Dear Verónica,
there are many earlier posts on this mailing list about how to proceed
here. The recommended way would be to specify several second level
models, for each question you have (main effect, interaction, ...).
Otherwise, what you can do here with the flexible factorial design is to
specify three factors (simplifying your two between-subject factors with
two levels each to a single factor with four levels):
* subject: equal, independent
* group: 4 levels, unequal, independent
* cond: 2 levels, equal, dependent
then enter:
* main effect: subject [1]
* interaction: group x cond [2 3]
The design matrix and covariance components will be much simpler and it
is likely that the results will be less dependent to the MATLAB version
you are using.
Best regards,
Guillaume.
On 05/10/16 14:21, Verónica García wrote:
> Dear Guillaume,
>
> Our model has three factors. Each factor has two levels. The first two
> factors are between subject and the third one is within subject. In the
> model, we have added four main effects (the three factors + the subject
> factor) and three interactions (the interactions two against two between
> those three factors).
>
> Thank you so much.
>
> Best regards,
>
> Verónica
>
> On 5 October 2016 at 12:57, Guillaume Flandin <[log in to unmask]
> <mailto:[log in to unmask]>> wrote:
>
> Dear Verónica,
>
> Thanks for providing extra details.
>
> I would question your use of the flexible factorial design here first.
> What are your factors? It seems that you have one or two between subject
> factors (with 4 levels or 2x2) and one within subject factor (with 2
> levels)?
>
> Best regards,
> Guillaume.
>
>
> On 05/10/16 11:30, Verónica García wrote:
> > Dear Marko and Guillaume,
> >
> > Thank you for your emails. We used the same preprocessed dataset as an
> > input for the statistics step (flexible factorial) in all cases
> > (MATLAB R2013a, MATLAB R2015a and MATLAB R2016b). I have attached the
> > screenshots of the results in the following email (I have some
> > problems with the message size)..
> >
> > The significant voxels are located in a small volume within the brain
> > because the image dataset doesn´t belong to human beings.
> >
> > Looking at the results, you can see for example:
> >
> > 1) Matlab R2013a: p(set-level) = 0.782 and the voxels of the
> > cluster-level are 1052 in total.
> > 2) Matlab R2015a: p(set-level) = 0.789 and the voxels of the
> > cluster-level are 1137 in total. So the difference compared to Matlab
> > R2013a is 0.9% in p(set-level) and 8.1% in number of voxels.
> > 3) Matlab R2016b: p(set-level) = 0.903 and the voxels of the
> > cluster-level are 888 in total. So the difference compared to Matlab
> > R2013a is 15.5% in p(set-level) and 15.6% in number of voxels.
> >
> >
> > Guillaume, tell me if it is really necessary to send you the copy of
> > the folders containing the SPM.mat because I am not be sure if I am
> > allowed to do that.
> >
> > Just in case it is important, Matlab was installed using its default
> > settings. I didn´t remove any Matlab toolbox from the default Matlab
> > installation list.
> >
> > Best regards!
> >
> > Verónica
> >
> > On 5 October 2016 at 11:24, Guillaume Flandin <[log in to unmask]
> <mailto:[log in to unmask]>> wrote:
> >> Hi Veronica,
> >>
> >> as Marko mentioned, it would be great if you could show us the
> >> differences you observe in the results you obtain when the only
> change
> >> is the MATLAB version (screenshot of the Result page and,
> possibly, copy
> >> of the folders containing the SPM.mat). I would not expect
> significant
> >> differences so it would be interesting to get to the bottom of this.
> >>
> >> Best regards,
> >> Guillaume.
> >>
> >> On 05/10/16 07:11, Marko Wilke wrote:
> >>> Veronica,
> >>>
> >>> while I do not have an answer, I believe it would be most
> helpful if you
> >>> could quantify the differences (are they on the order of .01%,
> 1%, or
> >>> 10%?), and report on how exactly you have made sure that the
> error is
> >>> only in the statistics steps (are these datasets identical, or
> >>> preprocessed identically in the different versions, or...). This may
> >>> also help to narrow it down to a real glitch or rounding
> differences or
> >>> whatever else.
> >>>
> >>> Cheers
> >>> Marko
> >>>
> >>> Verónica García wrote:
> >>>> Dear SPM experts,
> >>>>
> >>>> Which is the most appropriate version of MATLAB when using SPM12
> >>>> v6685? I am asking that question because we are comparing a set
> of PET
> >>>> images using a Flexible Factorial design (basic models) and the
> >>>> results are not the same if we use MATLAB R2013a, MATLAB R2015a or
> >>>> MATLAB R2016b.
> >>>>
> >>>> Thank you for any help you can provide.
> >>>>
> >>>> Best regards
> >>>>
> >>>> Verónica García Vázquez
> >>>>
> >>>
> >>
> >> --
> >> Guillaume Flandin, PhD
> >> Wellcome Trust Centre for Neuroimaging
> >> University College London
> >> 12 Queen Square
> >> London WC1N 3BG
>
> --
> Guillaume Flandin, PhD
> Wellcome Trust Centre for Neuroimaging
> University College London
> 12 Queen Square
> London WC1N 3BG
>
>
--
Guillaume Flandin, PhD
Wellcome Trust Centre for Neuroimaging
University College London
12 Queen Square
London WC1N 3BG
|