Print

Print
Hi Arsene,

Please see below:


On 7 June 2017 at 07:00, Henri Longin <[log in to unmask]> wrote:

Hi Anderson,


Thank you for your prompt answer. 


>> I think the topic in that thread is exactly what you need. I have just posted an example design there, please have a look.
>> There are 3 groups for that one, whereas yours have 2, so you only have to simplify it a bit.

Finally I found the thread dealing with groups and 4 time points. I changed the design to adapt it to my experiment involving 2 groups and 4 repeated measures (find it attached).

Could you please tell me:

1) if the attached design is correct for my case

Design and contrasts look fine to me.
 

2) On associated contrasts you mention linear increase and linear decrease.
But the changes that may occur throughout the 4 measures should not necessarily have to be linear. I wonder if this design still hold.

The design holds, only the fit will be poorer, with a higher (less significant) p-value.
 

3) Also, how could I set contrasts on this design matrix for example to test:
  • t2 > t1 in G1
  • t4 > t2 in G2
  • G2 > G1 for t1
  • ...


Yes, you can. Only note that, the more contrasts, the higher is the chance of error type I. To correct for multiple contrasts, use the option -corrcon in PALM.

All the best,

Anderson


 


Arsene




Von: FSL - FMRIB's Software Library <[log in to unmask]> im Auftrag von Anderson M. Winkler <[log in to unmask]>
Gesendet: Dienstag, 6. Juni 2017 02:36
An: [log in to unmask]
Betreff: Re: [FSL] Design for an experiment of 2 groups and 4 repeated measures... PALM?
 
Hi Arsene,

I think the topic in that thread is exactly what you need. I have just posted an example design there, please have a look. There are 3 groups for that one, whereas yours have 2, so you only have to simplify it a bit.

All the best,

Anderson


On 2 June 2017 at 08:32, Henri Longin <[log in to unmask]> wrote:

Hi Anderson..


Sorry about returning to you with non relevant questions 😞

But... concerning my experiment which was performed on 2 groups and 4 repeated measures (t1, t2, t3 and t4)...
Typically, I would like to compare
  • time points on each group
  • same time points between the 2 groups


You advised me the link https://www.jiscmail.ac.uk/cgi-bin/webadmin?A2=FSL;271bf4c3.1705...
I installed PALM, but:
1) I still don't see how to design this kind of experiment with FEAT
2) How could be a typical command example in PALM...


Best regards,


Arsene



Von: FSL - FMRIB's Software Library <[log in to unmask]> im Auftrag von Anderson M. Winkler <[log in to unmask]>
Gesendet: Donnerstag, 1. Juni 2017 02:06
An: [log in to unmask]
Betreff: Re: [FSL] Trouble with Two-Sample Paired T-test results using randomise TFCE
 
Hi Arsene,

In that case, please see the reply I've sent to the other ongoing topic in the list: https://www.jiscmail.ac.uk/cgi-bin/webadmin?A2=FSL;271bf4c3.1705

It seems an identical problem.

All the best,

Anderson


On 31 May 2017 at 12:50, Henri Longin <[log in to unmask]> wrote:

Hi Anderson,


Thank for answer

About the example design in the spreadsheet, it seems like you did it for 2 groups and 2 repeated measures (t1 and t2)...

Unfortunately, my experiment was done on 2 groups and 4 repeated measures (t1, t2, t3 and t4)...


Best regards,


Arsene


Von: FSL - FMRIB's Software Library <[log in to unmask]> im Auftrag von Anderson M. Winkler <[log in to unmask]>
Gesendet: Mittwoch, 31. Mai 2017 12:20

An: [log in to unmask]
Betreff: Re: [FSL] Trouble with Two-Sample Paired T-test results using randomise TFCE
 
Hi Henri,

TFCE will not give results that are one the opposite of another.

For the design, please see at this spreadsheet an example (it's the design1): https://dl.dropboxusercontent.com/u/2785709/outbox/mailinglist/design_arsene.ods

Note it's necessary one exchangeability block per subject.

All the best,

Anderson


On 31 May 2017 at 06:14, Henri Longin <[log in to unmask]> wrote:


Hi Anderson,



Thanks for the answer....


For the question 1) you said:

>> It can happen: TFCE uses not only the height (that is symmetric around zero for both contrasts) but also the spatial distribution (that is not symmetric). So, it differs.

So, what could be the signification of this for the results I obtained?

Also: I performed an experience with 4 repeated measures (condition 1, condition 2, condition 3, condition 4) over 2 groups (control and test).
So I want to examine:
1) the time effect in the control group
2) the time effect in the test group
3) the timeXgroup interaction...

How could I build my design matrix in FSL. I have look at the FEAT-UserGuide and GLM-FslWiki but all examples deal with 2 or 3 measures

Thanks in advance


Arsene



Von: FSL - FMRIB's Software Library <[log in to unmask]> im Auftrag von Anderson M. Winkler <[log in to unmask]>
Gesendet: Mittwoch, 31. Mai 2017 02:23
An: [log in to unmask]
Betreff: Re: [FSL] Trouble with Two-Sample Paired T-test results using randomise TFCE
 
Hi Arsene,

Please see below:


On 29 May 2017 at 19:14, Henri Longin <[log in to unmask]> wrote:

Dear all,


I'm performing a two-Sample Paired T-test VBM using a statistical analysis with randomise TFCE.

1) Results obtained by estimating  the 2 paired differences A-B and B-A showed to be different from each other (what was expected) but one was not the opposite of the other: why?

It can happen: TFCE uses not only the height (that is symmetric around zero for both contrasts) but also the spatial distribution (that is not symmetric). So, it differs.
 


 2) More generally, how can one optimally choose the number of permutations in randomise TFCE?

There is no rule for this. More permutations make the p-values more reliable. The default in randomise is 5000 and it's ok to stay with that.

All the best,

Anderson

 

Thanks in advance


Arsene