Hi,

I'm not sure what the benefit of an F-test here with these simple contrasts would be. If the reviewer is worried (rightfully) with multiple testing, the sole two contrasts that make up the F-test are orthogonal (only 1 needs to be selected), and the significance level could just be divided by two to guard against false positives.

In any case, to do what was requested, have you considered bringing up the COPEs for, respectively, contrasts C1, C3 and C5, and in the 2nd level, then run a 1-sample t-test with the corresponding simple F-test? I suspect this should answer the reviewer question and avoid the error you got.

This 2nd level can be performed in randomise (use sign-flips instead of permutations), and you'll have the TFCE then.

Hope this helps.

All the best,

Anderson



On 26 June 2014 23:53, Han-Gyol Yi <[log in to unmask]> wrote:
Hello all,

Apologies in advance if this is a duplicate question, but I could not find a similar question.

This is for a manuscript in which we report the results of an fMRI experiment. We wish to conform to the requests of a reviewer from the journal to whic the manuscript has been submitted.

Our experiment involved a single group of participants (N = 22), with only one scanning session ("run") per participant. During the run, the participants were presented with 80 stimuli. These stimuli were subdivided into 20 in each of the four conditions. The conditions were in a two-by-two design, two levels for each factor. This was of course coded into the first-level analysis that had four main EVs (not counting the ones not of interest):

EV1: A1B1; EV2: A1B2; EV3: A2B1; EV4: A2B2

The submitted draft included the results from the following six contrasts:

A1 - A2: [1 1 -1 -1]
A2 - A1: [-1 -1 1 1]
B1 - B2: [1 -1 1 -1]
B2 - B1: [-1 1 -1 1]
AXB: [1 -1 -1 1]
BXA: [-1 1 1 -1]

However, one of the reviewers asked for f-tests looking for:

Main effect of A
Main effect of B
Interaction between A and B

Which gives rise to the current question-- I do not know how to obtain these statistics. I tried two things. First, I ran f-tests in the first-level analysis:

F1: [1 0 0 0 0 0]
F2: [0 0 1 0 0 0]
F3: [0 0 0 0 1 0]

The corresponding f stats images were thereby generated, but these simply did not show up in a second-level analysis which took the feat directories as the input.

Then, I tried inputting fstatX.nii.gz images as the "cope image" inputs in three separate second-levels. This generated an error message:



$FSLDIR/bin/fsl_sub -T 60 -l logs -N feat2_pre   $FSLDIR/bin/feat $OUTPUTDIR/design.fsf -D $OUTPUTDIR/fstat1.gfeat -gfeatprep
syntax error in expression "1 + fstat1": variable references require preceding $
    while executing
"if { [ catch {



for { set argindex 1 } { $argindex < $argc } { incr argindex 1 } {
    switch -- [ lindex $argv $argindex ] {

        -I {
            incr argin..."
    (file "$FSLDIR/bin/feat" line 119)


*** note: $FSLDIR and $OUTPUTDIR are my edits replacing the actual path names in the error message.



Which made me think: hmm, something is not right. I also consulted the FSL wiki, but as far as I understand, these only concern multiple runs within each subjects.

In summary, I want to know how to obtain group-level f-test statistics on a single group, single run, two-by-two factorial design. Additionally, I need to run TFCE on these statsitics, so I want to know how the GLM should be set.

Thank you for reading this long message.

Best,
Han-Gyol Yi