Dear SPMers,
I want to perform an analysis in order to detect some activations
potentially scaling
linearly OR quadratically to one single parameter. As I mentioned in my
previous posting
(http://www.jiscmail.ac.uk/cgi-bin/webadmin?A2=ind0610&L=spm&D=1&O=D&P=12148),
I tried using the following model:
1)SPM.Sess(1).U(1).name = {'expectation'};
SPM.Sess(1).U(1).ons = [5.45 6.53 7.49 8.42 9.24]'; %ONSETS MAIN CONDITION
SPM.Sess(1).U(1).dur = 0;
2)SPM.Sess(1).U(1).P(1).name = 'linear_fit';
SPM.Sess(1).U(1).P(1).P = [0.2 0.5 0.4 0.6 0.9]'; % PARAMETER OF INTEREST
SPM.Sess(1).U(1).P(1).h = 1; % LOOKING FOR A LINEAR FIT
3)SPM.Sess(1).U(1).P(2).name = 'quadratic_fit';
SPM.Sess(1).U(1).P(2).P = [0.2 0.5 0.4 0.6 0.9]'; %(SAME VECTOR AS ABOVE)
SPM.Sess(1).U(1).P(2).h = 2; % LOOKING FOR A QUADRATIC FIT
4) control condition
However I get an error message, plus some parameters apparently are "not
uniquely specified"... When looking closer at the design matrix given by
SPM, there seems to be 5 regressors, when I would only expect 4. Is it
because SPM actually builds 2 regressors out of my regressor 3): one
linear term and one quadratic term ?
If this is true does it mean I have to get rid of my regressor 2) (which I
guess is the reason why there is a non uniquely specified parameter...)
and use only regressors 1), 3) and 4) ?
And very last question: with this last model, I guess the contrasts I
would have to use are:
- [0 1 0 0] to look for a linear correlation
- [0 0 1 0] to look for a quadratic correlation
BUT do they both have to be F-contrasts (or the 1st can be a T-contrast) ?
Once again these questions proably look very basic, but I would very much
appreciate some clarification... Thank you very much.
Guillaume Sescousse
|