Just to follow up, after a little more thought, this is not quite
similar to detrending and then picking out the time points.
To make that accurate, you would have to censor the time
points of interest, since they would not actually affect the
trend estimate. That would likely make a non-trivial
difference.
To put that into perspective, the regressors of interest
(for TR-locked TENTs with IM) are spike functions, which
behave like censoring with respect to the rest of the model,
meaning the data values at those time points would not
affect the rest of the model.
- rick