Hi. yes - this is the behaviour I would expect:
When you do a group test you ONLY get the group-wise contrast B-A (A is
specified as the "control" group). In general ALL thresholding and
colour-blob-rendering in FEAT (and similar methods) assumes you set a
threshold Z and only show parts of the image which exceed this. Thus when
you reverse the A and B members you will get the opposite test - and will
not get the same results.
So - there are two easy ways of testing both 2-1 and 1-2; either run two
group analyses, swapping the ordering, or, even more easily, just include,
in the original (first-level) analyses, extra ("negative") contrasts. To
explain in more detail - say that you have two EVs at the first level, and
have generated colour-blobs with the contrasts [1 0] and [0 1]; if you
also included [-1 0] [0 -1] then you would get the reverse group tests
included when you ran the initial 2-1 group analysis.
second thought-for-the-day: Note that this highlights the ambiguity always
present in such analyses:
a [1 0] contrast at first-level, fed into a 2-1 (A=1 B=2) group test means
"EV1 > rest for group 2 compared with group 1"
OR
"EV1 < rest for group 1 compared with group 2"
it is only possible to disambiguate these interpretations by
contrast-masking - for example, if you tested the [1 0] contrast at group
level for the 2 groups separately, and generated masks which show
+actvation, you can then mask the inter-group test with these results to
prove that interpretation 1 is the correct one.
final note - the next release of FSL, due out in December, will contain a
much more complete "general" second-level GLM - so you will be able to
create both test directions directly without either of the above
solutions.
Thanks, Steve.
Stephen M. Smith
Head of Image Analysis, FMRIB
Oxford University Centre for Functional MRI of the Brain
John Radcliffe Hospital, Headington, Oxford OX3 9DU, UK
+44 (0) 1865 222726 (fax 222717)
[log in to unmask] http://www.fmrib.ox.ac.uk/~steve
On Mon, 8 Oct 2001, Jim Ibinson wrote:
> I have 2 data sets I want to compare and analyze for differences in
> activation. I'll call them task 1 and task 2. When I enter task 1 as
> the "A" (control) data set and task 2 as the "B" (Test) set, I get a
> different answer than whe I enter task 2 as "A" and task 1 as "B", as
> displayed by the picts in the report. It seems as if the results are only
> showing positive Z values. I tried to enter a negative Z for the min value
> but couldn't. I also tried the auto-scale Z score option and that made no
> difference. Is there a way to display negative Z values as well? Or am I
> thinking about this all wrong?
>
|