Data Visualization

A chart is an argument. Choose the encoding that fits the question, keep it honest, and annotate so the reader sees what you see — in five seconds.

When to use

• Any chart destined for a report, dashboard, or decision.
• Reviewing charts for distortion or confusion.
• Dashboards: layout, hierarchy, what deserves space.

Method

1. Question → chart: comparison → sorted bars; trend → line (time on x); distribution → histogram/box; relationship → scatter; composition → stacked bars (sparingly); part-to-whole over time → small multiples, not pie-animations.
2. Honesty rules: y-axis at zero for bars; no dual axes implying causation; consistent scales when comparing panels; label data source + date; if an axis is truncated — flag it loudly.
3. Perception first: position/length beat angle/area/color; most important thing top-left (or largest); chart-junk minimized (faint gridlines, no 3D).
4. Color with purpose: colorblind-safe categorical palettes; sequential for magnitude; diverging around a real midpoint; never red/green alone; ≤7 categories.
5. Annotation that explains: title = the finding, not the topic ("Revenue fell 12% in Q3", not "Revenue"); mark events/thresholds; call out outliers directly.
6. Accessibility + text: legible sizes; alt text or table fallback; values readable without hover; contrast for print and screen.
7. Dashboards: one screen = one story; metric definitions + owners (product-analytics); refresh cadence visible; retire stale panels.
8. Tooling: matplotlib/plotly/vega/Excel — version + data snapshot with every exported chart, so it's reproducible.

Quality gates

• Chart type matches the question; zero-baseline rules respected (or flagged).
• Colorblind-safe palette; no red/green-only encoding.
• Title states the finding; key values annotated.
• Source + date + snapshot recorded; accessible alternative exists.

Red lines

• No misleading axes/scales; no chart-junk as branding; never chart data you can't reproduce.

Related

product-analytics · electrical-test-records (domain trending) · technical-writing.