Dealing with multiple sessions per participant in a long-duration ERP study

Hi everyone,

I am currently about to run an EEG study but before I actually start testing participants, I just wanted to make sure I've got a handle on the analysis before I start.

It's an ERP study where each participant will be completing three sessions. I will be computing and comparing the ERPs across the three sessions. In each session, it will have three conditions (so a within-subjects 3x3 design). However, because the task of each session is slightly different, I guess I have to worry about noise being different across sessions (e.g. different electrode contact with skin) not to mention the fact that the electrode cap might not be exactly placed across sessions. I will not be looking at any spatial aspect so will not have participant anatomy. It will also be quite a long experiment with many trials (about 2-2.5 hours per session including cap placement).

Based on reading a previous thread dealing with a similar issue (Help with scripting multiple sessions per subject) as well as the tutorials, here is what I guess I have to do. I would appreciate any advice with dealing with this issue:

For each individual session

  1. Standard preprocessing (ocular correction, rereferencing, filtering, baseline correction, bad channels)
  2. Compute noise covariance matrix for each session. I will be basing the matrix on the resting baseline at the start of each session (2 minutes in length) before the experiment starts.

Unfortunately, I'm not quite sure what to do here afterwards. Alternatively, I guess instead of comparing voltages across conditions and doing the above, I could just simply apply a z-score normalization and compare that instead.

Apologies, my EEG experience so far is limited to simple ERP comparisons using Brainvision Analyzer, so I hope my question makes sense.

Kind regards,

Hi Nathan,

In the case of EEG, averaging all epochs across runs is generally fine, as the electrode locations in the scalp should be the same across runs. This is also true for average across subjects.

If you still want to do the averaging for each run (because of the EEG cap concern), it is also possible to do, just keep in mind that if the different runs have different number of epochs per condition, the subject average needs to be weighted. Group average is never weighted as all subjects should have the same weight.

For z-score, it is not recommended to compute it in individual epochs, but in the Subject average.

You can check some general recommendations for event-related responses in the Workflow tutorial:

and in the example tutorial for group analysis of MEG/EEG event-related responses
(which follows the methods described in this article


1 Like