Mark,
It's very hard to tell which model/contrast is correct as you haven't
provided any details about whether this a group model or the first
level model. If these are group models, all of them are incorrect as
you have between-subject effect contrast of a within-subject design
(see previous emails on this topic). If these are first-level models,
then you should try to model each condition as its own EV. The one
limitation is when you don't have enough trials to estimate each
conditon, then you can collapse the conditions at the first level. You
should not be modeling the baseline at the first level, rather
baseline/condition 4 is the implicit baseline. Of course if
baseline/condition 4 is a stimuli that comes on and off, you should
model it.
On Sat, Aug 4, 2012 at 8:01 AM, Ming-Tsung Tseng <[log in to unmask]> wrote:
> Hi,
>
> In my fMRI paradigm, I have 4 conditions (1~4, and conditions 3 and 4 are the main EVs of interest) and 1 time slot of baseline of doing nothing.
>
> Model 1:
> I model the 4 conditions separately. To get activations of the main EV of interest, the contrast is (0 0 1 0) for condirion 3 and (0 0 0 1) for 4.
>
> Model 2:
> I model 1 and 2 conditions together as 1 EV (because they are both EV of no interest) and conditions 3 and 4 separately. To get activations of the main EV of interest, the contrast is (0 1 0) for condirion 3 and (0 0 1) for 4.
>
> I found the activations of both models are very identical, but it seems that activation in model 2 is slightly larger (and with slightly higher statistics) than that in model 2.
>
> Which model should I use!?
>
> Thanks.
>
> Mark
|