Dear Anderson,

thanks for your reply! I am sorry for the confusion I caused. In fact,
I only used the factor effects model and the references to the cell
means were actually references to the following text and table which
are realted to the factor effect model (citing from the web page):

"Obtaining inferences for cell means is not too difficult with this
model either and the contrast is actually contained in the design
matrix...."

Factor Effect PE1 PE2 PE3 PE4
mean(A) -1 -1 -1 1
mean(B) 1 0 0 1
mean(C) 0 1 0 1
mean(D) 0 0 1 1

To your answers:
1) OK, I will reconsider this and search for the related topics in the
mailing list.
2+5) As clarified above, I was referencing to the contrasts from
factor effects model I pasted above. I guess it should be equal to
what you suggested for my 5th question. The original question actually
was whether these contrasts represent the mean of each level + global
mean or just the mean of each factor level without the global mean?
Looking at my results, this is still not clear to me.

3) Ok, got it. Then the question is, can I build the following
contrast (just generally speaking and not necessarily with the kind of
data we are discussing here)?

Global mean 0 0 1

4) Do you mean corrected P values after voxel-based thresholding or
after TFCE? Would that also make sense after cluster mass thresholding
where the individual voxels in the clusters are not necessarily
significant? Anyway, the significant clusters in the "cell" means do
not overlap at all in my data, so I guess the global mean is what I
was really looking for.

I am aware that there might be another discussion about this, but do
you have any other suggestion how to compare voxel-wise the
tractography results beside comparing them visually on the
single-subject level? My research question is which parts of the brain
are (above-chance) more often connected with one class of the seeds
than with the other two across different subjects.

The second question
is which parts of the brain are (above-chance) connected with each of
the three classes. I did a lot of effort to make the individual
tractographies as comparable as possible and the factor level means
look actually very plausible (answering well my first hypothesis).

As an alternative to this approach I considered to run a
classification target tractography using atlas or cortical
parcellation. Then I would just compare the number (or proportion) of
samples reaching each target. But the values would be still the same,
it would only reduce the number of the "voxels" I am comparing with
each other, am I right?

Thanks again!

Kind regards

Pavel


On 10 February 2016 at 11:45, Anderson M. Winkler

<[log in to unmask]> wrote:
> Hi Pavel,
>
> The two designs (cell means and factor effects) model the same things, and
> will lead to identical results. The regression coefficients are different
> but the F-contrast will lead to the exact same conclusion.
>
> That said, I'm not sure a voxelwise test is the best way to compare
> tractography between subjects. This has been discussed in the list before.
>
> Also the questions are somewhat ill-posed, but trying to answer:
>
> 1) The design as in the manual is fine, though it may not be appropriate to
> do any voxelwise test with tractography.
>
> 2) The t-contrasts will be different, but the F-stat is the same. Choose
> only one of these designs (cell means is probably easier) and stick to it.
>
> 3) The intercept means the global mean.
>
> 4) Cannot construct a conjunction using the design directly. Instead, take
> the corrected p-values for the means of the three levels, then use fslmerge
> and fslmaths to keep only the largest p-value. The voxels where the largest
> (worst) p-value is significant are where the conjunction is significant.
>
> 5) In the cell means design, the contrasts for the mean of each level are:
>
> [1 0 0 ...]
> [0 1 0 ...]
> [0 0 1 ...]
>
> However, as you noticed, this contrast (and also the conjunction of question
> #4 above) make little sense as these quantities are all known to be positive
> and don't need to be tested. Likewise the global mean.
>
> All the best,
>
> Anderson
>
>
>
> On 9 February 2016 at 16:34, Pavel Hok <[log in to unmask]> wrote:
>>
>> Dear FSL experts,
>>
>> I would like to ask you for help with ANOVA design based on the 1-way
>> between-subject ANOVA from the GLM web page:
>>
>>
>> http://fsl.fmrib.ox.ac.uk/fsl/fslwiki/GLM#ANOVA:_1-factor_4-levels__.281-way_between-subjects_ANOVA.29
>>
>> The data are connectivity distribution maps from ProbtrackX, each
>> tractography was started from single vertex seed with sphere sampling.
>> The seeds were nearby sites, so their maps have roughly similar
>> distribution. Prior the group stats, maps were divided by waytotal,
>> transformed into standard space and smoothed by Gaussian kernel with
>> sigma 2.5 mm.
>>
>> There is one factor at three levels (with 6, 7 and 9 inputs per
>> factor, respectively). I used factorial design from the second example
>> and reduced the number of EVs and contrasts by 1. Then I was
>> interested in cell means, so I constructed their contrasts as
>> suggested in the example.
>>
>> I have several questions regarding the design:
>>
>> 1) Is the design valid for the described data?
>> 2) Do the cell means represent effects which are not in the two other
>> levels? I based this interpretation on the resulting data, which show
>> no overlap between any of the three cell means. This was quite
>> counter-intuitive to me given that the contrast includes the intercept
>> EV.
>> 3) Does the global mean or intercept in the factorial design mean
>> anything by itself?
>> 4) How can I construct conjunction of the three levels? (i.e. the common
>> effect)
>> 5) How can I test for mean effect of each level? One sample t-test
>> with Randomise produces very low number of permutations. Is there any
>> other way besides just flipping the signs? I don't think that makes
>> any sense in data which are by definition always zero or positive.
>>
>> Thanks a lot for any comments!
>>
>> Cheers
>>
>> Pavel
>>
>>
>> --
>>
>> -------------------------------------------------------------------------------
>> -- MUDr. Pavel Hok
>>
>> -------------------------------------------------------------------------------
>> -- Laboratoř funkční magnetické rezonance
>> -- Neurologická klinika
>> -- Lékařská fakulta
>> -- Univerzita Palackého v Olomouci
>> -- Fakultní nemocnice Olomouc
>>
>> -------------------------------------------------------------------------------
>> -- Laboratory of functional magnetic resonance imaging
>> -- Department of Neurology
>> -- Faculty of Medicine and Dentistry
>> -- Palacky University Olomouc
>> -- University Hospital Olomouc
>> -- Czech Republic
>>
>> -------------------------------------------------------------------------------
>> -- I.P. Pavlova 6, 775 20 Olomouc
>> -- web: fmri.upol.cz
>> -- tel.: +420 588 443 418
>> -- e-mail: [log in to unmask]
>>
>> -------------------------------------------------------------------------------
>
>



--
-------------------------------------------------------------------------------
-- MUDr. Pavel Hok
-------------------------------------------------------------------------------
-- Laboratoř funkční magnetické rezonance
-- Neurologická klinika
-- Lékařská fakulta
-- Univerzita Palackého v Olomouci
-- Fakultní nemocnice Olomouc
-------------------------------------------------------------------------------
-- Laboratory of functional magnetic resonance imaging
-- Department of Neurology
-- Faculty of Medicine and Dentistry
-- Palacky University Olomouc
-- University Hospital Olomouc
-- Czech Republic
-------------------------------------------------------------------------------
-- I.P. Pavlova 6, 775 20 Olomouc
-- web: fmri.upol.cz
-- tel.: +420 588 443 418
-- e-mail: [log in to unmask]
-------------------------------------------------------------------------------