Designing EEG experiments

Why do I need this step?

This lesson is optional, because the path works on data that has already been recorded. But it matters a lot if you will record your own data, or want to judge an open dataset before choosing it.

How do I do it in EEGLAB?

There are no EEGLAB steps in this lesson, because design happens before recording.

After recording, you will check the result of the design in EEGLAB in the events lesson: you list the events, their types and timing from Edit › then Event values, and see them as vertical lines over the signal in Plot › then Channel data (scroll).

How do I write it as code?

This script does not need EEGLAB. It simulates the most important design decision: the number of trials. Each trial contains the same wave (like a P3 component of 5 microvolts) plus random noise much larger than it. Then it averages different numbers of trials.

% How the number of trials changes the noise left in an average
srate = 250;                                         % samples per second
t     = -0.2:1/srate:0.8;                            % epoch from -200 to 800 ms
erp   = 5 * exp(-((t - 0.35).^2) / (2*0.08^2));      % P3-like wave, 5 microvolts
rng(1);                                              % same noise on every run

nTrials = [10 40 160];
labels  = cell(1, numel(nTrials));
figure; hold on;
for k = 1:numel(nTrials)
    n      = nTrials(k);
    trials = erp + 20 * randn(n, numel(t));          % same wave + noise on each trial
    avg    = mean(trials, 1);                        % the average across trials
    noise  = std(avg(t < 0));                        % noise left in the pre-stimulus period
    fprintf('%4d trials: noise left = %.2f microvolts\n', n, noise);
    plot(t*1000, avg);
    labels{k} = sprintf('%d trials', n);
end
plot(t*1000, erp, 'k', 'LineWidth', 2);
xlabel('Time (ms)'); ylabel('Amplitude (\muV)');
legend([labels, {'True wave'}]);

• A single trial does not show the wave, because the noise (20 microvolts) is four times larger than it.
• When you average, the constant wave stays and the random noise cancels out (Luck, Appendix 1).
• The remaining noise shrinks with the square root of the number of trials. Four times more trials only halves the noise.
• Simplification: real noise is not fully random, and the wave varies a little between trials.

Which value should I choose?

This is a checklist for any design, with the reason and source for each item.

An example from the path: eeglab_data.set has two events: square when the stimulus appears, and rt at the button press. The participant presses only when the stimulus appears at the attended location (EEGLAB tutorial, dataset description). This is what makes cutting epochs around square possible in station 5.

How do I check it worked?

• The script prints three numbers for the remaining noise, each about half of the one before (close to 6.3, 3.2 and 1.6).
• In the plot, the 160-trial curve is the closest to the black true wave.
• You can name two checklist items that cannot be fixed after recording: event recording and event timing.
• Given a design with two conditions, you can say whether they differ in one factor or more.

What mistakes do people make?

• Events saved only in the presentation software. After recording you find that the EEG file has no events, or events that do not match the timing. Test sending codes before the first participant.
• Not measuring screen delay. Every component will appear late by the same unknown amount, and the delay may vary between trials.
• Two conditions that differ in more than one factor. The ERP difference cannot be attributed to a single factor.
• A trial count with no margin. After rejecting bad trials in station 5, fewer trials than needed may remain.
• Recording without EOG channels. Telling blinks and eye movements apart becomes harder in station 4.

Going further

• EEGLAB Concepts guide: https://eeglab.org/tutorials/ConceptsGuide/Setting_up_your_EEG_lab.html (Maximizing EEG data quality; EEG synchronization)
• EEGLAB Tutorial: https://eeglab.org/tutorials/tutorial_data.html (Sample experiment description)
• Luck, Applied Event-Related Potential Data Analysis (1st edition): https://socialsci.libretexts.org/Bookshelves/Psychology/Biological_Psychology/Applied_Event-Related_Potential_Data_Analysis_(Luck)/12%3A_Appendix_1%3A_A_Very_Brief_Introduction_to_EEG_and_ERPs (Appendix 1 > Signal Averaging and Event-Related Potentials)
• 16 advices for conducting electro-encephalography experiments: https://www.youtube.com/watch?v=SSotxRqdFyY (Full video)