Hi all-- Getting a lot of different answers on this from my colleagues, curious what others might think.
So I am analyzing a new study and working on setting up the single-subject (i.e. first-level) analyses. The study works as follows: It is an event related design, and there are seven different types of stimuli presented to participants, including the null trials (fixation cross). Participants make a decision about all trials except fixation, and press either 1 or 2 on a button box. There is quite a bit of variability in what participants are pressing-- they aren't always pressing 2 when they see Stimulus A, for example. So button-press and stimuli type may be related, but they don't match entirely.
Because I hadn't prepared the button-press data yet, I started by entering the 7 types of stimuli as my EVs. All of the EVs were therefore completely unique. Now, however, I want to enter the behavioral data to see how neural response differs between 1 and 2, and also to account for the fact that button-press may be a confound with stimuli type (participants may end up pressing 1 more than 2 when they see Stimulus A, for example).
Can I enter the button-presses as additional EVs? (EV 7. Trials where participant pressed "1" EV 8. Trials where participant pressed "2"), and rely on FSL to "orthogonalise"? Or should I divide all of my existing EVs as follows:
EV1. Stimulus A-- participant pressed 1
EV2. Stimulus A-- participant pressed 2
EV3. Stimulus A-- participant ran out of time and pressed nothing (happened rarely)
EV4. Stimulus B--- participant pressed 1
etc. etc.
This gives me a LOT of EVs-- 19 of them-- and I don't know if that is more of an issue than using fewer but non-orthogonal EVs.
In case it is helpful to know, the task runs about 11 minutes and participants see 20-35 of each type of stimulus.
Ultimately the question boils down to this: Is it preferable to have fewer EVs, which are not entirely orthogonal, or more EVs which are orthogonal?
Thanks very much,
Lauren
|