[Eeglablist] GroupSIFT sessions

[Cedric Cannard]

Hi Erika,

Definitely don’t combine the sessions before ICA or dipole fitting. This is the same principle being discussed in the other thread “Single-trial principal components analysis of event-related potentials”: a decomposition only transfers cleanly when the data it’s applied to shares the same covariance structure as the data that produced the weights. Even if electrode positions are nearly identical (which is unlikely), day-to-day shifts in impedance, noise, and the participant’s state change the covariance structure of the recorded signals. So your ICs and dipoles come out different per session, and concatenating them into one ICA or borrowing weights across sessions would give you mushy components. 
So keep each session’s ICA and dipole fitting locked to its own data and combine them at the stats level. 

Not sure what SIFT can do at that level, but the standard EEGLAB solution  is component clustering: you cluster ICs across both sessions based on dipole location, scalp maps, and activity measures, and those clusters become your common “ROIs.” DIPFIT’s anatomical labeling can also map dipoles directly to atlas regions. Once both sessions’ components are expressed in the same set of clusters or anatomical labels, export your connectivity metric per subject, per session, and—crucially—per experimental condition.

Now, since you didn't provide much info on the study design and aims: assuming the two sessions are just a counterbalancing scheme (e.g., to control order effects) and your real research question is about some experimental conditions, then you can perform a linear mixed model (LMM) with Condition as the fixed effect of interest, Session as a fixed nuisance covariate (to absorb the day-to-day variance you already know exists), and Subject as a random intercept: Connectivity ~ Condition + Session + (1 | Subject)

--> This directly tests whether connectivity differs between your experimental conditions while controlling for session effects. 

Hope this helps,

Cedric


Sent with Proton Mail secure email.

On Thursday, June 18th, 2026 at 11:51 AM, Erika Nyhus via eeglablist <eeglablist at sccn.ucsd.edu> wrote:

> We are running GroupSIFT on data from a study that had two sessions of the same task that were run on separate days.  For both sessions we have 3D scans of the electrodes to improve dipole localization.  When should we combine the sessions since ICA and dipole localization differ by session?
> 
> 
> Erika Nyhus, Ph.D. [A button with "Hear my name" text for name playback in email signature]  <https://urldefense.com/v3/__https://www.name-coach.com/erika-nyhus__;!!Mih3wA!DQlBNqBX3QRYmnpG-QQ8G-826l26FuQCpMxFTJde5qmpMwvrdqGXyRxcVO2hXCtPBtxQvDB8xkmGBp97Rb8LY45g$ >
> 
> Department of Psychology and Program in Neuroscience
> 6900 College Station
> Bowdoin College
> Brunswick, ME 04011
> https://urldefense.com/v3/__https://www.bowdoin.edu/profiles/faculty/enyhus__;!!Mih3wA!DQlBNqBX3QRYmnpG-QQ8G-826l26FuQCpMxFTJde5qmpMwvrdqGXyRxcVO2hXCtPBtxQvDB8xkmGBp97RaNtnMyd$ <https://urldefense.com/v3/__https://www.bowdoin.edu/profiles/faculty/enyhus/index.html__;!!Mih3wA!DQlBNqBX3QRYmnpG-QQ8G-826l26FuQCpMxFTJde5qmpMwvrdqGXyRxcVO2hXCtPBtxQvDB8xkmGBp97RTPVAOx0$ >
> _______________________________________________
> To unsubscribe, send an empty email to eeglablist-unsubscribe at sccn.ucsd.edu or visit https://sccn.ucsd.edu/mailman/listinfo/eeglablist .
>