Dear Anderson,
Thank you very much for your reply. Using just one F-test indeed seems to make a lot more sense.
To come back to your suggestion to add three t-contrast with opposing sign, I guess my design matrix could look like this (when testing a 2x2 ANOVA having 2 factors with 2 levels):
Ax Ay Bx By F1 C1 A>B 1 1 -1 -1 X C2 B>A -1 -1 1 1 X C3 X>Y 1 -1 1 -1 X C4 Y>X -1 1 -1 1 X C5 interaction+ve 1 -1 -1 1 X C6 interaction-ve -1 1 1 -1 X
Do you think this design matrix looks as it should? Or am I misunderstanding something?
Yet again, thanks a lot for your help!
Kind regards,
Maya
From: FSL - FMRIB's Software Library [[log in to unmask]] on behalf of Anderson M. Winkler [[log in to unmask]]Sent: Thursday, January 24, 2013 2:05 PM
AndersonAll the best,About the F=15 being not significant, it's actually not so high (consider a t of ~3.8). Plus, here it's non-parametric, so the significances aren't identical to what would be expected from the F distribution (which also depends on the number of df, which I think you didn't mention).Dear Maya,Your F-tests are correct but perhaps they are not very useful. If you'd like to look at both tails, you may consider additional t-tests with the opposite sign to those you already have. If you'd like an F-test over all the three t-contrasts, then in the Glm GUI, define something as:
EV1 EV2 EV3 EV4 F1
C1 A 1 0 0 0 x C2 B
0 1 0 0 x
C3 AB
0 0 1 0 x
2013/1/24 Schutte, J.L. (Maya) <[log in to unmask]>
Dear Anderson,
Thank you very much for your help.
I run the analysis as you said, including the ' -f design.fts', but still have a question concerning the results and was wondering if you could give some advice on this.
The t-test for contrast 2 shows a large significant result, however this does not show up in the f-test of contrast 2. Also, the uncorrected f-stat image of C2 shows a maximum f-value of approximately 15, by which I would expect some clusters to survive the corrected threshold. Do you have any idea what causes this discrepancy?
Many thanks in advance!
Kind regards,
Maya
From: FSL - FMRIB's Software Library [[log in to unmask]] on behalf of Anderson M. Winkler [[log in to unmask]]
Sent: Tuesday, January 22, 2013 5:15 PM
To: [log in to unmask]
Subject: Re: [FSL] 2x2 ANOVA randomise
In case you are interested only on the F-test, and want to skip the t-tests to save time, include the option --fonly, but still keep the "-t design.con" there.Use the second (with the "-f design,fts"). The first will run only the t-tests (the contrasts C1 to C3).Dear Maya,
All the best,
Anderson
2013/1/22 Schutte, J.L. (Maya) <[log in to unmask]>
Dear FSL mailinglist,
I am analyzing DTI data and would like to use a 2x2 ANOVA having two factors with 2 levels. When using the FEAT manual, my design matrix would look like this:
EV1 EV2 EV3 EV4 F1 F2 F3 C1 A 1 0 0 0 x C2 B
0 1 0 0 x C3 AB
0 0 1 0 x
However, when using the randomise function I am not entirely sure about the input arguments -t and -f. Should I include the -f design.fts in the commandline because I included 3 f-test in my design matrix, or is including the -t design.con sufficient?
In other words, which of the command below should I use, and why?
randomise -i all_FA_skeletonised -o tbss -m mean_FA_skeleton -d design.mat -t design.con -n 5000 --T2 -V
or
randomise -i all_FA_skeletonised -o tbss -m mean_FA_skeleton -d design.mat -t design.con -f design.fts -n 5000 --T2 -V
Thanks a lot in advance for any suggestions.
Kind regards,
Maya Schutte