Downloading the data

Downloading the data#

Last reviewed: October 2026 (ABCD Release 7.0). Official sources: ABCD docs: Access, DEAP: My datasets, NBDC resource hub.

There are several ways to get ABCD data onto your machine, and which one is best depends on what you need. All of them require an approved DUC (see Getting access), and all downloaded data must live on a NIST 800-171–compliant system and be deleted when your project ends or your DUC expires.

Which route should I use?#

I want…

Use

Format

Notes

A handful of tabulated variables for one analysis

DEAP custom dataset

Parquet (recommended), CSV, TSV, RDS, Arrow, Stata, SPSS

Up to 10,000 variables; choose values vs. labels and missing codes vs. NA

A ready-made bundle of common tables

DEAP pre-assembled datasets

Parquet, CSV, TSV, RDS, Arrow

Compressed size shown; expect ≥ 5× when uncompressed

Whole instruments / tables as a CSV

NBDC Query Wizard

CSV (+ optional missingness reasons file)

Guide

Specific item-level variables found by searching the dictionary

NBDC Data Dictionary Query

CSV

Guide

The full tabulated release, to build datasets locally with NBDCtools

Globus (rawdata/phenotype/)

TSV + Parquet + JSON sidecars

Best for reproducible pipelines

Imaging (raw NIfTI, DICOM, FreeSurfer, ABCC), EEG, wearables, genetics, EARS/Fitbit

All Study Data Search & Filter → Globus

BIDS

Guide

An older release (6.0, 6.1)

Previous Releases → Globus

as released

Guide; one transfer at a time

Not to download at all

NBDC Sandbox

—

See Accessing the data

Route 1: DEAP (tabulated data, quickest start)#

  1. Log in to DEAP with your NBDC credentials.

  2. Find variables in the Data Dictionary (or with the semantic search tool).

  3. Add them to a custom dataset, or pick a pre-assembled dataset.

  4. Under My datasets, choose the format and options:

    • Format: Parquet keeps column types and labels and is much smaller than CSV.

    • Values vs. labels: labels are human-readable; values are easier to model. You can always map one to the other with the levels table.

    • Missing codes vs. NA: keeping the codes tells you why something is missing (e.g., “don’t know” vs. “refused”). Converting to NA is simpler for analysis.

  5. Each download includes data_dictionary_levels.xlsx. Keep it with your data.

Docs: Data dictionary · Create a dataset · My datasets · Video tutorial (DAIRC, 2025) · shorter overview

Route 2: NBDC Data Access Platform (Query tools)#

On the Data Access Platform:

  • Query Wizard returns whole instruments as a single CSV, optionally with a file explaining reasons for missingness.

  • Data Dictionary Query returns specific item-level variables found by searching the dictionary.

These work well for quick pulls when you don’t need Parquet.

Route 3: Globus (bulk, imaging, and file-based data)#

All large downloads go through Globus.

  1. Set up an endpoint.

    • Preferred: your institution’s Globus Connect Server (ask research computing). This usually points at a cluster that already meets your security requirements.

    • Otherwise: Globus Connect Personal on an approved machine. Do not tick “High Assurance” during setup (NBDC Globus guide).

  2. Select data.

    • For the full tabulated release, browse to rawdata/phenotype/.

    • For file-based data, use All Study Data Search & Filter (data wrangler) to filter by participant, session, modality, etc., then click Trigger Transfer.

    • Or browse directly in the Globus web app under Shared With You → “NBDC Datashare ABCD Release”.

  3. Download only what you need. Globus filter rules let you include/exclude by filename pattern (e.g., only *_T1w.nii.gz, or only certain sessions). Imaging data are very large.

  4. Globus verifies file integrity (checksums) during transfer by default, and failed transfers can be resumed.

Tip

For imaging, check the ABCC start page and the per-scan QC in scans.tsv before downloading. You may not need raw data at all if the tabulated imaging summaries (in DEAP) or ABCC derivatives cover your question.

Practical tips#

  • Use Parquet. It preserves types and labels, loads fast, and is much smaller. Read it in R with arrow or nanoparquet, or in Python with pandas, polars, or pyarrow.

  • Record exactly what you downloaded: release number, DOI, date, route, and the variable list. Save the DEAP dataset definition or your Globus filter rules in your project repo (they contain no participant data).

  • Don’t download into synced folders (Box, Dropbox, OneDrive, Google Drive, iCloud) unless your institution has approved them for controlled-access data. Also make sure data folders are in your .gitignore so they’re never pushed to GitHub.

  • Plan for deletion. Keep all ABCD data under one directory so it’s easy to delete at the end of the project/DUC.

  • Patch releases (e.g., 6.0 → 6.1) can change data. Re-download when you deliberately move releases, not mid-analysis.

Archived routes (NDA era)#

These applied to release 5.1 and earlier through the NIMH Data Archive. None of them distribute 6.0+ data.

Contents