Data Management Plan — Projet 25 (ADARP: EMA + Physio)

Version 0.1 — Phase 15a, data pipeline & governance layer.

This DMP follows the Science Europe / ANR core structure (six sections). It is intentionally concise; a fuller DMP would accompany a real SANPSY-style protocol with IRB, informed consent, and GDPR DPIA.

1. Data description

Origin

The project reuses the ADARP (Alcohol and Drug Abuse Research Program) public dataset, published by Sah et al. (Zenodo DOI 10.5281/zenodo.6640290, CC-BY 4.0). Data were collected 2019–2020 at Washington State University on adults in outpatient treatment for alcohol use disorder (AUD).

Components used in this project

Component Source file Format Size
Empatica E4 wristband recordings Sensor Data/Part NNNC/<device>_<date-time>/*.csv Empatica native CSV ≈ 4.56 GB
Stress events (EMA-like) Final Phone Survey Stress Events.xlsm Excel macro 26 kB
Alcohol use (not used here) PDF scans of paper calendars PDF ≈ 2 MB

What is not in the public release

The dataset description on Zenodo and the companion paper (Sah et al., arXiv 2206.14568) mention a 14-day EMA web survey with four daily prompts covering self-reported emotions, cravings, and stress. This component is not present in the public files as of April 2026. Our analyses therefore use the “Final Phone Survey Stress Events” xlsm as the EMA-like outcome source.

Known limitations of the Stress Events file (documented honestly)

  1. Sample size: 10 subjects, 99 stress events total (not the expected 4 prompts × 14 days × 15 subjects ≈ 840). Per-subject counts are highly unbalanced (range: 1 to 24 events).
  2. Temporal resolution of EMA prompts: the time submitted column has Excel number_format mmss.0 and its values correspond to survey-completion durations (minutes:seconds), not clock times. The clock time of each prompt is therefore not recoverable from the public data. We impute a canonical local hour per databurst code: 1 = morning → 09:00, 2 = midday → 12:00, 3 = afternoon → 15:00, 4 = evening → 20:00, in America/Los_Angeles timezone.
  3. Cravings: not captured. The outcome used is binary stress_event (derived from more_stressed) and continuous stress_composite (sum of more_stressed + more_overwhelm + more_anxious, range 0–3).

2. Standards and metadata

  • Signal timestamps: Empatica E4 stores the session start as Unix UTC seconds in row 1 and the sampling rate (Hz) in row 2. Subsequent samples are regularly spaced; per-sample timestamps are reconstructed as ts_start + i / sr_hz.
  • Subject folder naming: Part 101C, Part 102C, …, encoded directly into the subject_id field.
  • Session folder naming: <DeviceID>_<YYMMDD>-<HHMMSS>, parsed into device_id (string) and session_start_utc (UTC Timestamp). The time portion is the device’s wall clock when recording started.
  • Processed data format: Apache Parquet (columnar, language-agnostic, efficient). Read natively by Python (pyarrow) and R (arrow).

3. Storage, backup, security

Local storage (developer machine, data owner: F. Martin-Gomez)

  • Raw data (4.56 GB): C:\Users\fmarting\Documents\Travail_Data\6640290.
  • Not versioned with git; excluded by .gitignore.

Project repository (public, GitHub Pages)

  • Contains code only (Python modules, R helpers, Quarto sources).
  • Contains a slim demo subset (≤ 100 MB) derived from the full dataset, sufficient to run the entire pipeline on a few subjects for reproducibility. Published under the same CC-BY 4.0 licence as the original dataset, with explicit attribution.
  • No personally identifying information is redistributed; ADARP is already pseudonymous (numeric case IDs).

Backup

The raw dataset is always retrievable from Zenodo with fixed DOI; no custom backup needed. Processed intermediates can be fully regenerated by running python prepare_data.py.

5. Data sharing

  • Code: MIT licence, public on GitHub.
  • Demo data: redistributed under CC-BY 4.0 with the DEMO_README.md pointing to the original Zenodo DOI.
  • Processed parquets from the full 15-subject run: NOT redistributed (would duplicate the original dataset). Users reproduce them locally from the Zenodo download + this pipeline.
  • Notebooks and reports: rendered HTML from Quarto, public.

6. Responsibilities and resources

  • Data custodian: Francisco Martin-Gomez.
  • Software maintenance: same, with an issue tracker on the GitHub repository.
  • Reproducibility window: as long as Zenodo record 6640290 is online and the pinned dependency versions in requirements.txt remain available on PyPI.