running a t-test by combining different regressors

AFNI Message Board
1

Hello, I was wondering whether it was possible to run a 3dttest command by combining several regressors.

For example in my task I have 9 regressors. and I would like to combine 4 of them and compare them with 4 others (and not include the last one).
Is it possible with the 3dttest command or is it something I should have set up at the individual level?

Thank you!

2

Oh i saw a post from 2010 where it was suggested to use 3dcalc to combine each event; and weight them according to the number of trials. Given that I have the same number of trials per condition I guess I will just run 3dcalc like this
3dcalc
-a regressor1.tlrc
-b regressor2.tlrc
-c regressor3.tlrc
-d regressor4.tlrc
-prefix all_four.tlrc
-expr (a+ b + c + d)

and see if that works

3

Hi Lysianne,

3dttest++ does not have a direct mechanism to do that.

Such a GLT could be added to the regression model, but that would mean re-running the analysis. After the fact, it would be best to just run 3dcalc to extract the contrast that you want (per subject), and then feed those resulting contrast datasets to 3dttest++.

Specifically, it might make sense to get the average of 4 betas in on result and the average of the other 4 betas in another result, and then do a paired t-test between them.

Does that seem reasonable?

  • rick
4

Yes it seems that I'm on the right path then
and just to confirm I would use the regressor#0_Coef for each of them right ?

Thank you very much Rick!

5

Yes, that looks right, assuming there is only 1 basis function per stimulus class.
If there more, they would look likereg#0_Coef, reg#1_Coef, reg#2_Coef, ...

  • rick