Hello Marco,
Based on the design matrix that you attached, it appears (visually) that
your covariates of no interest consist of repeating an identical value
for the "pre" and "post" values of a given subject (e.g., as would be
the case for say gender, or handedness). In this case, your design is
indeed rank deficient, as you can model both of those covariates as a
linear sum of the EVs (EV6-59) that model the subject specific means.
You need to drop the covariates of no interest from your model, as
whatever they are modeling is already accounted for by the EVs modeling
the subject specific means.
cheers,
-MH
On Tue, 2011-07-12 at 18:27 +0100, Marco Loggia wrote:
> Dear all,
>
> I am setting up a design matrix (see attached) in which each of my 54 subjects has a 'pre-' vs 'post-' paired comparison.
> The subjects are divided into three groups (EVs 1-2-3), EVs 4 and 5 are covariates of no interest, and EV6-59 are the individual subjects.
>
> Feat is complaining that "at least one EV is (close to) a linear combination of the others" (see below for more details). However, when I double check the model, this does not seem to be the case. Or am I missing something? Also, editing the design matrix from the GUI is veeeery slow. Is it possible that it is an issue with the size of the design matrix?
>
> Thanks!
> Marco
>
> "Problem with processing the model: Warning: at least one EV is (close to) a linear combination of the others. You probably need to alter your design. (Design matrix is rank deficient - ratio of min:max eigenvalues in SVD of matrix is 2.28634e-17)
> with 54 subjects, divided into 3 groups, each with 2 conditions.
>
|