Re-referencing

What does this show?

The simulation needs JavaScript. Here is a still image of the same idea.

Why do I need this step?

If you change the reference, all channels change together. A channel close to the old reference looks small, and after the change it may grow a lot. There is no “true” channel without a reference.

Re-referencing computes the data as if they had been recorded with another reference. The two most common choices:

• Average reference: the mean of all scalp channels at each moment. The official tutorial recommends it for source analysis, and it is the reference on our path before ICA.
• Linked mastoids: the mean of two electrodes behind the ears. Common in ERP studies, such as Luck’s book.

How do I do it in EEGLAB?

Start with the main dataset after interpolation loaded, as in the Starting point. In D1 this is D1 badch from the bad channels lesson, because there was nothing to interpolate.

1. First make sure the channel count is complete: Channels per frame is 32 in D1.
2. Open Tools › then Re-reference the data. A window titled pop_reref - average reference or re-reference data appears.
3. The first line, Current data reference state is:, shows the current reference. In D1 it may show unknown, because the file does not store the reference name for each channel.
4. Keep Compute average reference checked; it is the default.
5. Leave Huber average ref. with threshold unchecked. The official tutorial says it should not be used before ICA.
6. Next to Exclude channel indices (EMG, EOG), press … and choose EOG1 and EOG2. The field fills with their names.
7. Leave Add old ref. channel back to the data empty, because the mastoid channel is not in D1.
8. Press OK. The command window prints Re-referencing data.
9. Name the new dataset D1 main and leave Overwrite it in memory (set=yes; unset=create a new dataset) unchecked.
10. Save it with File › then Save current dataset as. This is the “main dataset” that the ICA lessons start from.
11. Compare the two datasets with Plot › then Channel data (scroll): all scalp channels changed, and EOG1 and EOG2 stayed the same.

How do I write it as code?

% Keep a copy to compare before and after
EEG_before = EEG;

% Average reference, EOG channels excluded
eogIdx = find(ismember({EEG.chanlocs.labels}, {'EOG1', 'EOG2'}));
EEG = pop_reref(EEG, [], 'exclude', eogIdx);
EEG.setname = 'D1 main';

% Check: scalp channels now sum to (almost) zero at every time point
scalpIdx = setdiff(1:EEG.nbchan, eogIdx);
fprintf('Reference: %s\n', EEG.ref);
fprintf('Largest sum over scalp channels: %g uV\n', max(abs(sum(EEG.data(scalpIdx, :), 1))));

% Channel Oz over 5 seconds, before and after
oz = find(strcmp({EEG.chanlocs.labels}, 'Oz'));
t = 1:5*EEG.srate;
figure;
plot(t / EEG.srate, EEG_before.data(oz, t), t / EEG.srate, EEG.data(oz, t));
legend('Original reference (mastoid)', 'Average reference');
xlabel('Time (s)'); ylabel('Amplitude (uV)');

EEG = pop_saveset(EEG, 'filename', 'D1_main.set', 'filepath', pwd);

This is the same pop_reref line as in the Starting point scripts of the ICA lessons. The number the check prints is very small but not exactly zero, because the data are stored in single precision.

Which value should I choose?

How do I check it worked?

1. Main window: the Reference line shows average, and Channels per frame is still 32.
2. The script check: EEG.ref is average, and the sum over scalp channels is a very small number close to zero.
3. The eye channels did not change: in the scroll view, EOG1 and EOG2 look as they did before re-referencing.
4. The scalp channels changed: Oz, for example, looks clearly different from its shape with the mastoid reference.

What mistakes do people make?

• Including EOG or ECG channels in the average reference. These are non-brain channels, and their large signal enters every channel through the average. Exclude them in Exclude channel indices (EMG, EOG).
• Forgetting that the average reference reduces the data rank by one. After an average reference the scalp channels sum to zero, so any one of them equals minus the sum of the others. That is why in D1 the data rank becomes 31, not 32, and ICA gives 31 components.
• Re-referencing before interpolation. An average over missing channels leans toward the side with more channels. Interpolate first.
• Changing the reference after ICA. The weights are tied to the reference ICA was computed on. Re-reference before ICA and leave it afterward. EEGLAB warns you if you change it.
• Adding a reference channel that does not exist. In D1 the mastoid channel is not in the data. Do not fill Add old ref. channel back to the data unless you defined it in the channel editor.

How do I write it in my paper?

A template: edit the data, settings and versions to match what you ran.

Data were re-referenced to the average of all scalp channels (pop_reref, EEGLAB 2026.1); EOG channels were excluded from the reference.

Going further

• EEGLAB Tutorial: https://eeglab.org/tutorials/05_Preprocess/rereferencing.html (5. Preprocess data > b. Re-referencing)
• EEGLAB Tutorial: https://eeglab.org/tutorials/ConceptsGuide/rereferencing_background.html (Concepts guide: Re-referencing)
• Luck, Applied Event-Related Potential Data Analysis (1st edition): https://socialsci.libretexts.org/Bookshelves/Psychology/Biological_Psychology/Applied_Event-Related_Potential_Data_Analysis_(Luck) (Chapter 5: Referencing and Other Channel Operations)
• What is the best EEG reference?: https://www.youtube.com/watch?v=ioIETUX4G4k (Full video)