Print

Print
Hi Melanni,

NPC can be used to combine (a) multiple modalities and one design, (b) one modality and multiple designs, (c) multiple modalities and multiple designs and (d) multiple modalities and multiple designs when each design pair up with a single modality (option "-designperinput").

For the case you describe, it seems to be you'd compute the differences between the sessions, and these differences would be the inputs (or "modalities"), and with a single design, which is the same case that Qasim has.

All the best,

Anderson


On Mon, 30 Jul 2018 at 10:50, melanni nanni <[log in to unmask]> wrote:

Hello,


I have a similar design but instead of 4 groups  I have only 2 (and 3 session per subject), I was wondered which modality of NPC would be feasible, only one modality with multiple designs or multiple modalities with one design?


Thanks


De: FSL - FMRIB's Software Library <[log in to unmask]> en nombre de Anderson M. Winkler <[log in to unmask]>
Enviado: sábado, 30 de junio de 2018 05:41 p. m.
Para: [log in to unmask]
Asunto: Re: [FSL] Design Matrix with 4 groups and 3 sessions in fsl randomise
 
Hi Qasim,

There's information on NPC at this link: https://fsl.fmrib.ox.ac.uk/fsl/fslwiki/PALM/JointInference
Combining functions. The joint null hypothesis of the NPC is that the null hypotheses for all partial tests are true, and the alternative is that any is false (that is, the same as a union-intersection test, UIT).

In short: if you had just one test for a difference (say, post-pre), you'd compute such differences and run the 1 factor x 4 levels between-subjects ANOVA. This test can be repeated two more times, for the other differences (3 tests overall). NPC combines these 3 tests into a single, joint inference, that is more powerful than MANOVA (and MANOVA itself is an alternative over repeated measures ANOVA).

For single tests, randomise (permutation) or FEAT (parametric) can be used. For combination, would need to use PALM then.

All the best,

Anderson

On Wed, 20 Jun 2018 at 14:51, Qasim Bukhari <[log in to unmask]> wrote:
Hi Anderson
This is super useful. I will go with the 1 factor 4 levels then. Can you please also elaborate what is NPC ? I also wanted to reconfirm that this is in randomise and not in FEAT. That shouldnt be a problem, right ?

Thanks again and best
Qasim
From: FSL - FMRIB's Software Library <[log in to unmask]> on behalf of Anderson M. Winkler <[log in to unmask]>
Sent: Wednesday, June 20, 2018 8:03 AM
To: [log in to unmask]
Subject: Re: [FSL] Design Matrix with 4 groups and 3 sessions in fsl randomise
 
Hi Qasim,

You'd need a repeated-measures ANOVA (1 factor with 3 levels within-subject, 1 factor with 4 levels (groups) between-subjects). Another option, perhaps simpler, is to go with the 1 factor and 4 levels as in the link, and use as input differences between before and after. This can be done up to 3 times (post-pre, 4mo-pre, 4mo-post), and these can be combined with NPC.

Generally speaking, NPC tends to be more powerful than a repeated measures ANOVA, and that's the route I'd take. Even without the combination, running 3 separate smaller designs will be faster than just one big repeated measures design.

All the best,

Anderson



On 14 June 2018 at 07:39, Qasim Bukhari <[log in to unmask]> wrote:
Dear all,
We have a question regarding design matrix. We have 4 groups: sham, waiting list, Drug1 and Drug2. Each group has 3 session 3 in the scanner: pre, post and after 4 months. There are different subjects in each group, but the subjects are same within sessions.
I originally thought that a two sample paired t test would be best suited and I can test each session at a time for pre vs post. And if something is found significant in for example group 1 pre vs post and not significant in control group pre vs post, then i can consider it significant. However I have just realised that probably a better approach would be to use a 1 factor 4 level anova as described here. Is that the correct design matrix ? But here, the example considers 8 subjects, that means at each level there is a different subject. 
My question is, was my previous approach correct or should I follow the 1 factor 4 level approach as described below ?

Experimental Designs - No repeated measures. We start considering only designs where there is one scan per subject, that is, no repeated measures.
Any help will be very appreciated

Thanks and best
Qasim


To unsubscribe from the FSL list, click the following link:
https://www.jiscmail.ac.uk/cgi-bin/webadmin?SUBED1=FSL&A=1



To unsubscribe from the FSL list, click the following link:
https://www.jiscmail.ac.uk/cgi-bin/webadmin?SUBED1=FSL&A=1


To unsubscribe from the FSL list, click the following link:
https://www.jiscmail.ac.uk/cgi-bin/webadmin?SUBED1=FSL&A=1


To unsubscribe from the FSL list, click the following link:
https://www.jiscmail.ac.uk/cgi-bin/webadmin?SUBED1=FSL&A=1


To unsubscribe from the FSL list, click the following link:
https://www.jiscmail.ac.uk/cgi-bin/webadmin?SUBED1=FSL&A=1


To unsubscribe from the FSL list, click the following link:
https://www.jiscmail.ac.uk/cgi-bin/webadmin?SUBED1=FSL&A=1