Dear Donald,
thanks for the reply
> I am not aware of any rule that requires contrasts to sum to zero
Hmm, I'm not a stastitician, but for instance in the statistics textbook
"Discovering statistics using SPSS" by Andy Field it says on p.365 "Rule
3: If you add up the weigths for a given contrast the result should be
zero." Ok, this refers to planned contrasts in the ANOVA, but I always
thought that this assumption holds for SPM as well (with the only
exception of a contrast in which only one condition is set to 1 ant the
rest to 0, such as "1 0 0 0"). Maybe SPM will calculate a result, but
are you sure that can it be interpreted in a meaningful way? That means,
is the result of the contrast
A B C
1 -1 -1
really the subtraction of A - B - C? If the beta-values of a given voxel
are A=4, B=2, C=2, then this contrast would reveal 0, right? If this
works, I'd be more than happy :-)
> in some cases they do need to sum to zero (e.g. when parameters are
not uniquely estimable).
What does this mean exactly for the fMRI analysis? For instance, there
is mostly at least some correlation between regressors - is this already
a problem which requires the weights to sum to zero?
> A - (B+C) is the same as (A - B) - C
> I am not entirely sure why subtracting C from the difference of A and
B would be desirable.
Let's keep with my notation A - (B+C) or A - B - C. This is desired for
instance in the dual-task situation: B and C are two single tasks which
are performed in isolation or as a dual-task simultaneously (condition
A). If B activates left motor cortex (right hand response) and C
activates right motor cortex (left hand response), A is expected to
activate left and right motor cortex (as two responses, one with the
right and one with the left hand are performed in the dual-task). To
test for dual-task specific effects (i.e. over-additive effects), it
would be desirable to subtract both single-tasks from the dual-task.
> If you are interested in where A is different from B and C, use a
conjunction.
I'm interested in where A is different from B plus C. A conjunction can
be used for this question, but this specific question then needs further
assumptions which are often hard to meet (see e.g. D'Esposito et al.,
1995, Nature, who used a conjunction for this type of question).
> Use an F-test if your interested in where A is different from B or C.
Unfortunately, this is of no interest to me.
>> Now I had this idea: Is it possible to create an empty image (filled
>> with zeros), one for each participant, and then use this dummy image
>> during first-level contrast generation?
>> Then I would have (Dummy + A) - (B + C)
>> Dummy A B C
>> 1 1 -1 -1
>> As the dummy contains only zeros, (dummy + A) should be identical to
A
>> alone.
> This will produce the same results as 1 -1 -1.
That's what I hoped for. I just thought that "1 -1 -1" is an invalid
contrast which cannot be calculated. Unfortunately, I presently can't
check in SPM whether 1 -1 -1 gives a valid result.
>> A closely related question: I once read that each task can be
decomposed
>> into a task-specific effect and a task unspecific effect. In the
above
>> example, I would subtract the task-unspecific effect twice from A,
which
>> is not correct. Therefore, it was suggested to add the resting
baseline
>> as an estimator for the task-unspecific effect to A, i.e. (Base + A)
-
>> (B + C):
>> Baseline A B C
>> 1 1 -1 -1
>>
>> This is an interaction contrast and makes sense in my case[*].
However,
>> I would need a reference to bolster this assumption of a
task-unspecific
>> base effect which has to be added, but I can't find the reference. I
may
>> be wrong, but the faint recollection I have suggests a Karl Friston
>> paper from roughly the mid 1990s.
> Not sure about this last part. Modeling baseline is not necessarily
good
> though, especially in event-related designs.
I' aware that this approach has drawbacks, but I'm particularly looking
for the reference I can't find...
Department Psychologie
Neuro-Cognitive Psychology
Ludwig-Maximilians Universität
Leopoldstrasse 13
80802 München, Germany
Tel. +49-(0)89-2180 6778
Fax. +49-(0)89-2180 4866
www.psy.uni-muenchen.de/ncp
Office: Martiusstr. 4, Room 6
______________________________
|