the same question with `Results may be inaccurate. RCOND = 1.034803e-027` which was posted at Thu, 02/17/2011

Submitted by YukiSakai on Tue, 03/29/2011 - 16:43

Dear REST experts,

I encounter the same problem with `Results may be inaccurate. RCOND = 1.034803e-027` which was posted at Thu, 02/17/2011. I was not able to find the answer in the forum about that.

Would you mind explaining how can I solve this problem?

Thank you in advance.
Yours sincerely,
Yuki Sakai

Submitted by dongzy08 on Sat, 04/02/2011 - 09:27

This warning maybe occur when you try to detrend or regress out covariates. Anyway it means the covariates matrix is close to singular, and it means the covariates hardly affect the data, so just skip this step or try to reduce the kinds of covariates.

Submitted by YukiSakai on Sat, 04/02/2011 - 15:50

Thank you very much.

I encountered this caution message when I regressed out 9 kinds of covariates, such as 6 motion parameters and the fluctuation of the global signal, CSF, and white matter. I believe that these are common set of covariates when we do the functional connectivity analysis.

Which is the better way to ignore this caution and move to the subsequent steps or to reduce the kinds of covariate?

Thank you in advance.
Yours sincerely,
Yuki Sakai

Submitted by dongzy08 on Sat, 04/02/2011 - 22:18

It is same as dividing data by nearly zero. Only a few bits of significance are left for the
results. No body will be sure whether the results are accurate.
I propose to reduce the kinds of covariate, such as the fluctuation of the global signal, CSF, and white matter. Usually this problem will be solved.

Best,
DONG

Submitted by YukiSakai on Thu, 10/20/2011 - 05:32

Thank you very much. I am sorry for the delay in my response.

Although I reduced the 9 kinds of covariate into 7 by excluding the CSF and white matter, the same message still appeared. Is there any other ways to solve this problem?

Yours sincerely,
Yuki

Submitted by dongzy08 on Mon, 10/24/2011 - 18:51

The hypotheses for regression includes that the correlation among the regressors is smaller than the correlation between the regressors and the dependent variable. Maybe the correlation among the six parameters of head motion is so high to make the results inaccurate. So check the parameters of head motion, if the warning remains, I propose not to do the regression.

Best,

Dong