Data Visualization
Intermediate 12 min read Data Visualization

Color for Data Visualization

The third encoding axis, used with intent

Color carries meaning: categorical palettes for groups, sequential scales for rank, diverging scales for above/below. Chosen deliberately and colorblind-safe, it encodes without misleading.

palettesequentialdivergingcolorblindcategoricalscale
Prerequisites: Visualization Principles
Before we start

What you will learn

  • Choose categorical, sequential or diverging scales
  • Avoid rainbow and hue-after-hue traps
  • Verify colorblind accessibility
Introduction

What is this?

Color is a channel, not decoration: categorical palettes distinguish classes; sequential scales (light→dark) rank magnitude; diverging scales (two poles ± midpoint) show above/below. Rules: never rainbow for continuous, respect colorblind pairs (use blue/orange, not red/green), and give color one job per chart.

Motivation

Why do you need it?

  • Rainbow continuous scales imply order and distance that don't exist.
  • Red/green comparisons are illegible to a large share of readers.
  • Deliberate scales make the chart's structure self-explanatory.
  • Accessible charts are a professional floor, not a bonus.
Think about it

Mental model

Imagine

Road signs using one color code

Think of

Color in data

Traffic systems don't paint meanings randomly — red means stop, amber warns, green goes; one code, everyone reads it. Data color palettes are the same: sequential light→dark means 'more', diverging colors signal 'two poles', categorical colors merely 'different kinds'.

Mechanics

How it works

  • Categorical: distinct hues with equal lightness, ≤ ~7 classes.
  • Sequential: a single hue light→dark (perceptually uniform) for magnitude.
  • Diverging: two hues from a neutral midpoint (e.g., blue-white-red).
  • Rainbow: avoid — hue passes through unrelated colors between values.
  • Accessibility: Okabe-Ito or viridis palettes; test with a colorblind sim.
  • One role per chart color: category or magnitude or emphasis — never more.

Scale type by question.

Classes?
Magnitude?
Two poles?
Accessible?
One job
Hands-on

Examples

Choosing scales with seaborn

Choosing scales with seaborn
import seaborn as sns
cat = sns.color_palette("Set2")          # categorical
seq = sns.color_palette("viridis", 9)    # sequential, magnitude
div = sns.color_palette("RdBu_r")        # diverging with midpoint
print("categorical:", cat)
print("colorblind-safe-ish sequential and diverging ready")

Palette families map to questions: Set2 for classes, viridis for ordered magnitude, RdBu_r for above/below-midpoint.

Real world

In production

  • A KPI heatmap uses a single sequential viridis scale so cell darkness = size.
  • An experiment dashboard encodes lift with a blue-white-red diverging scale at zero.
  • A plan-mix report reuses a 3-hue categorical palette consistently across pages.
Guidelines

Best practices

  • Let the palette encode one role and force it consistently across a report.
  • Choose blue/orange over red/green for safe accessible comparison.
  • Sequential only light→dark; diverging always with an explicit midpoint definition.
  • Test palettes with a colorblind simulator before publishing.
Watch out

Common mistakes

  • Rainbow (jet) scales — order by hue is unreadable and unethical on maps.
  • Diverging where data is only positive — a sequential scale should be there.
  • Red/green for good/bad when blue/orange reads for far more people.
  • Evacuating hue so often that no single meaning survives.
When things break

Debugging tips

  • A chart 'unreadable' by a colleague → run the palette through a colorblind sim.
  • A heatmap confusing → check the scale type vs the question (mag vs class).
  • Frequent mid spectrum ambiguity → a perceptually ordered viridis usually rescues it.
  • Ask: if printed grayscale, does meaning survive? If not, color is load-bearing — encode elsewhere too.
Interviews

Interview checkpoint

Test yourself

Quick quiz

0/2
1

Best palette for a number getting larger (100→1000):

2

Red/Green is a safe universal comparison pair.

Go deeper

Official documentation

ColorBrewer
docs
Wilson Colorbrewer essay
docs