What a STUDY is and when you need one

Why do I need this step?

In EEGLAB, a STUDY is an index file for that question. It stores the list of participant files, and for each file the subject, condition, group and session. It also stores designs and precomputed measures. It does not merge the data, and it does not process it again.

So a STUDY trusts that your files match in everything that matters. If the sampling rate, the epoch window or the event names differ between participants, functions crash or silently give wrong results. This lesson teaches you to check before you create one.

How do I do it in EEGLAB?

About the data (D6): a Go-nogo experiment with 14 participants (Delorme et al.). A picture is flashed for 20 ms, and the participant lifts a finger off a button as fast as possible if it contains an animal, and keeps it down if not. The EEGLAB copy, ready as a STUDY, is already processed and epoched, with one file per participant for animal pictures (animal) and one for the other pictures (distractor). The raw data are on OpenNeuro in BIDS format, and we use them in the lesson on creating a STUDY from BIDS.

Why not continue with D1? Because D1 is one participant’s recording, and a STUDY needs several participants.

1. Download the data and open the folder

1. Run the “Starting point” script. It downloads animal_study.zip from the EEGLAB data page, unzips it once, and prints the folder location.
2. Open the folder in your file browser. It holds the ready file animal.study and one folder per participant named by its code, from cba to wpa. Each folder holds two datasets, such as sphanimal.set and sphdistractor.set. The data of each dataset are in a .dat file next to it (the older EEGLAB format; only cba keeps its data inside the .set). You will also see files such as sphanimal.daterp and sphanimal.icaerp: measures precomputed with EEGLAB 6 in 2007. We do not use them, and the precompute lesson explains why.
3. Do not open animal.study yet. We check the files before any STUDY.

2. Check one participant’s file

1. Open File › then Load existing dataset and choose animal_study/cba/cbaanimal.set, the animal-condition file of the first participant.
2. Read the main window and note these values on paper: Channels per frame, Frames per epoch, Epochs, Sampling rate (Hz), Epoch start (sec), Epoch end (sec), Reference, Channel locations, ICA weights.

3. Check another participant’s file and compare

1. Open animal_study/wpa/wpaanimal.set, the animal-condition file of the last participant, the same way.
2. Compare the values with your notes. All of them must match except Epochs, because the number of trials left after rejection naturally differs between people. In this copy you should read 32 channels, 750 frames per epoch, 250 Hz, and an epoch from -1 to 1.996 sec.
3. The script below compares the same two files, on the same values, plus channel names, event types, ICLabel results and dipoles, which the main window does not show.

4. STUDY or ALLEEG?

How do I write it as code?

% studyDir comes from the starting point script
files = dir(fullfile(studyDir, '*', '*.set'));
fprintf('%d dataset files under %s\n', numel(files), studyDir);

% Header only (no data) of the animal file of the first and the last participant
animalFiles = files(contains({files.name}, 'animal'));
A = pop_loadset('filename', animalFiles(1).name,   'filepath', animalFiles(1).folder,   'loadmode', 'info');
B = pop_loadset('filename', animalFiles(end).name, 'filepath', animalFiles(end).folder, 'loadmode', 'info');

% Side-by-side check of what must match before a STUDY
show = @(label, a, b) fprintf('%-20s %-14s %-14s\n', label, num2str(a), num2str(b));
fprintf('%-20s %-14s %-14s\n', 'Property', animalFiles(1).name, animalFiles(end).name);
show('Sampling rate (Hz)', A.srate,  B.srate);
show('Channels',           A.nbchan, B.nbchan);
show('Points per epoch',   A.pnts,   B.pnts);
show('Epoch start (s)',    A.xmin,   B.xmin);
show('Epoch end (s)',      A.xmax,   B.xmax);
show('Epochs',             A.trials, B.trials);
show('ICA weights',        ~isempty(A.icaweights), ~isempty(B.icaweights));
show('ICA components',     size(A.icaweights, 1), size(B.icaweights, 1));
show('ICLabel results',    isfield(A.etc, 'ic_classification'), isfield(B.etc, 'ic_classification'));
show('Dipoles',            isfield(A.dipfit, 'model'), isfield(B.dipfit, 'model'));
fprintf('Same channel labels in the same order: %d\n', ...
    isequal({A.chanlocs.labels}, {B.chanlocs.labels}));

% Event types and their counts in each file
eeg_eventtypes(A);
eeg_eventtypes(B);

Which value should I choose?

This is the checklist before any STUDY. “Must” means functions crash or go wrong if it differs.

How do I check it worked?

• The script prints the number of files in the folder, 28 if each of the 14 participants has two files.
• Every line matches between the two files except Epochs and ICA components, and the channel-name line prints 1. cba has 29 components and wpa 31: the number of components can differ between participants, and clustering does not need it to match.
• eeg_eventtypes prints the same event types for both animal files, Animal and rt, even if the counts differ. A distractor file has Distractor only, because there is no response on no-go trials.
• You know the answer to: do the files have ICA weights and dipoles? Here the authors’ ICA weights and dipoles are present (1 on both lines) and ICLabel results are not (0). The clustering lesson keeps the ICA and the dipoles, and runs ICLabel.

What mistakes do people make?

• Creating the STUDY before checking. A STUDY is created without error even if the files differ, and the problem shows up an hour later during precomputing.
• Trusting the file name alone. Two files both called clean may have gone through different filters. Look at EEG.history or the command history if in doubt.
• Mixing old and new processing. A participant processed a year ago with old settings stays different even if everything in the main window matches.
• Forgetting that ICLabel runs per file. If you ran it for some participants only, precomputing removes components from some and leaves the rest.
• Data with a different mains frequency. D6 was recorded in France, where mains is 50 Hz. Your data in Saudi Arabia have 60 Hz. This does not matter within one STUDY whose files all went through the same steps, but it does when you combine data from two labs.

How do I write it in my paper?

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

Before group analysis, the datasets of all [14] participants were checked for identical sampling rate, epoch limits, channel labels and event types (EEGLAB 2026.1).

Going further

• EEGLAB Tutorial: https://eeglab.org/tutorials/10_Group_analysis/multiple_subject_proccessing_overview.html (10. Group analysis > a. Multiple subject processing overview)
• EEGLAB Tutorial: https://eeglab.org/tutorials/10_Group_analysis/study_creation.html (10. Group analysis > b. STUDY creation)
• From raw data to group level analysis in EEGLAB: https://www.youtube.com/watch?v=-jL3PuHD3aY (Full video)
• Go-nogo categorization and detection task (OpenNeuro ds002680): https://openneuro.org/datasets/ds002680 (README; dataset_description.json)