ERPs & averaging

Why do I need this step?

The idea: the response to the stimulus repeats with the same shape and timing in every trial, while the other activity varies at random. If you average all epochs point by point, the response stays and the random activity cancels out.

The resulting waveform is the event-related potential (ERP). Everything after it builds on it: measurement, comparing conditions, and statistics.

How do I do it in EEGLAB?

Start with the dataset D1 epochs clean open, as in “Starting point”.

EEGLAB has no separate “compute average” menu. The average is computed inside the ERP plotting menus, or with one line in a script. Here we display it once for a first look; the plotting options in detail come in station 7.

1. Open Plot › then Channel ERPs › then In scalp/rect. array.
2. The Topographic ERP plot - pop_plottopo() window opens. Keep Channels to plot at 1:32, leave Plot single trials and Plot in rect. array unchecked, and press Ok.
3. The average of each channel appears at its place on the head. Click the Pz trace to enlarge it in its own window.
4. In the enlarged trace, look for two things after the vertical line at zero:
   • A small negative wave around 200 ms, clearest at posterior channels such as POz and Oz.
   • A large positive wave peaking around 400 ms, largest at central and parietal channels such as Cz and Pz.

How do I write it as code?

% The ERP is the mean over epochs: channels x time points
% (the same line pop_plottopo computes in the GUI step)
erp = mean(EEG.data, 3);
fprintf('Averaged %d epochs\n', EEG.trials);

% Plot the ERP at Pz, positive up
pz = find(strcmp({EEG.chanlocs.labels}, 'Pz'));
figure;
plot(EEG.times, erp(pz, :), 'k', 'LineWidth', 1.5);
xlim([-200 800]); grid on;
xline(0, '--'); yline(0);
xlabel('Time (ms)'); ylabel('Amplitude (\muV)');
title(sprintf('ERP at Pz, %d epochs', EEG.trials));

% Approximate latency of the largest positive peak between 250 and 600 ms
win = EEG.times >= 250 & EEG.times <= 600;
winTimes = EEG.times(win);
[peakAmp, iPeak] = max(erp(pz, win));
fprintf('Positive peak at Pz: %.1f uV at %.0f ms\n', peakAmp, winTimes(iPeak));

% Quick look at all channels, same as the GUI step
figure; pop_plottopo(EEG, 1:EEG.nbchan, EEG.setname, 0, 'ydir', 1);

Which value should I choose?

How do I check it worked?

• size(erp) returns 32 384: channels × time points, with no epoch dimension.
• At Pz there is a positive peak between roughly 300 and 500 ms, of about 10 µV. The official tutorial notes that the largest ERP variance in these data is at about 430 ms.
• The waveform before zero (the baseline period) is close to zero and fairly flat.
• If you plot the average of only 10 epochs, the waveform is rougher than the average of all epochs.

What mistakes do people make?

• Mixing conditions in one average. In D1 the square appeared at two positions (the position field). Averaging both positions together is fine as an example here, but comparing positions needs one average each. Comparing conditions is covered in the channel ERP plotting lesson in station 7.
• Reading the highest peak as “the component”. A peak at one channel can come from more than one overlapping ERP component. The approximate latency here is for recognizing the waveform, not for a scientific conclusion.
• Confusing the axis direction. A positive wave plotted downward is misread as negative. Check the axis.
• Averaging before rejection. One epoch with a large artifact is enough to change the average. Reject first, as in the previous lesson.
• Very different trial counts between conditions. A condition with 20 trials and another with 80 differ in noise, so their peaks look different even when the response is the same.

How do I write it in my paper?

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

ERPs were computed by averaging the [N] artifact-free epochs time-locked to the square stimulus (EEGLAB 2026.1).

Going further

• EEGLAB Tutorial: https://eeglab.org/tutorials/08_Plot_data/Data_Averaging.html (8. Plot data > a. ERPs: Plotting ERPs in a Topographic Map)
• Applied Event-Related Potential Data Analysis (Luck): https://socialsci.libretexts.org/Bookshelves/Psychology/Biological_Psychology/Applied_Event-Related_Potential_Data_Analysis_(Luck) (Chapter 6: Assigning Events to Bins, Averaging, Baseline Correction, and Assessing Data Quality)