[pymvpa] effect of signal on null distributions

[J.A. Etzel]

Ben Acland (who recently joined our lab) ran this in pyMVPA, and it made 
a big difference in the resulting null distributions: they now are 
pretty much the same (permute training-only and permute both) even 
though the true-labeled accuracy changes drastically with the amount of 
signal.

The post (with a bit more detail) is 
http://mvpa.blogspot.com/2013/02/pymvpa-null-distributions-different.html

And here is a direct link to the results image 
http://3.bp.blogspot.com/-tnxhXSC7EvQ/USU2SxFD32I/AAAAAAAAARo/1wwf7RUFF60/s400/way.png

So, yep, how we simulate the data really makes a difference.

Jo




On 2/13/2013 5:06 PM, J.A. Etzel wrote:
> 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)?


-- 
Joset A. Etzel, Ph.D.
Research Analyst
Cognitive Control & Psychopathology Lab
Washington University in St. Louis
http://mvpa.blogspot.com/