Hi.
On Fri, 28 Dec 2001, Appu wrote:
> Hi, Thank you for your prompt response which was very helpful. However, I
> would I'd like to clarify one thing. We present 4 conditions, each involving
> words with selected emotional features: 2 neutral conditions (A and C),
> a emotion1(B), and a emotion2(D). We present the conditions in this order:
> ABCDABCDABCDABCD. There's no fixation period. From your posting, when you
> refer to the global demeaning I gather that FEAT will automatically average
> all of my A, B, C, and D data together and substract that single global
> average from all of the data.
just a quick point in order to avoid possible confusion: I wouldn't use
the work "global" here - that implies something like the global intensity
normalisation (making all time points ie volumes have the same intensity,
which is turned off by default in FEAT as in general it's a bad thing). I
am referring to the demeaning of each voxel's timecourse - ie yes, you are
correct, that's equivalent to averaging ALL A, B, C and D blocks into a
single mean volume which then gets subtracted from ALL volumes. Then each
voxel's timecourse has zero mean.
another quick point - I assume that we are talking about block designs and
not single events (in the latter case then of course the conept of
"baseline" is much more straightforward).
> Then it seems to me that I then have available two potential "baselines":
>
> - the global mean, which essentially could serve as a baseline against
> whioch every EV is compared. Thus, if the global mean is the baseline then
> would it be correct to model the neutral conditions, A (EV3) and C (EV4) in
> the full model in addition to the emotional conditions EV1 (B) and EV2 (D)
> and the result of using each EV would be EV compared to global mean.
>
> - the neutral conditions - where EV1 could be B vs. (A+C)/2, and EV2
> could be D vs. (A+C)/2. In this case modeling a neutral condition makes no
> sense.
>
> Is that correct? If so, I'm not sure which you'd recommend, or whether
> that simply depends on what questions I want to ask.
neither case is quite right:
the main complication here is related to whether the two neutral
conditions are to be treated as the same condition or different - I
assumed that they were the same before. Is that not the case? If they are
definitely different then you will need to (arbitrarily) assign one as the
"experimental baseline" and then end up with three EVs. It doesn't make
any difference which - you will and up choosing different contrasts
depending on which neutral condition you treat as baseline, but the final
results will be the same.
SO - two cases with example contrasts:
A) A=C ie neutral conditions are the same:
* model EV1=B EV2=D
* to ask B>neutral use contrast 1 0
* to ask D>neutral use contrast 0 1
* to ask B>D use contrast 1 -1
B) A!=C ie neutral conditions are not the same:
* model EV1=B EV2=C EV3=D
* to ask B>neutralA use contrast 1 0 0
* to ask D>neutralC use contrast 0 -1 1
* to ask B>D use contrast 1 0 -1
Hope this all makes sense!
THanks, Steve.
|