permutation_t_test

External Email - Use Caution

Hi all,

I am wondering how the procedure behind "permutation_t_test" works. I have
used the example here
<https://mne.tools/stable/auto_examples/stats/plot_sensor_permutation_test.html#sphx-glr-auto-examples-stats-plot-sensor-permutation-test-py>
for EEG data (averaged over time window with size [subjects X electrodes])
and after plotting p-values, there are some areas that have red color (mean
significant p-value) but without any significant electrodes. How could this
be possible? Does this function average over p-values from each subject,
and because of the high variance, the corresponding electrodes won't be
significant?

Thank you,
-Maryam
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://mail.nmr.mgh.harvard.edu/pipermail/mne_analysis/attachments/20200219/f5c2018e/attachment.html

External Email - Use Caution

Does this function average over p-values from each subject, and because

of the high variance, the corresponding electrodes won't be significant?

No it does a permutation over the subjects (using sign flips), and the p
value is the proportion of time the (absolute) t score from the true data
was greater than the maximum (absolute) t value from the permutations. The
result for parametric data should be like what you get from just doing
`scipy.stats.ttest_1samp` on your data. For background, see:

https://mne.tools/dev/auto_tutorials/discussions/plot_background_statistics.html#non-parametric-tests

Eric