Hi Rujing,

However flattered I might be in being mistaken for him, it was Mark who answered your original question, not me... In any case, please, see below:

2014-02-07 15:38 GMT+00:00 charujing123 <[log in to unmask]>:

Hi Anderson,

Thanks Anderson,I see.

For the 1st question,did I need demean behavior data when I used -D option?

As MJ explained, the -D option will demean all regressors in the design, as well as the data. To "demean" means the same as "remove the mean" or "subtract the mean from", so you don't have to demean manually anything, be it behavioural data or not, if you use the -D option.

For the 2nd question,what I think below are appropriate?

1,Output of randomise is an inference based on non-paramatric method, and result from FEAT is an inference based on paramatric method.

Yes, that's correct.

2,For VBM and TBSS(for gray matter and white matter respectively),can I use the same GLM setup which is used for randomise previously to perform paramatric inference by FEAT at the end(not by randomise)? That is to say, I want to use FEAT not randomise to perform paramatric inference.

I'm afraid FEAT would need the lower level directories, which don't exist for VBM or TBSS, so you're probably better off sticking to randomise.

All the best,

Anderson

PS: If you aren't sure whether your email went through after sending, you can have a look at https://www.jiscmail.ac.uk/cgi-bin/webadmin?A0=fsl. It shows up there after just a few minutes. I say this because you sent the same email at least 4 times to the list, not to mention those sent directly to my inbox... thanks!

Thanks again, Anderson.

All the best.

Rujing Zha

2014-02-07

charujing123

发件人：Mark Jenkinson <[log in to unmask]>

发送时间：2014-02-07 17:45

主题：Re: [FSL] question about "randomise -D option" and "GLM design in Two-Group Difference"

收件人："FSL"<[log in to unmask]>

抄送：

Hi,

The -D option in recent versions of FSL removes the mean from both the data and the design matrix EVs.

Permutations and t-tests are not mutually exclusive. You are still using t-tests (or F-tests if you combine t-tests) in randomise, but the method by which it performs inference (calculating probabilities) uses a permutation-based non-parameteric method, as opposed to tools like FEAT where parametric methods are used. This is totally separate from the fact that you are performing a t-test.

All the best,

Mark

On 7 Feb 2014, at 03:12, charujing123 <[log in to unmask]> wrote:

Hi Anderson and other FSL experts,

I have a two questions about FSL statistics.

1,As we know, there is a -D option in randomise,which is "demeaning data temporally before model fitting". What I wonder is -D option demean what data, either only MR data or both for MR data and behavior data?

2,http://fsl.fmrib.ox.ac.uk/fsl/fslwiki/GLM#Two-Group_Difference_.28Two-Sample_Unpaired_T-Test.29 In method of this website, I can do two-sample unpaired t-test. In "randomsie details", I see word "permutations". So I didnot know whether it is a unpaired t-test or a permutaion test.There is same question in "Two-Group Difference Adjusted for Covariate". If they are all t-test, I want to know the functions of permutations and how to do permutation test.

Thanks Anderson and others.

All the best.

Rujing Zha

2014-02-07

charujing123