Digital Balance Index (DBI) Dashboard

---

A portfolio-grade analytics project that turns raw “hours spent” into composition-first behavioral insights.

Most screen-time analyses stop at totals (“people spend 8 hours/day”).
This project focuses on how that time is distributed across:

• Social
• Work/Study
• Entertainment

It ships:

• a reproducible scoring pipeline (python -m src.pipeline)
• clean exported datasets + figures
• a Streamlit dashboard for interactive exploration across age_group, primary_device, and internet_type

---

Dataset (source + thanks)

This project uses the Kaggle dataset:

• Daily Internet Usage Statistics by Age Group by jayjoshi37
• Dataset URL: https://www.kaggle.com/datasets/jayjoshi37/daily-internet-usage-statistics-by-age-group

Thanks to the dataset author and Kaggle for making the data available.

---

Why this project exists

The problem with “total hours”

Two users can both have 10 hours/day of screen time, but with completely different patterns:

• One: mostly Work/Study (structured use)
• Another: mostly Social/Entertainment (dominant categories)

Totals alone can’t describe behavior.
Composition metrics can.

The core idea

For each record, we compute:

1. Shares of total screen time for each category
2. DBI (Digital Balance Index): how evenly those shares are distributed
3. Dominance: how strongly one category “owns” the day
4. Practical tiers and a flag: High-load & Skewed

---

What’s inside (pipeline + dashboard)

Pipeline outputs

After running the pipeline you get:

• outputs/scored_rows.csv
  Original data + engineered features:

  • shares (p_social, p_work, p_entertainment)
  • dbi, dominance, dominant_category
  • tiers (dbi_tier, load_tier)
  • risk flag (flag_highload_skewed)

• outputs/segment_summary.csv
  Aggregated statistics by:

  • age_group × primary_device × internet_type

• outputs/daily_summary.csv
  Daily aggregates:

  • mean DBI / mean total / sample size (n) per day

• outputs/metric_cards.json
  KPI cards + thresholds and validation checks

• reports/figures/
  All figures used in this README

Dashboard

The Streamlit dashboard allows:

• filtering by age group, device, internet type, DBI tier
• exploring DBI distributions and segment comparisons
• safe interpretation notes (“decision safety”)

---

Metrics (clear + reproducible)

Let:

• total = total_screen_time
• social = social_media_hours
• work = work_or_study_hours
• ent = entertainment_hours

1) Shares (composition)

• p_social = social / total
• p_work = work / total
• p_entertainment = ent / total

These are proportions in [0, 1] and (when total > 0) they sum to ~1.

2) DBI (Digital Balance Index) 0 to 1

DBI uses normalized Shannon entropy over the three shares:

• H = - Σ (p_i * ln(p_i)) for i in {social, work, entertainment}
• DBI = H / ln(3)

Interpretation:

• DBI ≈ 1.00 → time is distributed evenly (balanced composition)
• DBI ≈ 0.00 → time is concentrated in one category (skewed composition)

Important note:

• DBI does not say “good” or “bad”
• DBI says “balanced” vs “dominated”

3) Dominance 0.33 to 1.00

• dominance = max(p_social, p_work, p_entertainment)

Interpretation:

• dominance close to 1.0 → one category dominates
• dominance close to 0.33 → near-even split

4) Practical tiers (used in outputs + dashboard)

DBI tiers:

• Balanced: DBI ≥ 0.80
• Mixed: 0.60–0.79
• Skewed: < 0.60

Load tiers (by quantiles of total screen time):

• Low / Medium / High based on P33 and P66

Flag:

• flag_highload_skewed = 1 if load_tier == High AND dbi_tier == Skewed

This is intentionally phrased as “attention-worthy,” not “harmful.”

---

Quickstart

1) Install

python -m venv .venv
# Windows:
#   .venv\Scripts\activate
# macOS/Linux:
#   source .venv/bin/activate
pip install -r requirements.txt

2) Run the pipeline

python -m src.pipeline --input data/raw/daily_internet_usage_by_age_group.csv

You should see a Done message and the output folders printed.

3) Launch the dashboard

streamlit run app/app.py

---

Project structure

digital-balance-index-dashboard/
  data/
    raw/
    processed/
  outputs/
  reports/
    figures/
  src/
    io.py
    validate.py
    scoring.py
    aggregates.py
    reporting.py
    pipeline.py
  app/
    app.py
  README.md
  requirements.txt

---

Figures (with interpretation)

Below are the key plots generated by the pipeline.

