Hi Cornelius,
Probably the most effective method is to have one regressor for each task
and an additional "speed" regressor. For each trial/block, enter an
intensity from 0 to 1 corresponding to the speed of execution. Since you
split the speeds into thirds, you will have values of [0, .5, 1]
corresponding to slow, medium, and fast. This regressor will identify
voxels that increase activity with execution speed. You can also include a
regressor that codes these trials/blocks [1, .5, 0] to find voxels
correlated to task "difficulty". This type of analysis assumes linearity
between voxel intensity and the cognitive process indexed by execution
speed. For this type of regressor, you won't have to do anything fancy at
the group level and you won't have any empty contrasts.
The one other thing to consider is if the three tasks are sufficiently
different that they tap into different cognitive processes. Then a single
"speed" regressor may not be measuring the same thing. In this case, you
should create three task regressors and three speed regressors.
cheers,
jack
On Tue, 3 Nov 2009 19:24:28 +0000, Cornelius Werner
<[log in to unmask]> wrote:
>Dear list,
>
>I've run into trouble with a pretty basic fMRI paradigm. We measured 10
patients
>suffering from a neuropsychiatric disorder and 10 controls. Both groups
performed three
>types of tasks plus a rest condition (A-Rest-B-Rest-C-Rest...3x). When
contrasting the
>tasks on the second level, there are meaningful differences across groups.
So far, so
>good.
>
>Now, however, we also acquired behavioral measurements according to the speed
>subjects performed the tasks with (difficulty C > B > A). I am interested
in activations
>correlating with the speed measurement, therefore I imagined a parametric
design would
>be in order. Here's what I did: first I classified each task condition as
being "fast",
>"medium" or "slow" as compared to the according group performance
quantiles. Then I
>designed EVs for each of the three categories and entered them into the
first level design
>matrix.
>
>Unfortunately, some subjects only exhibited "fast" speeds, leaving "slow"
and "medium"
>regressors empty, thus resulting in a rank deficient design matrix. Even
omitting the
>"medium" EV did not help, obviously. Others had two different speeds, no
one had three
>types.
>How can I specify a design matrix of this type, preserving all three EVs
for a higher level
>analysis? Or should I assign "slow", "medium" and "fast" to subjects on the
second level?
>
>Thanks for your help - any comment is appreciated!
>Cheers
>Cornelius
|