Precomputing measures

Why do I need this step?

Two points make this step more important than it looks:

• What you choose here enters every result. The baseline, interpolation and ICA component removal are applied at compute time. If you change your mind, you compute again.
• EEGLAB saves every trial separately, not only the average. This lets you select trials later in the design, but it takes a lot of space with ERSPs.

How do I do it in EEGLAB?

About the data (D6): the same animal_mine.study from the creation lesson, with design 2, image x gender, from the design lesson. The “Starting point” script rebuilds both if it cannot find them.

1. Open the compute window

1. Run the “Starting point” script. The STUDY appears in the main window.
2. Open Study › then Precompute channel measures. The top of the window reads Pre-compute channel measures for STUDY ‘Go-nogo animal (mine)’.

2. Data options (top part)

1. Spherical interpolation of missing channels (performed after optional ICA removal below): keep it ticked; it is ticked by default in the window. If all files have the same channels, as you checked in the first lesson, it changes nothing.
2. Remove ICA artifactual components pre-tagged in each dataset: leave it unticked for now. See “When do I remove ICA components?” below.
3. Remove artifactual ICA cluster or clusters (hold shift key): leave it unticked. It needs clusters from the component clustering lesson.

3. Measures (bottom part)

1. Tick ERPs, then press the … button next to it. The Select parameters for ERP window opens.
2. In Baseline (min max) (“rmbase”) type the epoch start in ms, then 0. Take the start from the Epoch start (sec) value you noted in the first lesson and multiply it by 1000. In this copy the start is -1 sec, so type -1000 0. Press Ok.
3. Tick Power spectrum and keep its default value 'specmode', 'fft', 'logtrials', 'off'.
4. Do not tick ERSPs and ITCs here. The window computes them at every channel, which takes a long time and a very large amount of space. With the default grid (100 frequencies and 200 time points) they take about 1.5 GB per participant. We compute them with the script on a lighter grid.
5. Keep Overwrite files on disk ticked, so that an old file with different settings is not used. If files already exist, the window asks WARNING: SOME DATAFILES ALREADY EXIST, OVERWRITE THEM? and warns that another STUDY on the same files may be affected. Press Yes: the ready animal.study and animal_mine.study share the folders, and we work on our own copy.
6. Press Ok. The computation takes one to two minutes for the 14 participants, and progress lines appear in the command window.

4. ERSP and ITC with the script

Run the script below. In one call it recomputes ERPs and spectra with the same choices as the window, and adds ERSP and ITC at every channel with 20 frequencies from 3 to 30 Hz and 50 time points, instead of the default 100 frequencies and 200 time points. Why every channel and not only the two you need? In EEGLAB 2026.1, an ERSP file computed for some channels only cannot be read back: plotting stops with Unrecognized field name "chan1", because the file names each channel by its number in the full list while the reader uses its place in the short list. A lighter grid at every channel avoids this and keeps the files small.

5. Look at the files

1. Open the first participant’s folder (cba). It has three new files named with the participant code only, such as cba.daterp: .daterp (ERP), .datspec (spectrum) and .dattimef (the time-frequency decomposition of each trial). ERSP and ITC are both computed from .dattimef at plot time, so they have no separate file.
2. The older files whose names include the condition, such as cbaanimal.daterp, are the EEGLAB 6 files from 2007. Their names differ from the new ones, so EEGLAB neither reads nor overwrites them: they stay unused. You can leave them or delete them.
3. Notice there is one new file per participant, not one per .set file. EEGLAB puts the trials of both conditions in one file, and separates them at plot time according to the design.

How do I write it as code?

% STUDY and ALLEEG come from the starting point script
% ERPs and spectra as in the GUI, plus ERSP and ITC at every channel on a light grid
% (20 frequencies from 3 to 30 Hz, 50 time points) to keep the files small
baseline = [round(ALLEEG(1).xmin*1000) 0];   % from the epoch start to 0 ms
[STUDY, ALLEEG] = std_precomp(STUDY, ALLEEG, {}, 'savetrials', 'on', 'interp', 'on', 'recompute', 'on', ...
    'erp', 'on', 'erpparams', {'rmbase', baseline}, ...
    'spec', 'on', 'specparams', {'specmode', 'fft', 'logtrials', 'off'}, ...
    'ersp', 'on', 'itc', 'on', 'erspparams', {'cycles', [3 0.8], 'freqs', [3 30], 'nfreqs', 20, 'ntimesout', 50});
STUDY = pop_savestudy(STUDY, ALLEEG, 'savemode', 'resave');

% Files written for the first participant, with their size on disk
first = STUDY.datasetinfo(1);
measureFiles = dir(fullfile(first.filepath, [first.subject '.dat*']));
for k = 1:numel(measureFiles)
    fprintf('%-24s %8.1f MB\n', measureFiles(k).name, measureFiles(k).bytes/1e6);
end
fprintf('Channel groups in the STUDY: %d\n', numel(STUDY.changrp));

Which value should I choose?

How do I check it worked?

• The first participant’s folder has three new files named with the code only, cba.daterp, cba.datspec and cba.dattimef, and no new .datersp or .datitc. The cbaanimal.* and cbadistractor.* files are the old ones.
• The script prints the size of each file. The three are close, tens of MB each (.dattimef about 75 MB, an estimate) , because we used a light grid for ERSPs. With the default grid, .dattimef would be about 1.5 GB.
• The script prints the number of channel groups, Channel groups in the STUDY, which equals the number of channels in the files.
• The command window shows no red errors and no IMPORTANT WARNING.

What mistakes do people make?

• Changing an option without recomputing. If you untick Overwrite files on disk, EEGLAB uses the existing files even if you computed them with another baseline.
• Computing ERSPs at every channel from the window. The disk fills up and the computation takes a long time.
• Computing ERSPs at a few channels only. The files are written, but plotting them stops with Unrecognized field name "chan1" in EEGLAB 2026.1. Compute at every channel with a lighter grid.
• Removing ICA components twice. If you removed components from each file before the STUDY, do not tick the option here. What is still flagged may be a clean component.
• Taking the 2007 files for your results. They were computed with settings you do not know, and EEGLAB does not read them. Use only the files named by participant code that you computed yourself.
• Moving the measure files (.daterp and the like) alone. They belong to the participant’s files. Move the whole folder.

How do I write it in my paper?

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

Channel ERPs and spectra were precomputed for each participant and condition (std_precomp), with baseline correction from [−1000] to 0 ms; with spherical interpolation of any missing channels enabled; event-related spectral perturbations were computed at every channel for [20] frequencies from [3] to [30] Hz.

Going further

• EEGLAB Tutorial: https://eeglab.org/tutorials/10_Group_analysis/study_data_visualization_tools.html (10. Group analysis > Precomputing ERPs)
• EEGLAB Tutorial: https://eeglab.org/tutorials/11_Scripting/command_line_study_functions.html (11. Scripting > Computing measures)
• Go-nogo categorization and detection task (OpenNeuro ds002680): https://openneuro.org/datasets/ds002680 (README)