 |
 |
Hi Graham, On 9 August 2010 16:45, Graham Murray <[log in to unmask]> wrote: > In an event related fMRI study I am interested in EVs where the value of the input (I am using the 3 column format to specify my EVs) varies across the experiment. > I decided to input one EV where the value of the input was always positive (varying between 0 and 1 as I'm looking for activations) and one EV where the value of the input was always negative (varying between 0 and -1 as I'm looking for deactivations). > However, on re-reading the Feat-in-detail webpage, I see that "ALL columns are demeaned before model fitting". > > So, it seems I can't obtain my original goal of finding areas where there are activations (relative to the unmodeled baseline) and those activations vary in magnitude in a way that proportional to a behavioural variable and other areas where there are deactivations (relative to the unmodeled baseline) and the deactivations are variable in magnitude proportional to a behavioural variable. I'm unclear whether your positive and negative EVs are the same (except for sign)? If so, you only need one: you can assess positive or negative correlations with a behavioural variable by assessing +1 and -1 contrasts. You may additionally want to have a mean response regressor (i.e. where the 3rd column is constant), otherwise you may not be able to show that the responses actually follow the variations in value, rather than there simply being an av non-zero response on average. You need to be a bit careful in interpreting modelling like this: you could have positive activations that are negatively correlated with value, for example. Hope that helps. Eugene > > In case it is of help in clarifying my study, my aim is to look for areas where signal strength correlates with positive and negative prediction error value (different value for every trial) in an associative learning experiment. > > Any comments most welcome > > Thanks > > Graham > > >