[pymvpa] effect of signal on null distributions

[Yaroslav Halchenko]

QED - H0 MC distribution is quite dependent on the data and not
only on design, i.e. how many classes and samples.  But then it could be
generalized to the known fact that different regions would have
differing H0 distributions.

On Fri, 15 Feb 2013, J.A. Etzel wrote:

> I posted images of some of the null distributions I get after
> changing the simulation code here: http://mvpa.blogspot.com/2013/02/comparing-null-distributions-changing_15.html
> - they now look a lot closer to Yaroslav and Michael's examples
> (though not exactly the same - I used different voxel counts, all
> voxels have the signal, etc.).

> So how we simulate the data really matters, more than I'd guessed.

> Do these conclusions seem sensible to all of you? Has anyone been
> able to take a look at the questions?

> thanks,
> Jo


> On 2/13/2013 5:06 PM, J.A. Etzel wrote:
> >Two questions:
> >Does my description seem accurate for normal_feature_dataset? In
> >particular, do you set the class B values as class A / snr, or generate
> >a separate set of normal values for A and B, then divide the B values?

> >What do the null distribution for different amounts of signal look like
> >if you try the simulations making the data as I describe (values from
> >two normal distributions with a small difference in mean but same
> >standard deviation)?
-- 
Yaroslav O. Halchenko
http://neuro.debian.net http://www.pymvpa.org http://www.fail2ban.org
Postdoctoral Fellow,   Department of Psychological and Brain Sciences
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