Dear all,
This question regards the interpretation of fitted timecourses that are
negative (i.e. go below zero). This has been discussed on this list a
number of times before but I cannot say I am sure of the answer looking at
the old posts.
In a second-level analysis in SPM2, I found a cluster (let's call it VS) of
significant activation for the contrast A-B. I took this to mean that A
activates this area significantly more than B.
Then I wanted to extract the time course from this cluster. This is an
event-related design with TR=2.0s and ISI=3s, with randomization and
jittering. There are 3 sessions, six conditions in session 1, and 11
conditions in sessions 2 and 3. I am modelling this using the canonical HRF
and its two derivatives, thereby getting three betas for each condition. To
get the contrast images A-B for each subjects, the contrast only looks at
the first beta for A and B (the two derivatives are only included to reduce
variance). Then these contrast images are used for the second-level
analysis, which gave significant activation in A-B for cluster VS.
The design parameters at the individual subject level are the following:
Basis functions: hrf (with time and dispersion derivatives)
Number of sessions: 3
Trials per session: 6 11 11
Interscan interval: 2.00s
High-pass filter: cutoff 128 s
Serial correlations: AR (0.2)
Global calculation: mean voxel value
Grand mean scaling: session specific
Global normalisation: none
I used Plot->Event-related response->Fitted Response to get the estimated
hrf at the voxel of interest for each subject. When I averaged them out
across the subjects, I got the following graph:
xx xxxxxxxxxxx
-- xx xx ------------
-- xxx xxx --
-- xxx --
-- ---
-- ----
-- --
--
Where A is x, and B is --.
So they are both going down! My question is what does this mean? Does this
genuinely mean that there is deactivation in both A and B and there is
simply less of it in A than B and so I get a significant difference? Is the
timecourse relative to anything? I seem to recollect from the SPM mailbase
that the fitted timecourses are somehow scaled to be relative to an overall
brain mean. Does it mean that if we somehow adjust for the overall brain
mean, this curves will become positive? How does one adjust for the overall
brain mean?
I will be very grateful for your input.
Thank you,
Martin
|