?

Thanks,
Joelle



On Fri, Jun 5, 2015 at 3:50 PM, H. Nebl <[log in to unmask]> wrote:

> Dear Joelle,
>
> > the PM should be represented by the 2nd row in your design matrix
> This should have read "by the 2nd column". Column = different regressors,
> row = different time points. Sorry for that.
>
> > Yes, I think I have 3 regressors as you suggest!
> Usually the realignment parameters (rp text file in the preprocessing
> folder) are added to the design, for that purpose select the rp file as
> "Multiple regressor". This should result in 6 additional columns for the
> three translations and the 3 rotations.
>
> > If I were interested to test both first (task activation) and second
> (behaviour PM) together, how can I do this? Something like [1 1 0]?
> [1 1 0] tests for the sum of beta estimates for the task regressor and
> beta estimates for the PM. This is a valid contrast, but it might be
> meaningless. Maybe you want to see voxels where any effect occurs, be it
> task or PM? This would be achieved with an F contrast [1 0 0; 0 1 0].
>
> > Do you mean after having done the first-level analysis, to forward the
> con images into a 1 sample t-test for the second-level (group) analysis?
> Yes. This is the two-stage summary statistics approach for mixed effects.
> Estimate beta values on indiviudal = single-subject = first level, forward
> these into a group = second-level model. There are special modules for
> group models, I'd suggest to look at the manual to get an impression of the
> procedure (again, you'll need a specification module, this time "Facorial
> design specification" to set up the design, followed by an estimation
> module).
>
> One note: The attached figure looks like unnormalised data. Of course it's
> correct, but within SPM the default procedure would be to normalise the
> realigned data, followed by smoothing (in contrast to e.g. FSL).
> Single-subject models would be based on the normalised, smoothed data,
> beta/con images from these single-subject models would be forwarded into
> group models.
>
> Best
>
> Helmut
>
>