[pymvpa] OneWayAnova for feature selection: how it works?

[Yaroslav Halchenko]

FWIW There was a recent MANOVA paper

http://www.sciencedirect.com/science/article/pii/S1053811913011920
Searchlight-based multi-voxel pattern analysis of fMRI by cross-validated MANOVA
Carsten Allefeld, John-Dylan Haynes

which hints on its usefulness in neuroimaging context ;) There is some matlab
code available from Allefeld's website: http://www.carsten-allefeld.de/code/
and Jo already bisected it a bit:
http://mvpa.blogspot.com/2014/03/allefeld-2014-searchlight-based-multi.html

would be nice to get it done "PyMVPA way"... any takers? ;)


On Tue, 06 May 2014, Vadim Axel wrote:

>    Thank you Brian, very helpful!
>    Obviously, if I run ANOVA for each voxel separately, I treat the trials as
>    "subject" *(based on my analogy). 

>    On Tue, May 6, 2014 at 3:21 PM, Brian Murphy <[1]brian.murphy at qub.ac.uk>
>    wrote:

>      Hello Vadim,

>      you're right that the Anova feature selection is univariate. And also
>      for a two-class problem it gives an identical result to that you would
>      get with a t-test (F statistic is different from t-statistic, but what
>      matters is the ranking, which will be identical). I suppose the reason
>      PyMVPA uses Anova rather than t or z statistics, is so that it also
>      works for more than two classes.
>      > In one-way ANOVA with data analysis we might have subjects as
>      > between-group factor and treatment as within-group factor. But
>      > treating voxels as "subjects" does not help me because there is no
>      > grouping for voxels (grouping is for condition).

>      Hmmm, not sure, but if I read your analogy correctly, I think you should
>      be thinking of *trials* as being equivalent to subjects.

>      best,

>      Brian

>      > Any ideas? Any links for explanation would be also appreciated!


>      > Thanks,
>      > Vadim
-- 
Yaroslav O. Halchenko, Ph.D.
http://neuro.debian.net http://www.pymvpa.org http://www.fail2ban.org
Research Scientist,            Psychological and Brain Sciences Dept.
Dartmouth College, 419 Moore Hall, Hinman Box 6207, Hanover, NH 03755
Phone: +1 (603) 646-9834                       Fax: +1 (603) 646-1419
WWW:   http://www.linkedin.com/in/yarik