Thank you so much for you response! I will submit a bug following the rule. However, I do think it is not a bug but a statistic misunderstand. The root of this problem is the variable C in two classes (e.g., A1B1 and A2B1) are all Zeros. If I add one non-zero number in C in these two classes, every thing is fine. However, 'all zeros' is my data and I cannot change it. I found the program make four regressors for C, each one contains only one class. So two regressors are all zeros. This caused the error information. I hope, in my analysis, the C is as only one regressor because it is only the co-variable. How can I do it? Thanks, again!
Xiaochu Zhang PhD
Visiting Research Fellow
Neuroimaging Research Branch
National Institute on Drug Abuse - IRP
Biomedical Research Center
251 Bayview Blvd.
Suite 200 (NIDA)
Baltimore MD
21224
Tel: 443-740-2619
-----Original Message----- From: Douglas N Greve [mailto:greve@nmr.mgh.harvard.edu] Sent: Thursday, July 16, 2009 11:30 AM To: Zhang, Xiaochu (NIH/NIDA) [F] Cc: freesurfer@nmr.mgh.harvard.edu Subject: Re: [Freesurfer] ANOVA and co-variable
Hi Zhang, when reporting a bug or problem, please refer to the following suggestions in terms of what information you should report: surfer.nmr.mgh.harvard.edu/fswiki/BugReporting
thanks!
doug
Zhang, Xiaochu (NIH/NIDA) [F] wrote:
Thank you so much for your answer! Now I meet another problem. Just as your see, the C in some classes (e.g., A1B1 and A2B1) is all zeros. When I run the mri_glm. Some error information was shown. I listed them below.
Matrix condition is 1e+08 ERROR: matrix is ill-conditioned or badly scaled, condno = 1e+08 Possible problem with experimental design: Check for duplicate entries and/or lack of range of continuous variables within a class.
How can I deal to this error?
Thanks again!
Xiaochu Zhang PhD
Visiting Research Fellow
Neuroimaging Research Branch
National Institute on Drug Abuse - IRP
Biomedical Research Center
251 Bayview Blvd.
Suite 200 (NIDA)
Baltimore MD
21224
Tel: 443-740-2619
-----Original Message----- From: Douglas N Greve [mailto:greve@nmr.mgh.harvard.edu] Sent: Wednesday, July 15, 2009 12:31 PM To: Zhang, Xiaochu (NIH/NIDA) [F] Cc: freesurfer@nmr.mgh.harvard.edu Subject: Re: [Freesurfer] ANOVA and co-variable
yes, that is correct. Make sure to handle the mean/offset of C in a rational way. The simple, though somewhat incorrect, way is to simply demean C. Better to pick a fixed value of C to test it at. If this is what you do, let's say you test at C=2, then your contrast would be:
0.5 -0.5 -0.5 0.5 2 2 2 2
ie, you would just replace the 0s in the C-columns with 2.
doug
Zhang, Xiaochu (NIH/NIDA) [F] wrote:
Hi Freesurfer experts, I want to do a group ANOVA analysis with one co-variable. I have read the corresponding wiki but still have some problems. I need your help.
I have 4 group subjects (A1B1, A1B2, A2B1, A2B2) and one co-vaiable (C). I want to get the interaction between A and B. However, because C is different in different groups, I need to rule out the influence from C in my ANOVA analysis. So the model is A*B+C.
The fsgd file is GroupDescriptorFile 1 Title SK-SZ Class A1B1 Class A1B2 Class A2B1 Class A2B2 Variable C Input S1 A1B1 0.0 Input S2 A1B2 2.0 Input S3 A2B1 0.0 Input S4 A2B2 3.0
The mtx file is 0.5 -0.5 -0.5 0.5 0 0 0 0
Is it correct? Thanks a lot!
Xiaochu Zhang PhD
Visiting Research Fellow
Neuroimaging Research Branch
National Institute on Drug Abuse - IRP
Biomedical Research Center
251 Bayview Blvd.
Suite 200 (NIDA)
Baltimore MD
21224
Tel: 443-740-2619
--
Freesurfer mailing list Freesurfer@nmr.mgh.harvard.edu https://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer
-- Douglas N. Greve, Ph.D. MGH-NMR Center greve@nmr.mgh.harvard.edu Phone Number: 617-724-2358 Fax: 617-726-7422
In order to help us help you, please follow the steps in: surfer.nmr.mgh.harvard.edu/fswiki/BugReporting
-- Douglas N. Greve, Ph.D. MGH-NMR Center greve@nmr.mgh.harvard.edu Phone Number: 617-724-2358 Fax: 617-726-7422
In order to help us help you, please follow the steps in: surfer.nmr.mgh.harvard.edu/fswiki/BugReporting