Charts That Persuade

You've found that two cafes are dragging down the business. You put it in a chart, and show it at a meeting. Nobody reacts. The chart was correct, but the message was buried. A good chart doesn't show data. It makes a point.

• Two cafes are the problem
• Nobody reacted
• A chart must make a point

Before and after

Here's the same data, in two charts. On the left, the default. On the right, we sort it, make it horizontal so that names are readable, colour only the bars that matter, and write the conclusion as the title. Run it. On the left, you have to work it out. On the right, your eye goes straight to the problem. The title says what to think.

import pandas as pd
import matplotlib.pyplot as plt

sales = pd.read_csv("sales.csv")
s = sales.groupby("city")["profit"].sum().sort_values() / 1000

fig, ax = plt.subplots(1, 2, figsize=(10, 4.2))
ax[0].bar(s.index, s.values); ax[0].tick_params(axis="x", rotation=60)
ax[0].set_title("Default chart")

colours = ["tab:orange" if v < s.quantile(0.2) else "lightgray" for v in s.values]
ax[1].barh(s.index, s.values, color=colours)
ax[1].set_title("Two cafes need attention (profit, $k)")
ax[1].spines[["top", "right"]].set_visible(False)
plt.tight_layout(); plt.savefig("story.png", dpi=110)
print("Lowest:", s.index[:2].tolist())

Lowest: ['Cairo', 'Abuja']

Rules for clear charts

Say what you want people to do. Use the title for the conclusion, not the topic. Remove anything that doesn't help, such as grid lines and borders. Use colour to focus attention, not to decorate. Sort your bars. Label directly on the chart. And start bars at zero, or you'll mislead.

Which chart?

Pick by what you're trying to show. Change over time, use a line. Compare categories, use sorted bars. Parts of a whole, use a stacked bar, and avoid pies with many slices. Relationships, use a scatter. Distributions, use a histogram or a box plot.

Summary