Question concerning statistics in fieldtrip

[Eric Maris]

Dear Markus,



Good to hear that you like our statistics routines. However, I'm afraid that
I have to disappoint you a little bit now. Permutation test are not well
suited for testing interaction effects. This is because the null hypothesis
(i.e., additive effects) is a constraint on the expected values of your
dependent variable, observed in the different conditions. In contrast, the
null hypotheses that can be tested with a permutation test all have to be
phrased in terms of equality of probability distributions in the different
conditions.

The closest you can get with a permutation test, is testing the effect of
one independent variable (say SIZE) within the different levels of AGE.
Hopefully, you will get significant effects for some levels of AGE, but not
for others. This result would be a sort of nonparametric interaction.


Good luck,

Eric Maris


> -----Original Message-----
> From: FieldTrip discussion list [mailto:FIELDTRIP at NIC.SURFNET.NL] On
Behalf Of
> Markus Werkle-Bergner
> Sent: Monday, February 12, 2007 8:40 PM
> To: FIELDTRIP at NIC.SURFNET.NL
> Subject: [FIELDTRIP] Question concerning statistics in fieldtrip
>
> Dear,
>
> in my PhD studies, I am working on lifespan differences on binding
> processes during visual perception. And for most parts of my ERP and TFR
> analyses, I use Fieldtrip - and it works fine.
>
> But currently, I'm a bit puzzled, how to set up appropriate statistical
> tests for my power and phase-locking analyses. Perhaps, I first describe
> my basic design:
> Subjects from  three age-groups (factor AGE, 3 levels, between subject)
> performed a simple visual discrimination task, while I varied the
> 'amount' of visual input in three levels (factor SIZE, 3 levels, whithin
> subject). The data was recored from 64 electrodes. I am mainly
> interested in the question, whether the effect of visual stimulation
> changes across age-groups (AGE x SIZE interaction), and where this
> interaction effect is located topographically (on the scalp level).
>
> Is it in general possible, to use the statistical interface of Fieltrip
> (e.g. clusterrandanalysis) to estimate - in one model - mixed designs
> like the one decribed above (with more than three levels per factor)? If
> this is the case, could anyone please give me a hint how to implement it?
>
> Thank you very much for your suggestions.
>
> Best regards,
> Markus