On 07/16/2012 09:30 PM, Meng Li wrote:
Dear professor, Thanks for your reply. You said that when drawing conclusions from both hemispheres then I need to use .025, but I found that in qdec interface, the default value of -cwpvalue is 0.05. So are they inconsistent?
This would be consistent for a single hemisphere, which is all that qdec analyzes.
I also want to find a solution about an ERROR. I tried to run a cortical thickness analysis to look for differences between two groups (patients and controls) with regressing out the gender factor, the patients group have 1 female and 17 male, after analysis, I got the results. But when I add another continuous variable as covariate, I received the message: "ERROR: matrix is ill-conditioned or badly scaled." Then I tried to change the gender composition the patients groups: 2 female/16 male, and the problem was solved. so is it right about what i did, or I need another way to solve this problem?
It is probably a bad idea to try to put a subject that is in one group into another group. In your case, you need more females. You can't have a single member in a group and have a covariate. You should try to have balanced groups. With only a single member of a group, you are going to have problems with power and be susceptible to that one member being an outlier.
doug
Thanks, Best wishes Meng
On 07/06/2012 10:00 PM, Meng Li wrote:
Hi, freesurfer expert, I performed the statistical analysis between two groups first in left hemisphere, and got some regions which showed different.
- so if I then do multiple correction, i don't need to run the
mri_glmfit-sim command, and just press the button of the monte carlo null-z simulation in the qdec interface, am i right?
correct
- Whether mri_glmfit-sim --cache command line and the mc-z in qdec
get the same results, or not? and as to the option --cwpvalthresh, should i choose 0.05 or 0.025?
If you are drawing conclusions from both hemispheres then you need to use .025
- And what is the difference between fdr and mc-z correction?
mc-z uses a cluster-wise correction and controls the rate of false positive clusters. FDR is voxel-wise and controls the false discovery rate, which is the number of false positives relative to the total number of positives (as compared to the number of false positives relative to the total number of tests). doug
Thanks,
--