Doug Burman wrote a very helpful message that concluded:
> So neuroimaging does not localize function. What it does do is
> indicate that the activity of some areas are more involved in some
> types of activities or functions than others, which allows
> inferences to be made. As such, unthresholded maps represent a sort
> of probability map of preferential activity, whereas a thresholded
> map specifies the region where we can reliably expect such
> preferential activity. The "better" map depends on what point you
> want to make, as addressed better (and more succinctly!) by others
> in this thread.
Doug, I'm in mostly complete agreement with this message. Russ
Poldrack made an important related point earlier in this thread, which
is that if you collect enough data, a point null hypothesis can always
be rejected. So whole-brain fishing with fMRI can't ultimately be an
exercise in figuring out where activity is vs. where it isn't. I
agree with your view that this isn't a purely academic point, and I
appreciate your careful articulation of this point.
In my admittedly narrow view, this kind of study is better
characterized as (a) an exercise in generating hypotheses about where
there's more vs. less activity; and (b) deciding where there's a
big/reliable enough difference to care about (i.e., testing a
non-point null hypothesis). Goal b is a little sketchy in practice,
and I think most researchers implicitly let their sample size and
thresholding approach make the decision, without giving it too much
thought.
I don't think either goal is uniformly better served by one map
vs. the other. But I would like to suggest that which is the better
map (or more generally what should be presented) depends not only on
what point you want to make, but also on the data.
I would take minor issue with your saying that the things we can
legitimately conclude shouldn't be counted as localizing, just because
I take "localizing" to encompass anything that differentiates between
regions. If "localizing" has to mean all-or-none, then I'd certainly
want to rephrase a few of the things I've posted to the list.
dan
|