Edu

A field guide to dashboards

Learn the trade, then break ground.

What kinds of dashboards exist, which tool to reach for, how this site parses its own data — and the best free places to learn. Opinionated on purpose; every link is a primary source.

🗂 A field guide to the world’s surveys and data programmes

81 research programmes — attitude surveys, household and health surveys, panels, school assessments and indices — with who runs each one, what it covers, how to get the data, and whether BullDozer uses it.

💻 AI on a student laptop

No GPU budget? We installed 22 open language models from 17 makers on one ordinary 16 GB MacBook, asked each the same 24 questions in seven languages, and compared them with the cloud. Which ones fit, which answer well, and when the cloud is worth paying for — every answer published.

🎬 Images and video with AI — a student’s map

The cloud generators and the open models you can run yourself, what is free, whose licence lets you publish the result, and the EU’s labelling rules — every fact linked to its source.

📱 App analytics — where the numbers come from

The App Store, Google Play and a website count the same people differently. Which source counts what, how to join them without personal IDs, and the pitfalls that make honest numbers disagree — checked against official docs.

Six kinds of dashboard

Most dashboard failures are type errors — a real-time layout answering a strategic question. Decide the type first, then the tool.

📌 KPI / executive

A handful of headline numbers with trend and target — revenue, share, NPS. Read in ten seconds, updated on a schedule.

Example: Board one-pager, weekly business review

Typical tools: Tableau, Power BI, Looker Studio

⚡ Operational / real-time

Live state of a system or campaign: traffic, conversions, error rates. Optimised for monitoring and alerting, not analysis.

Example: Grafana wall in a NOC, campaign launch monitor

Typical tools: Grafana, Superset, Metabase

🔬 Analytical / exploratory

Many linked filters and cross-sections so an analyst can slice the data themselves and hunt for the "why" behind a number.

Example: Survey explorer, cohort browser — most of BullDozer’s Charts page

Typical tools: Tableau, Superset, Plotly Dash, Streamlit

🧭 Strategic

Long horizons and few updates: market sizing, country scoring, scenario comparison. Depth over freshness.

Example: Market-entry scorecard, annual planning deck

Typical tools: Tableau, Power BI, hand-built D3

📖 Data story / scrollytelling

A fixed narrative with charts in a deliberate order. The author controls the path; the reader scrolls, not filters.

Example: FT and The Pudding longreads, Our World in Data articles

Typical tools: D3/SVG, Observable, Flourish, Datawrapper

🗺️ Geospatial

The map is the interface: choropleths, point clusters, flows. Works when location itself carries the signal.

Example: BullDozer’s Geo page, election maps, store networks

Typical tools: Kepler.gl, Mapbox, deck.gl, QGIS

Tooling

BI tools, honestly compared

There is no best tool, only a best fit. Rough rule: Metabase to give a team self-serve answers, Superset or Grafana to self-host seriously, Tableau or Power BI when the company already pays for them, Dash/Streamlit when the logic lives in Python, and D3 when the output must be exactly what you designed.

ToolLicenceStrongest atKeep in mind
Tableau ↗ Commercial Deep visual exploration; the analyst drag-and-drop standard. Free publishing via Tableau Public.
👀 See it live: the World Happiness Report’s own viz ↗
Desktop app; steep licence, gentle learning curve
Power BI ↗ Commercial Corporate reporting in Microsoft shops; tight Excel/Azure integration, DAX for modelling.
👀 See it live: Microsoft’s interactive sample report ↗
Cheap per seat; Windows-centric
Apache Superset ↗ Open source Self-hosted BI over SQL databases; dashboards, SQL Lab, row-level security. The main OSS Tableau alternative.
👀 See it live: dashboard walkthrough with the real UI ↗
Needs a server and some ops
Grafana ↗ Open source Time series and monitoring: metrics, logs, alerting. Less suited to ad-hoc table analytics.
👀 See it live: live time-series dashboard on Grafana Play ↗
Try live demos at play.grafana.org
Metabase ↗ Open source The fastest "questions over a database" tool for non-analysts; five-minute setup.
👀 See it live: a live survey dashboard ↗
Simpler charts than Superset
Looker Studio ↗ Free (Google) Free shareable dashboards over Google Analytics, Sheets and BigQuery.
👀 See it live: the report gallery — every report opens live ↗
Weak outside the Google ecosystem
Plotly / Dash ↗ Open source Analytical web apps in pure Python — full control over interactivity and layout.
👀 See it live: the Dash app gallery, apps run live ↗
You write code, not drag charts
Streamlit ↗ Open source Turning a Python script into an interactive app in an afternoon; ideal for prototypes and ML demos.
👀 See it live: the app gallery, apps run live ↗
Reruns whole script per interaction
Observable ↗ Freemium JavaScript notebooks for bespoke D3 visuals; home of Observable Plot.
👀 See it live: a World Happiness Report notebook ↗
For people who like code
Datawrapper ↗ Freemium Newsroom-grade single charts and maps with zero code; excellent defaults.
👀 See it live: Weekly Charts — live charts by their team ↗
Charts, not full dashboards
Flourish ↗ Freemium Animated templates — bar chart races, scrollytelling — publishable in minutes.
👀 See it live: the template gallery, all interactive ↗
Template-bound
D3 / hand-written SVG ↗ Open source Total control, zero dependencies, any design. Everything on BullDozer is built this way.
👀 See it live: our happiness story — this site is the demo ↗
Slowest to build, most durable
Under the hood

How BullDozer is built

This whole site is one worked example of a parsing-to-publication pipeline — no database, no BI licence, no server.

1 · Parse

Every dataset starts as a public file — IMF and World Bank APIs, survey CSVs from WVS, Afrobarometer or V-Dem, Excel sheets from research institutes. A parser script per source downloads it and reads it as-is.

One source = one script, so a broken source never blocks the rest.

2 · Normalise

Everything is reshaped into one long format: country · period · value, with ISO-3 codes and a metadata block (source, licence, method, link). 304 datasets, one schema.

The long format is why every chart on the site works with every indicator.

3 · Publish

A static site generator (Astro) bakes the JSON into pages and hand-written SVG charts at build time. The result is plain HTML on a CDN — fast, free to host, nothing to hack.

Interactivity is a few KB of vanilla JS per chart; no framework ships to the reader.

Practical parsing notes: prefer an official API or bulk-download over scraping HTML; keep the raw file and parse a copy; re-run on a schedule (GitHub Actions cron) rather than by hand; and record the source URL and access date next to the data — that is what makes “every number sourced” possible. Browse the results on Polls and Macro, or check definitions in the Glossary.

Learning

Free places to learn

A short list that is actually short — each entry earns its place.

Chart choice & dataviz craft

Inspiration — the best public data journalism

SQL & analytics practice

Scraping & data pipelines