DATA CLEANING AND ANONYMISATION PROCEDURES

This document summarises how the study data was prepared (cleaned) and anonymised prior to analysis, with specific procedures for each data source used in the project.

1) Logbook (Google Form / CSV)
- Participant identification:
  All participants initially entered logbook responses using their real first names. This was done to avoid errors that can occur when participants self-enter an ID code incorrectly.
  Before analysis began, each participant was assigned an anonymised participant code (Participant 1–Participant 5). Real names were removed from the analytical dataset, and all subsequent data collection and file labelling followed the same participant codes to ensure consistency across the study.

- Timestamp handling (reducing inconsistencies / bias):
  To avoid inconsistencies and to prevent highlighting or shaping behaviour through time-based questioning, participants were not asked to manually enter the date or time as part of the logbook questions.
  Instead, the submission timestamp was automatically recorded by Google Sheets at the moment each log was submitted. This timestamp was then used to derive two additional columns for clarity and ease of interpretation:
   • Day of week (e.g., Monday, Tuesday)
   • Week of study (Week 1 or Week 2)
  No other logbook responses were edited or changed. The substantive content of each entry remained exactly as submitted by the participant.

2) Interviews (Audio → Transcripts)
- Transcription method and accuracy checks:
  Interviews were transcribed using Otter.ai. Following this automated transcription step, the researchers listened through each recording while reading the transcript to check for any transcription errors (e.g., misheard words, missing phrases, or incorrect speaker content).
  Where inaccuracies were found, the transcript was corrected to match what the participant actually said.

- Transcript integrity and light cleaning:
  Transcripts were otherwise kept unabridged and verbatim in relation to the recordings. The only exception was the removal of brief filler utterances (e.g., repeated “um”) when they did not add meaning, to improve readability and prevent the transcript from becoming unnecessarily cluttered.

- Anonymisation and file labelling:
  No identifying information was discussed during the interviews, so no redaction was required. All interview-related files (audio, transcripts, coding files) were labelled using the participant number system (Participant 1–Participant 5) to maintain consistent anonymised referencing across the dataset.

3) Screen Time Usage (Representative Evidence + Raw Exports)
- Screenshots (representative evidence):
  Screen time screenshots included in the repository were anonymised. They are presented only to demonstrate the medium and method through which this data was collected (i.e., phone-reported screen time).

- Raw screen time spreadsheets:
  The raw screen time spreadsheets did not require additional processing for anonymisation within the analytical workflow described here.
  Categorisation (e.g., Entertainment, Social Media, Productivity, Misc.) was based on the phone operating system’s built-in grouping of app usage, and this system-level categorisation was retained when screen time data was grouped by category.

4) Consent Forms
- Handling of identifiable consent materials:
  Because this study is not intended for publication, signed consent forms were retained unredacted to demonstrate that consent was received from all participants.
  These consent forms contain signatures and therefore remain identifiable documents; they are treated as separate administrative evidence of consent rather than part of the anonymised analytical dataset.

END OF DOCUMENT