The sign of t-test in frequency space and in connectivity tests

Discussions
1

Hello Francoise!

We have a question about the signs (+&-) in t-test in different scenarios.
We are wondering the meaning of the signs when we run independent permutation t-test to PSDs that are calculated from source localized EEGs in 3D cortex models.
We think that the signs are meaning increase or decrease in power on specific frequency band, due the nature of the input-files (PSD), where all frequencies have only positive values, opposite to EEG where the values are above and below 0 constantly (As explained on tutorial).
And how does this apply to permutation t-tested functional connectivity-graphs, as there we can have both negative and positive values to demonstrate the power of connectivity between certain anatomical locations. (Slider for t-value). Functional connectivities are calculated from source-localized EEG so is it so, that these sings are not reliable then?

Thank you for your help!

T & H

2

Correct.

as there we can have both negative and positive values to demonstrate the power of connectivity between certain anatomical locations

It depends on the measure your use. Coherence is a value normalized between 0 and 1, therefore testing for increase or decrease of coherence with a non-parametric t-test works exactly like the PSD.

For values that can be positive or negative, the T-statistic would still give you meaningful values: the values of your samples A can be significantly higher than the values in your samples B, regardless of their sign.
What is more complicated is the interpretation: what does it mean to have a significantly "more positive" or a "more negative" value for the measure of interest?

For EEG, the sign is meaningful: at a given electrode, some ERP waves are always negative, other are always positive, and it is documented in the EEG literature that a given experimental condition can lead to "more negative" or "more positive" values.

For constrained minimum norm maps, the sign is difficult to interpret, it depends on the orientation of the normal to the cortex surface. This is explained in the tutorials:
https://neuroimage.usc.edu/brainstorm/Tutorials/SourceEstimation#Sign_of_constrained_maps
https://neuroimage.usc.edu/brainstorm/Tutorials/Difference#Constrained_sources

For connectivity measures, it depends on which measure you want to use. For example, with Granger Causality, it corresponds to the direction of the interaction (A=>B or B=>A).

3

Yes, this helped.
At the moment we are using Coherence measure for the functional connectivities,
so it's also now clear how it can be statistically compared in a reliable way.
Thank you for help!
T & H