---

1) Composition by age group (mean shares)

What this chart shows

Each bar is an age group. The bar stacks to 1.0 (100% of total screen time), split into:

• Social
• Work/Study
• Entertainment

This answers:

• Do age groups differ more by total time, or by how time is distributed?

How to read it correctly

• Look for relative differences in shares (e.g., Work/Study share slightly higher in one group).

• A similar composition across groups suggests:

  • differences may lie more in total screen time than usage mix.

Common pitfalls

• Don’t interpret this as “who uses more.” This plot is composition, not absolute hours.
• A group can have the same composition but different total hours.

---

2) DBI distribution

What this chart shows

A histogram of DBI values across all records.

Why it matters

This is your “macro fingerprint”:

• If DBI is mostly high → usage is generally mixed/balanced in composition.
• If DBI has a heavy low tail → many users have dominated usage (one category owns most of the day).

Practical interpretation

• High DBI dominance suggests many “mixed days,” not necessarily “low usage.”

• A low-DBI tail is where you’d investigate dominant categories:

  • dominated by Social?
  • dominated by Entertainment?
  • or dominated by Work/Study?

---

3) Total screen time vs DBI (composition vs load)

What this chart shows

Each dot is a record:

• x-axis: total screen time (hours)
• y-axis: DBI (0–1)

The key insight

This separates two different questions:

• How much? (load)
• How distributed? (composition)

The useful “quadrants”

Even without drawing lines, you can think in quadrants:

• High total + High DBI Heavy usage, but spread across categories (mixed day)

• High total + Low DBI Heavy usage, dominated by one category (attention-worthy)

• Low total + High DBI Light usage, balanced composition

• Low total + Low DBI Light usage, but concentrated (short + focused)

Decision safety note

This plot is descriptive: it suggests where to look, not what to diagnose.

---

4) Daily trends: mean DBI with sample size

What this chart shows

• Solid line: daily mean DBI
• Dashed line: sample size (n) that day

Why sample size is plotted

Because daily means are only meaningful if daily n is stable.

• If n swings sharply, the mean can swing even if behavior didn’t change.

What you can and cannot conclude

Reasonable:

• “DBI is fairly stable across dates.”
• “Some days show deviations, but sample size should be checked.”

Not reasonable:

• “Behavior changed over time for the same people.” This dataset is not a per-person time series; it’s a collection of records by date.

---

5) Mean DBI by primary device

What this chart shows

Average DBI per primary device category.

What it’s useful for

• Quick comparison: are some devices associated with more balanced composition?

How to interpret cautiously

• Small differences can be real or just sample noise.
• Device does not “cause” DBI differences; it’s an association.

A strong follow-up is to check:

• DBI distribution by device (not just means)
• segment breakdown: age group × device

---

6) Mean DBI by internet type

What this chart shows

Average DBI for WiFi vs Mobile Data.

What it suggests (carefully)

If values are similar:

• connectivity context may not be a major driver of composition balance

If values differ:

• could reflect behavior contexts (e.g., mobile data used on the go)

Again: this is association, not causation.

---

7) Mean DBI heatmap: age group × primary device

What this chart shows

Mean DBI for each (age_group, primary_device) combination.

Why it’s powerful

This reveals interaction patterns that “mean by device” hides:

• a device might look balanced overall,
• but show different balance depending on age group.

How to use it

• Spot extremes (highest and lowest cells)

• Use segment_summary.csv to confirm:

  • n per segment
  • composition means
  • high-load & skewed rate

---

Decision Safety (how to present this responsibly)

This project is designed for safe analytics:

• DBI is not a mental health score.
• DBI is not a productivity score.
• DBI is not a diagnosis.

Recommended language:

• “composition pattern”
• “dominant category”
• “balanced vs skewed distribution”
• “attention-worthy segment”

Avoid language like:

• “addicted”
• “harmful”
• “unhealthy”
• “caused by WiFi/device”

---

Reproducibility & validation

The pipeline includes:

• schema validation (required columns)
• identity validation: total ≈ social + work + entertainment
• consistent export paths and figures

---

Roadmap (optional upgrades)

If you want to level this up further:

• Add bootstrap confidence intervals for segment means (DBI/total)
• Add “What-if composition simulator” (move minutes between categories → new DBI)
• Add dominance + category-specific deep dives
• Add segment stability checks (min n threshold)