Hi,
If you have two regressors that are the same then the fitting
mathematics will be poorly conditioned and you will not be able to run
such contrasts as:
[1 0]
[1 -1]
[0 1].
In this case you will have to drop the performace EV I suspect.
However, if the subject performs exacrtly the same for every task it
doesn't make sense to add it anyway. Maybe instead you maybe want to
be adding a performance regressor at the group level, in which case
you wouldn't need to have it like the in the first level.
Cheers, Steve.
On 24 Jan 2008, at 14:32, Hanne Lehn wrote:
> Hi again,
>
> I have now run the first level analyses discussed in previous e-mails,
> with an added regressor that models performance effects. For some
> subjects, I get a warning about my design being rank deficient. It
> seems
> it occurs whenever the performance regressor (EV2) is (too close to
> being)
> colinear to the main regressor (EV1).
>
> Example: The onsets and durations of EV1 and EV2 are always
> identical, as
> they model the same event; EV1=main effect, EV2= performance
> modulation.
> EV1 is always weighted with 1's (3rd column of onset file) Sometimes
> EV2
> is also weighted with only 1's (if performance equaled 1 point on
> every
> trial), or perhaps only 2's. In these cases my design is regarded rank
> deficient, and the resulting statistics don't make any sense (EV2 z-
> scores
>> 50).
>
> Does anyone know how to deal with this problem?
>
> From the past correspondence I understand there is no need to demean
> the
> performance scores before entering them as weights for EV2, as long
> as EV2
> is orthogonalized wrt EV1. What, then, could I do to get a proper
> design?
>
> Thanks a lot in advance!
>
> Hanne
>
---------------------------------------------------------------------------
Stephen M. Smith, Professor of Biomedical Engineering
Associate Director, Oxford University FMRIB Centre
FMRIB, JR Hospital, Headington, Oxford OX3 9DU, UK
+44 (0) 1865 222726 (fax 222717)
[log in to unmask] http://www.fmrib.ox.ac.uk/~steve
---------------------------------------------------------------------------
|