I'm attempting to gain a better understanding of PPI and have come across
two possible strategies and need some input into the advantages and
disadvantages of each.
Method 1: The PPI.ppi column is modeled as usual using 1 or -1 for each
event type to estimate the task interaction.
Method 2: Compute 2 different PPI analyses; using 1s and 0 and then 1s and 0
for each task independently. Then do a subtraction of the two PPI maps.
Second -level statistics would then be the same.
Method 1, at least the way I've learned how to do PPI, seems easier to
implement. However, Method 2 seems to have the advantage that there is no
assumption about the parameteric relationship between the tasks (e.g. 1 and
-1 as the multiplier), especially if the relationship not of that form (e.g.
one task is has twice the amplitude of the other.
Any thoughts would be appreciated.
|