Plotting Best Practices
The 20% that makes charts trusted
Naming axes, unit labels, consistent palettes, colorblind-safe encodings and small multiples turn competent charts into trustworthy ones.
What you will learn
- Craft informative titles/labels/units
- Choose accessible palettes
- Design for the smallest holdout that reads it
What is this?
Best-practice plotting is a checklist: every axis named with units, titles stating the takeaway, palettes that don't mislead (sequential vs diverging, colorblind-safe), and small multiples instead of mega-charts. Good charts are legible in grayscale print, at a glance, by someone who didn't build them.
Why do you need it?
- A titled-but-unitless axis can't be interpreted — the chart fails silently.
- Colorblind audiences (~8% of men) can't read red/green encoding.
- Small multiples compare better than crowded single charts.
- Self-describing charts get to the point in reviews — less meeting time, fewer misread decisions.
Mental model
Road sign design: clear type, consistent color language, one message
Production chart design
Drivers read signs at speed without instructions; stakeholders read charts the same way. Consistency of color, type and structure is what makes the message land.
How it works
- Title + axis labels + units (revenue ($) vs revenue) — three non-negotiables.
- Sequential palette (light→dark) for magnitudes; diverging (blue↔red) for deviations; categorical (distinct hues) for groups.
- Colorblind-safe sets (e.g., Okabe-Ito) and stop relying on red/green alone — add patterns/markers.
- Small multiples — same axes across panels — align comparisons.
- Remove non-data ink: gridlines turned down, legends deduplicated, 3D and shadows banned.
The chart checklist.
In production
- A BI style guide standardizes the company palette to colorblind-safe hues, so embedded dashboards stay consistent.
- A financial report labels every revenue axis in $ and annotates the one inflection point stakeholders must discuss.
- An experiment comparison uses small-multiple funnel plots instead of a 12-series megachart.
Best practices
- Put the message in the title ('Conversion fell 3pp after the March price change').
- Keep one palette per variable family across a report.
- Design for 100% grayscale legibility — print it once.
- Limit series per chart to 4-6; facet instead.
Common mistakes
- Rainbow palettes that imply ordering.
- Red/green-only categorical comparisons.
- Auto-generated legends full of 'series 1, series 2'.
- Missing units so '10' could be $10k or 10%.
Debugging tips
- A grayscale print of the chart should still encode the message — if not, channel choice or palette is wrong.
- If two groups share a hue, the palette or the legend mapping broke.
- Blurred text at view size → annotations were too dense; cut to one message.
Interview checkpoint
Quick quiz
Which palette is appropriate for a sequential metric like revenue?
Red/green encoding alone is safe for all audiences.