Data Ethics, Visualization and Storytelling Concepts
What this domain tests
Ethics items cover privacy, consent and honest reporting; visualization items cover encoding accuracy and accessibility; storytelling items cover audience, message and structure.
Ethics rules
- Collect only what the question needs (data minimisation) and state the purpose up front.
- Personal data (names, emails, IP addresses, precise location) needs consent and protection, including combinations that re-identify people.
- Anonymisation removes identifiers irreversibly; pseudonymisation replaces them with a key and is reversible. Swapping a name for a customer ID is pseudonymisation, not anonymisation.
- Small cells and rare combinations (ZIP code + birth date + gender) re-identify individuals - aggregate or suppress them.
- Report methods and limitations. Do not cherry-pick time windows, quietly drop inconvenient groups, or redraw an axis to fit a message.
Visualization and storytelling
- Encoding accuracy: position and length are read most precisely, area and angle least - which is why bars beat pie slices.
- Use a colourblind-safe palette, sufficient contrast, and never colour as the only channel; label directly where possible.
- Remove chart junk, maximise data ink, and give each chart one message.
- Story structure: audience, question, key message, evidence, recommended action. Annotate the event that matters (price change, outage) instead of making readers hunt for it.
Worked example
A team wants to publish average session time by cohort alongside cohort sizes, and two cohorts have three users each. Ethical handling: report medians, disclose the small n, or suppress those cohorts entirely. Publishing a leaderboard built on n=3 is mathematically correct and still misleading, because readers will treat noise as a ranking.
Traps
- Calling keyed identifiers "anonymised data".
- Red/green only palettes, or rainbow scales applied to ordered data.
- Stacked bars when the question is comparing individual series.
- Figure titles that say nothing, such as "Figure 1".
- A truncated axis presented without disclosure.