[pymvpa] Analysing data from a 1-run block design

[Jan Derrfuss]

Thanks, Yaroslav. This is a very good point. I'll keep it in mind for 
the actual experiment. Right now, I'm just analysing old data as a proof 
of principle.

Interesting article, by the way. I had no idea that turtle brains are 
so, well, robust. If somebody else is interested: "Because of the 
turtle's resistance to anoxia, it is not only possible to record 
apparently normal brain electrical activity in vitro, but to record this 
activity from an intact brain in which most, if not all, of the blood 
has been removed following cardiac perfusion with aCSF. Moreover, since 
the eyes can be left attached to this “bloodless” brain in vitro, it 
becomes possible to study neuronal current MRI signals evoked by natural 
sensory stimulation."

Best,
Jan


> phew -- better, otherwise you would have not known either you are
> classifying A vs B or a condition-after-the-R vs condition-before-the-R.
>
> But in any case, with such non-randomized design you must be more
> cautious/careful about interpreting the results.  Any physiological
> rhythm of the same 60 sec period is your confound now.  My tale of
> caution would be the analysis of the generously shared T1 MR data from
> turtle brains [1] which had similar non-randomized design:  whole
> brain classification analysis resulted in significantly above chance
> performance.  But 1. it was very unlikely, 2.  there were neither
> corresponding clear localization of the effects nor correspondence to
> ERP recording site which showed the relevant activation.
>
> So unfortunately I had to conclude that it indeed just was a
> confound driving the results.  Not sure how any magical
> filtering/detrending or alternative analysis could help here really,
> besides may be classification across subjects.  But for that, depending
> on what effects you are looking after, you might need fancier alignment
> procedures etc.  But I will be happy to be proven wrong.