Connectivity and epoch length

Dear Mike,

I have a question regarding the length of data used for connectivity analysis.

I am interested in running a connectivity analysis on EEG collected during a working memory task (n-back task). My plan is to take the EEG data after the onset of the stimulus, compute the power spectrum across this window for each frequency band of interest (to return the complex Fourier spectra), and then calculate the connectivity (DwPLI). Due to the nature of the task, I can only use epochs of around 1000ms for the analysis (due to the frequency of the stimuli). I understand that trial length can influence connectivity values. I would typically take epoch lengths which are longer than this (e.g., ~3000ms), however, in this instance, I don’t have that luxury. Just wondering if I can expect to still achieve reliable connectivity results using these shorter epochs?

Thank you.

Hi Brian. It’s difficult to say exactly how much time you need for a connectivity analysis (or any other kind of analysis). If you have a huge effect, then you only need a little bit of data.

I usually recommend having a time window of at least 3 cycles of the lowest frequency. But that’s a rough rubric based on my intuition and experience, not anything I could justify with math or statistics.

Anyway, I would guess that 1 second should suffice, particularly for frequencies above ~2-3 Hz.