[Mne_analysis] relating dSPM to p

[Matt Panichello]

Hi All,


I am interested in relating the grandaveraged dSPM values for a set of
subjects to a particular p-value. I know this has come up a few times on
both this and the old mne_analysis listserv, but I am confused about a few
issues in particular:


First, I have seen several users mention using the function inv_fpdf. As I
understand it, this function relates a p-value to dSPM by treating dSPM as
an f statistic with degrees of freedom (3* number of  subjects) / (number of
time points used to compute the noise covariance matrix).  How can I grab
the number of time points in the noise covariance matrix? I think
“mne_show_fiff [file] --tag 3531 –verbose” (where [file] is the covariance
matrix for one of my subjects) may return this value, but I’m not completely
clear on this even after looking at the help.


Also, to correct for multiple comparisons, I was considering doing a
bonferroni correction with the correction factor equal to (n timepoints) x
(# vertices). However, I was told this isn’t a good idea because 1) results
are correlated across both time and space and 2) the forward solution is
only calculated for a max of 20,000 vertices (not every single one).  Can
anyone recommend a better correction?


Finally, in addition to finding the threshold for the grandaverage of
conditions, I’d also like to visualize grandaveraged difference waves.
Should these be treated any different statistically?

Any help would be greatly appreciated!


Thanks,


Matt
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://mail.nmr.mgh.harvard.edu/pipermail/mne_analysis/attachments/20110411/6aee563b/attachment.html