Hello list!
I have a design where I use visual stimuli (6ms long), tactile stimuli (2ms), and combined, visuo-tactile stimuli, composed of the two previous ones (6ms and 2ms, common onset). I have questions regarding the modeling of such short events, and also regarding the best solution to model the different length events with respect to the contrasts I want to analyse.
Regarding how to model very short events such as mine, I've read a few threads, but did not get a clear answer - to me, that is. Steve Smith has said that " the sampling of the "high resolution temporal model" generated initially by feat_model when the model is processed is 0.05 seconds, so anything shorter than this may get lost" (https://www.jiscmail.ac.uk/cgi-bin/webadmin?A2=ind04&L=fsl&D=0&1=fsl&9=A&I=-3&J=on&d=No+Match%3BMatch%3BMatches&z=4&P=274675). He then advised to modify and recompile feat_model.c, which I'm willing to do if that's the best solution, but I'll definitely need help with this... :-)
On the other hand, in another thread (that I can't find any more,sorry!), I think it was Steve again who advised to multiply actual durations by a common factor so that they would all be superior to 100ms or so, saying that this should not modify the model sensitivity, if I undestood correctly. I've done this, using 200ms and 600ms to model my tactile and visual events, respectively. However, I clearly get stronger activations for tactile than visual events, even in areas related to common components of the task (i.e. V1 related to the permanent fixation LED - my visual stimuli activate more V5). This leads to artificial activations when I contrast tactile vs. visual stimuli, whereas the opposite contrasts give empty stat maps. Could this be because 600ms is really to much longer than the actual duration?
By contrast, if I model all events with a 400ms duration, which is what I had decided to use for the visuotactile events anyway (the average of tactile and visual events duration, x100ms), then contrasts against rest don't change, but the contrasts between tactile and visual events are much "cleaner".
Having said that, I was wondering whether it would be worth it to model everything with real durations, and thus maybe modify feat_model to have a better temporal resolution? Would it make sense to model all events with 100ms or 200ms durations, to use durations >100ms that would not be so much longer than the actual ones?
The second question is already somewhat embeded in the first: should I model my different events with different durations, knowing that I would like to run contrasts between them? If so, what duration should I pick for the visuotactile events?
I hope my message isn't too confusing... Any help with those issues woujld be greatly appreciated, as always!
Best regards,
Stephane
|