FVPD DATASET README
Version: 1.0
Date: 28 August 2026

Dataset title
FVPD: Facial Action Unit, Personality and Depression Measures Derived from
Naturalistic Free-Speech Video in an Arabic-Speaking Sample (Tartous, Syria)

Repository record
University of Glasgow Enlighten: Research Data
Data reference: 2400
Reserved DOI: 10.5525/gla.researchdata.2400

Access status
The participant-level dataset is pseudonymised, not anonymous, and is deposited
under controlled access. Access is subject to the conditions specified on the
Enlighten record and the applicable University of Glasgow data-sharing agreement.

The original facial video and audio recordings are not included in the deposit
and are not available for reuse.

1. FILES INCLUDED

FVPD_participant_level.xlsx
    Controlled-access participant-level dataset. The workbook contains three
    worksheets:
      - README: summary information about the dataset and disclosure controls.
      - Data: one row per participant (103 rows), with 37 participant-level
        variables.
      - Data dictionary: variable names, data types, descriptions and ranges.

FVPD_README.txt
    This file. Public documentation for the dataset.

No folder structure is required for the deposited dataset.

2. SOFTWARE REQUIRED

FVPD_participant_level.xlsx can be opened using Microsoft Excel, LibreOffice
Calc, or other software capable of reading .xlsx files.

No specialist software is required to inspect the deposited dataset.

The analysis code associated with the study is archived separately on Zenodo:
    Version-specific DOI: 10.5281/zenodo.22121607
    Concept DOI: 10.5281/zenodo.22120732

The Zenodo archive documents the Python environment and dependencies used to
reproduce the primary analyses from FVPD_participant_level.xlsx.

3. DATA STRUCTURE AND NAMING

Each row in the Data worksheet represents one participant.

Participant identifiers use the format:
    FVPD_001 to FVPD_103

These are pseudonymous identifiers assigned in random order. The identifiers
do not encode participant identity, recruitment order, age, sex, PHQ-9 score,
or any other participant characteristic. The linkage key is held separately by
the corresponding author and is not deposited.

Exact age is not included. Age is represented using the following bands:
    18-20
    21-22
    23-25

No real names, contact details, video URLs, raw video, raw audio, or linkage
information are included in the deposited dataset.

The FVPD cohort overlaps with participants in a related acoustic-feature
research record (DOI 10.5525/gla.researchdata.2332), but participant identifiers
are not shared across the two deposits.

4. VARIABLE DEFINITIONS

Demographic and questionnaire variables

participant_id
    Pseudonymous participant identifier, FVPD_001 to FVPD_103.

age_band
    Participant age in years, supplied as one of three bands:
    18-20, 21-22, or 23-25.

sex
    Self-reported sex. Values: F or M.

phq9_total
    Total Patient Health Questionnaire-9 (PHQ-9) score.
    Possible range: 0-27.
    Observed range in this dataset: 2-26.

bfi_openness
    BFI-2-XS Openness trait score. Sum of three items scored 1-5.
    Possible range: 3-15.

bfi_conscientiousness
    BFI-2-XS Conscientiousness trait score.
    Possible range: 3-15.

bfi_extraversion
    BFI-2-XS Extraversion trait score.
    Possible range: 3-15.

bfi_agreeableness
    BFI-2-XS Agreeableness trait score.
    Possible range: 3-15.

bfi_neuroticism
    BFI-2-XS Neuroticism trait score.
    Possible range: 3-15.

Derived facial constructs

DFI
    Depressive Facial Index:
        DFI = mean(au01, au04, au15, au17)
    Higher values represent greater activity across the selected
    depression- and sadness-related facial action units.

PAS
    Positive Affect Suppression:
        PAS = 1 - mean(au06, au12)
    Higher values represent lower smile-related activity.

BAD
    Between-AU Dispersion:
        BAD = sample standard deviation across the 20 participant-level
        mean AU intensities (ddof = 1).
    BAD describes dispersion across action units. It is not a measure of
    within-video temporal variability.

Facial action unit variables

The following variables contain the participant-level mean py-feat output for
each facial action unit, averaged across all analysed frames in which a face
passed the detection threshold:

au01    Inner brow raiser
au02    Outer brow raiser
au04    Brow lowerer
au05    Upper lid raiser
au06    Cheek raiser
au07    Lid tightener
au09    Nose wrinkler
au10    Upper lip raiser
au11    Nasolabial deepener
au12    Lip corner puller
au14    Dimpler
au15    Lip corner depressor
au17    Chin raiser
au20    Lip stretcher
au23    Lip tightener
au24    Lip pressor
au25    Lips part
au26    Jaw drop
au28    Lip suck
au43    Eyes closed

The AU values are the continuous outputs produced by the documented py-feat
pipeline. They were retained as exported, without post-hoc clipping or
rescaling.

Recording and extraction metadata

duration_s
    Measured recording duration in seconds.
    Observed range: 58.14-60.67 seconds.

fps
    Measured recording frame rate.
    Observed range: 29.84-30.06 frames per second.

n_frames_total
    Total number of frames contained in the recording.
    Observed range: 1744-1820.

n_frames_analysed
    Number of frames in which a face passed the detection threshold and was
    included in participant-level AU averaging.
    Observed range: 1743-1820.

detection_rate
    n_frames_analysed / n_frames_total.
    Observed range: 0.9994-1.0000.

5. FACIAL FEATURE EXTRACTION

Facial action units were extracted using:
    py-feat version 0.5.1
    Face detector: RetinaFace
    Face-detection confidence threshold: 0.85
    AU model: XGBoost AU intensity regression

Every available frame of all 103 recordings was processed.

Across the complete dataset:
    Total frames: 184,104
    Analysed frames: 184,025
    Overall face-detection rate: 99.96%
    Mean frame rate: approximately 29.97 frames per second

Participant-level AU variables are means across analysed frames.

6. DISCLOSURE CONTROL AND DATA PROTECTION

The dataset is pseudonymised rather than anonymous.

Disclosure-control measures applied before deposit include:
    - removal of real names and direct identifiers;
    - random assignment of FVPD participant identifiers;
    - replacement of exact age with age bands;
    - removal of video URLs and other links to source media;
    - exclusion of the linkage key;
    - exclusion of all original facial video and audio recordings.

Using exact age together with sex and PHQ-9 severity band left 19 of the 103
participants in a unique combination within the study sample. Age banding
reduced this number to 8 but did not eliminate uniqueness. The combination of
mental-health, personality, demographic and high-dimensional facial-feature
variables creates residual disclosure risk. Controlled access is therefore
required.

Users must not attempt to re-identify participants.

7. RELATIONSHIP TO THE ASSOCIATED ANALYSIS

FVPD_participant_level.xlsx contains the full-frame-rate participant-level
features used for the primary analyses reported in the associated manuscript.

The separately archived Zenodo code can reproduce the primary analyses using
this workbook as input.

8. LICENSING

This README may be shared publicly under the Creative Commons Attribution
4.0 International (CC BY 4.0) licence.

The participant-level data are not licensed for open reuse. Access and reuse of
the controlled dataset are governed by the University of Glasgow Enlighten
record and the applicable data-sharing agreement.

END OF README