Skip to content

basseat/attention-paradox

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

6 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

The Attention Paradox

Gen Z supposedly can't focus. So why are they watching 3-hour Twitch streams?

This is a data investigation into one of the stranger contradictions in how we talk about young people and technology. The mainstream narrative says shorter content wins — TikTok, Reels, Shorts. But live streaming, an unedited, unscripted, hours-long format with no algorithm-friendly hooks, is growing fastest among exactly the generation we're told can't pay attention.

That gap between the story and the data is what this project is about.

→ View the Tableau dashboard on Tableau Public


What the data actually shows

Three hypotheses. Three verdicts.

# Hypothesis Verdict
H1 Younger generations show shorter measurable attention spans Supported (via proxies)
H2 Live streaming is growing faster among Gen Z than older generations Supported
H3 Despite shorter attention spans, Gen Z shows longer and more engaged live-streaming sessions Supported — the paradox is real

H1 — The short attention evidence

Gen Z spends the most time online of any generation (349 minutes/day, Ofcom 2022) and has the highest social media daily use rate (84%, Pew 2023). Their preferred content formats — TikTok videos and Instagram Reels — average just 0.5 minutes per piece. The Microsoft Canada (2015) study put average attention span at 8.25 seconds, down from 12 in 2000 — though that study has since been critiqued for methodology. The behavioural signal is clearer: Gen Z favours the shortest content formats available.

Caveat: online time and content format preference are proxies, not neurological measurements. Self-reported data has known limitations.

H2 — Live streaming is a Gen Z phenomenon

  • 45% of 18–24 year olds watch live streams weekly on YouTube, vs 28% for 35–49 (Statista/YouTube, 2023)
  • 73% of Twitch's audience is under 35
  • Total Twitch hours watched grew from 9.4 billion (2019) to 22.4 billion (2021) — a 139% increase driven largely by Gen Z audience expansion
  • Post-COVID normalisation has brought Twitch back to ~19.3B hours (2023), but the demographic dominance of younger viewers has held

H3 — The paradox confirmed

This is the finding that makes the project interesting.

From a sample of 634 YouTube live streams and completed VODs:

  • Streams with duration data show a mean session length of ~2.3 hours
  • ~49% of streams exceeded 1 hour; ~36% exceeded 2 hours
  • Median engagement rate across live streams: 1.8% — comparable to or exceeding VOD benchmarks
  • The platform most dominated by Gen Z viewers (Twitch) hosts streams with a typical duration of ~3 hours, frequently running 6–8+ hours

For comparison: TikTok videos average 0.5 minutes. A standard Twitch session is 360× longer.

The same demographic that popularised ultra-short content is the primary audience for the longest unedited format in mainstream media. That is the paradox — and the data holds it up.


Why this matters

The "goldfish generation" narrative has shaped content strategy, education policy, and workplace design. But it's built on shaky evidence — a single 2015 Microsoft report that has since been scrutinised — and it ignores an enormous counter-signal sitting in plain sight.

A more precise framing: Gen Z doesn't have a short attention span. They have a high bar for earning sustained attention. Content that doesn't capture them in the first seconds gets skipped. Content that does — a streamer with parasocial pull, a competitive game, a live event — holds them for hours.

That's a different problem than the one most of the narrative assumes.


Data sources

Platform & viewership data

Source What's captured
Twitch API + StreamElements State of the Stream (2019–2023) Hours watched, growth trends, audience age distribution
YouTube Data API v3 Live streams vs VOD: views, likes, comments, duration, engagement rate
SullyGnome / TwitchTracker Historical Twitch viewership (2018–2025)

Research reports & surveys

Source Data
Ofcom Online Nation (2022) Daily online minutes by age group (UK)
Pew Research Center (2023) Social media daily use by generation
Statista / Backlinko (2023) Platform demographics by age
Microsoft Canada (2015) Average attention span (widely cited; methodology disputed)
Dscout (2016) Daily phone interaction frequency

Project structure

attention_paradox/
├── README.md
├── requirements.txt
│
├── data/
│   ├── raw/                         ← Unmodified source data
│   │   ├── twitch/
│   │   ├── youtube/
│   │   ├── surveys/
│   │   └── reports/
│   ├── processed/                   ← Cleaned, transformed data
│   └── exports/                     ← Final CSVs for Tableau
│
├── etl/
│   ├── config.py
│   ├── twitch_collector.py
│   ├── youtube_collector.py
│   ├── research_scraper.py
│   ├── kaggle_loader.py
│   └── pipeline.py
│
├── eda/
│   └── attention_paradox_eda.ipynb
│
├── tableau/
│   ├── tableau_guide.md
│   └── calculated_fields.txt
│
└── report/
    └── final_report.md

Running the project

Only have a YouTube API key? The quick start below runs a full demo on YouTube data only — Twitch can be added later.

# 1. Install dependencies
pip install -r requirements.txt

# 2. Set up environment
cp .env.example .env
# Add your YOUTUBE_API_KEY (and optionally TWITCH_CLIENT_ID / SECRET)

# 3. Run the demo
python demo.py

For the full pipeline with Twitch:

python etl/pipeline.py
jupyter notebook eda/attention_paradox_eda.ipynb

API keys:

Then load any CSV from data/exports/ into Tableau Desktop and follow tableau/tableau_guide.md.


Limitations

  • Proxy problem — no direct neurological measure of attention span is publicly available at scale. All H1 evidence is behavioural or self-reported.
  • Platform data gaps — Twitch and YouTube do not expose viewer age directly. Demographic breakdowns come from third-party cross-sectional reports, not longitudinal tracking.
  • Survivorship bias — the streaming dataset captures people who watched long streams; non-watchers aren't represented.
  • Context-switching — a viewer "watching" a 3-hour stream may still be multitasking. We observe platform sessions, not individual attention states.
  • Gen Alpha — very limited data for under-13s due to COPPA/GDPR restrictions.

Related research

  • Microsoft Canada (2015) — Attention Spans (the "goldfish" study; later critiqued)
  • Rosen et al. (2013) — The war for your attention: multitasking and media use
  • StreamElements — State of the Stream reports (2019–2023)
  • Pew Research Center — Social Media Use by Generation (2021, 2023)
  • Ofcom — Online Nation (annual)
  • Nielsen — The Gauge streaming reports

Basit Ayoade · Data Analytics Portfolio · 2026

About

Is the short attention span narrative true? A data investigation into Gen Z's contradictory content habits.

Resources

Stars

0 stars

Watchers

0 watching

Forks

Releases

No releases published

Packages

 
 
 

Contributors