Annotation and small multiples

Intermediate

Introduction

A correct chart is not yet a clear one. Readers spend a few seconds on a chart, so it should tell them what to see: a title that states the finding, labels where the eye already is, and notes that explain the bumps. When there are too many series for one chart, splitting them into small multiples keeps each readable while still allowing comparison, provided the panels share a scale.

How it works

Active titles. "Monthly sales, 2024" names the data; "Sales rose 20% from January to February" states the finding. An active title helps every reader, and a subtitle can give the details: units, period and source.

Direct labelling. A legend forces the eye to jump back and forth matching colours. Put each series' name at the end of its line or on its bar instead. When the ends of several lines are close, labels collide; a simple fix sorts the labels by position and pushes each one down until it is at least a minimum gap below the previous label, then draws a short leader line if a label moved far.

Annotations. Mark events that explain changes ("price increase", "website outage"), reference lines such as targets or averages, and the specific values the text discusses. Keep annotations short and few, and make them visually lighter than the data.

Small multiples. With eight countries, one line chart becomes spaghetti. A grid of eight small charts, one per country, each with the same axes, lets readers see every country's shape and compare across panels. The key rule is a shared scale: if each panel has its own y range, a small rise in one country looks as big as a large rise in another. Libraries do this with options such as sharey=True in Matplotlib or facets in Vega-Lite and seaborn. Order the panels meaningfully, for example by the latest value, not alphabetically.

Example

# Small multiples: one tiny chart per series, all on the same scale.
series = {"North": [3, 5, 8, 10], "South": [2, 2, 3, 4], "West": [9, 7, 4, 2]}
top = max(max(values) for values in series.values())
levels = " .:-=+*#"
for name, values in series.items():
    marks = "".join(levels[round(v / top * (len(levels) - 1))] for v in values)
    change = values[-1] - values[0]
    print(f"{name:6} {marks}  {'up' if change > 0 else 'down'} {abs(change)}")
print("shared scale: 0 to", top)
Open in compiler

Line by line

  1. Three regions, four periods each.
  2. top is the largest value of all series, so every panel uses the same scale.
  3. Each value becomes one character from a ladder of increasingly dense marks, a tiny chart per region.
  4. A short note gives the direction and size of the change, a mini annotation.

Expected result

North  :=*#  up 7
South  ..:-  up 2
West   *+-.  down 7
shared scale: 0 to 10

In the real world

Newspapers print grids of small charts for election results by state and for infection curves by country. Business dashboards use direct labels and target lines, and the best report charts have titles that could be read aloud as the conclusion.

Common mistakes

  • Titles that only name the data, leaving readers to find the point.
  • Legends with many colours when direct labels would fit.
  • Overlapping labels at the ends of lines.
  • Small multiples with independent scales, which make different changes look alike.

Best practices

  • Write the title last, as the one sentence the chart proves.
  • Label directly and resolve collisions with a minimum gap.
  • Use small multiples with a shared scale and a meaningful panel order.

Summary

Active titles state the finding, direct labels replace legends, and short annotations explain events and targets. Labels that collide are pushed apart by a minimum gap. Small multiples split many series into panels that must share one scale so they can be compared.

Quick check

1. Which is an active title?
2. Why label lines directly instead of using a legend?
3. Small multiples must share…
4. Two line labels overlap at the right edge. A simple fix is…

Key takeaways

  • Say the finding in the title and annotate what explains it.
  • Label lines directly and move labels apart rather than letting them overlap.
  • Small multiples must share their scale, or comparisons between panels are wrong.

Practice problems for this lesson

My notes

Write down what you want to remember from this lesson. Notes are saved on this device as you type, and appear on your dashboard.