[Eeglablist] ERP analyses across groups with different latencies

PERNET Cyril cyril.pernet at ed.ac.uk
Fri Mar 22 01:47:13 PDT 2019

Hi Kim, Mate,

I have quite a different take on this - let me ask why would you want to capture mean amplitude / peak? from a physiological perspective, peak index a change in neural computation so in your case whatever before / after N2 peak are likely different computational things.

What I would recommend is using a hierarchical linear model with a group level focusing either on a virtual electrode or better an IC (if there is one that captures well the N2). In short, compute for each subject estimate of each effect and at the group level perform a comparison of the whole time course of the IC or of the electrodes showing the best fit per subjects (which make one 'virtual' electrode). This way slight differences in location between groups/subject are accounted for. By analyzing the whole time course, a difference in peak latency is captured via differences in amplitudes telling you when the two N2 start to diverge. The whole thing is available via the LIMO EEG toolbox (https://github.com/LIMO-EEG-Toolbox/limo_eeg) and if you use the beta version of eeglab, straightforwardly available from the eeglab STUDY.



Dr Cyril Pernet,
Senior Academic Fellow
Neuroimaging Sciences

Centre for Clinical Brain Sciences
Chancellor's Building, Room GU426D
The University of Edinburgh
49 Little France Crescent
Edinburgh BioQuarter EH16 4SB

cyril.pernet at ed.ac.uk
tel:  +44 (0)131 465 9530

From: Whitehead, Kimberley <k.whitehead at ucl.ac.uk>
Sent: 21 March 2019 06:33
To: Mate Gyurkovics; eeglablist at sccn.ucsd.edu
Subject: Re: [Eeglablist] ERP analyses across groups with different latencies


A Scientific Reports paper from our lab used a similar study design to your own (a within-subject condition and then the between-subjects factor is child/adult), you could have a look for ideas: DOI: 10.1038/srep28642

We often use Woody filtering (DOI: 10.1007/BF02474247) to align traces and correct for intra- and inter-subject small differences in latency, but your latency differences might be too big for this.

But we’ve also recently started using Thomas Koenig’s Ragu software. This software encourages you to check topographic similarity before comparing magnitudes (GFP in the case of Ragu). Because if the children and adults don’t have the same cortical source configuration of the ‘frontocentral N2’, then that is the more relevant way to compare them, rather than comparing amplitudes which actually derive from different sources.


Kimberley Whitehead

Research Associate

UCL Dept. of Neuroscience, Physiology and Pharmacology

Tel: 020 7679 3533 (internal 33533)

From: eeglablist <eeglablist-bounces at sccn.ucsd.edu> On Behalf Of Mate Gyurkovics
Sent: 20 March 2019 12:26
To: eeglablist at sccn.ucsd.edu
Subject: [Eeglablist] ERP analyses across groups with different latencies

Hi all,

I am looking for "best practice" suggestions on how to deal with analyzing group differences in component magnitude when the two groups have different latencies and probably different latency jitter.

I have EEG data from two age groups (kids and adults), and I would like to look at the magnitude of the frontocentral N2 component across the two groups. I have a within-subject condition (conflict vs. no conflict trial) too, so what I'm mostly interested in is a Trial Type by Age Group interaction. I was originally planning on using mean amplitude but I ran into a problem, namely that the latency of the component is different between the two groups, so using the same time window to capture the N2 in mean amplitude in both groups seems difficult or impossible because if the window is broad enough to capture the wider peak of the younger group, it's too wide for the adults, and if it suits the adults, it's too narrow for the kids. I am reluctant to switch to peak amplitude because trial numbers are slightly different across conditions thus noise level differs too.

One option I considered is to create multiple shorter time windows and get the mean from each of those, and then add "Time" as a variable to my analyses, however the choice of time window lengths and number of time windows feels very arbitrary. I was also considering using an adaptive mean approach as that hopefully would be more robust to slight differences in noise level than a simple peak amplitude measure, but I'm not sure. What do you think? Any suggestions are welcome.

Thank you.



The University of Edinburgh is a charitable body, registered in Scotland, with registration number SC005336.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://sccn.ucsd.edu/pipermail/eeglablist/attachments/20190322/983a7ae4/attachment.html>

More information about the eeglablist mailing list