Dear SPMers,
I try to setup a flexible factorial design with a single factor
and one covariate for e.g. two subject.
Since the values of the covariate arrange in 'scan' order,
I implicitly model the factor of replication
(Without replication, spm automatically rearrange the images in condition order.)
There are three factors, the first factor is 'subject', the second
one is 'repl', and the third one is 'Task'.
All measurements are assumed to be independent between levels which have equal variance.
The first main effect is 'Task' and the second one is 'subject'.
I also model a covariate whose values arrange in scan order.
There are assumed to be the interaction between the covariate and the subject.
When the spm_config_factorial_design was run,
I got the design matrix as I wished.
However, the covariance component of this design was wrong as shown below,
and hence, my design couldn't be estimated.
The problem is that the length of Vi matrix is doubled.
I guess that there's something wrong with spm_eeg_get_vc.m, where
the extra iteration occurs in VC setup when the factor of 'repl' is in the design matrix.
Is this bug, or is there somthing wrong with my procedures?
SPM.xX.X =
Task, Subject, Cov#1, Cov#2
10 10 4.5 0
01 10 3.5 0
10 10 2.5 0
01 10 1.5 0
10 10 0.5 0
01 10-0.5 0
10 10-1.5 0
01 10-2.5 0
10 10-3.5 0
01 10-4.5 0
10 01 0 4.5
01 01 0 3.5
10 01 0 2.5
01 01 0 1.5
10 01 0 0.5
01 01 0-0.5
10 01 0-1.5
01 01 0-2.5
10 01 0-3.5
01 01 0-4.5
SPM.xVi =
Qidentical: {[1] [1] [1]}
Qindependent: {[1] [1] [1]}
Vi: {[40x40 double]}<-Error!
I: [20x4 double]
Many thanks in advance for you help,
Takanori Kochiyama.
-------------------------------------------------------------
Takanori Kochiyama, Ph.D., Research Associates
Faculty of Engineering, Kagawa University
Hayashi-cho 2217-20,Takamatsu 761-0396, JAPAN
Phone: +81-87-864-2337,Fax: +81-87-864-2369
e-mail: [log in to unmask]
-------------------------------------------------------------
|