> I'm not exactly sure what you mean by a "baseline response" to a condition.
> Variations in the baseline signal level of individual sessions will not
> directly affect fits to a normal response regressor nor an FIR regressor,
> because the mean signal level is accounted for in the first level
> modelling.
Hi, Eugene. I thought about this, but in my experiment I have multiple
conditions (six, actually). While the first-level modeling will take
out the overall session baseline for each subject, it's
possible/likely that subjects will come still come in with different
baselines for a given condition, since these are a subset of the whole
session. That's the problem that I was hoping to account for by
specifying the subject-specific regressor.
>
>>> our f-test, which is just asking if the 5 FIR bases are different from
>>> each other, regardless of the individual subject baselines.
> If you're interested in this question, just drop one of the 5 FIR evs in
> your group level analysis, and do an F-test across the remaining four
> (keeping the subject evs in). Each subject ev will model a steady level of
> signal across the 5 timepoints, while a significant fit of any the FIR evs
> will reflect a significant deviation of that FIR basis from the one left
> out. I'm still not sure that this is an interesting test - as it tests
> deviations from a rectangular response. I guess you could make the subject
> EV match a standard HRF, and then your F-test might show regions
> significantly deviating from this?
Let me be a little more clear on what I was hoping to do, just in case
someone can point out that it's totally absurd:
I was hoping to use the f-test to identify areas of the brain that
were "responding" in some way to the condition of interest. From
there, I was hoping to merge the five individual COPEs into a
time-course for that region that would show me the average HRF for the
group during that condition.
If I drop the 5th EV, I'm prevented from doing the second step. Also,
I'm not entirely clear that it's asking the question I want to ask,
though I haven't yet thought this all the way through...
Does that make any sense, or is this just a ridiculous way to analyze data? :)
Thanks for your responses,
Todd
|