Dear Kevin,
>First, I have computed various contrasts for an experiment just conducted
>and am wondering what might be gained by examining directly the beta
>values for particular regions that show activation in the contrasts I've
>performed. Let's say that one contrast shows area X to be more active in
>condition A than condition B, whereas another contrast shows area Y to be
>more active in condition C than condition B. What would looking at the
>beta weights for area X and area Y in contions A, B, and C tell me?
>i.e., are the betas by themselves a valid dependent measure of the
>functional activity of a brain region?
The 'beta weights' or 'parameter estimates' are an indication of the sign
(+ve or -ve) and the size of the relationship between a column of your
design matrix and BOLD signal change at each voxel, considered after
variance explained by the rest of the columns of your design matrix (the
rest of your model) has been accounted for. If we assume that your
regressors (the columns of your design matrix that represent A,B and C)
model your evoked heamodynamics well, then your parameter estimates/beta
weights are a direct indication of the strength of the neurovascular
response in that voxel (assuming, as we do, that the height alone of the
hrf is a good metric of the underlying signal change).
So, to take your example, looking at the betas for X and Y in conditions
A,B and C is an anecdotal way of looking at a question that you have tested
more formally in your contrasts. In your particular example, you would get
some idea of what is going on in X in condition C, and in Y in condition A
(since your test did not explicitly examine this). In addition, you are
able to characterise the relative differences between A/B and B/C in
greater detail. So, anecdotally, the betas are a measure of the activity in
a particular voxel. By themselves, however, they may not be 'valid', as
only 'constrasts' are valid in SPM. See the following
http://www.mailbase.ac.uk/lists/spm/1999-10/0207.html for a fuller
explanation of contrasts.
>Second, if betas can be used as a dependent measure, can you then perform
>statistics on them? In the above example, could I extract betas for areas
>X and Y across my 3 conditions, and then use t-tests or other contrasts to
>compare the magnitudes of the betas?
In each of your contrasts, you are effectively performing t-tests when you
compare A vs. B and B vs. C. The values entering into your t-tests are the
parameter estimates/beta weights at each voxel. All of the tests that you
perform on your design matrix will involve some kind of comparison of these
weights.
I may be missing something, but I think the answer really is just this simple.
>Muchas gracias, many thanks, and danke schoen.
De nada, nae problem, and bitte schon.
Dave McG.
|