ERP image

Why do I need this step?

The brain does not respond with the same timing on every trial. If a component comes with the button press, and the press time varies between 330 and 730 ms, that component is smeared across the average and looks smaller and wider than it really is.

An ERP image shows each trial as one colored row and stacks the rows on top of each other. Sort them by reaction time and you can see what follows the response and what follows the stimulus, before the difference disappears in the average.

How do I do it in EEGLAB?

Start with the D1 epochs clean dataset loaded, as in “Starting point”. Each epoch has a square event at time 0, and most have an rt event at the button press.

1. The raw image

1. Open Plot › then Channel ERP image.
2. In Channel, click the ... button and choose Pz.
3. In Smoothing, enter 1. The default is 10, but first we want to see each trial on its own.
4. In Time limits (ms), enter -200 800.
5. Make sure Plot scalp map, Plot ERP and Plot colorbar are checked. Click Ok.
6. An image appears. The horizontal axis is time, and each row from bottom to top is one trial in recording order. The averaged ERP is below it.

2. Sort trials by reaction time

1. Open Plot › then Channel ERP image again. The fields remember your previous values.
2. In Smoothing, enter 10.
3. Click the Epoch-sorting field button and choose latency.
4. Click the Event type(s) button and choose rt.
5. In Event time range, enter 0 1000. EEGLAB then only takes the response that follows the stimulus of that epoch.
6. Leave Rescale at no and Align empty. Click Ok.
7. The trials are now sorted, fastest at the bottom and slowest at the top, with a black curve through the response time on each row. Look in the command window for the line Removing ... trials with NaN sortvar values: these are trials without a response, left out of the image.

3. View it from the response side

1. Open the window a third time, enter Inf in Align, and clear Time limits (ms). Click Ok.
2. All trials are now shifted so the response falls on a straight line at the median reaction time, and the slanted curve now marks the stimulus time.
3. Compare: what stays straight in the second image is tied to the stimulus, and what becomes straight here is tied to the response.

4. Read the image

How do I write it as code?

EEG = pop_loadset('D1_epochs_clean.set', pwd);
chan = find(strcmp({EEG.chanlocs.labels}, 'Pz'));

% Response latency per epoch, NaN when no response follows the stimulus
rt = eeg_getepochevent(EEG, 'type', {'rt'}, 'timewin', [0 1000]);
fprintf('%d of %d epochs have a response, median RT %.0f ms\n', ...
    sum(~isnan(rt)), EEG.trials, median(rt, 'omitnan'));

% Trials in recording order, no smoothing
figure; pop_erpimage(EEG, 1, chan, [], 'Pz', 1, 1, {}, [], '', ...
    'erp', 'on', 'cbar', 'on', 'limits', [-200 800], ...
    'topo', {chan EEG.chanlocs EEG.chaninfo});

% Trials sorted by response latency, 10-trial moving average
figure; pop_erpimage(EEG, 1, chan, [], 'Pz sorted by RT', 10, 1, {'rt'}, [0 1000], 'latency', ...
    'erp', 'on', 'cbar', 'on', 'limits', [-200 800], ...
    'topo', {chan EEG.chanlocs EEG.chaninfo});

% Same trials aligned to the median response latency
figure; pop_erpimage(EEG, 1, chan, [], 'Pz aligned to response', 10, 1, {'rt'}, [0 1000], 'latency', ...
    'align', Inf, 'erp', 'on', 'cbar', 'on', ...
    'topo', {chan EEG.chanlocs EEG.chaninfo});

Which value should I choose?

How do I check it worked?

• The black curve rises from left to right between about 330 and 730 ms, a little narrower with smoothing, and never passes 800 ms.
• The exclusion message in the command window reports 6 trials or fewer, equal to the number of epochs minus the count with a response from the script’s fprintf line.
• With Align set to Inf, the black curve becomes a straight line at about 400 ms and the stimulus-time curve appears slanted.
• With Smoothing set to 1, you see more detailed and less regular rows than with 10.

What mistakes do people make?

• Leaving Event time range empty. EEGLAB takes the first rt anywhere in the epoch, which may belong to another stimulus. The sorting is then wrong, with no error message.
• Smoothing too much. With a 40-trial width out of about 75 trials, each row averages more than half the data, and the image becomes a repeated ERP that adds nothing.
• Forgetting the excluded trials. Trials without a sorting value are removed from the image and from the waveform below it. That waveform may therefore differ slightly from the ERP in the averaging lesson. Read the Removing ... trials message and report the count.
• Reading a slanted band as a new component. The image shows a relation to the response, not its source. Confirming it takes a group-level analysis.
• Plotting an ERP image for every channel. The tool handles one channel (or the mean of several) at a time. Choose the channel from the channel ERP plot first.

How do I write it in my paper?

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

Single-trial ERP images at Pz were sorted by reaction time and smoothed with a 10-trial moving average (pop_erpimage, EEGLAB 2026.1); [N] epochs without a response between 0 and 1000 ms were excluded from the image.

Going further

• EEGLAB Tutorial: https://eeglab.org/tutorials/08_Plot_data/Plotting_ERP_images.html (8. Plot data > b. ERP-image: Basic plotting options; Sorting trials by reaction time; Using the Align parameter)
• EEGLAB Concepts Guide: https://eeglab.org/tutorials/ConceptsGuide/erp_image_background.html (ERP image)