Dear Mariam,
In the batch you sent, the model seems to be equivalent to a one-way
anova with 123 levels. With the non-sphericity options you used (equal
variance and dependent data), this corresponds to 7504 variance
components so I'm not too surprise you came across estimation problems.
The next update of SPM12 will contain changes of spm_reml to make it
more robust using Bayesian model reduction but anyway you have here to
think of a different way to model your data (what are your first level
models and what are you trying to test at the second level?).
Best regards,
Guillaume.
On 05/05/17 12:06, Mariam Kostandyan wrote:
> Dear SPM users,
>
>
> I use a mixed block/event-related design with big first-level design
> matrices. Moreover, the number of regressors is different for each
> participants because I have an unbalanced design. I want to run a
> flexible factorial analysis for my 2nd level. However, I get the warning
> messages (see the attachement) that the results might be inaccurate.
> Does anyone know where the problem can be hidden? I have also attached
> the batch that I am using for the 2nd level analysis (it is a test
> batch, so I have added only two participants).
>
>
> Kind regards,
>
> Mariam
>
>
>
> ----------------------
>
> Mariam Kostandyan
> PhD student
> University of Gent
> Department of Experimental Psychology
> Henri Dunantlaan 2, 9000 Gent, Belgium
> ----------------------
>
> Room 140.021
> Tel.: +32 9 264 6398
> E-mail: [log in to unmask] <mailto:[log in to unmask]>
> E-mail: [log in to unmask] <mailto:[log in to unmask]>
--
Guillaume Flandin, PhD
Wellcome Trust Centre for Neuroimaging
University College London
12 Queen Square
London WC1N 3BG
|