Hi Vadim
> I have one scan with two conditions.
> When I model it as HRF I get two regressors and my t-contrast is [1 -1]
> When I model it as FIR I get 10 regressors for each condition.
> Would the t-contrast now be: [1 1 1 1 1 1 1 1 1 1 -1 -1 -1 -1 -1 -1 -1 -1 -1
> -1] ?
> I tried this one, but the activations were drastically reduced so I doubt
> whether this is a correct one.
The t-contrast you described asks whether the average activation
(across all 10 bins) for the first condition is bigger than the
average activation for the second condition (again, across all 10
bins). This isn't 'wrong', but may not be what you really want to
know...and you're not taking advantage of having measurements at 10
separate time points. :)
What you probably want to ask is whether these conditions differ at
any time point; i.e., at time point 1, or time point 2, or time point
3...thus, you want an F test in which each row does one of these
comparisons. Let's say you just had 3 time bins * 2 conditions, it
would be something like:
1 0 0 -1 0 0
0 1 0 0 -1 0
0 0 1 0 0 -1
For 10 bins each, it would be larger, which you could write in Matlab
doing something like:
[eye(10) -1*eye(10)]
Because it's an F test, the sign doesn't matter; i.e. you will pick up
voxels where condition 1 > condition 2, but also where condition 2 >
condition 1. And, you can't tell which bin it was that is driving the
difference. So, it is important to plot the effects to be able to
interpret them.
Hope this helps!
Jonathan
|