When you construct your model, the mean is included—this is the last
column in the design matrix. This means that all other regressors
describe activity relative to the mean across all scans. A "main
effect" of a condition, such as in your example of 5 conditions (and
one column for the mean)
1 0 0 0 0 0
simply tells you that the activity described by the first column is
significantly greater than zero—in other words, significantly greater
than the mean level of activity over the entire analysis.
You could imagine an experiment in which many events were presented
with very little pause and no baseline trials. In this case brain
activity may be quite high relative to "rest" for your entire
experiment, for every condition. In such a situation, if you did a 1
0 0 0 0 0 contrast and did not find any significant voxels, it would
be wrong to conclude that there was no activity during this condition;
you could only conclude that the activity was not significantly
greater than the mean level (which would be high).
If you now imagine the same situation of many trials, but include some
baseline trials or null events, it may be that the activity following
the baseline trials will be lower than the average level of activity.
In this case you may get a negative weight for the beta value
associated with your baseline trials, because these events result in
less than average activity. Thus, it is possible that a condition
does not differ from the mean, because the mean level of activation is
1 0 0 0 0 0 0
(where you have 5 conditions of interest, baseline, and mean = 7 columns)
but does differ from your baseline events:
1 0 0 0 0 -1 0
because these activities have less-than-average activation.
Although the above explains why these two approaches might sensibly
provide different results, in my experience it's not uncommon for the
results to be quite similar (but not identical). A lot will depend on
the specifics of your design and stimuli.
Hope this helps!
On Mon, Jan 26, 2009 at 6:53 PM, Liliana Demenescu
<[log in to unmask]> wrote:
> Dear users,
> I have an fMRI event-related design with 6 conditions = 5 conditions of
> interest + a baseline condition and an interstimulus interval, varied from
> 0.5 sec. to 1.5 sec. I want to model my design using the HRF.
> My question is: what are the differences in case I model the baseline
> condition in my design, and if I don't model the baseline condition in my
> As I understood,
> 1) if I do model my baseline condition, then my contrast will be (for
> example condition1>baseline) defined in spm as [1 -1]. In this case the
> baseline is different from zero.
> 2) if I do not model the baseline condition in the design, then this is set
> to zero and my contrast should be (e.g. condition1) [1 0].
> As I saw, the results of the main effect of condition1 do not differ between
> the two designs. Is all this correct? Should I see/expect differences, and
> if yes, why?
> Thank you!