Dear experts,
this topic has been discussed partially before, but for my particular case, there seems to be no clear consensus.
I have a relatively simple auditory category learning design, where participants are presented with tones that differ in duration and frequency, with different spreads of duration and frequency values. Participants have to attribute each stimulus to one of two possible categories, and receive immediate feed-back. The task is done while we recorded 2-seconds of EPI volumes in a sparse design, i.e. after the participants button presses.
In 25% of all cases, instead of tones, we would present silent trials (null trials) that should serve as baseline for the subsequent analyses.
Behaviorally, we find that the Euclidean distance of each tone to a most ambiguous 'mid-point' in the acoustic space (of duration and frequency) modulates accuracy: The closer a given tone to this ambiguous point, the less accurate the response.
For this reason, we wanted to modulate the Euclidean distance in SPM, on the first level, as parametric modulator of 'tone' (this distance is available for each tone).
If modeling null trials, the resulting design matrix has three columns, one for 'tone', one for the PM 'distance' and one for 'nulltrial'.
While it seems straightforward to test 'tone' against 'nulltrial' (with the vector 1 -1), it is less clear whether the same is allowed for the PM 'distance'. I am not asking whether it is possible (that I figured out already), the question is, is it legal to compare the PM 'distance' against the 'nulltrial' (using the vector 1 -1)? Is it just a different baseline (e.g. if specifying the PM 'distance' with 1 and the nulltrial with 0, I understand that this tests against an implicit baseline)?
Thanks for comments & help,
Mathias
----------------------------------------------------------
|