Start with the decision, not the chart

A visualization is an interface between evidence and a reader. Before drawing, write down the question, comparison, population, time range, units, uncertainty, intended audience, and action the chart may influence. Then choose and test the encoding.

1. Match the visual encoding to the comparison

  • Compare categories: position or bar length on a common scale.
  • Show change over ordered time: lines or aligned points, with gaps and irregular intervals represented honestly.
  • Show a distribution: dot plots, histograms, box/violin plots, or empirical distributions, depending on sample size and audience.
  • Show a relationship: scatterplots or related encodings, with uncertainty and overplotting addressed.
  • Show composition: stacked bars or other part-to-whole views; use direct labels and prefer a common baseline when precise comparison matters.

Cleveland and McGill's graphical-perception experiments are a useful reason to prefer common-position and length judgments for precise comparisons, but no ranking replaces user and task context.

2. Keep context that changes interpretation

Remove decoration that competes with the data, but retain meaningful baselines, reference values, sample size, uncertainty, targets, annotations, and methodology. “High data-ink ratio” is a prompt to edit, not a score to maximize.

3. Use color by data meaning—and never alone

  • Use qualitative palettes for unordered categories.
  • Use sequential lightness for ordered low-to-high values.
  • Use a diverging scale only when the midpoint has meaning, such as zero or a target.
  • Keep the same category mapping across related charts.
  • Add text, shape, pattern, or position so color is not the only carrier of meaning.
  • Test text and essential graphical-object contrast, including adjacent regions.

4. Label the evidence

Give the chart a title that states a supported finding or question. Label axes, units, population, time period, transformations, and uncertainty. Cite the source and link methodology. Direct-label series when space permits; otherwise make the legend easy to associate with marks.

An annotation should explain a relevant event or caveat, not imply causation from temporal coincidence.

5. Make scale choices visible

  • Start bar-length comparisons at a meaningful zero; otherwise length no longer represents magnitude proportionally.
  • A restricted line-chart range can be valid for showing variation, but disclose it and consider a full-range companion when magnitude matters.
  • Keep axes aligned across small multiples intended for comparison.
  • Use logarithmic axes for appropriate positive, multiplicative relationships; label ticks in readable values and explain the transformation.
  • Do not hide missing intervals, suppressed values, changed definitions, or breaks in a series.

6. Design for a named audience and task

Specify who will use the chart, what they need to find, their domain knowledge, the device and setting, and the consequence of an incorrect reading. Test with representative users. Provide definitions and a summary for unfamiliar audiences; provide filters or downloadable detail only when those features serve a real task.

7. Simplify without deleting necessary evidence

Use hierarchy, alignment, whitespace, and restrained emphasis. Split genuinely different questions into aligned views. Do not remove denominators, uncertainty, subgroup behavior, source notes, or exceptions merely to make the page look cleaner.

8. Make interaction optional, operable, and fast

The default state should communicate a useful result without interaction. Filters, zoom, brushing, drill-down, and tooltips should have clear instructions, keyboard access, visible focus, named controls, persistent state, and an equivalent table or downloadable representation where needed. Test touch, keyboard, screen-reader, small-screen, slow-network, empty, error, and large-data states.

Publication checklist

  1. Can every number be traced to a source and transformation?
  2. Does the encoding answer the stated comparison?
  3. Are scale, units, denominator, time period, uncertainty, and missingness visible?
  4. Can the chart be understood without color?
  5. Do text and essential graphical objects meet applicable contrast requirements?
  6. Is there a meaningful text alternative or data table?
  7. Can all interactions be completed without a mouse?
  8. Did representative readers interpret the chart as intended?